Landslides from Massive Rock Slope Failure (NATO Science Series: IV: Earth and Environmental Sciences)

  • 78 275 5
  • Like this paper and download? You can publish your own PDF file online for free in a few minutes! Sign Up

Landslides from Massive Rock Slope Failure (NATO Science Series: IV: Earth and Environmental Sciences)

Landslides from Massive Rock Slope Failure NATO Science Series A Series presenting the results of scientific meetings

1,476 72 59MB

Pages 659 Page size 336 x 571.2 pts Year 2009

Report DMCA / Copyright


Recommend Papers

File loading please wait...
Citation preview

Landslides from Massive Rock Slope Failure

NATO Science Series A Series presenting the results of scientific meetings supported under the NATO Science Programme. The Series is published by IOS Press, Amsterdam, and Springer in conjunction with the NATO Public Diplomacy Division Sub-Series I. Life and Behavioural Sciences II. Mathematics, Physics and Chemistry III. Computer and Systems Science IV. Earth and Environmental Sciences

IOS Press Springer IOS Press Springer

The NATO Science Series continues the series of books published formerly as the NATO ASI Series. The NATO Science Programme offers support for collaboration in civil science between scientists of countries of the Euro-Atlantic Partnership Council. The types of scientific meeting generally supported are “Advanced Study Institutes” and “Advanced Research Workshops”, and the NATO Science Series collects together the results of these meetings. The meetings are co-organized bij scientists from NATO countries and scientists from NATO’s Partner countries – countries of the CIS and Central and Eastern Europe. Advanced Study Institutes are high-level tutorial courses offering in-depth study of latest advances in a field. Advanced Research Workshops are expert meetings aimed at critical assessment of a field, and identification of directions for future action. As a consequence of the restructuring of the NATO Science Programme in 1999, the NATO Science Series was re-organised to the four sub-series noted above. Please consult the following web sites for information on previous volumes published in the Series.

Series IV: Earth and Environmental Sciences – Vol. 49

Landslides from Massive Rock Slope Failure edited by

Stephen G. Evans Department of Earth Sciences, University of Waterloo, ON, Canada

Gabriele Scarascia Mugnozza Department of Earth Sciences, Università degli Studi di Roma “La Sapienza”, Rome, Italy

Alexander Strom Institute of the Geospheres Dynamics, Russian Academy of Sciences, Moscow, Russia and

Reginald L. Hermanns Geological Survey of Canada, Vancouver, BC, Canada

Published in cooperation with NATO Public Diplomacy Division

Proceedings of the NATO Advanced Research Workshop on Massive Rock Slope Failure: New Models for Hazard Assessment Celano, Italy 16–21 June 2002 A C.I.P. Catalogue record for this book is available from the Library of Congress.


1-4020-4036-9 (PB) 978-1-4020-4036-8 (PB) 1-4020-4035-0 (HB) 978-1-4020-4035-1 (HB) 1-4020-4037-7 (e-book) 978-1-4020-4037-5 (e-book)

Published by Springer, P.O. Box 17, 3300 AA Dordrecht, The Netherlands.

Printed on acid-free paper

All Rights Reserved © 2006 Springer No part of this work may be reproduced, stored in a retrieval system, or transmitted in any form or by any means, electronic, mechanical, photocopying, microfilming, recording or otherwise, without written permission from the Publisher, with the exception of any material supplied specifically for the purpose of being entered and executed on a computer system, for exclusive use by the purchaser of the work. Printed in the Netherlands.


EDOARDO SEMENZA (1927-2002) Professor Edoardo Semenza passed away on May 31st, 2002 shortly before the NATO Advanced Research Workshop on Massive Rock Slope Failure. He was one of Italy’s leading landslide researchers and put his distinctive mark on the understanding of the Alpine chain structure, bringing original and important contributions to the geology, tectonics and geomorphology of the Dolomites. His main scientific interest was the understanding of geological processes with reference to landscape evolution and slope stability conditions. His life and work were deeply influenced by the 1963 Vaiont rockslide. Professor Semenza was the geologist who realized, well before the first recorded movements, that an ancient landslide mass was present on the left slope of the Vaiont River valley and he was aware of the very high hazard posed to the dam then under construction. His studies and interpretation were considered of major interest and accuracy by successive Vaiont researchers such as Hendron and Patton. He spent more than 40 years of his academic career at Ferrara University as Professor of Engineering Geology. His geological insight, humanity and culture (he also wrote many poems both in Italian and the Latin language) remain in the memory of colleagues, students and the Italian scientific community. One of his fundamental teachings was the importance of the careful collection of field data as a basis for understanding slope stability conditions, for reliable modelling, and for designing effective remedial works. He strongly believed in the role of both geology and geomorphology as fundamental support to any civil engineering works and in the importance of effective communication among the various specialists involved in large civil engineering projects. This volume is dedicated to his memory.

TABLE OF CONTENTS1 PREFACE ........................................................................................................................xi ACKNOWLEDGEMENTS............................................................................................ xv PART 1: INTRODUCTION 1. Landslides from Massive Rock Slope Failure and Associated Phenomena S.G. Evans, G. Scarascia Mugnozza, A.L. Strom, R.L. Hermanns, A. Ischuk and S. Vinnichenko.................................................................................. 3 2. *Single-Event Landslides Resulting from Massive Rock Slope Failure: Characterising their Frequency and Impact on Society S.G. Evans ............................................................................................................ 53 PART 2: ANALYSIS OF INITIAL ROCK SLOPE FAILURE 3. On the Initiation of Large Rockslides: Perspectives from a New Analysis of the Vaiont Movement Record D.N. Petley and D.J. Petley .................................................................................. 77 4. *From Cause to Effect: Using Numerical Modelling to Understand Rock Slope Instability Mechanisms E. Eberhardt ......................................................................................................... 85 5. Gravitational Creep of Rock Slopes as Pre-collapse Deformation and some Problems in its Modelling A.A. Varga........................................................................................................... 103 6. Models Available to Understand Failure and Pre-failure Behaviour of Large Rock Slope Movements: The Case of La Clapière, Southern Alps, France V. Merrien-Soukatchoff and Y. Gunzburger ....................................................... 111 7. Numerical Modelling of Rock Slopes Using a Total Slope Failure Approach D. Stead and J.S. Coggan ................................................................................... 129 8. *The Role of Topographic Amplification on the Initiation of Rock Slopes Failures During Earthquakes W. Murphy........................................................................................................... 139 PART 3: MONITORING OF ROCK SLOPE MOVEMENT 9. Application of Ground-Based Radar Interferometry to Monitor an Active Rockslide and Implications for Emergency Management N. Casagli, P. Farina, D. Leva and D. Tarchi ................................................... 157 10. Monitoring and Assessing the State of Activity of Slope Instabilities by the Permanent Scatterers Technique C. Colesanti, G.B. Crosta, A. Ferretti and C. Ambrosi ...................................... 175


* denotes Keynote Lecture vii

viii PART 4: ANALYSIS OF POST-FAILURE BEHAVIOUR 11. *Forecasting Runout of Rock and Debris Avalanches R.M. Iverson........................................................................................................ 197 12. *Continuum Numerical Modelling of Flow-like Landslides G.B. Crosta, S. Imposimato and D.G. Roddeman .............................................. 211 13. Landslide Mobility and the Role of Water F. Legros ............................................................................................................. 233 14. *Rock Avalanche Occurrence, Process and Modelling O. Hungr ............................................................................................................. 243 15. Mobility of Rock Avalanches Triggered by Underground Nuclear Explosions V.V. Adushkin ..................................................................................................... 267 16. *Rapid Rock Mass Flow with Dynamic Fragmentation: Inferences from the Morphology and Internal Structure of Rockslides and Rock Avalanches M.J. McSaveney and T.R.H. Davies ................................................................... 285 17. *Morphology and Internal Structure of Rockslides and Rock Avalanches: Grounds and Constraints for their Modelling A. Strom............................................................................................................... 305 PART 5: CASE STUDIES OF MASSIVE ROCK SLOPE FAILURE 18. The Flims Rockslide: History of Interpretation and New Insights A. von Poschinger, P. Wassmer and M. Maisch................................................. 329 19. *Rock Avalanche and Mountain Slope Deformation in a Convex Dip-Slope: The Case of the Maiella Massif, Central Italy G. Scarascia-Mugnozza, G. Bianchi Fasani, C. Esposito, S. Martino, M. Saroli, E. Di Luzio and S.G. Evans ............................................ 357 20. Slow-Moving Disintegrating Rockslides on Mountain Slopes R. Couture and S.G. Evans ................................................................................. 377 21. Edoardo Semenza: The Importance of Geological and Geomorphological Factors in the Identification of the Ancient Vaiont Landslide M. Ghirotti .......................................................................................................... 395 22. *Failure Mechanisms and Runout Behaviour of Three Rock Avalanches in the North-Eastern Italian Alps R. Genevois, C. Armento and P.R. Tecca .......................................................... 407 PART 6: VOLCANIC LANDSLIDES 23. Large Modern Collapses on the Active Volcanoes of Kamchatka: Causes and Mechanism of Formation I.V. Melekestsev .................................................................................................. 431 24. Assessing Massive Flank Collapse at Stratovolcanoes Using 3-D Slope Stability Analysis M.E. Reid and D.L. Brien ................................................................................... 445 25. Catastrophic Volcanic Landslides: The La Orotava Events on Tenerife, Canary Islands M. Hürlimann and A. Ledesma........................................................................... 459

ix PART 7: REGIONAL STUDIES OF MASSIVE ROCK SLOPE FAILURE 26. *Rock Slope Failures in Norwegian Fjord Areas: Examples, Spatial Distribution and Temporal Pattern L.H. Blikra, O. Longva, A. Braathen, E. Anda, J.F. Dehls, and K. Stalsberg.................................................................................................. 475 27. *Rock Avalanching in the NW Argentine Andes as a result of Complex Interactions of Lithologic, Structural and Topographic Boundary Conditions, Climate Change and Active Tectonics R.L. Hermanns, S. Niedermann, A. Villanueva Garcia, and A. Schellenberger......................................................................................... 497 28. Rock Avalanches with Complex Run Out and Emplacement, Karakoram Himalaya, Inner Asia K. Hewitt ............................................................................................................. 521 29. Dissected Rockslide and Rock Avalanche Deposits, Tien Shan Kyrgyzstan K. Abdrakhmatov and A. Strom ......................................................................... 551 PART 8: INFLUENCES ON GEOMORPHOLOGICAL EVOLUTION 30. *Landslide-driven Erosion and Topographic Evolution of Active Mountain Belts N. Hovius and C.P. Stark .................................................................................... 573 31. *Impacts of Landslide Dams on Mountain Valley Morphology R.L. Schuster ....................................................................................................... 591 PART 9: STATE-OF-THE-ART 32. *Massive Rock Slope Failure: Perspectives and Retrospectives on State-of-the-Art J.N. Hutchinson................................................................................................... 619

PREFACE This volume contains contributions by experts who participated in the NATO Advanced Research Workshop on “Massive Rock Slope Failure: New Models for Hazard Assessment” which was held in Celano, Italy, in June 2002. This event has become known as the Celano Workshop. Landslides resulting from large-scale rock slope failures are a major hazard in mountainous regions. In the 20th century, disasters caused by massive rock slope failures have killed more than 50,000 people on a global basis with 20,000 of this total being killed in the NATO and Partner Countries (for details of NATO structure and Partner country designation see In addition, hazardous conditions prevail at previous rock slope failure sites that pose considerable risk to communities downslope (in the case of cracked slopes adjacent to landslide scars, e.g., Frank, Canada) or downstream (in the case of potentially unstable landslide dams, e.g., Lake Sarez, Tajikistan) from initial failure sites. Furthermore, slopes that have not undergone major failure may show movement or surface features that indicate potential for catastrophic landslides. As an illustration of the hazard posed by such processes it is noted that during the preparation of this book a combined massive rock-ice avalanche and mudflow, tens of millions of cubic meters in volume, occurred in the Caucasus mountains of Russia’s Ossetia Republic. It ran over 30 kilometres downvalley killing over a hundred of people. The Celano Workshop was purposely organized to bring together earth scientists, engineering geologists and geotechnical engineers actively involved in a range of research topics related to massive rock slope failure, in order to undertake a critical assessment of the state-of-the-art concerning catastrophic rock slope failure and for defining future research directions. It was specifically dedicated to an examination of major mountain rock slope hazards, combining different experiences from different mountain environments of the world, as a contribution to the “Year of the Mountains”, declared by the United Nations in 2002. A highlight of the Celano Workshop was provided by the fact that the NATO Science Programme afforded a unique opportunity to solicit the participation of scientists from the former Soviet Union who introduced little known case-histories of massive rock slope failure in Russia, and the Central Asia Republics which are published in this volume for the first time. This book is not intended as a simple “proceedings volume”, i.e. a raw collection of papers summarising oral presentations in Celano concerning different aspects of largescale rock slope failures. Rather, it was our objective to produce a self-standing volume whose contents display the most recent research within the international scientific community involved in the field of bedrock slope failure, framed in a logical order and with linkages to each other. The structure of the book broadly reflects the main issues addressed by the Celano Workshop, even though there are some topics not as fully covered as was originally our expectation and plan. Topics related to risk assessment, such as consideration of vulnerability were intentionally not addressed, since at this stage we have focussed on


xii basic knowledge and models of massive rock slope failure processes and mechanisms. It is our belief that such specialized knowledge of hazard forms the foundation for any effective and successful risk management activity. Hopefully it will be the task of future meetings to address such themes, completing the loop that we have initiated in Celano. According to our objectives, the book is divided in nine parts. The first includes an introduction to the phenomena associated with massive rock slope failure and an analysis of the frequency of catastrophic rock slope failures and their impact on society. Part two deals with processes and mechanisms which lead to the onset of failure in rock slopes, with particular reference to the contribution of numerical modelling. The third part is dedicated to recent monitoring techniques of rock slope movement specifically based on radar interferometry. In part four are presented different approaches and models for analysing post-failure behaviour and mobility of bedrock landslides, including results from monitoring of rock avalanches triggered by underground nuclear explosion. Part five is an overview of both well known landslide events and new case records, with a particular emphasis on the influence of geological factors on failure mechanism. A very broad topic, such as volcanic landslides, is partially treated in part six through contributions on recent collapses of active volcanoes, landslides along volcano flanks and use of 3D slope stability analysis in assessing flank collapse at stratovolcanoes. In part seven, regional studies from various areas of the world which are particularly concerned with spatial and temporal pattern of the events as well as the features of landslide deposits and related implications for the emplacement processes, are reported. Effects on landscape evolution induced by massive rock slope failures, either at local or regional scale, are discussed in part eight. The last part, containing a single chapter, highlights some of the most pressing problems in the understanding of massive rock slope failure and suggests new directions for future research activities in this field. In all, the volume contains 32 papers by 63 authors from 16 countries. They consist of 15 invited key-note papers and 17 papers by invited discussants. All papers in the volume have been carefully reviewed by the editors and all have been thereby improved. We note that some of the chapters report the results of different research teams that agreed in producing joint papers which encompass different aspects of some crucial events. We are grateful to these authors for having accepted our encouragement in doing so. Last but not least, the Workshop was held in the dramatic landscape of the Italian Appennines. This was significant because recent work, examined during the Workshop field trip, has uncovered the widespread presence of prehistoric rock avalanche deposits in the region. This situation exemplifies the poor state of knowledge of the distribution and occurrence of massive rock slope failure that exists in many populated mountain areas. It reminds us that a large amount of research work remains to be done in this field in order to better assess an underestimated mountain hazard.

Stephen G. Evans (Canada), Gabriele Scarascia Mugnozza (Italy), Alexander Strom (Russia), Reginald L. Hermanns (Germany)

xiii Rome, May 2003

L’acqua disfa li monti e riempie le valli e vorrebbe ridurre la Terra in perfetta sfericità s’ella potesse “

Leonardo da Vinci (1452-1519)

ACKNOWLEDGEMENTS At the end of a long process such as the publication of this volume it is a pleasure and a duty to recognize all who contributed to the success of this enterprise, which was generated by the NATO Advanced Research Workshop on “Massive Rock Slope Failure: New Models for Hazard Assessment”. First of all, we want to express our thanks for the financial support of the NATO Science Programme under whose auspices was possible to organize the ARW and thus to realize this book. In particular, we wish to acknowledge the precious help of the Scientific and Environmental Affairs Division - Environmental and Earth Science Technology Programme, namely Dr. Alain Jubier and his assistant Lynne Nolan. In addition we thank the NATO Science Programme Representative, Prof. Francesco Mulargia, who attended the workshop and further encouraged us in undertaking the publication of this book. Of course, our deepest thanks go to all the authors who contributed to this volume. We are indebted to them for their personal commitment to this part of the NATO project and for their efforts in the Celano Workshop itself. We are particularly thankful to the keynote speakers who not only delivered stimulating presentations in Celano but produced excellent keynote papers for this volume. The financial support of Italian Scientific Institutions and Local Administrations, who contributed significantly to a successful and effective workshop are gratefully acknowledged. In particular, we recognized the Italian National Research Council (CNR), the National Institute for the Scientific and Technological Development of the Mountain (INRM), the University of Rome “La Sapienza” and the Faculty of Sciences of the same University. We also acknowledge support by the Regione Abruzzo, the Comunità Montana Aventino Medio Sangro and the ARSSA (Regional Agency for Agriculture). Special thanks are due to the Municipality of Celano namely the mayor, Italo Taccone, and the Deputy for the Environment, Loreto Ruscio, who gave fundamental support in carrying out the workshop in the town of Celano. The workshop itself could not have been held without the help of Gianluca Bianchi Fasani, Carlo Esposito, Alfredo Maffei, Salvatore Martino, Beatrice Salvati and Mario Floris. In addition, Gian Paolo Cavinato and Marco Petitta gave an important contribution in organizing and leading the field trip during the workshop. Finally, a special thanks to our respective institutions for their support and encouragement. The GeoForschungsZentrum in Potsdam and the University of Rome “La Sapienza” hosted the meetings of the editorial committee during the preparation of this book. SGE GSM AS RLH Rome, May 2003



LANDSLIDES FROM MASSIVE ROCK SLOPE FAILURE AND ASSOCIATED PHENOMENA S.G. EVANS1 Department of Earth Sciences University of Waterloo, 200 University Avenue West, Waterloo, Ontario, Canada, N2L 3G1 G. SCARASCIA MUGNOZZA Department of Earth Sciences University of Rome “La Sapienza”, P. le A. Moro, 5, 00185 Roma, Italy A.L. STROM Institute of the Geospheres Dynamics, Russian Academy of Sciences, Leninskiy Avenue, 38-1, 119334, Moscow, Russia R.L. HERMANNS Geological Survey of Canada 101-605 Robson Street Vancouver, British Columbia, Canada V6B 5J3 A. ISCHUK Institute of Earthquake Engineering and Seismology, Academy of Sciences, Dushanbe, Tajikistan S. VINNICHENKO The Focus Organisation, Dushanbe, Tajikistan Abstract Landslides from massive rock slope failure (MRSF) are a major geological hazard in many parts of the world. Hazard assessment is made difficult by a variety of complex initial failure processes and unpredictable post-failure behaviour, which includes transformation of movement mechanism, substantial changes in volume, and changes in the characteristics of the moving mass. Initial failure mechanisms are strongly influenced by geology and topography. Massive rock slope failure includes rockslides, rock avalanches, catastrophic spreads and rockfalls. Catastrophic debris flows can also be triggered by massive rock slope failure. Volcanoes are particularly prone to massive rock slope failure and can experience very large scale sector collapse or much smaller partial collapse. Both these types of failures may be transformed into lahars which can travel over 100 km from their source. MRSF deposits give insight into fragmentation and emplacement processes. Slow mountain slope deformation presents problems in interpretation of origin and movement mechanism. The identification of thresholds for the catastrophic failure of a slow moving rock slope is a key question in hazard assessment. Advances have been made in the analysis and modeling of initial failure and post-failure behaviour. However, these studies have been retrodictive in nature and their true predictive potential for hazard assessment remains uncertain yet promising. 1

E-mail of corresponding author; [email protected]

3 S.G. Evans et al. (eds.), Landslides from Massive Rock Slope Failure, 3–52. © 2006 Springer. Printed in the Netherlands.

4 Secondary processes associated with MRSF are an important component of hazard. These processes, which can be instantaneous or delayed, include the formation and failure of landslide dams and the generation of landslide tsunamis. Both these processes extend potential damage beyond the limits of landslide debris. The occurrence of MRSF forms orderly magnitude and frequency relations which can be characterized by robust power law relationships. MRSF is increasingly recognized as being an important process in landscape evolution which provides an essential context for enhanced hazard assessment. 1.




Landslides from massive rock slope failure (MRSF) are a major geological hazard in many parts of the world [e.g., 2, 29, 85, 96, 110, 267, 282, 325, 326, 342, 303] and have been responsible for some of the most destructive natural disasters of recent history [54; Evans, this volume]. They involve the initial rock failure mass but may also incorporate a variety of earth materials entrained in its path [2, 3, 148], and exhibit a range of landslide volumes covering at least five orders of magnitude between 105 and 1010 m3. We use the descriptor “massive” in this context not only to describe large or unusually large massive rock slope failure but also with reference to the resultant geomorphic and socio-economic impact. This hybrid definition of massive therefore encompasses a wide range of primary landslide and secondary phenomena. Landslides resulting from massive rock slope failure are frequently multiple phase landslides in which, for example, a disintegrating rock mass involved in an initial rockslide and subsequent rock avalanche becomes transformed into a massive, rapid debris flow which travels well beyond expected limits [e.g., 45, 134, 275, 276]. It is thus difficult to classify landslides from MRSF using recently proposed landslide classification schemes [e.g. 78, 72], although the scheme proposed recently by Hungr at al. [163] appears to be useful for the classification of flow-type landslides resulting from MRSF. Secondary processes associated with massive rock slope failure are an important aspect of the phenomena. They include landslide-generated waves and displaced water effects [e.g., 124; Blikra, this volume; Ghirotti this volume] and those associated with landslide dams [e.g., 204, 350]. Catastrophic secondary effects may be instantaneous or delayed and extend the impact of a landslide beyond the boundaries of the primary landslide debris. 1.2


The complexity of the massive rock slope failure phenomenon is best exemplified by a field example. Figure 1 shows a typical field situation encountered in the mountainous regions of the world. It shows rock avalanche deposits which originated in a dip-slope of folded 33-41° east dipping Cretaceous conglomerates in the Tonco valley, NW-


Figure 1. Aerial photograph of massive rock slope failure in the Tonco valley, northwest Argentina (see text for discussion).

Argentina [see Hermanns et al. this volume]. Different ages of rock avalanches can be inferred from boulder size, landslide- and breakaway-scarp morphology. In the north, at least two rock avalanche deposits overlie each other. The older larger rock avalanche, which originated from the collapse of a block 1.4 km wide, 1 km long and 50 m thick, dammed the Tonco valley. While the main river (flowing from NE to SW, entering from the right of the photograph) could erode through this natural dam, the tributary stream (from the N, entering from the top of the photograph) could not , thus causing the filling up of the small landslide-dammed basin by sediments. This deposit is overlain by a smaller rock-avalanche deposit which originated from the collapse of the northern break-away scarp of the older rock-avalanche and can be distinguished from the older deposits by larger and more abundant mega-blocks on its surface.

6 In the south the remains of an older rock-avalanche deposit exists. At this site the break-away scarp is much more deeply eroded and no mega-blocks are preserved at the surface of the strongly eroded deposit. A possible eroded break-away scarp suggests that at least one further landslide occurred in the wall of the deeply incised canyon which represents the outlet of the Tonco valley. Along this canyon, in some segments not more than ~1 m wide, possible landslide deposits would have been rapidly removed by erosion and thus whether a rock slope failure took place at this location is uncertain. Figure 1 illustrates several issues related to MRSF; the role of structural control (dip slope) on initial failure and rock avalanche recurrence, deposit morphology, the interpretation of possible sources of rock slope failure, landslide damming with related fill-up of basins and lake sediments, and the problems of interpreting the sequence of rock slope failure events. The example and the dating of this sequence of failures are described in more detail by Hermanns et al (this volume). 1.3.


Our objectives in this introductory chapter are to review the range of landslides associated with massive rock slope failure with an emphasis on recent research, new events, and new data on already well known rockslides, rock avalanches and landslides involving volcanoes. Where possible, examples will be drawn from the countries of the former Soviet Union, particularly Russia, Kyrgyzstan and Tajikistan in an attempt to fill an information gap on important rock slope failures from this region. We also briefly touch on recent thematic developments in the analysis and modeling of initial and postfailure behaviour of massive rock slope failure. Finally, we highlight recent research on spatial/ temporal patterns of occurrence which have led to new insights into the role of MRSF in landscape evolution. 2.

Initial Rock Slope Failure (Excluding the Failure of Volcano Slopes)

The mode of initial failure in bedrock slopes is strongly controlled by slope geometry and geologic structure, including rock mass fabric and lithology contrasts in the source slope [e.g. 50, 65, 69, 123, 127, 265, 299]. In sedimentary rocks and bedded volcaniclastic sequences, sliding frequently takes place along persistent planar discontinuities such as bedding planes (Figure 2), faults, joint surfaces, or lithologic contacts. In dip slopes, sliding takes place on bedding planes [e.g., 31, 48, 66, 84, 102, 115, 136, 140, 156, 259]. Sliding is frequently facilitated by the presence of bedding plane shears resulting from tectonic processes [e.g., 98, 333], gouge zones, or weak primary interlayers such as tuffaceous zones [104], shale, marl, or clay interbeds [138]. Dip-slope sliding may be facilitated by buckling [e.g., 312, 333] or by shear across bedding [104]. Initial failure in steep underdip slopes, and reverse slopes in bedded rock sequences is more complex and may involve buckling [154], toppling [68, 230] or break-out across bedding [100]. The dip of key discontinuities may vary in a given dip slope and the sliding surface may thus be concave [e.g., 100] or convex [e.g. 118, 140, 333]. Buckling and break-out are important failure mechanisms in these cases [Scarascia-Mugnozza et al., this volume].

7 Planar or gently curved discontinuities are also important in determining failure mode in plutonic [e.g., 140, 344] and strongly foliated metamorphic rocks [94, 95, 124, 140, 332]. In structurally complex rocks, failure is controlled by impersistent but closely spaced discontinuities independent of lithology [e.g., 217, 248] and may be complex in detail consisting of single or multiple wedges combined with local toppling.

Figure 2. The breakaway scarp and sliding surface of the Avalanche Lake rock avalanche, Mackenzie Mountains, North West Territories, Canada [102]. The sliding surface consists of a bedding plane in Paleozoic carbonates that form a dip slope, evident in background.

In very steep slopes, in steep mountain peaks or coastal cliffs for example, high angle detachment may occur along surfaces consisting of vertical to high angle tension cracks below which tension failure may follow suitably oriented discontinuities [e.g., joints] and at the base, failure occurs by shear through an intact wedge of material [e.g., 19, 80, 164]. In coastal cliffs the development of an erosional notch at the base of the cliff can reduce or eliminate the resistance of the passive wedge [19, 211] which may also result in a toppling failure [Stead, this volume]. Initial failure may be preceded by observable slope deformation. This is manifested in developing and widening tension cracks, increased rockfall activity and increasing disaggregation of the initial failure mass on the slope [e.g. 124, 169, 305, 306]. The description by Leopold Muller [228] of the development of the perimetral crack in the Vaiont slope as early as 1960 remains a chilling testimony to the need for correct interpretation of such movements [see Petley and Petley, this volume]. Time-to-failure

8 calculations may be made on the rate of movement of survey stations on the moving slope, as recently reviewed by Crosta and Agliardi [64]. The complexity of prefailure creep movements is described by Varga (this volume). Initial rock slope failure may occur without warning, however, as a result of a sudden earthquake trigger [e.g., 135, 217, 249], conventional and/or nuclear explosions [e.g., 5; Adushkin this volume] or as a result of sudden heavy rains [124, 275]. Some of the largest and most destructive massive rock slope failures in recent history have been triggered by seismic forces [181, 183, 264], such as the Tsao-Ling, Chi-fen-erh-shan, and Usoi rockslides, and the Khait and Mount Cook rock avalanches. Strong earthquakes may also simultaneously trigger a large number of massive rock slope failures over a large area as in the case of the M= 8.5 Great Alaska earthquake of 1964 [255], the M=7.0 1991 Racha earthquake in Georgia [24] and the 2002 M=7.9 Denali Fault earthquake in Alaska [83, 131].


Catastrophic Rockslides

A rockslide results from initial bedrock slope failure if distance of travel is limited, disintegration of the slide mass is incomplete, and if a significant amount of debris remains on the initial sliding surface [e.g., 32, 130, 247]. Catastrophic rockslides are characterised by high velocity despite the fact that the vertical displacement of the centre of gravity may be relatively small. High velocity requires some type of dramatic initial strength loss through such processes as brittleness of internal shears [165], or passive failure of intact rock in the toe region of the landslide [76, 296]. Although emplaced rapidly, rockslide debris frequently contains massive transported blocks of relatively undisturbed bedrock as at KĘfels, Flims, Usoi and Vaiont. In the case of the Vaiont rockslide, catastrophic failure occurred on a sharply curved (concave) chair-like pre-existing sliding surface in Cretaceous limestones interbedded with clay layers on which sliding took place [138, 277, 278,279, 311, Ghirotti, this volume]. The Vaiont debris reached a peak velocity of 20-30 m/s in ca. 25 s of movement during which the centre of gravity was displaced a vertical distance of only 130 m. The brittleness implicit in this high velocity continues to be the subject of discussion [e.g., 165; Petley and Petley, this volume]. 3.1.


The world’s largest known historical rockslide occurred in February 1911 in the Pamir Mountains of Tajikistan. The landslide was triggered by the M~7.4 Pamir Earthquake [117, 256, 271, 284, 351].


Figure 3. Sliding surface and breakaway scarp of the massive Rockslide Pass rockslide, Mackenzie Mountains, N.W.T., Canada. Sliding surface is a bedding plane in Paleozoic carbonates that dips at only 14 degrees.

The rockslide of ca. 2.2 B m3 in volume, blocked the Murgab valley and formed 75 km long Lake Sarez [272]. It also blocked its left tributary – the Shaddau creek and formed a minor lake of the same name (Figures 4 and 5). Sliding took place in Carboniferous and Triassic sedimentary rocks and during failure the centre of gravity was displaced in a vertical distance of about 500 m. The source zone of landslide is composed of two major rock units: 1) its upper part – PermoTriassic dolomite, limestone, gypsum, and anhydrite) and 2) its lower part – Carboniferous Sarez Formation (sandstone, schist, and quartzite). The stratigraphic bedding mainly dips 30 to 45° towards NNW, with marked local variations due to internal folding and effects of small local faults. These formations are separated by the Usoi Thrust dipping 60 to 80° towards SE. In addition, there is a secondary (eastern) shear zone dipping 50° towards NW and the collapsed block was formed by the wedge bounded by these two fault planes (Figure 6). The main part of the landslide body (Figures 5, 6, and 7) is composed of the debris of the Sarez Formation and its proximal part - of marble and shale with some subordinate gypsum, anhydrite and dolomite debris. Since the uppermost part of the source zone affected moraines of local hanging glaciers it is expected that some glacier ice may be buried in the proximal part of the landslide body. However, the internal composition of the Usoi Dam is not known. According to surface observations,


Figure 4. The Usoi rockslide shortly after it occurred. Scar is arrowed (white arrow at right). View is downstream. Lake Shaddau is to the left (S) and Lake Sarez is at the right (LS). Photograph was taken before the complete filling of Lake Sarez behind the rockslide dam (Plate I in Preobrajensky, 1920 [256]).

Figure 5. Satellite photograph of the Pamir Mountains, Tajikistan, showing the 1911 Usoi Landslide (U) which blocked the Murgab River to form Lake Sarez (LS) and Shaddau Lake (S).

significant variations in granulometry can be expected, with the grain-size composition of different parts of debris ranging from sandy-silty fines to blocks tens and hundreds of cubic meters in volume. Three main parts of the dam body can be identified;

11 1.



The southern part is the highest part of the dam with a maximum height of about 250-270 m above the present-day lake level. Its surface is covered by angular blocks of Carboniferous rocks from 2 to 20 m and no fines are visible on the surface. Just along the southern border of the dam body one can see the moraine deposits composed of boulders and pebbles with fine loamy matrix. Similar deposits can be seen near the Shaddau Lake. Since they contain some granite boulders it is clear that the moraine material, at least that, resting at the Shaddau mouth, was scraped by the moving rockslide from the Murgab valley bottom. The Central part of the dam body rises up to 100 m above lake level on the average and is bounded by an expessive escarpment facing to the downstream slope of the dam. It is clearly seen on the space images that this part of the dam was formed by a tremendous block of Carboniferous deposits (Figure 6) with preserved stratigraphy, though intensively fractured. At some places its surface is the "natural slope surface" displaced from its original position. Northern part of the dam is the lowest one and is only 38-45 m above the lake level. It is covered by the large blocks of sandstone and shirts with the diameters from 2 to 20 m without fines. It can be expected that the frontal and proximal blocky zones represent outer parts of the huge block which central part remained more intact though fractured.

The dams surface abut to the foot of the scar and right flank of the Murgab River is covered by the deposits of the subsequent rockfalls, debris flows and mudflows with limestone, marble, gypsum and fine material of moraine deposits that came from the glacier valleys above the scar. This secondary deposits extents along the right part of the downstream slope of the dam up to the head of the erosional canyon. The latter is forming by the water filtrating through the dam. Previously it was also eroded by the mudflows from the glacier valleys above the scar that went in this direction. But since 1934-35, when rockfall from the scar wall diverged mudflows towards the lake, canyon is eroded only by springs. Several north-south trending arcuate escarpments are clearly visible in the central and southern parts of the dam (Figure 6). The first step from the upstream looks like the ridge with very steep slope (60-70 degrees). Investigations performed in the 1960s and 1970s show that infiltration in the dam takes place mainly in the narrow zone 100-130 meters below water level and that the dam' body below is impervious. The level of the lake is characterised by slow but gradually increase with annual variation about ± 6 m. Filtrating water first appeared as the lake level reached about 3100 masl (about 160 m below the present day level) and at present forms 57 powerful springs in the erosional canyon 140-150 m below the lake level (Figure 6). The discharge from all springs is 4575 m3/s (season variation) and depends on the lake level. No evidence of internal erosion has been found to date. Filtration takes place through the opened fractures in the uppermost blocky unit. No vertical deformations have been directly measured on the landslide’s surface since 1947. However, according to measurements performed in 1915 and in 1967, differences in elevations have reached 1-20 meters.


Figure 6. Satellite photograph of Usoi rockslide dam. Approximate distal boundary of debris is indicated. . A = Massive intact slab mentioned in text; LS = Lake Sarez; S = Shaddau Lake

Analysis of the Usoi Landslide body and source zone allow a reconstruction of the mechanism of the earthquake-triggered slope failure. It is hypothesised that the huge wedge of the rocks slid down rapidly as a single block. When it struck the valley bottom and the opposite slope it expanded along the valley axis, mainly in a downstream direction, which led to the formation of the above mentioned arcuate escarpments,

13 downthrown downvalley. Lack of seepage through the lower part of the blockage can be explained as a result of the intensive comminution of the debris that compose the internal part of this natural dam forming a substantial impermeable core [239, 300, 301] as observed at the numerous dissected rockslide dams elsewhere in Central Asia [Abdrakhmatov & Strom, this volume, Hewitt, this Volume].

4800 Distal limit of debris Top of Source Zone

4400 4000 3600

Level of Lake Sarez

3200 0




Figure 7. Profile of Usoi rockslide. Level of Lake Sarez (~ 3255 m.a.s.l.) is indicated. Profile suggests a fahrboschung of ~ 12 degrees.



Importantly, the debris of some rockslides exhibits partial mobility in which some of the debris disaggregates completely and is transformed into a fast moving debris flow or even a secondary rock avalanche, thus extending the distal reach of the movement [e.g. 98, 188, 231, 302]. An example is illustrated in Figure 8. The mechanism of distal flows is thought to involve undrained loading of valley floor sediments (see Section 7 below).


Catastrophic Spreads

Other styles of catastrophic bedrock slope instability develop in situations where a thickness of hard resistant caprock overlies weaker softer ductile rocks, such as tuffs, shales, or flysch sediments. This style of instability may involve toppling [89] and/or spreading of the subjacent weak layer and has catastrophic potential. Such movements are common in layered volcanic successions [90] where the caprock is lava, and in the thrust and nappe belts of the Alps [e.g., 199] and the Rocky Mountains of North America [174] where the cap rock is frequently overthrust Proterozoic or Paleozoic limestone and the subjacent material is Cretaceous shale.


Figure 8. Vertical aerial photograph of North Nahanni rockslide, N.W.T. Canada (98). The landslide was triggered by the October 1985 North Nahanni earthquake. Note tongue-shaped distal flow that ran out from main debris deposit (98).

Spreads can be catastrophic. On June 24, 1765, a catastrophic spread destroyed the village of Roccamontepiano in central Italy resulting in more than 500 deaths [11, 62]. At the site, 30 – 40 m of resistant Pleistocene travertine forms a rigid cap rock which overlies weaker subjacent Upper Pliocene marine clays. Failure occurred as deformation in the subjacent clay led to tensile failure in the travertine cap which then was free to load a soft clay foundation leading to a catastrophic collapse of the sequence [75]. The catastrophic collapse took place only hours after the first signs of instability in the travertine cap. 5.


Rockfall involves the fall of a rock mass following initial detachment from a very steep rock slope, its disintegration and subsequent movement which may involve bouncing,

15 rolling, or sliding generally down the steep source rock slope. Massive rock falls frequently transform into highly mobile debris flows or rock avalanches as in the case of the 1970 Huascaran, the 1959 Pandemonium Creek, and several recent events in the European and New Zealand Alps [14, 99, 217, 249]. Rockfalls retain their identity as a landslide type when the initial failure volume is less than the threshold volume where mass flow does not result. This process has been termed fragmental rock fall [97]. The threshold volume for this transition varies with the source material. In hard non-porous rocks it is ca. 1 M m3 and in soft porous rocks it is about 500,000 m3. Although rockfalls, thus defined, are much smaller than rock avalanches and rock slides they are more frequent and may be highly destructive over a limited area. Rockfalls can be a major hazard in mountain communities [8, 33, 97]. At Lecco, in the Italian Alps, for example, about 15,000 m3 of dolomite broke away from a cliff above the town in 1969 [8]. The mass broke up and rockfall fragments smashed into homes below claiming 8 lives [8]. In the case of the Malpa rockfalls, which occurred in the Kumaun Himalaya of India in 1998, rockfalls dammed a steep watershed and the resulting dam-break debris flow overwhelmed Malpa Village claiming the lives of 221 people [243]. Rockfalls are also a common hazard in transportation corridors that traverse rocky terrain [e.g., 27, 161]. Rockfalls may attain high velocities [8, 97]. In 1996, a remarkable rock fall occurred in Yosemite National Park [318, 346] in which a large block of granite (est. volume 27-62,000 m3) detached from a cliff, slid down a 50º slope for about 185 m and then launched itself into the air. The rock then fell in free-fall before impacting on a talus slope 500 m below its launch point. The impact after the free fall was measured on seismographs within a 100 km radius and based on these records it is estimated that the rock impacted with a velocity of 146 m/s (525 km/hr) (318). The rockfall generated a wind blast which knocked down trees for a distance of 300 m. Rockfalls are the most frequent landslide type triggered by earthquakes [181].


Rock Avalanches



Initial bedrock failure results in a rock avalanche when the rockmass disintegrates, leaving the source surface completely, and travels a downslope distance far from its origin (e.g., 103; Figure 9). The term was first used to describe the 1903 Frank Slide in the Alberta Rocky Mountains by R.G. McConnell and R. W. Brock [214] in 1904.


Figure 9. Perspective view to the southeast of the 1997 Mount Munday rock avalanche on Ice Valley Glacier, Waddington Range, southern Coast Mountains. This digital image was prepared from aerial photographs flown on August 20, 1997, and consists of a DEM with an orthophoto drape. Note flow lines in the debris. Elevation of the top of the source area is 3000 m.a.s.l. and the lower tip of the debris is at 2100 m.a.s.l. The length of the rock avalanche path is 4.7 km [95].

17 Rock avalanches are extremely rapid movements. At the 1970 Huascaran event statements of eye witnesses suggested a mean velocity for the movement of 75 m/s (270 km/hr) with peak velocities perhaps as high as 277 m/s (997 km/hr) being indicated by the analysis of the ballistic trajectory of huge granodiorite blocks [249]. As recently reviewed by Wieczoreck et al. [346] some rock avalanches generate destructive winds. The 1984 Mount Cayley rock avalanche traveled so rapidly that it generated winds that not only felled mature trees but drove spear-shaped wood fragments into solid tree trunks along its margins (Figure 17 in [71]). Wind velocities in excess of 30 m/s (108 km/hr) are required to inflict this type of damage on mature pine trees [71]. During the course of a rock avalanche the debris may exhibit dramatic mobility effects in one or more of the following ways; 1. the surface of the moving debris shows superelevation as it passes through bends in its path (Figure 9; e.g., 99, 134). 2. abrupt changes in the direction of travel; the debris may run at right angles (Figure 9; e.g., 84, 94, 95), or even turn a full 180° to the original movement direction [99]. 3. the debris may run over significant obstacles in its path [132 ; Hewitt, this volume], or run-up a considerable distance on opposing valley sides [91]. At the prehistoric Avalanche Lake rock avalanche, for example, debris ran up the opposite slope to a height of 640 m above the pre-landslide valley floor (Figure 10) [91, 102]. Excessive run-up has also been observed at rock avalanches in the Karakoram Himalayas as described by Hewitt [148; this volume]. Rock avalanches with volumes above about 1 M m3 generally show a decrease in fahrböschung with volume [e.g., 55, 74, 123, 153, 186, 198, 270; Legros, this volume]. Landslides resulting from rock slope failures below this threshold volume may also show similar mobility behaviour when mobility is enhanced by such mechanisms as impact collapse, undrained loading, or fluidisation [e.g., 3, 105, 168, 266]. Interesting quantitative data on the dynamics of rock avalanche motion has been recorded during the study of the artificial rock avalanches (with volumes up to 4 M m3), triggered by underground nuclear explosions [5, 6, Adushkin this volume] on Nova Zemylaya. 6.2.


One of the most destructive rock avalanches in recent history occurred on July 10, 1949 in Tajikistan when a large rock avalanche (Figure 11) caused by the Ms ~ 7.6 Khait earthquake destroyed the town of Khait with the loss of as many as 24,000 inhabitants [200]. The initial rock slope failure that caused the Khait rock avalanche occurred on the northern slope of the Chohrak Mountain (el. ~ 3076 m) in the upper reaches of the Obi-Khaus-Dara River, a tributary of the Obi-Kabud river, which, in turn, flows into the Surkhob River. Initial failure involved about 80 Mm3 of Paleozoic


Figure 10. Excessive run-up at Avalanche Lake rock avalanche, Mackenzie Mountains, N.W.T. The landslide moved from the right and ran up the steep slope of the wall running out on the surface of The Shelf, 640 m above the pre-landslide valley floor [84, 91, 102].

gneiss. The left margin of the source zone coincided with the fault plane of the large fault and the collapsing mass slid along this smooth plane. It is thought that this fault was ruptured during the earthquake and displaced up to 0.7-0.8 m. The 1949 rock avalanche was not the first massive rock slope failure in this area. Just upstream from the 1949 failure, the Obi-Khaus-Dara River is blocked by an ancient rockslide, several Mm3 in volume, that dams a small lake (Figure 11). According to lichenometric dating, this failure occurred during the period from 1740 to 1880 AD [233]. One of the characteristics of the Khait area is the wide distribution of loamy loess deposits on the mountain slopes and river terraces. According to eyewitness reports, the earthquake was preceded by several days of rain (Leonov, personal communication) and the loess mantle appeared to be significantly saturated. Intensive seismic shaking led to widespread sliding of saturated loess downslope. As a result, the initial rockslide entrained a large amount of saturated loess, also mixed with water from the Obi-KhausDara River, and finally its volume increased to about 400 Mm3, approximately 5 times the original failure volume. All this mass rapidly moved downstream in the Obi-Khaus-Dara River and entered the Obi-Kabud valley. It is estimated that the velocity was in excess of 30 m/s; a powerful air blast pushed ahead of the landslide which destroyed buildings, uprooted trees and threw them for hundreds of metres [292, 293, 294]. The debris spread over the valley floor forming a fan-shaped apron (Figure 11) from 12 to 75 m in thickness


Figure 11. Aerial photograph mosaic of Khait rock avalanche, Tajikistan, which was triggered by 1949 Khait earthquake. The landslide source and path is outlined approximately. Town of Khait destroyed at A and village of Khisarak destroyed at B. As many as 24,000 people may have perished in these settlements. D is a prehistoric landslide dated by lichenometry to the 18th Century and C is a lake dammed by the landslide. Inset: aerial view of the Khait landslide toward the source area.

with numerous mounds (molards) formed by water spouting (“ground fountains”) in the consolidating debris [294]. The debris buried the town of Khait and the neighbouring small village of Khisarak (Figures 11 and 12). As many as 24,000 people may have been killed in these settlements. The total runout of the Khait rock avalanche was about 11 km over a

20 vertical distance of about 1500 m (H/L = 0.14; fahrboschung ~ 8 degrees); the final 5700 m of travel was over an average slope of about 3 degrees.

Figure 12. Aerial view of the Khait rock avalanche showing the source massive rock slope failure (top right) the middle path, where massive entrainment of loess took place, and part of the deposit which buried the town of Khait (bottom left) (USGS photograph).


Debris Flows Induced by Rockslope Failure

When debris from a rockslope failure impacts on channel or valley floor sediments a destructive debris flow may be mobilised that travels well beyond the margins of the initial landslide debris [3]. In the Swiss Alps, for example, the Bonaduz Gravels, which are about 50 m thick and extend up to 12 km upstream from the margins of the prehistoric Flims rockslide (est. vol. 9 B m3) have been interpreted as resulting from a debris flow generated by the mobilisation of saturated valley fill by the rockslide [3, Poschinger et al., this volume]. Splash zones are sometimes formed around the debris by fluidised material displaced from beneath the debris [214] and compressional deformation structures may be formed in sediments beneath or adjacent to the debris sheet [3, 52, 254]. Similar effects may result from impacts of rockfall debris on saturated colluvium or talus forming the lower part of a valley side slope [Hungr, this volume]. At Fidaz, Switzerland a catastrophe occurred in 1939 when a rock mass fell (est. vol. 100,000 m3) from the scarp of the prehistoric Flims rockslide and impacted a colluvial slope below. The rockfall entrained colluvial material increasing the volume of the debris to about

21 400,000 m3. The rockfall/debris flow travelled rapidly downslope reaching a velocity of over 40 m/s. The debris reached a point 1.3 km from the source cliff and in its travel the landslide overwhelmed a children’s sanitorium located below the cliffs causing 18 deaths [160, 232]. A similar event occurred in 1953 in Modalen, Norway, when a rockfall (est. volume 10,000 m3) fell about 200 m onto a 30° talus slope [21, 192]. The impact triggered a flowslide in the talus (est. volume 100,000 m3) which swept downslope and across level farmland, destroying two farm houses. The debris reached 200 m beyond the foot of the talus. A well documented case occurred in Hawaii in 1981 [178]. In this case 500,000 m3 of volcanic rock fell from a steep lava cliff, impacted on the canyon floor below and mobilized a large volume of material thus generating a rapidly moving debris flow that ran out a distance of 4.6 km. The volume of the debris flow was approximately 2.5 M m3, about 5 times the volume of the initial rockfall [178]. More recently, a similar event occurred on Vancouver Island, Canada in 1999 (Figure 13). Detailed mapping has indicated that the volume of the initial rock failure was about 300,000 m3. Impacting a saturated colluvial slope below, the rockfall entrained a further 350,000 m3 of material. The debris flow, with a volume approaching 700,000 m3, then turned sharply (Figure 13) and ran down the Nomash Valley for a further 1.75 km on an average slope of only 3 degrees. Such responses of valley sediments and valley side deposits suggest that the impact loading of saturated materials can generate pore pressures which reduce the frictional resistance at the base of the moving debris; undrained loading generated by rapidly moving debris may thus be an important mechanism in explaining the anomalous mobility of certain rock avalanches and rockfall events [3, 266]. Further, the volume of entrained material from the landslide path may significantly enhance the volume involved in initial rock slope failure.


Landslides from Volcanoes

Volcanoes are highly unstable piles of ejecta and eruptive products and as such are the locus of frequent episodic landslide activity [7, 215, 219, 269, 327, 328], either in association with volcanic eruptions or during periods of qiuescence. Landslides from volcanoes include large-scale flank collapse and smaller scale landslides (volcanic debris avalanches) involving part of the edifice. Sometimes these massive rock slope failures are transformed into lahars (volcanic debris flows). 8.1.


Massive catastrophic failure of the volcanic edifice itself has produced some of the largest sub-aerial landslides on earth; the Mount Shasta debris avalanche deposit, for example, is estimated to have a volume of 45 B m3 [59] and the 18,520 y BP Nevado de Colima debris avalanche of Central Mexico is estimated to have a volume


Figure 13. The 1999 Nomash River rockfall/debris flow, Vancouver Island, Canada. A rock mass failure originated in the top right of the image and entrained a large volume of material from the valley side. The resultant debris flow traveled a further 1.75 km downstream on an average slope of only 3 degrees. Image consists of DEM draped with orthophoto.

in the range 22 - 33 B m3 [298]. At Socompa volcano in northern Chile, a total of 53 B m3 was displaced during the collapse of its northwestern flank at ca. 7,000 y BP [331] including a massive debris avalanche (est. vol. 26 B m3) that flowed up to 40 km from the cone. Many volcanoes in the world have now been shown to have experienced major flank collapses during their existence [42, 112, 285, 286, 287, 289, 317, 320]. Multiple flank collapses have been documented at a number of volcanoes including Shiveluch volcano, Kamchatka, Russia where 8 collapses have been documented in the last 10,000 years [16]. Flank collapses may be triggered by magma emplacement, local tectonic displacements, oversteepening and/or overloading by the deposition of eruptive

23 products, oversteepening and incision of the edifice by stream erosion, the generation of pore pressures generated as a result of magma intrusion or seismic shaking, and can also be induced by the degradation in the shear strength of materials within the edifice by low temperature hydrothermal alteration [260, 261, 286, 287, 202, 328, Reid and Brien, this volume]. Spreading of weak volcano foundations, first noted at some Indonesian volcanoes by van Bemmelen [321], may also lead to the deformation of a volcanic edifice and its subsequent catastrophic collapse [322, 323]. As with non-volcanic rock avalanches, the mobility of volcanic debris avalanches is directly related to volume. However, for a given volume, volcanic landslides resulting from flank collapse are more mobile, i.e., a lower fahrboschung, than their non-volcanic counterparts [198, 285, 330]. This may reflect amongst other factors, the common presence of large amounts of low strength hydrothermally-altered material rich in smectite in the landslide debris, the frequent inclusion of summit snow or ice caps in the failed mass which results in an increased water content, and a high degree of entrainment of material from the path of the landslide. At least 11 flank collapses have taken place in the world since 1850 [286, 287, 288] including the massive eruption-related landslides at Mount Augustine, Alaska in 1883 (300 M m3; [288]), Bandai-San, Japan, in 1888 (1.5 B m3), Bezymianney, Kamchatka Peninsula, Russia, in 1956 (800 M m3), Shiveluch, Kamchatka Peninsula, Russia in 1964 (1.5 km3 ) and Mount St. Helens, Washington, U.S.A. (2.8 B m3) in 1980. The 1888 Ritter Island lateral collapse, off the coast of New Guinea [177, 334] involved the mobilization of as much as 5 B m3. 8.2.


Smaller scale landslides also occur on the slopes of volcanoes without involving the failure of a large part of the volcano’s superstructure. Initial failure volumes are typically less than 100 M m3 and commonly involve mechanically weak pyroclastic debris or hydrothermally altered rocks. Because of this, the “rock” mass involved in initial failure is easily fragmented and quickly becomes transformed into a rapidly moving rock/debris avalanche. Volcanic landslides of this type may be triggered by an eruption, small steam explosions, earthquake shaking, heavy rains, or glacier unloading [e.g., 60, 101, 125, 225, 339]. A typical example is the large prehistoric debris avalanche (est. volume 91 M m3) on the slopes of Lastarria volcano in the Chilean Andes [229]. The initial slip surface parallels pyroclastic bedding in the volcano and the maximum travel distance of 6.7 km yields a fahrboschung of 8.5°. The pyroclastic debris is highly fragmented; near its distal limit, the debris ran up and over a small cone 125 m in height suggesting a minimum emplacement velocity of 50 m/s (180 km/hr). In historic times, the Ontake debris avalanche/debris flow, triggered by the 1984 Nagano earthquake (M=6.8) in central Japan, is the best documented volcanic landslide of this type [e.g., 86, 225, 266, 329]. Initial sliding (est. vol. ~ 36 M m3) involved failure of welded and non-welded scoriaceous pyroclastic rocks and lavas along a partially altered clayey pumice layer on the southeastern flank of Mount Ontake. The rockslide was quickly transformed into a rapid, highly-mobile debris flow which swept

24 down the Denjo, Nigori, and Ohtaki rivers. Entrainment of deposits in these river valleys augmented the volume of the moving mass to an estimated 56 M m3 [86]. The maximum travel distance was about 13 km over a vertical distance of 1600 m yielding a fahrboschung of 7º. Mobility of the debris avalanche may have been enhanced by the undrained loading of torrent deposits in the Denjo valley [266]. The average velocity of the landslide, based on eyewitness accounts, was 22 m/s with a maximum velocity estimated from super-elevation in bends of 36 m/s. Volcanic rock/debris avalanches are frequently transformed into debris flows that travel beyond the limit of the debris deposited by the primary event [e.g., 101, 276] either by an immediate direct transformation of part of the debris into a distal mass flow or by the breach of a landslide dam formed by the debris sometime after it is deposited. Debris avalanches from volcanoes have caused significant disasters in recent times. Most recently, in October 1998, a rock avalanche (est. vol. ~ 200,000 m3) from Casita Volcano, Nicaragua, triggered by the rains of Hurricane Mitch, initially travelled about 3.2 km, [184, 275, 276, 283]. A debris flow originated from the rock avalanche debris, transforming the rock avalanche deposit into a rapidly moving fluid mass, which overwhelmed the villages of El Porvenir and Rolando Rodriguez 3 km further downslope. The debris flow travelled further downstream up to a distance of about 18 km from the limit of the initial rock avalanche deposit partially destroying many other villages. The immediate and delayed consequences of the event resulted in the deaths of approximately 2,500 people. 8.3.


Lahars are commonly associated with eruptions and may be triggered by a variety of processes including the melting of snow or glacier ice by hot ejecta, the ejection, or breaching, of the waters of a crater or caldera lake, the transformation of glowing ash avalanches, pyroclastic surges, or the transformation of an eruption-triggered flank collapse – debris avalanche [e.g., 338]. Some prehistoric lahars are enormous. Mothes et al. [227] have described the Chillos Valley lahar (est vol. 3.8 B m3) in Ecuador that originated on the northern slopes of Cotopaxi volcano and travelled 326 km to the Pacific at about 4500 y BP. The lahar was generated by an ash flow following a small sector collapse which melted part of the volcano’s ice cap and was transformed rapidly into the lahar. Further examples include a massive lahar (est. volume 1.8 B m3), dating from the late Pleistocene, which originated from the edifice collapse of Citlaltepetl (5675 m) an ice-capped stratovolcano (43) in the trans-Mexican volcanic belt. In the Cascade Volcanic Belt of the Pacific Northwest of the United States, work by Vallance and Scott [319] built on the classical descriptions of Crandell and Waldron [61] and Crandell [58] of the equally enormous Osceola Mudflow (est. vol. 3.8 B m3), a lahar that originated in a sector collapse from the summit of Mount Rainier volcano about 4800 y BP. The Osceola Mudflow is estimated to have had a velocity of 19 m/s (68 km/hr) at 40-50 km downstream from its source. The 1980 collapse of Mt St Helens generated a massive lahar (est. vol. 108m3) in the North Fork of the Toutle River by the delayed transformation of the flank collapse rock avalanche [175]. Scott [274] notes that it was formed by the dewatering of the avalanche deposit and slumping and

25 erosion of its surface after a delay of about 5 hours. Flow velocity ranged from 6-12 m/s. Significant volumes of ice and snow may be contained in the intitial failure volume on snow/ice clad volcanoes and their melting contributes to the downstream transformation of flank collapse avalanches into long run-out lahars that may travel more than 100 km in valleys draining the source volcano [207].


Morphology, Internal Structure and Sedimentology of Massive Rock Slope Failure Deposits

The characteristics of MRSF deposits (morphology, internal structure and sedimentology) are important from a number of perspectives. First, they are essential in the initial identification of MRSF deposits. The MRSF literature is replete with examples of MRSF deposits that were initially interpreted as glacial deposits [e.g., 145, 238, 253]. As noted by Watson and Wright [335] even the colossal Saidmarreh rock avalanche, in the Zagros Mountains of Iran, was initially mapped as being the result of Pleistocene glaciation. Erroneous interpretations such as this led to underestimation of the extent of the occurrence of MRSF in the mountainous areas of the world especially in the Alps and Himalayas, which has been corrected only relatively recently. Second, they give important evidence on the processes of fragmentation, transport and final emplacement/consolidation [e.g., 70, McSaveney and Davies, this volume; Poschinger et al., this volume] as well as the mechanism of the interaction with the substrate that the debris travels over [e.g., 254]. Morphological features of the deposit surface give indications of flow patterns [e.g., 216], emplacement sequence [e.g., 102] and post-depositional movement associated with consolidation of the debris. Sedimentological studies of prehistoric MRSF deposits have also shed light on the precise sequence of initial failure and possible secondary processes, such as distal debris flows initiated by impact loading of valley sediments or by dam break. In some cases this has led to a reinterpretation of prehistoric events as in the case of the Pleistocene debris-avalanche deposit from Nevado de Colima volcano, Mexico, which is now thought to largely consist of a 10 B m3 dam breakout flow [41]. Third, knowledge of massive rock slope failure deposits and their internal structure is a key requirement in assessing the present stability of deposits that form landslide dams, either with respect to slope stability, piping, and/or resistance to overtopping [e.g., 44, 341]. Fourth, MRSF deposits are important foundation materials. Many built dams are constructed on natural landslide dams [e.g., 136, 308] and their material properties are required for geotechnical foundation design. In addition, many settlements in mountainous regions are situated on MRSF deposits and knowledge of their geotechnical characteristics is important in assessing their seismic response, particularly with respect to site amplification, and thus is important in seismic zonation. Lastly, as the number of pits excavated in MRSF deposits around the world testifies, they are important sources of borrow material and thus a knowledge of their gradation has important consequences for their potential exploitation as a mineral resource.

26 Although a number of studies have been carried out on dissected landslide deposits in the present landscape, an important contribution to the knowledge of MRSF deposits has been made by a study of them in the geological record [e.g., 1, , 20, 113, 226, 349] where they have been termed “megabreccias” [e.g., 35, 195, 201]. 10.

Mountain Slope Deformation, Non-Catastrophic Rockslides, Catastrophic Failure Thresholds, and Problems of Origin

Evidence of mountain slope deformation is widespread in the mountain regions of the world [e.g., 9, 25, 26, 49, 50, 51, 63, 133, 212, 307, 324]. Mountain slope deformation consists of slow, deep-seated movement of a large rock mass that commonly exhibits loosening and fracturing in the sub-surface and signs of displacement on the surface of the slope itself slope surface. The process is termed “gravitational spreading” by Varnes et al. [324] and "sagging” by Hutchinson [166] who regards it as an early phase in the development of deep-seated landsliding. This type of slope movement may involve movement along discrete shear surfaces and/or deep seated mass creep [166]. It is commonly manifested in topographic features such as cracks, fissures, trenches, antislope (counter) scarps at mid or upper slope locations, and, in some cases, slope bulging at lower slope locations [e.g., 25, 26]. These linear geomorphic features may be collectively termed “sackungen”, after the German word for sagging [e.g., 213]. Frequently, these surface features occur without well defined headscarps, lateral scarps, or lateral shear zones suggesting that slope movement is occurring without the formation of well defined shear surfaces in contrast to rockslides described earlier. Mountain slope deformation features (or sackungen) present difficulties in landslide hazard assessment in that (a) the precise movement mechanism is difficult to establish and thus to analyse, (b) the potential for the development of catastrophic detachment is difficult to evaluate, (c) the relationship to tectonic processes may be complex [e.g., 348] and (d) the origin of the linears themselves may be problematical, i.e. whether they represent a tectonic fault formed by an earthquake or a response to mountain slope deformation, is not always clear [e.g., 212, 310]. This issue is complicated by the fact that many examples of deep seated slope deformation occur in close association with pre-existing faults which may themselves be active. The problem of origin is also of concern in seismic hazard assessments for major engineering facilities in many countries such as Norway, Japan, the Cordillera of western Canada and neighbouring areas of the United States [e.g., 309]. The precise mechanism of mountain slope deformation is difficult to establish even at sites that have been extensively studied [235]. In the case of the slope above B.C. Hydro's Wahleach Power Station, in the Fraser Valley of southwestern British Columbia [224] an extensive slope stability investigation was began in January 1989 when the steel lining of a conduit tunnel leading down to the generating station was ruptured by slope movement. A major subsurface investigation established that no discrete sliding surface pre-existed or developed as a result of the movements. Some sagging slopes are transitional to slow-moving rockslides. At Dutchmans Ridge, 1.5 km upstream from the Mica Dam in the Columbia Mountains, Canada, for example, detailed investigations by B.C. Hydro have shown that movement has taken place in the lower portion of the slope [170, 218, 222, 223]. Shortly after reservoir

27 filling, downslope movement of about 10 mm/year was detected. It has taken place on a tectonic fault that dips toward the valley at 29º and involves 115 M m3 of fractured gneissic and schistose rock. As Steidl and Riedmuller [297] recently illustrated, deep-seated gravitational slope deformation can quite complex in detail. In their study, involving a slope in metamorphic rocks in the Austrian Alps, translational sliding at the head of the slope movement was driving deep-seated toppling at its base. Indeed, where the geological structure is favourable mountain slope deformation frequently involves some degree of toppling [e.g., 50, 120, 251, 263] and/or sliding movement of a slope that has been previously disturbed by toppling. At the Clapière site in the French Alps, a moving slope consisting of migmatitic gneiss suddenly accelerated beginning in 1977 to create a major public safety issue [22]. The movements appeared to involve the slow sliding of a previously toppled slope in which the originally-vertical foliation in the slope decreased in dip by toppling forward thus creating a sliding surface for later movement [106, 107]. A major problem in the interpretation of massive rock slope movements is the prediction of the future behaviour of the slope and the establishment of conditions that determine the transition to possible catastrophic failure once ongoing mountain slope deformation has been detected [e.g., 120, 345, Bhasin and Kaynia, In press]. For example, the 1987 Valtellina rock avalanche took place in a slope that had undergone significant non-catastrophic deformation in the post-glacial period [124]. The boundary between non-catastrophic ductile flexural toppling and catastrophic brittle block toppling has been examined by Nichol et al. [230]. Large-scale non-catastrophic rockslides are common in many mountainous areas of the world, particularly in metamorphic rocks [e.g., 87, 119, 155, 171]. Rates of movement may be in the range of 1-2mm/yr. 11. Analysis and Modeling of Initial Rock Slope The quantitative analysis of rock mass discontinuities coupled with the analysis and modeling of initial rock slope failure was slow to develop in engineering geology. One of the earliest quantitative attempts to relate the kinematics of massive rock slope failure to geological structure was that of Fuganti [115]. A classic early paper in the use of limit equilibrium analysis in the collapse of a rock slope was by Hutchinson [164]. The development of stability analyses based on limiting equilibrium for rock slopes and the use of stereographic projection techniques for kinematic analysis is summarized in Hoek and Bray [149] and more recently by Norrish and Wyllie [234]. Examples of complex limit equilibrium analyses of rock slopes are the work of Voight et al. [330], who analysed the 1980 failure of Mt. St. Helens, Hendron and Patton [138] who analysed the Vaiont rockslide, and Kaiser and Martin [208] who analysed a rock slope at the Revelstoke damsite, in British Columbia. In extending analytical methods beyond limit equilibrium, early attempts in the use of finite element analysis in the analysis of large-scale natural rock slope movements were made by Kalkani and Piteau [180] and Krahn and Morgenstern [194]. Kohlbeck et al. [191] used a finite element model to analyse stresses in an alpine valley. Examples of later work using finite element techniques to analyse natural rock slope movements includes work by Hutchinson [167].

28 With the development of the Distinct-Element Method [73, 172] and the advent of UDEC and FLAC sophisticated numerical analyses of rock slopes became possible. Early work such as Pritchard and Savigny [257, 258] demonstrated their use in the analysis of movements in open pit slopes and natural rock slopes. Although extensively used in the analysis of excavated slopes in mines [e.g., 296] and construction sites little has been published, until recently, on the application of numerical methods to natural rock slope movements. Recent examples include work by Kimber et al. [187] on Portland limestone cliffs, Benko and Stead [17] on the 1903 Frank Slide, Voight [327] on the failure of andesitic volcanoes and lava domes, and others [9, 230, Bhasin and Kaynia, In Press]. These works have demonstrated the power of such methods to analyse kinematic scenarios as well as to model the actual failure mechanisms of the case in question. Such analysis requires detailed geological models and the accurate laboratory characterisastion of the materials involved in the movement [209]. In this volume, the numerical analysis of initial failure of natural slopes is reported in chapters by Scarascia Mugnozza et al., Eberhardt, Stead and Coggan, Genevois et al., and Merrien-Soukatchoff and Gunzberger. The reader is referred to a recent detailed review of numerical modeling for rock mechanics and rock engineering by Jing [176]. 12.

Analysis and Modeling of Post-Failure Behaviour of Rock Slopes

Substantial success has been achieved in the development of analytical models that simulate the bouncing and rolling of rockfall and rock-release fragments, the run-out distance and the velocity profile of rockfall [e.g., 28, 13, 158, 190, 246]. Calibration against observations of actual rockfalls has allowed a more rigorous selection of model parameters which govern rockfall motion [e.g., 12, 46, 97]. Transport mecahnisms of long run-out rock avalanches have been the subject of considerable research over the last twenty five years [e.g., 39, 40, 74, 157, 295]. Numerous mechanisms have been proposed to account for long run-out. These include various types of bulk fluidisation of all or part of the debris sheet, the generation of low friction (air, melted rock or liquefied substrate, snow ice) at the base of the mobile debris sheet, and those that invoke changing mass/rheology. Attempts to model these processes have met with a degree of success and have ranged from simple slide-block frictional models to those involving complex rheology [e.g., 157, 159, 162, 173, 268, 295, Crosta, this volume]. Retroactive simulation, or hindcast analyses, of individual catastrophic events utilising these models, in which model parameters are adjusted by trial and error, has successfully replicated run out distances, elapsed emplacement time, mean and peak velocities, and even debris sheet width and thickness [e.g., 157, 162, 315, Hungr, this volume]. Sousa and Voight [295], however, illustrate the uncertainties involved in the use of simulation models in real predictions of future landslide behaviour for hazard assessment.

29 13.

Nature of Secondary Processes Resulting from Massive Rock Slope Failure – Instantaneous and Delayed

13.1. LANDSLIDE GENERATED WAVES (LANDSLIDE TSUNAMIS) AND DISPLACED WATER EFFECTS Extremely rapid bedrock landslides that enter the sea, rivers, natural lakes, or artificial reservoirs generate displacement waves (landslide tsunamis) that may have catastrophic effects beyond the limits of the debris of the initiating landslide. Landslide-generated waves in the sea; One of the greatest landslide disasters of the last millennium occurred on May 21st , 1792, and resulted from the flank collapse of Mayuyama, a dacitic dome in the Unzen volcanic complex, on Kyushu Island, Japan [236; 240, 287, 289]. The massive landslide (estimated volume 340 M m3) entered the Ariake Sea generating a tsunami which swept across and around the narrow enclosed sea [10]. The tsunami reached a height of 22.5 m at Musumi in Kumamoto Prefecture [316]. Other major tsunamigenic landslides from volcanoes in the Pacific occurred [196, 289] at Komagatatake, Japan (in the year 1640 – over 700 deaths in coastal villages), Oshima-Oshima, Japan (in the year 1741- 1475 deaths), Augustine, Alaska (in 1883), Ritter Island, New Guinea (in 1888 [177, 334], Harimkotan, Kurile Islands (in 1933) and Iliwerung, Java (in 1979 – 500 deaths). Destructive displacement waves are also generated by landslides entering the waters of narrow fjords or confined bays [291]. Several examples have been documented in Norway [179; Blikra et al., this volume] Perhaps the most incredible landslide-generated wave is that which occurred in Lituya Bay in 1958 as a result of an earthquake triggered rock avalanche [114, 206, 220]. The rock avalanche (estimated volume 30 M m3) impacted on the waters at the head of the bay and generated a wave run-up of 524 m, the highest recorded in history. The 1958 wave destroyed forest over an area of 10 km2 and was the fourth or fifth to have occurred in Lituya Bay in the last two centuries [220]. More recently, in November 2000, a rockslide into the sea on the west coast of Greenland generated a tsunami that swept along the shores of the narrow Vaigat Strait in the vicinity of Paatuut. No lives were lost but 10 boats were destroyed and the abandoned coal-mining village of Qullissat was severely damaged [244]. Landslide-generated waves and displaced water effects in lakes: Water displaced by landslides entering lakes may have two effects, i.e., along the shoreline of the lake or downstream of the lake outlet. Huge waves may be generated by landslides entering confined lakes. On Vancouver Island, for example, a rock avalanche debris generated a wave in Landslide Lake that ran up a vertical distance of 51 m on the opposite shore [92] snapping mature cedar trees like matchsticks. At Mt St Helens in 1980, a lobe of the rockslide-avalanche resulting from the sector collapse of the volcano, entered Spirit Lake and generated a displacement wave that ran up 260 m above original lake level [330]. Landslide-generated waves may cause considerable destruction along the lakeshore, particularly at locations directly opposite the landslide source. A number of historical examples in Norway are described by Jorstad [179] including devastating events at Loen Lake in 1905 and 1936. In January 1905, a rockfall (est. vol. 50,000 m3)

30 occurred on the west side of the lake; it incorporated till and talus from the valley side at the base of the rock slope forming a rapidly moving mass of about 300,000 m3 which entered the lake below. The resulting waves caused widespread destruction along the lakeshore resulting in 61 deaths; the maximum vertical wave height was estimated to be 40.5 m. In 1936 another rockslide occurred on the west side of Lake Loen, just to the north of the 1905 scar. On this occasion, 1 M m3 of debris entered the lake causing a huge wave which killed a further 73 people; the wave travelled at approximately 25 – 30 m/s (90 – 108 km/hr) along the lake shore [179]. The highest wave runups were directly opposite the landslide source area where a maximum of 74.2 m was recorded. In March 1971, a major disaster occurred at Chungar in the Peruvian Andes. A wave, generated by a small rock avalanche (est. vol. 100,000 m3), struck a mining camp located on the shore of a small lake and killed an estimated 400 people [250]. The wave washed up the opposite shore to a vertical height of 30 m and reduced bunkhouses constructed with concrete blocks to rubble [250]. A large landslide entering a small lake may also displace a large volume of lake water which subsequently travels downstream as a high velocity flood wave. In doing so material from the valley floor may be mobilised and entrained resulting in sediment concentrations approaching that of a debris flow [92, 121, 336]. A more complex case in which displaced lake waters augment the moving mass of a rapidly moving landslide occurred in Iceland [189] in 1965. A rock avalanche, consisting of rock and glacier ice (1 M m3), fell onto a glacier. Part of the debris together with snow from the surface of the glacier entered a lake. Waters of the lake were displaced downstream in what could be termed a rockfall wave-outburst flood. The flood wave reached more than 25 km downstream. Similar events have been documented in Alaska by Wiles and Calkin, [347] and in New Zealand by McSaveney [217]. In the case of the 1987 Valtellina rock avalanche, the only fatalities from the landslide occurred when the rock avalanche displaced waters in a debris-dammed lake generating a catastrophic water-mud wave which overwhelmed part of the village of Aquilone claiming 27 lives [124]. Wavetrains generated by rockslope failure along the shorelines of landslidedammed lakes may result in the overtopping of the landslide dam and its subsequent breaching [e.g., 337]. Further, wavetrains generated by rockfall or glacier avalanching are the major cause of the breaching of moraine dammed lakes [93] and other natural dams. Waves overtop the dam and initiate irreversible erosion leading to the catastrophic release of the water from the lake. These result in the generation of waterfloods, debris floods or debris flows downstream which themselves may be catastrophic as in the case of the 1941 Huaraz disaster in the Cordillera Blanca, Peruvian Andes [137, 205]. Landslide-generated waves and displaced water effects in artificial reservoirs; Artificial reservoirs form conditions which are especially vulnerable to catastrophic landslide-generated downstream water displacement because of the high volume of water in the reservoir and the substantial head difference at the dam. The 1963 Vaiont disaster in the Italian Alps is by far the worst recorded disaster caused by displacement waves in reservoirs. On October 9, 1963 a rockslide with a volume of approximately 270 M m3 [138, Ghirotti, this volume] slid into the reservoir ponded behind the Vaiont Dam [185; 277], which at the time was the world’s second highest dam (261.6 m high; crest elevation 725.5 m.a.s.l). The rockslide occurred very rapidly (20-30 m/s; [138])

31 and displaced a massive amount of water in the reservoir impounded behind the dam. The reservoir was two thirds full at the time and contained 115 M m3 of water. A displacement wave ran up the opposite shore to el. 930 m., 230 m above the reservoir level and displaced about 25 M m3 (ca. 22 % of the reservoir water) over the top of the Vaiont Dam [277]. As the wave overtopped the dam the maximum water depth was estimated to be 100 m. Seismograph data indicates that water was displaced from the reservoir over a period of 11 minutes with an average discharge of approximately 105 m3/s (calculated from Selli and Trevisan, [277]). As is now well known the concrete arch dam survived the overtopping and was left intact, but the displaced water ran down the Vaiont gorge and into the Piave river valley where it ran upstream and downstream, overwhelming several towns and villages including Longarone, Pirago, Villanova, Rivalta, and Fae. Approximately, 2000 inhabitants of the Piave valley lost their lives. The main loss of life was in Longarone (el. 473 m.a.s.l) which was overwhelmed 6 minutes after the frontal wave of the displaced water overtopped the Vaiont dam. Landslide-generated waves may also damage facilities along the reservoir shoreline and damage the dam, possibly leading to a breach if waves overtop the structure. Many hydro-electric dams have been retrofitted with hardened crests to resist landslidegenerated wave induced overtopping. 13.2.


Landslide dams form as a result of the blocking of drainage by landslide debris [4, 30, 57, 88, 136, 156, 193, 221, 340, 341, 350; Schuster, this volume]. Secondary effects resulting from landslide dams occur upstream as a result of water ponding up to the height of the dam (upstream flooding), or downstream due to the sudden release of the impounded water due to the catastrophic failure of the natural dam resulting in outburst floods [e.g., 280, 287, 350, 357]. Stable landslide dams may impound lakes that may become permanent features of the Holocene landscape [273] or form lakes that eventually disappear because of sediment infilling (147, 148, Schuster, this volume). Landslide dams may fail by overtopping, piping, or slope failure of the downstream face [e.g., 57, 88]. Outburst floods [56, 57, 143, 210] from landslide dams have been responsible for considerable death and destruction [Evans, this volume] in historical times. They are also of considerable geomorphic significance since they frequently reset fluvial systems downstream from the breach location and are typically orders of magnitude greater than the maximum probable “normal” hydrologic flood [e.g., 197]. The formation and failure of landslide dams during the travel of landslide debris may also augment the travel distance of landslide debris as in the case of the 1987 Parraguirre event, Chile [134]. Outburst floods from landslide dams have caused some notable disasters in China [203, 204] including what is probably the worst single-event landslide disaster of the millennium, the outburst flood on the Dadu River, Sichuan in 1786. [203]. The landslide, triggered by the Kangding-Louding earthquake, occurred at Momianshan, near Luding and was. It dammed the Dadu River for 10 days; when the dam was overtopped the dam failed and a great flood swept down the Dadu River as far as Yibin City, 1,400 km downstream. According to Li and Wang [203] the flood took as many as

32 100,000 lives along its path. One hundred and fourty seven years later, the Deixi landslide, triggered by the highly destructive Deixi earthquake, formed a 250 m high dam across the Minjiang River also in Sichuan Province, in 1933. The dam impounded the river for 45 days and failed rapidly after overtopping. Approximately 400 M m3 of water was suddenly released in the flood which traveled a distance of 253 km downstream with an average velocity of 5.5 – 7.0 m/s; [204]. At least 2,423 people lost their lives. Landslide dams have also been documented from many parts of the Himalayas [36, 108, 111, 144, 145, 146, 147, 148, 210, 332]. Three dramatic examples of landslide dam formation and failure are documented from the nineteenth century, viz. landslide dams on the Indus (1840 - 1841), which created the First Great Indus Flood of Mason [210], 1929, Hunza (1858), which created the Second Great Indus Flood of Mason [210], 1929, and the Gohna landslide (1893-94; [122, 150]). In the Indus case a large bedrock landslide, apparently triggered by an earthquake, blocked the mighty Indus, downstream from Gilgit, in what is now Pakistan, in January, 1841 [e.g., 15, 38, 53, 210, 241]. The landslide occurred on the left bank of the Indus at the western foot of the Nanga Parbat massif, in an area which is subject to an uplift of 7mm/yr and active neotectonics [37] and formed a lake which extended 60 km upstream, almost to Gilgit [el. ca. 1490 m.a.s.l.]. River elevation at the landslide location is 1178 m, and the full pool elevation of the impounded reservoir was about 1402 m.a.s.l [83]; the landslide lake filled in approximately 6 months and may have contained as much as 10 B m3 of water, one of the largest bodies of water impounded by a natural dam documented on earth in the Holocene. In May 1841, warnings written on birch-bark were sent downstream urging inhabitants to flee from the rivers edge as it became obvious that the landslide dam would burst when it was overtopped [15, 139]. Early in June 1841, the blockage was breached. An enormous flood wave was released into the Indus as the contents of the lake apparently drained “off in a day” [79] and swept downstream (The Great Indus Flood of Mason, [210]). As summarised by Mason [210], the flood caused considerable destruction and great loss of life in the lower Indus valley, including soldiers of a Sikh army encamped at Attock who were overwhelmed some 320 km downstream from the landslide dam. Flooding also occurred in some Indus tributaries because of the hydraulic obstruction created by the flood waters. The rise in the Indus at Attock (>25 m) was the highest recorded and a conservative estimate of peak discharge is 2 M m3/s [143]. The 1858 landslide dam on the Hunza River existed for 6 months before draining catastrophically. During this time at least 10-11 m of silt was deposited in the lake [242]. Typical deposits of outburst floods are terraces as described in the Indus valley and tributaries by Hewitt [144, 147]. A sequence of terraces defended by landslide barriers at different levels above the actual valley indicate therefore intermittent breaching of the barrier. Of high importance to dam stability is where the first outlet gorge forms. In the Karakoram 21 landslide dams of 96 cross-valley deposits have gorges within the bedrock and therefore at different position than the pre-landslide river course [144]. At such “epigenetic” gorges fluvial incision is controlled by the bedrock properties significantly different from the strongly shattered landslide deposits. Landslide dams with such type of outlets are less likely to fail catastrophically.

33 14.

Magnitude and Frequency of Massive Rock Slope Failure and Its Role In Landscape Evolution

14.1. BACKGROUND TO LANDSLIDE MAGNITUDE AND FREQUENCY RELATIONS Analysis and use of the magnitude and frequency (m/f) relations of landslides, is a comparatively recent development in massive rock slope failure hazard assessment (Evans, this volume). The approach is based to some extent on the well-known Gutenberg-Richter power-law relation [126] for earthquakes (1) log N(m) = aM-b


where N(m) = cumulative frequency equal to or greater than M, M is earthquake magnitude, a and b are constants. Because of its scale-invariance and universal characteristic (1) has formed the basis for seismic hazard assessment methodologies world-wide based on the analysis of earthquake occurrences recorded in historical earthquake catalogues supplemented by geological evidence for prehistoric earthquakes. In its application to landslides, magnitude (m) has been taken to be some measure of landslide size based on area (A) or volume (V). Frequency (f) may be expressed in a simple cumulative (or rank-ordering), in a non-cumulative manner (see discussion in Guzzetti et al. [128]), or in terms of frequency density, i.e., the number of landslides in any given magnitude bin divided by the bin size [129]. Frequency may also be expressed directly as an annual frequency (cumulative number per year) if, as discussed below, the dataset is time constrained. The methodology is applied utilizing spatial datasets (landslide inventories) for a region representing landslide occurrence in various types of temporal records; 14.2. THE STRUCTURE OF LANDSLIDE MAGNITUDE/FREQUENCY RELATIONS Early work by Fuji [116] analysed the m/f relationship for 650 rainfall-triggered events and found that the frequency of landslides is inversely related to their volume and can be defined by a power law similar to the Gutenburg-Richter relation in (1). Whitehouse and Griffiths [343] found a similar relationship for rock avalanches in New Zealand. Later work by Ohmori and Hirano [237] and Sugai et al. (304) further showed that landslide m/f relations are power law functions of magnitude. However, it was the work of Hovius et al. [151] and Pelletier et al. [245] that initiated the current interest in landslide magnitude and frequency by deriving what can be described as the characteristic form of the magnitude/frequency relation. Hovius et al. [151] analysed multiple sets of air photos between 1948 and 1986 in the western Southern Alps of New Zealand. They found that m/f relations for the area of landslide scars (As) are scale invariant and had a robust power law m/f distribution over

34 approximately two orders of area magnitude with a flattening of the curve at lower magnitudes. Pelletier et al. [245] analysed three data sets in which magnitude was expressed in terms of area (A); a data set of landslides in Japan, in which landslide area included the run-out zone, a data set of landslides in Bolivia, and a record of 11,111 landslides triggered by the 1994 Northridge earthquake over an area of 10,000 km2. All three m/f plots showed a linear segment characterized by a power law and a flattening of the curve at small landslide magnitudes. Thus the m/f log-log plots of Hovius et al. [151] and Pelletier [245] show two characteristics; first a liner segment at small to large magnitudes and second, a flattening of the curve at small magnitudes which has been termed “rollover”. The linear portion of the m/f plot obeys a power law of general form in (2) N (A) ~ A-b


where A is landslide area, N(A) the number of events greater than V and b is a constant. Subsequent studies on different types of landslides in different geological environments have found similar results and m/f plots of similar shape. Hungr et al. [161] analysed maintenance records for the volume and frequency of rockfall along transportation routes in British Columbia and found m/f relations characterised by a power law. Other studies of the m/f of rockfall and rock slope failure, using volume as magnitude, were carried out by Chau et al. [47] in Hong Kong, Guzzetti et al. [129] in Yosemite, and Singh and Vick [290] in British Columbia. All these studies found broadly similar m/f relations. In a comprehensive study, Dussauge-Pressier et al. [82] and Dussauge et al. [81] found that datasets of rockfalls from Yosemite and the Grenoble area as well as rockslides and rock avalanches from a global data set followed a m/f relation characterised by a power law in (2). Thus the structure of landslide m/f relations is characterized by scale invariance (i.e., the power-law segment of the m/f plot is linear over several orders of landslide magnitude); similar shaped m/f plots are obtained for various measures of landslide magnitude consisting of area and volume, different landslide types, in different geological environments both in space and time, and with different triggers. The characteristic relation is obtained from the analysis of various types of temporal records. The characteristic m/f relation also applies to landslides from natural and artificial slopes in natural and human-modified terrain. The power law structure of the m/f relation makes it possible to predict the frequency of larger landslides (for which a record may not exist) based on the slope of the linear part of the m/f plot derived from the occurrence of smaller landslides, assuming that the record of smaller landslides is complete [Evans, this volume]. These apparently universal characteristics of landslide m/f relations result in their extreme usefulness for landslide hazard assessment; they form a type of hazard model which may be used in the quantification of landslide hazard which can serve as input into a quantitative risk calculation. Once a magnitude and frequency relation has been established for a region or a site, it may be used to estimate the probability of occurrence of a landslide of a certain magnitude providing the length of the record is known [e.g., 81, 161]. This gives a quantitative estimate of hazard which when combined with vulnerability data can give a quantitative estimate of landslide risk. The

35 first application of landslide m/f relations to formal landslide hazard and risk assessment was by Hungr et al. [161]. In this study, m/f relations were derived for rockfalls from natural and artificial slopes in transportation corridors in southwestern British Columbia utilizing a set of very complete Type 3 records. These data gave the probability of rockfalls of a given size occurring in the narrow linear road and rail corridors. Combined with traffic density data for a segment of a corridor, the risk of a fatal accident due to rockfall impact was calculated [161]. It is also possible to use the m/f relations derived by Hungr et al. [161] to calculate the probability of such scenarios as total blockage of a given corridor by a large landslide involving massive rock slope failure.

14.3. ASSUMPTIONS AND LIMITATIONS IN THE MAGNITUDE/FREQUENCY APPROACH TO LANDSLIDE HAZARD ASSESSMENT Despite the accumulating evidence of a characteristic, possibly universal landslide m/f signature there are some assumptions and limitations that should be kept in mind in the application of the methodology to landslide hazard assessment. A major difficulty is the assumption of invariance in the occurrence of landslides in time, i.e., that the rate of occurrence implicit in the temporal records will persist at the same rates into the future, at least in terms of engineering time scales. Landslide occurrence reflects to some extent the frequency of landslide triggers (e.g., rainstorms and earthquakes). In this regard climate change may affect the frequency and intensity of rainfall triggers such that regional landslide events may be more frequent in the future. This could increase the frequency and thus the hazard of rainfall-triggered landslide events. At geological time scales, Cruden and Hu [67] have argued that the probability of occurrence of large rockslides in the Rocky Mountains of Canada is decaying with time as the number of rockslide sites conditioned by Pleistocene glaciation becomes exhausted by Holocene rockslide occurrence. This is in contrast to the implicit assumption of steady state landslide occurrence in landslide m/f relations. Further limitations are associated with the quality of the landslide dataset. The accuracy of landslide magnitude measurement, whether it is expressed in terms of area or volume, is an important consideration. A related problem is that of record completeness. Erosion censoring of large magnitude landslides over long time periods removes them from long-time-period records [343]. In addition, the scale of spatial inventories determines the resolution of the record and the frequency count of smaller landslides. Despite these assumptions and limitations, the analysis of landslide m/f relations, increasingly provides a key element in landslide hazard assessment and subsequent risk evaluation at regional and site scales.

14.4. MASSIVE ROCK SLOPE FAILURE AND QUATERNARY CLIMATE CHANGE Ages of landslide dammed lakes and ages of landslide deposits have been compared to climatic periods in the Quaternary. For example, massive rock slope failures have been related in strongly glaciated regions with glacial retreat [e.g., 2, 67]. In Western Canada

36 formation of recent rock avalanches has been shown to be a consequence of declining lateral support of valley walls by thinning of glaciers after the Little Ice Age [93, 95]. In a similar way, ages of landslide deposits in postglacial colluvium in western Norway coincides with the main deglaciation phase ~ 11 – 10 ka (23). Recent studies of rock avalanches in fjords of Norway show that older events (>10 ka) took place in the outer fjord areas while younger rock avalanches (3-6 ka) occurred in the inner parts of the fjords indicating the close link of slope collapse with deglaciation after the last Ice Age (Blikra, this volume). In semi-arid regions, run off along deeply incised valleys more likely influences slope stability. For example, along Rio Grande, New Mexico, large slumps damming the White Rock Canyon only occurred in the Late Pleistocene [77, 262]. No such mass movements are recorded from the Holocene along the Rio Grande, suggesting that slope destabilization was the consequence of downcutting of the river related to both glacial melt and enhanced pluvial activity in the late Pleistocene. In NW-Argentina rock avalanche ages depend strongly on their geomorphic setting. While rock avalanches along mountain fronts [103] bordered by wide piedmont areas are relatively old and have recurrence intervals of several tens ka; those in narrow valleys are young, have recurrence intervals of a few ka and apparently cluster during periods characterized by more humid climate conditions in subtropical South America [141, 142, 313, 314; Hermanns et al, this volume] suggesting at these sites that stream power directly influence slope stability. These examples highlight the impact of Quaternary climate change on rock slope failures. In the same way future climatic changes resulting in further deglaciation and/ or more humid conditions with a higher frequency of extreme meteorological events, are likely to influence the occurrence of massive rock slope failures in high mountain regions. 14.5. ROLE OF MASSIVE ROCK SLOPE FAILURE IN LANDSCAPE EVOLUTION It has been recognized that, in tectonically active areas, bedrock landsliding is an important contributor to denudation and a major mechanism controlling mountain slope formation, valley incision and sediment flux [e.g., 34, 108, 109, 110, 151, Hovius and Stark, this volume]. Keefer [182], showed that erosion rates from earthquake-induced landslides vary significantly from region to region and that in some seismically active mountains such as western New Guinea, seismically-triggered landslides are the predominant agent of slope erosion. Hovius et al. [151] quantified denudation rates related to landsliding on the western slopes of the Southern Alps of New Zealand by analyzing aerial photographs spanning 60 years. Calculated erosion rates average 9 mm/yr and range from 5 to 12 mm/yr within individual catchments. Although these erosion rates are extremely high they coincide with areas of extreme rainfall and also high rock uplift, measured by fission track ages, thus a steady-state landscape can be envisioned in this mountain belt. In contrast, the Finisterre Mountains in Papua New Guinea are a young mountain belt in a pre-steady-state. Here, watersheds expand by large-scale landsliding controlled by ground-water seepage after their initiation as single gorges [152].

37 Landslide scars by themselves control further fluvial incision. These landslide controlled drainage patterns later only rarely are modified by further major landslides. In the northwestern Himalaya, another mountain belt with high tectonic activity landslides from massive rock slope failure are widespread. Burbank et al. [34] could show that the Indus river surrounding mountains have average steep hillslope angles with a mean of 32 +/- 2°, which are independent of local erosion rates, suggesting control by a common threshold process where landsliding adjusts hillslopes efficiently between bedrock uplift and river incision. Although these examples highlight the influence of landslides on hillslope formation, valley incision and sediment flux on regional scales, the volumetric impact of these processes by landslides are still poorly understood in many geologic and climatic settings.



Landslides from massive rock slope failure (MRSF) are a major geological hazard in many parts of the world and have been responsible for some of the world’s major natural disasters. New mapping, and the reinterpretation of surficial materials previously mapped as glacial deposits, has led to a new understanding of the extent and frequency of massive rock slope failure in such regions as the Himalayas and the mountains of Central Asia in the countries of the former Soviet Union. Hazard assessment is made difficult by a variety of complex initial failure processes and unpredictable post-failure behaviour, which includes transformation of movement mechanism, substantial changes in volume by deposition and/or entrainment, and changes in the characteristics of the moving mass. Initial failure mechanisms are strongly influenced by geology and topography, and, because of this, the development of geological models is essential for the analysis of these mechanisms. Massive rock slope failure includes extremely rapid movements such as rockslides, rock avalanches, catastrophic spreads and rockfalls. However, catastrophic debris flows can also be triggered by massive rock slope failure and may present a significant hazard in themselves. Volcanoes are particularly prone to massive rock slope failure and can experience very large-scale sector collapse, such as the 1980 flank collapse of Mt St Helens, or much smaller partial collapse, such as the 1984 earthquake-triggered failure at Mt. Ontake. Both these types of failures may be transformed into volcanic mudflows (lahars) which can travel over 100 km from their source. Many cases of historical MRSF were preceded by precursory signs; movement monitoring data can be used to estimate time to failure. The study of MRSF deposits gives insight into fragmentation and emplacement processes. These studies are also important in the stability analysis of landslide dams. Slow mountain slope deformation presents problems in interpretation of both origin and movement mechanism. Geomorphic features related to slope sagging have been interpreted as tectonic faults. The problem of interpretation is made more difficult when mountain slope movement is related to the presence of tectonic faults. The identification of thresholds for the catastrophic failure of a slow moving rock slope is a key remaining question in

38 landslide hazard assessment. Advances have been made in the analysis and modeling of initial failure and post-failure behaviour. However, these studies have been retrodictive in nature and their true predictive potential for hazard assessment remains uncertain yet promising. Secondary processes associated with MRSF are an important component of hazard. These processes, which can be instantaneous or delayed, include the formation and failure of landslide dams and the generation of landslide tsunamis. Both these processes extend potential damage beyond the limits of landslide debris. The occurrence of MRSF in space and time forms orderly magnitude and frequency relations which can be characterized by robust power law relationships. Uncertainty about the temporal controls on MRSF occurrence condition the application of these relationships to hazard assessment. MRSF is increasingly recognized as being an important process in landscape evolution which provides an essential context for enhanced hazard assessment. References 1.

2. 3. 4. 5. 6.


8. 9. 10. 11. 12. 13. 14. 15. 16.

Abbott, P.L., Kerr, D.R., Borron, S.E., Washburn, J.L., and Rightmer, D.A. (2002) Neogene sturzstrom deposits, Split Mountain area, Anza-Borrego Desert State Park, California, in S.G. Evans and J.V. DeGraff (eds.), Catastrophic landslides: effects, occurrence, and mechanisms, Geol. Soc. Am. Rev. Eng. Geol., v. XV, pp. 379-400. Abele, G. (1974) Bergstürze in den Alpen, ihre Verbeitung, Morphologie und Folgeerscheinungen, Wissensch. Alpenverein 25, 1-230. Abele, G. (1997) Rockslide movement supported by the mobilization of groundwater-saturated valley floor sediments, Zeits. für Geomorph. 41, 1-20. Adams, J. (1981) Earthquake-dammed lakes in New Zealand, Geology 9, 215-219. Adushkin, V.V. (2000) Explosive initiation of creative processes in Nature, Combustion, Explosion, and Shock Waves, 36, 695-703. Adushkin, V.V. and Spungin, V.G. (1998) The influence (of) granular structure of rockfalls on their spreading along mountain slopes, in H-P. Rossmanith (ed.) Proc. 3rd Int. Conf. on Mechanics of Jointed and Faulted Rock, Vienna, Austria p. 541-546., A.A. Balkema, Rotterdam. Adushkin, V.V., Zykov, Y.N., and Fedotov, S.A. (1995) Mechanism of volcanic slope failure; assessment of potential collapse and debris avalanches at Klyuchevskoi volcano, Volc. Seism. 16, 667-684. Agliardi, F. and Crosta, G.B. (2003) High resolution three-dimensional numerical modeling of rockfalls, Int. Jo. of Rock Mech. and Mining Sci. 40, 455-471. Agliardi, F., Crosta, G., and Zanchi, A. (2001) Structural constraints on deep-seated slope deformation kinematics, Eng. Geol. 59, 83-102. Aida, I. (1975) Numerical experiments of the tsunami associated with the collapse of Mt. Mayuyama in 1792, Jo. Seismol. Soc. Jpn 28, 449-460 (in Japanese). Almagià, R. (1910) La grande frana di Roccamontepiano (prov. Di Chieti) (24 giugno 1765), Riv. Abruzz. di Sci. Lett. ed Arti, 25, 337-349. Azzoni, A. and de Freitas, M.H. (1995) Experimentally gained parameters decisive for rock fall analysis, Rock Mech Rock Eng, 28, 111-124 Azzoni, A., La Barbera, G., Zaninetti, A. (1995) Analysis and prediction of rock falls using a mathematical model, Int. Jo. Rock Mech. Min. Sci., 32, 709-724. Barla, G., Dutto, F., and Mortara, G. (2000) Brenva Glacier rock avalanche of 18 January 1997 on the Mount Blanc Range, northwest Italy, Landslide News 13, 2-5. Becher, J. (1859) Letter addressed to R.H. Davies, Esquire, Secretary to the Government of the Punjab and its ependencies, Jo. Asiatic Soc. of Bengal, 28, 219-228. Belousov, A., Belousova, and Voight, B. (1999) Multiple edifice failure, debris avalanches and asssociated eruptions in the Holocene history of Shiveluch volcano, Kamchatka, Russia, Bull. Volcanol. 61, 324-342.

39 17. 18. 19.

20. 21. 22.

23. 24. 25. 26. 27. 28. 29. 30. 31. 32. 33. 34.

35. 36. 37. 38. 39. 40. 41.


Benko, B. and Stead, D. (1998) The Frank Slide: a reexamination of the failure mechanism, Can. Geotech. Jo. 35, 299-311. Bhasin, R. and Kaynia, A.M. In Press. Static and dynamic simulation of a 700-m high rock slope in western Norway. Eng. Geol. Birch, G.P., and Warren, C.D. (1996). The cliffs behind the Channel Tunnel workings, in C.S. Harris, M.B. Hart, P.M. Varley, and C.D. Warren, (eds.), Engineering Geology of the Channel Tunnel, Thomas Telford, London, pp. 76-87. Bishop, K. (1997) Miocene rock-avalanche deposits, Halloran/Silurian Hills area, southeastern California, Env. Eng. Geosci. 3, 501-512. Bjerrum, L. and Jorstad, F.A. (1968). Stability of rock slopes in Norway. Norwegian Geotechnical Institute Publication No. 79, pp. 1-11. Blanc, A., Durville, J-L., Follacci, J-P., Gaudin, B., and Pincent, B. (1987) Methodes de surveillance d’un glissement de terrain de tres grande ampleur : La Clapiere, Alpes Maritimes, France, Bull. Int. Assoc. of Eng. Geol., 35, 37-46. Blikra, L.H., and Nemec, W. (1998) Postglacial colluvium in western Norway: depositional processes, facies and palaeoclimatic record, Sedimentology 45, 909-959. Bogachkin, B.M. and Rogozhin, Y.A. (1994) Rock avalanches on the southern slope of the Greater Caucasus Range produced by the Racha earthquake, Trans. Russ. Acad. Sci., 326, 29-33. Bovis, M.J. and Evans, S.G. (1995) Rock slope movements along the Mount Currie “fault scarp”, southern Coast Mountains, British Columbia, Can. Jo. Earth Sci. 32, 2015 – 2020. Bovis, M.J. and Evans, S.G. (1996) Extensive deformations of rock slopes in southern Coast Mountains, southwest British Columbia, Canada, Eng. Geol. 44, 163-182. Bozzolo, D. and Pamini, R. (1986) Simulation of rock falls down a valley side, Acta Mech., 63, 113130 Bozzolo, D., Pamini, R., and Hutter, K. (1988) Rockfall analysis-a mathematical model and its test with field data, Proc. 5th Int. Symp. on Landslides, Lausanne, v. 1, pp. 555-560. Brabb, E.E. and Harrod, B.L. (Editors) (1989) Landslides: extent and economic significance. A.A. Balkema, Rotterdam. Bromhead, E.N., Coppola, L., and Rendell, H.M. (1996) Valley-blocking landslide dams in the eastern Italian Alps, Proc. 7th Int. Symp. on Landslides, Trondheim, v. 2, pp. 655-660. Bruce, I.G. and Cruden, D.M. (1980) Simple rockslides at Jonas Ridge, Alberta, Canada, Proc. 3rd. Intl. Symposium on Landslides, New Delhi, v. 1, p. 185-190. Brückl, E., Brückl, J., and Heuberger, H. (2001) Present structure and prefailure topography of the giant rockslide of Kofels, Zeits. fur Gletscherkunde und Glazialgeol., 37, 49-79. Budetta, P. and Santo, A. (1994). Morphostructural evolution and related kinematics of rockfalls in Campania (southern Italy): a case study, Eng. Geol. 36, 197-210. Burbank, D.W., Leland, J., Fielding, E., Anderson, R.S., Brozovic, N., Reid, M.R., and Duncan, C. (1996) Bedrock incision, rock uplift and threshold hillslopes in the northwestern Himalayas, Nature 379, 505-510. Burchfiel, B.C. (1966) The Tin Mountain landslide, southeastern California, and the origin of megabreccia, GSA Bull., 77, 95-100 Burgisser, H.M., Gansser, A., and Pika, J. (1982) Late Glacial lake sediments of the Indus valley area, northwestern Himalayas, Eclog. Geol. Helv. 75, 51-63 Butler, R.W.H., and Prior, D.J. (1988) Tectonic controls on the uplift of the Nanga Parbat Massif, Pakistan Himalayas, Nature 333, 247-256. Butler, R.W.H., Owen, L., and Prior, D.J. (1988) Flashfloods, earthquakes and uplift in the Pakistan Himalayas, Geology Today 4, 197-201. Campbell, C.S. (1989) Self-lubrication for long run-out landslides. Jo. Geol., 97, 653-665 Campbell, C.S., Cleary, P.W., and Hopkins, M. (1995) Large-scale landslide simulations, global deformation, velocities and basal friction, Jo. Geophys. Res. 100B, 8267-8283. Capra, L. and Macías, J.L. (2002) The cohesive Naranjo debris-flow deposit (10km3): a dam breakout flow derived from the Pleistocene debris-avalanche deposit of Nevado de Colima volcano (Mexico), Jo. of Volc. And Geotherm. Res., 117, 213-235. Capra, L., Macías, J.L., Scott, K.M., Abrams, M., and Garduño-Monroy, V.H. (2002) Debris avalanches and debris flows transformed from collapses in the Trans-Mexican Volcanic Belt, Mexico – behaviour, and implications for hazard assessment. Jo. of Volc. and Geotherm. Res., 113, 81-110.

40 43.


45. 46. 47. 48. 49. 50. 51. 52. 53.

54. 55. 56. 57. 58. 59.

60. 61. 62.


64. 65. 66. 67.

Carrasco-Nunez, G., Vallance, J.W., and Rose, W.I. (1993) A voluminous avalanche-induced lahar from Citlaltépetl volcano, Mexico: implications for hazard assessment, Jo. Volc. Geotherm. Res. 59, 35-46 Casagli , N., Ermini, L., and Rosati, G. (2002) Determining grain size distribution of the material composing landslide dams in the Northern Appennines: sampling and processing methods, Eng. Geol. 69, 83-97. Cassassa, G. and Marangunic, C. (1993) The 1987 Río Colorado rockslide and debris flow, Central Andes, Chile, Bull. Ass. Eng. Geol. 30, 321-330. Chau, K.T., Wong, R.H.C., and Wu, J.J. (2002) Coefficient of restitution and rotational motions of rockfall impacts, Int. Jo. Rock Mech. Min. Sci., 39, 69-77. Chau, K.T., Wong, R.H.C., & Lee, C.F. (2003). Rockfall hazard analysis for Hong Kong based on rockfall inventory. Rock Mech. and Rock Eng. 36, 383-408. Chen, T-C., Lin, M-L., and Hung, J-J. (2003) Pseudostatic analysis of Tsao-Ling rockslide caused by Chi-Chi earthquake, Eng. Geol. 71, 31-47. Chigira, M. (1992) Long-term gravitational deformation of rocks by mass rock creep, Eng. Geol., 32, 157-184. Chigira, M. (2000) Geological structures of large landslides in Japan, Jo. Nepal Geol. Soc. 22, 497504. Chigira, M. and Kiho, K. (1994) Deep-seated rockslide-avalanches preceded by mass rock creep of sedimentary rocks in the Akaishi Mountains, Central Japan, Eng. Geol. 38, 221-230. Choffat, P. (1929) L’écroulement d’Arvel (Villeneuve) de 1922, Bull. de la Soc. Vaud. Sci. Nat. 57, 5-28. Code, J.A. and Sirnhindi, S. (1986) Engineering implications of impoundment of the Upper Indus river, Pakistan, by an earthquake-induced landslide. In R.L. Schuster (Ed.) Landslide Dams :processes, risk, and mitigation, Geotechnical Special Publication No. 3, American Society of Civil Engineers, New York, p. 97-109. Cornell, J. (1982) The great international disaster book, 3rd. Edition, Charles Scribner, New York, 472 p. Corominas, J. (1996) The angle of reach as a mobility index for small and large landslides, Can. Geotech. Jo. 33, 260-271. Costa, J.E. (1988) Floods from dam failures In V.R. Baker, R.C. Kochel, and P.C. Patton (Editors) Flood Geomorpholgy, J. Wiley and Sons, New York, p. 439-463. Costa, J.E. and Schuster, R.L. (1988) The formation and failure of natural dams, GSA Bull. 100, 1054-1068. Crandell, D.R. (1971) Postglacial lahars from Mount Rainier Volcano, Washington. United States Geol. Surv. Prof. Paper 677, 73 p. Crandell, D.R. (1989) Gigantic debris avalanche of Pleistocene age from ancestral Mount Shasta volcano, California, and debris-avalanche hazard zonation, United States Geol. Surv. Bulletin 1861, 32 p. Crandell, D.R. and Fahnestock, R.K. (1965) Rockfalls and avalanches from Little Tahoma Peak on Mount Rainier, Washington, United States Geol. Surv. Bull 1221-A, 30 p. Crandell, D.R. and Waldron, H.H. (1956) A recent volcanic mudflow of exceptional dimensions from Mt. Rainier, Washington, Am. Jo. Sci. 254, 349-362 Crescenti, U., D’Alessandro, L., and Genevois, R. (1987) La Ripa di Montrepiano (Abruzzo): un primo esame delle caratteristiche geomorfologieche in rapporto alla stabilità, Mem. della Soc. Geol. Ital. 37, 775-787. Crescenti, U., Dramis, F., Prestiniinzi, A., and Sorriso-Valvo, M. (1994) Deep-seated gravitational slope deformations and large-scale landslides in Italy, Dip. di Sci., Univ G. D’Annunzio, Pescara, Italia, 71 p. Crosta, G.B. and Agliardi, F. (2003) Failure forecast for large rock slides by surface displacement measurements, Can. Geotech. Jo., 40, 176-191. Cruden, D.M. (2000) Some forms of mountain peaks in the Canadian Rockies controlled by their rock structure, Quat. Int. 68-71, 59-75. Cruden, D.M. and Antoine, P. (1984) The slide from Mt. Granier, Isere and Savoie, France, on November 24, 1248, Proc. 4th Int. Symp. on Landslides, v.1, p. 475-481. Cruden, D.M. & Hu, X.Q. (1993) Exhaustion and steady state models for predicting landslide hazards in the Canadian Rocky Mountains. Geomorphology 8, 279-285.

41 68. 69. 70. 71. 72.


74. 75.

76. 77. 78. 79. 80.




84. 85. 86.


88. 89. 90. 91.

Cruden, D.M. and Hu, X.Q. (1994) Topples on underdip slopes in the Highwood Pass, Alberta, Canada, Quart. Jo. Eng. Geol. 27, 57-68. Cruden, D.M. and Hu, X.Q. (1996) Hazardous modes of rock slope movement in the Canadian Rockies, Env. Eng. Geosci. 2, 507-516. Cruden. D.M., and Hungr, O. (1986) The debris of the Frank Slide and theories of rockslideavalanche mobility, Can. Jo. Earth Sci. 23, 425-432. Cruden, D.M. and Lu, Z.Y. (1992) The rockslide and debris flow from Mount Cayley BC in June 1984, Can. Geotech. Jo. 29, 614-626. Cruden, D.M. and Varnes, D.J. (1996) Landslide types and processes. In A.K. Turner and R.L. Schuster (Editors) Landslides:Investigation and Mitigation. Transportation Research Board Special Report 247, p. 36-75. National Academy Press, Washington, D.C. Cundall, P.A. (1987) Distinct element models of rock and soil structure, In E.T. Brown (Ed.) Analytical and Computational Methods in Engineering Rock Mechanics, Allen and Unwin, London, p. 129-163. Dade, W.B. and Huppert, H.E. (1998) Long-runout rockfalls, Geology 26, 803-806. D’Alessandro, Genevois, R., Berti, M., Urbani, A., and Tecca, P.R. (2002) Geomorphology, stability analyses and the stabilization works on the Montepiano travertinous cliff (Central Italy), In R.J. Allison (ed.) Applied Geomorphology: Theory and Practice, John Wiley and Sons, pp. 21-38. Del Prete, M. and Hutchinson, J.N. (1988) La frana di Senise del 267-1986 nel quadro morfologico del verante meridionale della collina Timpone, Riv. Italiana di Geotec., 22, 7-33. Dethier, D.P., and Reneau, S.L. (1996) Lacustrine chronology links late Pleistocene climate change and mass movements in northern New Mexico, Geology 24, 539-542. Dikau, R., Schrott, L., Brunsden, D., and Ibsen, M.-L. (1996) Landslide Recognition. J. Wiley, New York, 251 pp. Drew, F. (1875) The Jummo and KashmirTerritories: a geographical account. Stanford, London, 568 p. Duperret, A., Genter, A., Mortimore, R.N., Delacourt, B., and De Pomerai, R.D. (2002) Coastal rock cliff erosion by collapse at Puys, France : the role of impervious marl seams within Chalk of NW Europe. Jo. Coastal Res. 18, 52-61. Dussauge, C., Grasso, J-R., & Helmstetter, A. (2003) Statistical analysis of rockfall volume distributions: implications for rockfall dynamics. Journal of Geophysical Research 108, B6, ETG 21 to ETG 2-11. Dussauge-Peisser, C., Helmstetter, A., Grasso, J-R., Hantz, D., Desvarreux, P., Jeannin, M., & Giraud, A. (2002) Probabilistic approach to rock fall hazard assessment: potential of historical data analysis. Natural Hazards and Earth System Sciences 2: 15-26. Eberhart-Phillips, D., Haeussler, P.J., Freymueller, J.T., Frankel, A.D., Rubin, C.M., Craw, P., Ratchovski, N.A., Anderson, G., Carver, G.A., Crone, A.J., Dawson, T.E., Fletcher, H. Hansen, R., Harp, E.L., Harris, R.A., Hill, D.P., Hreinsdottir, S. , Jibson, R.W., Jones, L.M., Kayen, R., Keefer, D.K., Larsen, C.F., Moran, S.C., Personius, S.F., Plafker, G., Sherrod, B., Sieh, K., Sitar, N., and Wallace, W.K. (2003) The 2002 Denali Fault Earthquake, Alaska: a large magnitude, slip-partitioned event, Science 300, 1113-1118. Eisbacher, G.H., (1979) Cliff collapse and rock avalanches (sturzstroms) in the Mackenzie Mountains, northwestern Canada, Can. Geot. Jo. 16, 309-334. Eisbacher , G.H.and Clague, J.J. (1984) Destructive mass movements in high mountains: hazard and management, Geological Survey of Canada Paper 84-16, 230 p. Endo, K., Sumita, M., Machida, M., and Furuichi, M. (1989) The 1984 collapse and debris avalanche deposits of Ontake Volcano, central Japan. In Latter, J.H. (Ed) Volcanic Hazards, Springer-Verlag, Berlin, p. 210-229. Enegren, E.G. and Imrie, A.S. (1996) Ongoing requirements for monitoring and maintaining a large remediated rockslide, Proceedings, 7th. International Symposium on Landslides, Trondheim, Norway, v. 3, p. 1677-1682. Ermini, L. and Casagli, N. (2003) Prediction of the behaviour of landslide dams using a geomorphological dimensionless index, Earth Surf. Processes Landforms 28, 31-47. Evans, R.S. (1981) An analysis of secondary toppling rock failures-the stress redistribution method. Quart. Jo. Eng. Geol. 14: 77-86. Evans, S.G. (1984) Landslides in Tertiary basaltic successions. Proceedings, IV International Symposium on Landslides, Toronto, v.1, 503-510. Evans, S.G. (1989a) Rock avalanche run-up record, Nature 340, 271.

42 92. 93. 94. 95.

96. 97. 98.

99. 100.



103. 104.

105. 106. 107. 108. 109. 110.

111. 112. 113. 114. 115. 116.

Evans (1989b) The 1946 Mount Colonel Foster rock avalanche and associated displacement wave, Vancouver Island, British Columbia. Can. Geotech. Jo., 26, 447-452. Evans, S.G., and Clague, J.J. (1994) Recent climatic change and catastrophic geomorphic processes in mountain environments, Geomorphology 10, 107-128. Evans, S.G., and Clague, J.J. (1998) Rock avalanche from Mount Munday, Waddington Range, British Columbia, Canada, Landslide News 11, 23-25. Evans, S.G. and Clague, J.J. 1999. Rock avalanches on glaciers in the Coast and St. Elias Mountains, British Columbia. In Slope stability and landslides, Proceedings, 13th Annual Vancouver Geotechnical Society Symposium, p. 115-123. Evans, S.G. and DeGrafff, J.V. (Editors) (2002) Catastrophic landslides: effects, occurrences and mechanisms. Geol. Soc. Am. Rev. in Eng. Geol. v. 15, 412p. Evans, S.G. and Hungr, O. (1993) The analysis of rockfall hazard at the base of talus slopes. Can. Geotech. Jo. 30, 620-636. Evans, S.G., Aitken, J.D., Wetmiller, R.J., and Horner, R.B. (1987) A rock avalanche triggered by the October 1985 North Nahanni Earthquake, District of Mackenzie, N.W.T., Can. Jo. Earth Sci. 24, 176-184. Evans, S.G., Clague, J.J., Woodsworth, G.J., and Hungr, O. (1989) The Pandemonium Creek rock avalanche, British Columbia, Can. Geotech. Jo. 26, 427-446 Evans, S.G., Couture, R., Locat, J., Hadjigeorgiou, J.,Antoine, P., and Brugnot, G. (1997) Two complex cataclinal slope failures in Paleozoic limestones, Rocky Mountains, Jasper National Park, Alberta, Proc. 50th Can. Geotech. Conf., Ottawa. Preprint Volume v. 1, pp. 102-109. Evans, S.G., Hungr, O., and Clague, J.J. (2001) Dynamics of the 1984 rock avalanche and associated debris flow on Mount Cayley, British Columbia, Canada; implications for landslide hazard assessment on dissected volcanoes, Eng. Geol. 61, 29-51. Evans, S.G., Hungr, O. and Enegren, E. G. (1994) The Avalanche Lake rock avalanche, Mackenzie Mountains, Northwest Territories, Canada; description,dating, and dynamics, Can. Geotech. Jo. 31, 749-768. Fauque, L. and Strecker, M.R. (1988) Large rock avalanche deposits (Sturzsröme, sturzstroms) at Sierra Aconquija, northern Sierra Pampeanas, Argentina, Eclog. Geol. Helv. 81, 579-592. Fauque, L. and Tciligharian, P. (2002) Villavil rockslides, Catamarca Province, Argentina, in S.G. Evans and J.V. DeGraff (eds.), Catastrophic landslides: effects, occurrence, and mechanisms, Geol. Soc. Am. Rev. Eng. Geol., v. XV, pp. 303-324. Finlay, P.J., Mostyn, G.R., and Fell, R. (1999) Landslide risk assessment: prediction of travel distance, Can. Geot. Jo. 36, 556-562. Follacci, J-P. (1987) Les mouvements du versant de la Clapière à Saint-Etienne-de-Tinée (AlpesMaritimes). Bull. liason Lab. Pont et Chaussee, 150, 31-46. Follacci, J-P., Guardia, P., and Ivaldi, J-P., 1988. Le glissement de la Clapière (Alpes Maritimes, France) dans son cadre géodynamique, Proc. 5th Int. Symp. on Landslides, v. 2, pp. 1323-1327. Fort, M. (1987) Sporadic morphogenesis in a continental subduction setting : an example from the Annapurna Range, Nepal Himalaya. Zeits. fur Geomorph. Suppl.-Bd. 63, 9-36. Fort, M. (2000) Glaciers and mass wasting processes: their influence on the shaping of the Kali Gandaki valley (higher Himalaya of Nepal). Quat. International, 65/66, 101-119 Fort, M. and Peulvast, J-P. (1995) Catastrophic mass-movements and morphogenesis in the PeriTibetan Ranges: examples from West Kunlun, East Pamir, and Ladakh, in O. Slaymaker (ed.) Steepland Geomorphology J.Wiley & Sons, New York, pp. 171-198. Fort, M. Burbank, D.W., and Freytet, P. (1989) Lacustrine sedimentation in a semiarid alpine setting: an example from Ladakh, northwest Himalaya, Quat. Res., 31, 332-350. Francis, P.W. and Wells, G.L. (1988) Landsat Thematic Mapper observations of debris avalanche deposits in the Central Andes, Bull. Volc., 50, 258-278. Friedman, S.J. (1997) Rock-avalanche elements of the Shadow Valley Basin, eastern Mojave Desert, California: processes and problems, Jo. Sed. Res. 67, 792-804.S Fritz, H.M., Hager, W.H., and Minor, H-E. (2001) Lituya Bay case: rockslide impact and wave runup, Science of Tsunami Hazards 19, 3-22. Fuganti, A (1969) Studio geologico di sei grandi frane di roccia nella regione Trentino-Alto Adige, Mem. del Museo Tridentino di Scienze Naturali, 17, 63-128. Fuji, Y. (1969) Frequency distribution of the magnitude of the landslides caused by heavy rainfall. Journal of the Seismological Society of Japan 22, 244-247.

43 117. 118. 119. 120. 121. 122. 123.


125. 126. 127. 128. 129. 130. 131.

132. 133. 134.


136. 137. 138.


140. 141.

Gaziev, E. (1984) Study of the Usoi Landslide in Pamir. Proc. 4th Int. Symp. on Landslides, Toronto, v. 1, p. 511-515. Ghirotti, M. (1995) L’Antica frana del M. Borgá e primi resultati del monitoraggio dell’area di prá di salta (Casso, Pn), Quad. di geologia Applic. Riccione, 1, 123-130. Gillon, M.D. and Hancox, G.T. (1992) Cromwell Gorge Landslides – a general overview, Proc. 6th Int. Symp. on Landslides, Balkema, Rotterdam, v. 1, 83-120. Giraud, A., Rochet, L., and Antoine, P. (1990) Processes of slope failure in crystallophyllian formations. Eng. Geol., 29, 241-253. Glancy, P.A. and Bell, J.W. (2000) Landslide-induced floodiing at Ophir Creek, United States Geol. Surv. Prof. Paper 1617. Glass, J.H. (1896) The great landslip at Gohna, in Garhwal, and the measures adopted to prevent serious loss of life, Jo. of the Society of Arts, 44, 431-445. Glastonbury, J. and Fell, R. (2000) Report on the analysis of “rapid” natural rock slope failures, University of New South Wales School of Civil and Environmental Engineering UNICIV Report No. R-390, 154 p. and Appendices. Govi, M., Gulla, G., and Nicoletti, P.G. (2002) Val Pola rock avalanche of July 28, 1987, in Valtellina (Central Italian Alps), in S.G. Evans and J.V. DeGraff (eds.), Catastrophic landslides: effects, occurrence, and mechanisms, Geol. Soc. Am. Rev. Eng. Geol., v. XV, pp. 71-89. Griggs, R.F. (1920) The Great Mageik Landslide, Ohio Jo. of Sci. 20, 325-354. Gutenberg, B. and Richter, C.F. (1954) Seismicity of the Earth, Princeton University Press, Princeton, 440 p. Guzzetti , F., Cardinali, M., and Reichenbach, P. (1996) The influence of structural setting and lithology on landslide type and pattern, Env. Eng. Geosci. 2, 531-555. Guzzetti, F., Malamud, B.D., Turcotte, D.L., & Reichenbach, P. (2002) Power-law correlations of landslide areas in central Italy, Earth and Planetary Science Letters 195, 169-183. Guzzetti, F., Reichenbach, P. & Wieczorek, G.F. (2003) Rockfall hazard and risk assessment in the Yosemite valley, California, USA, Natural Hazards and Earth System Sciences 3, 491-503. Hancox, G.T. and Perrin, N.D. (1994) Green Lake Landslide: a very large ancient rock slide in Fiordland, New Zealand. Proc. 7th IAEG Cong., v. 3, 1677-1689. Harp, E.L., Jibson, R.W., Kayen, R.F., Keefer, D.K., Sherrod, B.L., Carver, G.A., Collins, B.D., Moss, R.E.S., and Sitar, N. (2003) Landslides and liquefaction triggered by the M 7.9 Denali Fault earthquake of 3 November 2002, GSA Today 13, 4-10. Harrison, J.V. and Falcon, N.L. (1938) An ancient landslip at Saidmarreh in southwestern Iran, Jo. Geol., 46, 296-309. Hauge, T.A. (1985) Gravity-spreading origin of the Heart Mountain allochton, northwestern Wyoming; GSA Bulletin, 96, 1440-1456. Hauser, A. (2002) Rock avalanche and resulting debris flow in Estero Parraguirre and Rio Colorado, Región Metropolitana, Chile, in S.G. Evans and J.V. DeGrafff (eds.), Catastrophic landslides: effects, occurrence, and mechanisms, Geol. Soc. Am. Rev. Eng. Geol., v. XV, pp. 135-148. Havenith, H-B., Strom, A., Jongmans, D., Abdrakhmatov, K., Delvaux, D., and Trefois, P. (2003) Seismic triggering of landslides; Part A: Filed examples from the northern Tien Shan. Nat. Haz. Earth Syst. Sci. 3, 135-149. Heim, Albert. (1932) Bergsturz und Menschenleben, Fretz and Wasmuth Verlag, Zurich, 218 p. Heim, Arnold (1948) Wunderland Peru, Verlag Hans Huber, Bern, 196 p. Hendron, A.J. and Patton, F.D. (1985) The Vaiont Slide: a geotechnical analysis based on new geological observations of the failure surface: U.S. Army Engineer Waterways Experiment Station, Technical Report GL-85-5, 2 volumes. Henderson, W. (1859) Memorandum on the nature and effects of the flooding of the Indus on 10th August, 1858, as ascertained at Attock and its neighbourhood, Jo. Asiatic Soc. of Bengal 28, 199219. Hermanns, R.L. and Strecker, M.R. (1999) Structural and lithological controls on large Quaternary rock avalanches (sturzstroms) in arid northwestern Argentina, GSA Bulletin 111, 934-948. Hermanns, R.L., Niedermann, S., Villanueva Garcia, A., Sosa Gomez, J. and Strecker , M.R. (2001) Neotectonics and catstrophic failure of mountain fronts in the southern intra-Andean Puna Plateau, Argentina, Geology 29, 619-623.

44 142.

143. 144. 145. 146. 147.


149. 150. 151. 152.

153. 154. 155. 156.

157. 158. 159. 160.



163. 164.


Hermanns, R.L., Trauth, M.H., Niedermann, S., McWilliams, M. and Strecker, M.R. (2000) Tephrochronologic constraints on temporal distribution of large landslides in northwest Argentina, Jo. of Geol. 108, 35-52. Hewitt, K. (1968) Records of natural damming and related events in the Upper Indus Basin, Indus: Jo. of Water and Power Dev. Auth., 10, 11-19. Hewitt, K., (1998) Catastrophic landslides and their effects on the upper Indus streams, Karakoram Himalaya, northern Pakistan, Geomorphology 26, 47-80. Hewitt, K. (1999) Quaternary moraines vs catastrophic rock avalanches in the Karakoram Himalaya, northern Pakistan, Quat. Res. 51, 220-237. Hewitt, K. (2001) Catastrophic rockslides and the geomorphology of the Hunza and Gilgit River valleys, Karakoram Himalaya, Erdkunde, 55, 72-93. Hewitt, K, (2002a) Postglacial landform and sediment associations in a landslide-fragmented river stsytem: the transHimalayan Indus streams, Central Asia, In K. Hewitt, M-L. Byrne, M. English, and G. Young (Eds.) Landscapes in Transition, Kluwer Academic Publishers, Dordrecht, p. 63-91. Hewitt, K. (2002b) Styles of rock avalanche depositional complexes conditioned by very rugged terrain, Karakoram Himalaya, Pakistan. In S.G. Evans and J.V. DeGraff (Editors) Catastrophic Landslides: Effects, Occurrence, and Mechanisms, Geol. Soc. Am., Reviews in Engineering Geology, v. XV, 345- 377. Hoek, E. and Bray, J.W. (1981) Rock Slope Engineering 3rd Ed. Inst. Min. Metall., London, Holland, T.H. (1894) Report on the Gohna Landslip, Garhwal. Records Geol. Surv. of India 27, 5565. Hovius, N., Stark, C.P., and Allen, P.A. (1997) Sediment flux from a mountain belt derived by landslide mapping, Geology 25, 231-234. Hovius, N., Stark, C.P., Tutton, M.A., and Abbott, L.D., (1998) Landslide-driven drainage network evolution in a pre-steady-state mountain belt: Finisterre Mountains, Papua New Guinea, Geology 26, 1071-1074. Hsu, K.J. (1975) Catastrophic debris streams (sturzstroms) generated by rockfalls, GSA Bull. 86, 129-140. Hu, X.Q. and Cruden, D.M. (1993) Buckling deformation in the Highwood Pass, Alberta, Canada, Can. Geotech. Jo. 30, 276-286. Huder, J. (1976) Creep in Bunder Schist, Laurits Bjerrum Memeorial Volume, Norwegian Geotechnical Institute, Oslo, p. 125-153. Hung, J-J., Lee, C-T., and Lin, M-L. (2002) Tsao-Ling rockslides, Taiwan, in S.G. Evans and J.V. DeGraff (eds.), Catastrophic landslides: effects, occurrence, and mechanisms, Geol. Soc. Am. Rev. Eng. Geol., v. XV, pp. 91-115. Hungr, O. (1995) A model for the runout analysis of rapid flow slides, debris flows, and avalanches, Can. Geotech. Jo. 32, 610-623. Hungr, O. and Evans, S.G. (1988) Engineering evaluation of fragmental rockfall hazards. Proceedings, V International Symposium on Landslides, v. 1, p. 685-690. Hungr, O. and Evans, S.G. (1996) Rock avalanche runout prediction using a dynamic model. Procceedings, 7th International Symposium on Landslides, Trondheim, Norway, v. 1, p. 233-238. Hungr, O. and Evans, S.G. (1997) A dynamic model for landslides with changing mass. Proceedings of International Association of Engineering Geology Symposium on Engineering Geology and the Environment, Athens, Greece, v. 1, p. 719-724. Hungr, O., Evans, S.G., and Hazzard, J. (1999) Magnitude and frequency of rock falls and rock slides along the main transportation corridors of southwestern British Columbia, Can. Geotech. Jo. 36, 224-238 Hungr, O., Dawson, R.F., Kent, A., Campbell, D., and Morgenstern, N.R. (2002) Rapid flow slides of coal-mine waste in British Columbia, Canada, in S.G. Evans and J.V. DeGraff (eds.), Catastrophic landslides: effects, occurrence, and mechanisms, Geol. Soc. Am. Rev. Eng. Geol., v. XV, pp. 191-208. Hungr, O., Evans, S.G., Bovis, M.J., and Hutchinson, J.N. (2001) A review of the classification of landslides of the flow type, Env. Eng. Geosci. 7, 221-238. Hutchinson , J.N. (1971). Field and laboratory studies of a fall in Upper Chalk cliffs at Joss Bay, Isle of Thanet. In Stress-Strain behaviour of soils, Proceedings, Roscoe Memorial Symposium, Cambridge, pp. 692-706. Hutchinson, J,N, (1987) Mechanisms producing large displacements in landslides on pre-existing shears, Mem. Geol. Soc. of China 9, 175-200.

45 166.

167. 168.

169. 170.


172. 173. 174.


176. 177. 178. 179. 180. 181. 182. 183. 184.

185. 186. 187. 188. 189. 190. 191. 192.

Hutchinson, J.N. (1988) General report : morphological and geotechnical parameters of landslides in relation to geology and hydrogeology, Proceedings of the 5th International Symposium on Landslides, Lausanne, Switzerland, v. 1, 3-35. Hutchinson, J.N. (1995) Deep-seated mass movements on slopes, Mem. Soc. Geol. It., 50: 147-164 Hutchinson, J.N. (2002) Chalk flows from the coastal cliffs of northwest Europe, in S.G. Evans and J.V. DeGraff (eds.), Catastrophic landslides: effects, occurrence, and mechanisms, Geol. Soc. Am. Rev. Eng. Geol., v. XV, pp. 257-302. Hutchinson, J.N. and Kojan, E. (1975) The Mayunmarca Landslide of 25th April 1974, Peru, Report Ser. No. 3124/RMO..RD/SCE, UNESCO, Paris. Imrie, A.S. and Moore, D.P.(1997) BC Hydro’s approach to evaluating reservoir slope stability from a risk perspective, In D.M. Cruden and R. Fell (Editors) Landslide Risk Assessment, A.A. Balkema, Rotterdam, pp. 215-226. Imrie, A.S., Moore, D.P., and Enegren, E.G. (1992) Performance and maintenance of the drainage system at Downie Slide, Proceedings, 6th International Symposium on Landslides, Christchurch, New Zealand, v. 1, p. 751-757, A.A. Balkema, Rotterdam. Ishida, T., Chigira, M., and Hibino, S. (1987) Application of the Distinct Element Method for analysis of toppling observed on a fissured rock slope. Rock Mech. and Rock. Eng. 20, 277-283. Iverson, R.M., Reid, M.E., and LaHusen, R.G. (1997) Debris-flow mobilization from landslides, Ann. Rev. Earth Plan. Sci., 25, 85-138. Jackson, L.E. (2002) Landslides and landscape evolution in the Rocky Mountains and adjacent Foothills area, southwestern Alberta, Canada, in S.G. Evans and J.V. DeGraff (eds.), Catastrophic landslides: effects, occurrence, and mechanisms, Geol. Soc. Am. Rev. Eng. Geol., v. XV, pp. 325344. Janda, R.J., Scott, K.M., Nolan, K.M., and Martinson, H.A. (1981) Lahar movement, effects, and deposits, In Lipman, P.W. and Mullineaux, D.R. (Editors) The 1980 eruptions of Mount St Helens, Washington, United States Geol. Surv., Prof. Paper 1250, p. 461-478. Jing, L. (2003) A review of techniques, advances and outstanding issues in numerical modelling for rock mechanics and rock engineering, Int. Jo. of Rock Mech. and Mining Sci., 40, 283-353. Johnson, R.W. (1987) Large-scale volcanic cone collapse: the 1888 slope failure of Ritter volcano, and other examples from Papua New Guinea, Bull. Volcanol., 49, 669-679. Jones, B.L., Chinn, S.S.W., and Brice, J.C. (1984) Olokele rock avalanche, island of Kauai, Hawaii, Geology 12, 209-211. Jorstad, F. (1968) Waves generated by landslides in Norwegain fjords and lakes. Norwegian Geotechnical Institiute Publication No. 69, p. 13-32. Kalkani, E.C. and Piteau, D.R. (1976) Finite element analysis of toppling failure at Hell’s Gate Bluffs, British Columbia, Bull. Assoc. Eng. Geol., 13, 315-327 Keefer, D.K., (1984) Landslides caused by earthquakes, Geol. Soc. Am. Bull. 95, 406-421. Keefer, D.K., (1994) The importance of earthquake-induced landslides to long-term slope erosion and slope-failure hazards in seismically active regions, Geomorpholgy 10, 265-284. Keefer, D.K. (2002) Investigating landslides caused by earthquakes – a historical review, Surv. in Geophysics 23, 473-510. Kerle, N. and van Wyk de Vries, B., (2001) The 1998 debris avalanche at Casita volcano, Nicaragua – investigation of structural deformation as the cause of slope instability using remote sensing, Jo. Volc. Geotherm. Res. 105, 49-63. Kiersch, G.A. (1964) Vaiont reservoir disaster, Civil Engineering, ASCE 34: 32-39 Kilburn, C.R.J., and Sorensen, S-A. (1998) Runout lengths of sturzstroms: the control of initial conditions and of fragment dynamics Jo. Geophys. Res. 103B, 17877-17884. Kimber, O.G., Allison, R.J., and Cox, N.J. (1998) Mechanisms of failure and slope development in rock masses. Trans. Inst. Br. Geogr., NS 23, 353-370. King, J.., Loveday, I., and Schuster, R.L. (1989) The 1985 Bairaman landslide dam and resulting debris flow, Papua New Guinea, Quat. Jo. Eng. Geol. 22, 257-270. Kjartansson, G. (1967) The Steinsholtshlaup, central-south Iceland on January 15th , 1967, Jokull 17, 249-262. Kobayashi, Y., Harp, E.L., and Kagawa, T. (1990) Simulation of rockfalls triggered by earthquakes, Rock Mech. and Rock. Eng. 23, 1-20. Kohlbeck, F., Scheidegger, A.F. and Sturgul, J.R. (1979) Geomechanical model of an alpine valley. Rock Mech., 12, 1-14. Kolderup, N-H. (1955) Raset I Modalen 14. August 1953, Norsk Geol. Tidssk. 34, 211-217.

46 193. 194. 195. 196. 197. 198. 199. 200. 201. 202. 203. 204.


206. 207. 208. 209.

210. 211. 212.

213. 214. 215.

216. 217.

Korup, O. (2002) Recent research on landslide dams – a literature review with special attention to New Zealand. Prog. in Phys. Geog., 26, 206-235. Krahn, J. and Morgenstern, N.R. (1976) Mechanics of the Frank slide. In Rock Engineering for Foundations and Slopes; American Society of Civil Engineers, v. 1, p. 309-331. Krieger, M.H. (1977) Large landslides, composed of megabreccia, interbedded in Miocene basin deposits, southeastern Arizona, United States Geol. Surv. Prof. Paper 1008, 25 p. Latter, J.H. (1981) Tsunamis of volcanic origin: summary of causes, with particular reference to Krakatoa, 1883, Bull. Volcanol. 44, 467-499. Lee, K.L. and Duncan, J.M. (1975) Landslide of April 25, 1974 on the Mantaro River, Peru. National Acad. Sci., Washington, DC, 72 p. Legros, F. (2002) The mobility of long run-out landslides, Eng. Geol. 63, 301-331. Lehmann, O. (1926) Die verheerungen in der Sandlinggruppe (Salzkammergut), Akad. Der Wiss. In Wien, Denk. 100, 259-299. Leoniv, N.N. (1960) The Khait, 1949 earthquake and geological conditions of its ocurrence. Izvestia of the Academy of Sciences of the USSR, Geophysical Series, No 3, 409-424. Longwell, C.R. (1951) Megabreccia developed downslope from large faults, Am. Jo. Sci., 249, 343355. Lopez, D.L. and Williams, S.N. (1993) Catastrophic volcanic collapse: relation to hydrothermal processes, Science 260, 1794-1796. Li, T. and Wang, S. (1992) Landslide hazards and their mitigation in China, Science Press, Beijing, 84 pp. Li, T., Schuster, R.L., and Wu, J. (1986) Landslide dams in south-central China, R.L. Schuster (ed.) Landslide Dams:processes, risk, and mitigation, Geotechnical Special Publication No. 3, American Society of Civil Engineers, New York, p. 146-162. Lliboutry, L., Morales, B., Pautre, A. and Schneider, B. (1977) Glaciological problems set by the control of dangerous lakes in Cordillera Blanca, Peru. 1. Historical failures of morainic dams, their causes and prevention, Jo. of Glaciol., 18, 239-254. Mader, C.L. (1999) Modelling the 1958 Lituya Bay mega-tsunami, Science of Tsunami Hazards 17, 57-67. Major, J.J. and Newhall, C.G. (1989) Snow and ice perturbation during historical volcanic eruptions and the formation of lahars and floods, Bull. Volcanol. 52, 1-27. Martin, C.D. and Kaiser, P.K. (1984) Analysis of a rock slope with internal dilation. Can Geotech. Jo. 21, 605-620. Martino, S., Prestininzi, A. and Scarascia-Mugnozza, G. (2001) Mechanisms of deep seated gravitational deformations: parameters from laboratory testing for analogical and numerical modeling, in P. Sarkka P. Eloranta (eds.) Rock mechanics – a challenge for society, Balkema, Rotterdam, p. 137-142. Mason, K. (1929) Indus Floods and Shyok Glaciers, Himal. Journal 1, 10-29. Matsukura, Y. (2001) Rockfall at Toyohama Tunnel, Japan, in 1996: effect of notch growth on instability of a coastal cliff, Bull. Eng. Geol. Env. 60, 285-289. McCalpin, J.P. (1999) Criteria for determining the seismic significance of sackungen and other scarplike landforms in mountainous regions. Appendix A, p. A-122 to A-142 in Techniques for identifying faults and determining their origins. U.S. Nuclear Regulatory Commission NUREG/CR5503. McCalpin, J.P. and Irvine, J.R. (1995) Sackungen at the Aspen Highlands ski area, Pitkin County, Colorado, Env. Eng. Geosci. 1, 277-290. McConnell, R.G. and Brock, R.W. (1904) Report on the Great Landslide at Frank, Alberta, Annual Report of the Department of Interior for the year 1902-1903, Part VIII, 17 p. McGuire, W.J. (1996). Volcano instability; a review of contemporary themes. In McGuire, W.J., Jones, A.P. and Neuberg, J. (Editors) 1996. Volcano instability on the earth and other planets. Geological Society of London Special Publication No. 11, p. 1-23. McSaveney, M.J. (1978) Sherman Glacier rock avalanche, Alaska, U.S.A. in B. Voight (ed.) Rockslides and avalanches, 1. Natural Phenomena, Elsevier, Amsterdam, p.197-258. McSaveney, M.J. (2002) Recent rockfalls and rock avalanches in Mount Cook National Park, New Zealand, in S.G. Evans and J.V. DeGraff (eds.), Catastrophic landslides: effects, occurrence, and mechanisms, Geol. Soc. Am. Rev. Eng. Geol., v. XV, pp. 35-70.

47 218.

219. 220. 221. 222. 223. 224. 225.

226. 227. 228. 229. 230. 231. 232. 233. 234.

235. 236. 237.



240. 241. 242. 243. 244.

Meidal, K.M. and Moore, D.P. (1996) Long-term performance of instrumentation at Dutchman’s Ridge, Proceedings, 7th. International Symposium on Landslides,Trondheim, Norway, v. 3, p. 15651570. Melekestsev, I.V., and Braitseva, O.A. (1988) Gigantic collapses on volcanoes, Volcanology and Seis-mology 6, 495-508 (in Russian). Miller, D.J. (1960) Giant waves in Lituya Bay, Alaska, United States Geol. Surv. Prof. Paper 354-C, p. 51-86. Montadon, F. 1933. Chronologie des grands éboulements alpins du début de l’ère chrétienne à nos jours. Matériaux pour l’étude des calamités, 32, 271-340. Moore, D.P. (1990) Stabilization of Dutchman's Ridge, Geotechnical News, 8, 47. Moore, D.P. and Imrie, A.S. (1995) Stabilization of Dutchman’s Ridge, Proceedings of 6th International Symposium on Landslides, Christchurch, New Zealand, v.3, p. 1783-1788. Moore, D.P., Ripley, B.D., and Groves, K.L. (1992) Evaluation of mountainslope movements at Wahleach; in Geotechnique and natural hazards Bitech Publishers, Vancouver, B.C. p. 99-107. Moriwaki, H., Yazaki, S., and Oyagi, N. (1985) A gigantic debris avalanche and its dynamics at Mount Ontake caused by the Nagano-ken-seibu earthquake, 1984. Proc. 4th Int. Conf. Field Workshop on Landslides, Tokyo, p. 359-364. Morris, T.H. and Hebertson, G.F. (1996) Large-rock avalanche deposits, eastern Basin and Range, Utah: emplacement, diagenesis, and economic potential, AAPG Bulletin 80, 1135-1149. Mothes, P.A., Hall, M.L., and Janda, R.J. (1998) The enormous Chillos Valley Lahar: an ash-flowgenerated debris flow from Cotopaxi Volcano, Ecudaor, Bull. Volcanol. 59, 233-244. Muller, L. (1964) The rock slide in the Vaiont valley, Rock Mech. and Eng. Geol., 6, 148-212. Naranjo, J.A. and Francis, P. (1987) High velocity debris avalanche at Lasstarria volcano in the north Chilean Andes, Bull. Volc. 49, 509-514. Nichol, S.L., Hungr, O., and Evans, S.G. (2002) Large-scale brittle and ductile toppling of rock slopes, Can. Geotech. Jo. 39, 773-788. Nicoletti, P.G. and Parise, M. (1996) Geomorphology and kinematics of the Conturrana rockslidedebris flow (NW Sicily), Earth Surf. Proc .and Landforms, 21, 875-892. Niederer, J., (1941) Der Felssturz am Flimserstein. Jahresbericht der Naturforschenden Gesellschaft Graubünden, Chur Nikonov, A.A., and Shebalina, T.Y. (1979) Lichenometry and earthquake age determination in central Asia, Nature 280, 675-677. Norrish, N.I. and Wyllie, D.C. (1996) Rock slope stability analysis. In A.K. Turner and R.L. Schuster (Editors) Landslides:investigation and mitigation, Special Report 247 Tran. Res. Bd., Nat. Res. Council, National Academy Press, Washington, p. 391-425 Noverraz , F. (1996) Sagging or deep-seated creep: fiction or reality? Proc. 7th International Symposium on Landslides, Trondheim, Norway, p, 821-828, A.A. Balkema, Rotterdam Ogawa, T. and Homma, F. (1926) The geology of the Unzen Volcanoes, Pan-Pac. Sci. Congr. Guidebook for Excursion E-1, 3, 4, 30 p. Ohmori, H. and Hirano, M. (1988) Magnitude, frequency and geomorphological significance of rocky mud flows, landcreep, and the collapse of steep slope, Zeits für Geomorph, Suppl. –Bnd 67, 55-65. Orombelli, G. and Porter, S.C. (1988) Boulder deposit of upper Val Ferret (Courmayeur, Aosta valley) : deposit of a historic giant rockfall and debris avalanche or a late-glacial moraine ? Eclog. Geol. Helv. 81, 365-371. Osipov, V.I., Mamaev, Yu.A. (1998) Peculiarities of formation and structure of large-scale blockages in mountain valleys. Geoecology, Engineering Geology, Hydrogeology, Geocryology, 6, 94-99 (in Russian). Ota, K. (1969) Study on the collapses in the Mayu-yama, Sci. Rpt. Shimabara Volc. Obs., Fac. Sci., Kyushu University, 5: 6-35 (in Japanese) Owen, L. (1989) Neotectonics and glacial deformation in the Karakoram Mountains and Nanga Parbat Himalaya, Tectonophysics, 163, 227-265. Owen, L. A. (1996) Quaternary lacustrine deposits in a high-energy semi-arid mountain environment, Karakoram Mountains, northern Pakistan, Jo. Quat. Sci. 11, 461-483. Paul, S.K., Bartarya, S.K., Rautela, P., and Mahajan, A.K. (2000) Catastrophic masss movement of 1998 monsoons at Malpa in Kali Valley, Kumaun Himalaya (India), Geomorphology 35, 169-180. Pedersen, S.A.S., Larsen, L.M., Dahl-Jensen, T., Jepsen, H.F., Pedersen, G.K., Nielsen, T., Pedersen, A.K., von Platen-Hallermud, F., and Weng, W. (2002) Tsunami-generating rock fall and landslide on



246. 247. 248.

249. 250. 251. 252. 253. 254.

255. 256. 257. 258. 259.

260. 261. 262. 263.

264. 265. 266. 267. 268. 269. 270.

the south coast of Nuussuaq, central West Greenland, Geology of Greenland Survey Bulletin, 191, 73-83. Pelletier, J.D., Malamud, B.D., Blodgett, T., & Turcotte, D.L. (1997) Scale-invariance od soil moisture variability and its implications for the frequency-size distribution of landslides, Eng. Geol. 48, 255-268. Pfeiffer, T.J. and Bowen, T.D. (1989) Computer simulation of rockfalls, Bull. Assoc. Eng. Geol. 26, 135-146. Philip, H. and Ritz, J-F. (1999) Gigantic paleolandslide associated with active faulting along the Bogd fault (Gobi-Altay, Mongolia), Geology 27, 211-214. Plafker, G. (1968) Source areas of the Shattered Peak and Pyramid Peak Landslides at Sherman Glacier. In The Great Alaska Earthquake of 1964. v. 3, p. 374-382. National Academy of Sciences, Washington, D.C. Plafker, G. and Eriksen, G.E. (1978) Nevados Huascarán avalanches, Peru. In B. Voight (Ed.) Rockslides and Avalanches v. 1, Elsevier, Amsterdam, p. 277-314. Plafker, G. and Eyzaguirre, V.R. (1979) Rock avalanche and wave at Chungar, Peru. In B. Voight (Ed.) Rockslides and Avalanches, v. 2. Engineering Sites, Elsevier, Amsterdam, p. 269-279 Poisel, R. (1998) Kippen, Sacken, Gleiten, Felsbau, 16, 135-140. Porter, S.C. and Orombelli, G. (1980) Catastrophic rockfall of September 12, 1717 on the Italian flank of the Mont Blanc Massif, Zeits. fur Geomorph. 24, 200-218. Porter, S.C. and Orombelli, G. (1981) Alpine rockfall hazards, American Scientist, 69, 67-75. Poschinger, A. von. (2002) Large rockslides in the Alps: A commentary on the contribution of G. Abele (1937-1994) and a review of some recent developments, in S.G. Evans and J.V. DeGraff (eds.), Catastrophic landslides: effects, occurrence, and mechanisms, Geol. Soc. Am. Rev. Eng. Geol., v. XV, pp. 237-255. Post, A. (1967) Effects of the March 1964 Alaska Earthquake on Glaciers, United States Geol. Surv. Prof. Paper 544-D, 42 p. Preobrajensky, J. (1920) The Usoi Landslide, Geol. Comm., Papers on Applied Geol., 14, 21 p. (In Russian) Pritchard, M.A. and Savigny, K.W. (1990) Numerical modeling of toppling, Can. Geotech. Jo., 27, 823-834. Pritchard, M.A and Savigny, K.W. (1991) The Heather Hill landslide: an example of a large scale toppling failure in a natural slope, Can. Geotech. Jo. 28, 410-422 Read, S.A.L., Beetham, R.D., and Riley, P.B. (1992) Lake Waikaremoana barrier – a large landslide dam in New Zealand, Proc. 6th Intern. Symposium on Landslides, Christchurch, New Zealand, v. 2, pp. 1481-1487, Balkema, Rotterdam. Reid, M.E., Christian, S.B., and Brien, D.L. (2000) Gravitational stability of three-dimensional stratovolcano edifices, J. Geophys. Res. 105, 6043-6056. Reid, M.E., Sisson, T.W., and Brien, D.L. (2001) Volcano collapse promoted by hydrothermal alteration and edifice shape, Mount Rainier, Washington, Geology 29, 779-782. Reneau, S.L., and Dethier, D.P., (1996) Late-Pleistocene landslide-dammed lakes along the Rio Grande, White Rock Canyon, New Mexico, GSA Bull. 108, 1492-1507. Riemer, W., Locher, T., and Nunez, I. (1988) Mechanics of deep seated mass movements in metamorphic rocks in the Ecuadorian Andes, Proceedings of the 5th International Symposium on Landslides, Lausanne, Switzerland, v. 1, 307-310. Rodriguez, C.E., Bommer, J.J., and Chandler, R.J. (1999) Earthquake-induced landslides: 19801997, Soil Dyn. and Earthquake Eng. 18, 325-346. Sartori, M., Baillifard, Jaboyedoff, M., and Rouiller, J-D. (2003) Kinematics of the 1991 Randa rockslides (Valais, Switzerland) Nat. Haz. and Earth Syst. Sci., 3, 423-433 Sassa, K. (1988) Geotechnical model for the motion of landslides (Special lecture). Proceedings, 5th International Symposium on Landslides, Lausanne, Switzerland, v. 1, p. 37-56. Sassa, K. (Editor) (1999) Landslides of the world. Kyoto University Press, 413 pp. Savage, S.B. and Hutter, K. (1991) The dynamics of avalanches of granular materials from initiation to runout. Part 1. Analysis, Acta Mech. 86, 201-223. Scarpa, R. and Tilling, R.I. (Editors) (1996) Monitoring and Mitigation of Volcano Hazards, Springer-Verlag, Berlin. 841 pp. Scheidegger, A.E., (1973) On the prediction of the reach and velocity of catastrophic landslides. Rock Mechanics, 5, 231-236.

49 271.

272. 273. 274. 275.


277. 278. 279. 280.

281. 282. 283.

284. 285. 286.


288. 289. 290.


292. 293. 294. 295.

Scheko, A.I. and Lekhatinov, A.M. (1970) Current state of the Usoi Blockage and tasks of future studies. In Materials of Scientific-technical meeting on the problems of study and forecast of the mudflows, rockfalls and landslides, Dushanbe, 219-223 (in Russian) Schuster, R.L. (2002) Usoi landslide dam, southeastern Tajikistan, Proc. Int. Symp. on Landslide Risk Mitigation and Protection of Cultural and Natural Heritage, Kyoto, p. 489-505 Schuster, R.L., Logan, R.L., and Pringle, P.T., (1992) Prehistoric rock avalanches in the Olympic Mountains, Washington, Science 258, 1620 – 1621. Scott, K.M. (1988) Origins, behaviour and sedimentology of lahars and lahar-runout flows in the Tooutle-Cowlitz river system, United States Geol. Surv. Prof. Paper 1447-A, 74 p. Scott, K.M. (2000) Precipitation-triggered debris flow at Casita Volcano, Nicaragua: implications for mitigation strategies in volcanic and tectonically active steeplands, in G.F. Wieczorek and N.D. Naeser (eds.) Debris-flow hazards mitigation: mechanics, prediction and assessment, A.A. Balkema, Rotterdam, p. 3-13 Scott, K.M., Macías, J.L., Naranjo, J.A., Rodriguez, S., and McGeehin, J.P. (2001) Catastrophic debris flows transformed by landslides in volcanic terrains: mobility, hazard assessment, and mitigation strategies. United States Geol. Surv. Prof. Paper 1630, 67 p. Selli, R. and Trevisan, L. (1964) Caratteri e interpretazione della frana del Vaiont, Giornale di Geologia 32, 7-104. Semenza, E. and Ghirotti, M. (1998) Vaiont-Longarone 34 Anni Dopo La Catastrophe. Annali dell’Univsita di Ferrara (Nuovo Serie) Scienze della Terra 7, 63-94. Semenza, E. and Ghirotti, M. (2000) History of the 1963 Vaiont slide: the importance of geological factors, Bull. Eng. Geol. Env. 59, 87-97. Sevilla, J.H. (1994) The Josefina landslide and its implications in the electrical service for the Republic of Ecuador; Proceedings, 7th International Congress, International Association of Engineering Geology, Lisbon, v. 3, p. 1801-1810. Shang,Y., Yang, Z., Li, L., Liu, D., Liao, Q. and Wang, Y. (2003) A super-large landslide in Tibet in 2000 : background, occurrence, disaster, and origin, Geomorphology 54, 225-243. Sharpe, C.F.S. (1938) Landslides and related phenomena, Columbia University Press, New York, 137 pp. Sheridan, M.F., Bonnard, C., Carreno, R., Siebe, C., Strauch, W., Navarro, M., Calero, J.C., and Trujillo, N.B. 1999. Report on the 30 October 1998 Rock Fall/ Avalanche and breakout flow of Casita Volcano, Nicaragua, triggered by Hurricane Mitch, Landslide News, v. 12, p. 2-4. Shpilko, G.A. (1915) New data on the Usoi blockage and the Sarez Lake, Proc. of the Turkestan Department of Russian Geographical Society, 11, 11-17 (in Russian). Siebert, L. (1984) Large volcanic debris avalanches: characteristics of source areas, deposits, and associated eruptions, Jo. Volc. Geotherm. Res. 22, 163-197. Siebert, L. (1996) Hazards of large volcanic debris avalanches and associated eruptive phenomena, in R. Scarpa and R.I. Tilling (eds.) Monitoring and mitigation of volcano hazards, Springer-Verlag, Berlin, p. 541-57. Siebert, L. (2002) Landslides resulting from structural failure of volcanoes, in S.G. Evans and J.V. DeGraff (eds.), Catastrophic landslides: effects, occurrence, and mechanisms, Geol. Soc. Am. Rev. Eng. Geol., v. XV, pp. 209-235. Siebert, L., Beget, J.E., and Glicken, H. (1995) The 1883 and late-prehistoric eruptions of Augustine volcano, Alaska, Jo. of Volc. and Geotherm. Res. 66, 367-395. Siebert, L., Glicken, H., and Ui, T. (1987) Volcanic hazards from Bezymianny- and Bandai-type eruptions, Bull. Volcanol. 49, 435-459. Singh, N.K. & Vick, S.G. (2003) Probabilistic rockfall hazard assessment for roadways in mountainous terrain. Proceedings, 3rd Canadian Conference on Geotechnique and Natural Hazards, Canadian Geotechnical Society, Edmonton, p. 253-260. Slingerland, R.L. and Voight, B. (1979) Occurrences, properties and predictive models of landslidegenerated impulse waves. In B. Voight (Ed.) Rockslides and avalanches v. 2, p. 317-397, Elsevier, Amsterdam. Solonenko, V.P. (1970) Scars on the Earths' face, Priroda (Nature), 9, 17-25 (in Russian). Solonenko, V.P. (1972) Seismogenic destruction of mountain slopes, Proc. Int. Geol. Congr., Section 13, Montreal, p. 284-290. Solonenko, V.P. (1977) Landslides and collapses in seismic zones and their prediction, Bull. Int. Assoc. Eng. Geol., 15, 4-8 . Sousa, J, and Voight, B. (1991) Continuum simulation of flow failures, Geotechnique, 41, 515-538.

50 296. 297. 298.

299. 300.

301. 302.

303. 304. 305. 306. 307. 308. 309. 310.

311. 312.

313. 314.

315. 316.

317. 318. 319. 320. 321. 322.

Stead, D. and Eberhardt, E. (1997) Developments in the analysis of footwall slopes in surface coal mining, Eng. Geol. 46, 41-61. Steidl, A. and Riedmuller, G. (2003) Deep-seated gravitational slope deformation – a case study. Felsbau 21: 55-61. Stoopes, G.R. and Sherida\n, M.F. (1992) Giant debris avalanches from the Colima Volcanic Complex, Mexico: implications for long-runout landslides (>100 km) and hazard assessment, Geology 20, 299-302. Strecker, M.R., and Marrett, R. (1999) Kinematic evolution of fault ramps and its role in development of landslides and lakes in the northwestern Argentine Andes, Geology 27, 307-310. Strom, A.L. (1994a) Mechanism of stratification and abnormal crushing of rockslide deposits. Proceedings, 7th International Association of Engineering Geology Congress, Lisbon, Portugal, v. 3, p. 1287-1295. Strom, A.L. (1994b) Formation of structure of large rockslide deposits, Geoecology, Engineering Geology, Hydrogeology, Geocryology., No 5, 64-77 (in Russian). Strom, A.L. (1996) Some morphological types of long-runout rockslides: effect of the relief on their mechanism and on the rockslide deposits distribution. In K. Senneset (ed.) Landslides, Proceedings, 7th International Symposium on Landslides, Trondheim, Norway, p. 1977-1982. Strom, A.L. (1998) Giant ancient rockslides and rock avalanches in the Tien Shan Mountains, Kyrgyzstan, Landslide News, 11, 20-23. Sugai, T., Ohmori, H., & Hirano, M. (1994) Rock control on magnitude-frequency distribution of landslides. Transactions Japanese Geomorphological Union 15, 233-251. Suklje, L. and Vidmar, S. (1961) A landslide due to long term creep. Proc. 5th Int. Conf. Soil Mech. and Found. Eng., v. 2, 727-735. Suwa, H. (1991) Visually observed failure of a rock slope in Japan, Landslide News, 5, 8-10. Tabor, R.W. (1971) Origin of ridge top depressions by large-scale creep in the Olympic Mountains, Washington. GSA Bulletin 82, 1811-1822. Terzaghi, K. (1960) Storage dam founded on landslide debris, Journal of the Boston Society of Civil Engineers, 47, 64-94. Thorsen, G.W. (1989) Splitting and sagging of mountains, Washington Geology, 17, 3-13. Thompson, S.C., Clague, J.J., and Evans, S.G. (1997) Holocene activity of the Mt. Currie scarp, Coast Mountains, British Columbia, and implications for its origin. Environmental and Engineering Geoscience, 3, 329-348. Tika, T.E. and Hutchinson, J.N. (1999) Ring shear tests on soil from the Vaiont landslide slip surface, Geotechnique 49, 59-74. Tommasi, P. and Galadini, F. (1996) Rock slides and buckling phenomena on a homoclinal mountain slope in south-eastern Alps (Italy) Proc. 7th Int. Symp. on Landslides, Trondheim, v. 2, 1391-1396, Balkema, Rotterdam. Trauth, M.H., Alonso, R.A., Haselton, K.R., Hermanns, R.L. and Strecker, M.R. (2000) Climate change and mass movements in the Argentine Andes, Earth Planet. Sci. Lett. 179, 243-256. Trauth, M.H., Bookhagen, B., Marwan, N., and Strecker, M. (2003) Multiple landslide clusters record Quaternary climate changes in the northwestern Argentine Andes, Palaeogeog., Palaeoclim., Palaeoec., 194, 109-121. Trunk, F.J., Dent, J.D., and Lang, T.E. (1986) Computer modeling of large rock slides, Jo. Geotech. Eng., 112, 348-360 Tsuji, Y. and Hino, T. (1993) Damage and inundation height of the 1792 Shimabara landslide tsunami along the coast of Kumamoto Prefecture, Bulletin of the Earthquake Research Institute, University of Tokyo, 68, 91-176. Ui, T., Yamamoto, H., and Suzuki-Kamata, K. (1986) Characterisation of debris avalanche deposits in Japan, Jo. Volcanol and Geothermal. Res. 29, 231-243. Ulrhammer, R.A. (1996). Yosemite rock fall of July 10, 1996, Seism. Res. Lett. 67, 47-48. Vallance, J.W. and Scott, K.M. (1997) The Osceola mudflow from Mount Rainier: sedimentology and hazard implications of a huge clay-rich debris flow, GSA Bulletin, 109, 143-163. Vallance, J.W., Siebert, L., Rose Jr., W.I., Girón, J.R., and Banks, N.G., (1995) Edifice collapse and related hazards in Guatemala, Jo. of Volc. and Geoth. Res. 66, 337 – 355. van Bemmelen, R.W. (1970) The geology of Indonesia. General geology of Indonesia and adjacent archipelago, 2nd. Edition, vol. 1A, Matinhus Nijhiff, The Hague. van Wyk de Vries, B., and Francis, P.W. (1997) Catastrophic collapse at stratovolcanoes induced by gradual volcano spreading, Nature 387, 387-390.

51 323. 324.

325. 326. 327. 328. 329. 330. 331. 332. 333.

334. 335.

336. 337. 338. 339. 340. 341.

342. 343. 344.



347. 348.

van Wyk de Vries, B., Self, S., Francis, P.W., and Keszthelyi, L. (2001) A gravitational spreading origin for the Socompa debris avalanche, Jo. Volc. Geotherm. Res. 105, 225-247. Varnes, D.J., Radbruch-Hall, D.H., and Savage, W.Z. (1989) Topographic and structural conditions in areas of gravitational spreading of ridges in the western United States, United States Geol. Surv. Prof. Paper 1496, 28 p. Voight, B. (Editor) (1978) Rockslides and Avalanches 1: natural phenomena. Elsevier, Amsterdam, 833 pp. Voight, B. (Editor) (1979) Rockslides and Avalanches 2 : engineering sites. Elsevier, Amsterdam, 850 pp. Voight, B. (2002) Structural stability of andesite volcanoes and lava domes, Phil. Trans. R. Soc. Lond. Ser. A 258, 1663-1703. Voight, B. and Elsworth, D. (1997) Failure of volcano slopes, Géotechnique 47, 1-31. Voight, B. and Sousa, J. (1994) Lesson from Ontake-san: a comparitive analysis of debris avalanche dynamics. Eng. Geol. 38, 261-297. Voight, B., Janda, R.J., Glicken, H., and Douglass, P.M. (1983) Nature and mechanics of the Mount St. Helens rockslide-avalanche of 18 May, 1980, Géotechnique 33, 243-273. Wadge, G., Francis, P.W., and Ramirez, C.F. (1995) The Socompa collapse and avalanche event, Jo. Volc. Geoth. Res. 66, 309-336. Waltham, T. (1996) Very large landslides in the Himalayas, Geology Today 12, 181-185. Wang, W-N., Chigira, M., and Furuya, T. (2003). Geological and geomorphological precursors of the Chiu-fen-erh-shan landslide triggered by the Chi-Chi earthquake in central Taiwan, Eng. Geol. 69, 1-13 Ward, S.N. and Day, S. (2003) Ritter Island Volcano-lateral collapse and the tsunami of 1888, Geophys. J. Int. 154, 891-902. Watson. R.A. and Wright, H.E. (1969) The Saidmarreh Landslide, Iran, In S.A. Schumm and W.C. Bradley (eds.) United States Contributions to Quaternary Research, Geol. Soc. Am. Special Paper 123, p. 115-139. Watters, R.J. (1983) A landslide induced waterflood-debris flow, Bull. Int. Ass. Eng. Geol. 28, 177182. Wayne, W.J. (1999) The Allemania rockfall dam: a record of a mid-holocene earthquake and catastrophic flood in northwestern Argentina, Geomorphology, 27, 295-306. Waythomas, C.F. and Wallace, K.L. (2002) Flank collapse at Mount Wrangell, Alaska, recorded by volcanic mass-flow deposits in the Copper River lowland, Can. Jo. Earth Sci., 39, 1257-1279. Waythomas, C.F., Miller, T.P., and Beget, J.E. (2000) Record of Late Holocene debris avalanches and lahars at Iliamna Volcano, Alaska, Jo. Volc. Geoth. Res. 104, 97-130. Weidinger, J. T. (1998) Case history and hazard analysis of two lake-damming landslides in the Himalayas, Jo. of Asian Earth Sci.16, 323-331. Weidinger, J.T., Wang, J. and Ma, N. (2002) The earthquake-triggered rock avalanche of Cui Hua, Qin Ling Mountains, P.R. of China – the benefits of a lake damming prehistoric natural disaster. Quat. Internat. 93-94, 207-214. Whitehouse, I.E. (1983) Distribution of large rock avalanche deposits in the central Southern Alps, New Zealand, New Zealand Jo. Geol. Geophys., 26, 271-279. Whitehouse, I.E. and Griffiths, G.A. (1983) Frequency and hazard of large rock avalanches in the central Southern Alps, New Zealand, Geology 11, 331-334. Wieczorek, G.F. (2002) Catastrophic rockfalls and rockslides in the Sierra Nevada, USA, In S.G. Evans and J.V. DeGraff (eds.), Catastrophic landslides: effects, occurrence, and mechanisms, Geol. Soc. Am. Rev. Eng. Geol., v. XV, pp. 165-190. Wieczorek, G.F., Jakob, M., Motyka, R.J., Zirnheld, S.L., and Craw, P. (2003) Preliminary landslide-induced wave hazard assessment: Tidal Inlet, Glacier Bay National Park, Alaska, United States Geol. Surv. Open-File Report 03-100 Wieczorek, G.F., Snyder, J.B., Waitt, R.B., Morrisey, M.M., Uhrhammer, R.A., Harp, E.L., Norris, R.D., Bursik, M.I., and Finewood, L.G. (2000) Unusual July 10, 1996, rock fall at Happy Isles, Yosemite National Park, California, GSA Bulletin 112, 75-85. Wiles, G.C. and Calkin, P.E. (1992) Reconstruction of a debris-slide-initiated flood in the southern Kenai Mountains, Alaska, Geomorphology. 5, 535-546. Wise, D.U. (1963) Keystone faulting and gravity sliding driven by basement uplift of Owl Creek Mountains, Wyoming. Bull. Am. Assoc. of Petrol. Geol. 47, 586-598.

52 349.

350. 351.

Yarnold, J.C. and Lombard, J.P. (1989) A facies model for large rock-avalanche deposits formed in dry climates, In Colburn, I.P., Abbott, P.L., and Minch, J. (eds.) Conglomerates in Basin Analysis: A Symposium Dedicated to A.G. Woodford. Pacific Section S.E.P.M. v. 62, p. 9-31. Zhou, C.H., Yue, Z.Q., Lee, C.F., Zhu, B.Q., and Wang, Z.H. (2001) Satellite image analysis of a huge landslide at Yi Gong, Tibet, China, Quat. Jo. Eng. Geol. 34, 325-332. Zolotarev, G.S., Fedorenko, V.S., and Vinnichenko, S.M. (1986) Ancient seismogenic and recent slides and slumps in Tajikistan (Central Asia), Geol. Applic. e Idrogeol., 21, 149-157.


Abstract Landslides resulting from massive rock slope failure (rockslides, rock avalanches and the failure of volcano slopes (including edifice collapse)) are an important geological hazard in many regions of the world and have been responsible for some of the most destructive natural disasters in world history. Massive rock slope failure occurs with measurable frequency in the mountains of the world; based on twentieth century data massive rock slope failure involving volumes equal or greater than 20 M m3 occur every 2.7 years. Historical data indicates that the frequency in the Alps is one every century. The magnitude and frequency of massive rock slope failure from three datasets were analysed and robust power law scaling relations were obtained which had similar negative values of the exponent. A database of 38 landslide disasters involving 1000 (or more) fatalities in the period A.D. 1000-1999 indicates that 75% of the total fatalities were due to massive rock slope failure. 68% of these fatalities were due to the indirect effects of such failure, viz. bursting of landslide dams and the occurrence of landslidegenerated waves. The Landslide Destructiveness Index (LDI) expresses losses as loss (in this case life loss) per unit volume of a landslide. LDI varies inversely landslide volume. The magnitude and frequency of fatalities resulting from landslide disasters shows power law scaling with the value of the exponent ~ -1. Whilst the frequency of landslide disasters (in the period 1000-1999) with a given number of fatalities is comparable to disasters related to volcanic eruptions, it is an order of magnitude less frequent for comparable magnitude disasters involving destructive earthquakes.


E-mail; [email protected]

53 S.G. Evans et al. (eds.), Landslides from Massive Rock Slope Failure, 53–73. © 2006 Springer. Printed in the Netherlands.

54 1.




Landslides resulting from massive rock slope failure (rockslides, rock avalanches and the failure of volcano slopes (including edifice collapse)) are an important geological hazard in many regions of the world [3, 5, 6, 7, 26, 38, 39, 57, 58, 63] and have been responsible for some of the most destructive natural disasters in world history [12]. Despite this fact no clear view of the real frequency of such landslides, nor the impact of massive rockslope failure on society, has emerged in over 150 years of study. In this context it is appropriate that the International Strategy For Disaster Reduction (ISDR), a successor to the International Decade of Natural Disaster Reduction (IDNDR) of the 1990s, has launched the concept of living with risk, having at its core a concern with the processes involved in the “awareness, assessment and management of disaster risks” [65]. The recent formalization of quantitative frameworks for characterizing the occurrence and impact of natural catastrophes [70] has led to an increased understanding of disaster systems [47] which promises to form the basis of enhanced crisis management and risk assessment in regions exposed to massive rock slope failure hazard. A wide range of primary processes are involved in massive rock slope failure and reflect the complex interaction between the properties of the materials involved, the geological structure of the rock mass, the geomorphic setting of the event (including slope, relief and position in a drainage basin), and the nature of stress changes induced by possible triggering events. Primary processes include mechanisms of initial failure, the progressive fragmentation of the rock mass and its subsequent rapid movement in which debris may be transported many kilometres away from its source. In some cases, run-out distances may exceed 20 km as a result of the transformation of the intial failure, and resulting rock avalanche, into a rapid large magnitude debris flow [e.g., 9, 37]. Massive rock slope failures are often associated with a trigger, a response to a sudden change in ambient geophysical stress (volcanic and/or seismic) (e.g., the 1792 Mayuyama collapse [63], the 1911 Usoy rockslide, Tajikistan [28], the 1970 Huascaran event, Peruvian Andes [53] and/or hydrometeorologic variables (e.g., the 1987 Valtellina event, Italain Alps, [29]. They may also occur in response to human forcing, particularly subsurface or surface mining (or quarrying) [4]. (Figure 1). Secondary processes associated with massive rock slope failure events may themselves result in catastrophic consequences. They include landslide-generated waves and displaced water effects and those processes associated with the formation and failure of landslide dams. Catastrophic secondary effects may be instantaneous or delayed. They extend the impact of a rock slope failure in space beyond the boundaries of the primary landslide debris and in time beyond the timing of the landslide event itself. Further, massive rock slope failure is increasingly recognized as an important geomorphic process in the evolution of mountain landscapes [8, 16, 17, Hovius and Stark, this volume].

55 1.2.


Risk from natural hazards has been recently defined [United Nations 2002, p. 341] as “the probability of harmful consequences, or expected loss (of lives, people injured, property, livelihoods, trauma, economic activity disrupted or environment damaged) resulting from interactions between natural or human induced hazards and vulnerable/capable conditions”. Risk (R) is commonly expressed as the product of hazards (H) and vulnerability (V), where vulnerability may be seen as the ratio of susceptibility to resistance [24].

Figure 1. Aerial oblique photograph, taken in September 1944, of Turtle Mountain, Alberta, Canada and the 1903 Frank rock avalanche (NAPL T31L-213). About 70 people died in the rock avalanche, which involved the initial failure of 30 M m3 of Paleozoic limestone.

In turn, hazard (H) is defined [65] as “a potentially damaging physical event, phenomenon and /or human activity, which may cause the loss of life (and other elements of risk noted above)”. Hazards include latent conditions that may represent future threats. Importantly, hazards can be a) combined (e.g., the creation of floods and landslides by heavy rainfall as in the case of Hurricane Mitch [60, 61] or b) sequential, as for example, in the case of the formation and failure of landslide dams [72] or in the case of landslide-generated waves [29]. In a unit region, each hazard is characterized by its location, intensity (cf. [41]), frequency, and probability of occurrence.

56 Vulnerability (V) is defined by the United Nations (2002, p. 342) as a set of conditions and processes resulting from physical, social, economic, and environmental factors which increase the susceptibility of a community (or infrastructure) to the impact of hazards.” Positive factors that decrease susceptibility (e.g., public education and awareness, warning systems, or built defences), and thus increase resilience may be considered as resistance, which in turn may decrease vulnerability and therefore total risk. Landslide risk has been the subject of much discussion in recent years [13, 15, 45]. It has centred on estimates of landslide hazard, assessment of vulnerability, and risk acceptability criteria, particularly in Hong Kong where landslide risk management is highly developed.


Occurrence in Time; an Estimate of Global Hazard



Massive rock slope failure occurs with measurable frequency in the mountain regions of the world. In the Coast Mountains of British Columbia, for example, rock avalanches, with volumes in excess of 1 M m3, occurred with a frequency of 1 every 3.5 years in the period 1955 to 1999 (Figure 2) [25]. 4 major rock avalanches are known to have occurred in the mountains of northern British Columbia since 1999 (Figure 3) [59]. Globally, recent examples of massive rock slope failure include the earthquake-triggered failures at Tsao-Ling (ca. 120 M m3) and Chiufengershan (ca. 50 M m3) during the 1999 Chi-Chi earthquake, Taiwan [11, 62, 66], the extraordinary Zhamulongba rock avalanche (300 M m3) in Eastern Tibet which dammed the Yigongzangbu River occurred in 2000 [72], and the Paatuut rock avalanche (90 M m3) and displacement wave which occurred on the west coast of Greenland also in the year 2000 [50]. The Yigongzangbu River landslide dam failed 2 months later releasing a massive debris flow which traveled over 500 km downstream reaching the main valley of the Brahmaputra in northern India. Most recently in November 2002, the Denali Fault Earthquake (M=7.9) triggered a large number of massive rock slope failures (Figure 4), mainly in the Alaska Range [21, 35]. 2.2.


Based on methods of seismic hazard assessment, and in particular the GutenbergRichter recurrence law for earthquakes [30] a number of studies have analysed spatial data sets of landslide geometry from discrete triggering events such as heavy rainfall [27, 32] and earthquakes [51] or have analysed spatial data sets [32, 40] of landslide geometry. These analyses together with the analysis of well constrained temporal sequences [10, 19, 20, 42] have shown that despite different triggering mechanisms, landslide styles, tectonic settings, the frequency-size distribution of landslides in a variety of geologic environments are quite similar and follow a robust power law indicating a scale invariant distribution. Size, or landslide magnitude, may be expressed

57 in terms of landslide volume (initial failure volume or debris volume) or landslide area (scar/source area or total area of landslide including debris). Landslide frequency may

Figure 2. Perspective view to the southeast of the 1997 Mount Munday rock avalanche (volume = 3.2 M m3) on Ice Valley Glacier, Waddington Range, southern Coast Mountains. This digital image was prepared from aerial photographs flown on August 20, 1997, and consists of a DEM with an orthophoto drape. Note flow lines in the debris. Elevation of the top of the source area is 3000 m.a.s.l. and the lower tip of the debris is at 2100 m.a.s.l. The length of the rock avalanche path is 4.7 km.


Figure 3. The Zymoetz River rock avalanche (est. volume of initial failure ~ 1.5 M m3) which occurred in the Coast Mountains of British Columbia in 2001. The initial rock avalanche was transformed into a long run-out debris flow as it entrained snow and debris from the valley bottom. The debris flow traveled approximately 3.5 klm to the Zymoetz river where the debris formed a temporary dam [59].

Figure 4. Aerial oblique photograph of two rock avalanches that covered the Black Rapids Glacier, Alaska Range, Alaska. The massive rock slope failures were triggered by the November 2002 Denali Fault Earthquake. (USGS Photograph).

59 be expressed as cumulative frequency, as in the Gutenberg-Richter formulation, or a non-cumulative frequency. With respect to landslide volume (V), the power law has the form; N(V) = aVb


where N(V) is the frequency (or number) of rock avalanches occurring with volume V or greater and a and b are constants. To explore the temporal magnitude and frequency of massive bedrock failure landslides three data sets were analysed; 1) the record of massive rockslides that occurred in the twentieth century (1900-2000), 2) the record of massive historical rock slope failure in the Alps in the period 883-1987 A.D. (extracted from Abele (1974) and Eisbacher and Clague, 1984) and, 3) the record of massive rockslides in the Southern Alps, New Zealand in postglacial time [68, 69]. A lower magnitude threshold volume was selected as being equal to or exceeding a threshold of 20 M m3, a lower volume threshold that withstands erosion censoring for the duration of the longest sample interval, 10,000 years, the approximate length of the post-glacial period (cf. [69]). The Usoi landslide, which dammed Lake Sarez in 1911 [28; Evans et al., this volume] (Figure 5), was the largest (2 x 109 m3) non-volcanic rock slope movement of the twentieth century. For the twentieth century record, a review of the world landslide literature as well as a search of newspaper indexes (e.g., New York Times Index) has been undertaken by the author in recent years. The search and review found that at least 37 rock avalanches over 20 M m3 occurred in the world in the period 1900-2000 (Table 1) suggesting an annual global frequency of massive rock slope failure of at least 0.31, an occurrence rate of 1 every 2.7 years. This compares to regional data from the European Alps for the period 883 to 2002 A.D. [e.g., 1, 22] which shows that 11 rock avalanches with volumes in excess of 20 M m3 occurred during the period 883 to 1987, a frequency of occurrence of 1 every 101 years. The New Zealand record consists of 14 massive rockslides over a post-glacial period of ca. 10,000 years, suggesting a frequency of 1 event in 714 years. The magnitude-frequency relation based on cumulative frequency and with landslide magnitude expressed as volume, for both twentieth century global events and historical Alpine landslides shows a strong power law relationship (Figure 6; Table 2). A similar result was obtained for the data of Whitehouse and Griffiths (1983), a record of rock slope failure in the Southern Alps of New Zealand during the post-glacial period (10,000 years; Figure 6; Table 2). For all three power law fits, similar values of the constant b and high coefficients of determination were obtained (Table 2). Thus, despite widely different geologic environments, different time periods covering three orders of temporal magnitude (10,000 years /millennium/ and century), a variety of triggering mechanisms, and uncertainties of volume estimation, the three records show remarkable similarity sufficient to suggest a universal magnitudefrequency relation for massive bedrock derived landslides above a threshold volume of 20 M m3.


Figure 5. Oblique photograph taken from Space Station Alpha in 2001 of the Usoi Landslide, Tajikistan, which dammed the Murgab River in 1911 to form Lake Sarez. The Usoi Landslide was the largest nonvolcanic subaerial landslide (est. volume 2 B m3) to have occurred in the twentieth century (NASA photograph ISS002-E-7771).

The results indicate that massive bedrock landslides join other styles of landslides such as rockfalls [20, 33, 42], debris flows [49] and rain-induced landslides [14, 27] together with other catastrophic geological processes such as earthquakes [30] and volcanic eruptions [54] in forming orderly empirical magnitude and frequency relations that are of use in assessing the probability of catastrophic events occurring over time and provides a basis of assessing massive rock slope failure hazard [19]. 2.2.


In evaluating these types of magnitude and frequency data, the upper limiting volume of massive rock slope failure is provided by the largest subaerial non-volcanic landslides documented on earth including those at Baga Bogd, Mongolia (5 x 1010m3 ) [52] dated to the late Pleistocene), Green Lake, New Zealand (2.7 x 1010m3; [34], dated to 1213,000 y BP), and Saidmarreh, Iran (ca. 2.0 x 1010m3; [36]; dated to ca. 10,350 y BP by Watson and Wright, 1969)). The Flims rockslide, in the Swiss Alps has a volume of ca. 8 x 109m3 and has been dated to ca. 8,200 y BP (Poschinger et al., this volume). These events provide the maximum credible landslide for the analysis of data sets of landslide geometry and are transitional to even greater tectonic-scale gravitational movement such as the giant tilted blocks and collapses described in [71] on the western Andean slope of northern Chile which moved volumes in excess of 5 x 1010 m3 during the Tertiary.

61 Table 1. Listing of 37 known massive rock slope failures that occurred worldwide between 1900 and 2000 inclusive (sources discussed in text; full database available on request from author). No.

1 2 3 4 5 6



Earthquake Trigger (E)

Usoy Mayunmarca Pufu Ravine, Luquan Yigong Vajont Bairaman

Tajikistan Peru China

E -

Landslide Dam Formed (LD) LD LD LD

China Italy Papua New Guinea Japan China Taiwan China Taiwan Greenland Taiwan Peru Tajikistan China New Zealand Papua New Guinea Taiwan


7 8 9 10 11 12 13 14 15 16 17 18

Hiedayama Diexi Tsao-Ling 4 Zana Tsao-Ling 2 Paatuut Tsao-Ling 1 Huascaran, II Khait Tanggudong Falling Mountain Ok Tedi Mine


Chiu-Fen-ErhShan Hope Sale Mountain Gros Ventre Mount Ontake Valtellina Frank Xintan Schwan Fairweather Tsao-Ling 3 Ancash 2 Allen 4 La Josefina Madison Canyon Steller 1 Ancash 1 Bualtar Glacier Costantino

20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37

Canada China USA Japan Italy Canada China Alaska USA Alaska USA Taiwan Peru Alaska USA Ecuador USA Alaska USA Peru Pakistan Italy


Estimated Massive Rock Slope Failure Volume (x 106 m3)

1911 1974 1965

2000 1000 450


2000 1963 1985

300 292 180

E E E E E -


1911 1933 1999 1943 1942 2000 1941 1970 1949 1967 1929 1989

150 150 125 120 100 90 84 75 75 68 57 50





E E E E E E E E -


1965 1983 1925 1984 1987 1903 1985 1964 1964 1979 1946 1964 1993 1959 1964 1946 1986 1973

47 45 40 36 35 30 30 27 26 26 25 23 23 20 20 20 20 20


Figure 6. Magnitude and frequency relations for massive single event massive rock slope failure (with volumes equal or greater than 20 Mm3) for three data sets; A – World events (n=37), B – European Alps (A.D. 883-1987) (n=11), and the Southern Alps, New Zealand (68, 69) (n= 14).

It is noted that these limiting events have been dated to the Late Pleistocene and the early part of the Holocene suggesting that the frequency of giant landslides is less than the 1 in 1,000 years indicated by the extrapolation of the regression line for the global 20th Century dataset in Figure 6 to limiting magnitudes. Table 2. Summary power law fits for the three data sets in Figure 6. DATA SET A B

World European Alps


Southern (N.Z.)


TIME PERIOD 1900-1999 A.D. 8831987 Last 10,000 yrs


POWER LAW EQUATION 151384.56V-0.770 27194.48V-0.881





63 3.

The Record of Landslide Catastrophes A.D. 1000-1999; An Estimate of Global Risk



Following the approach of Evans (1997) and Guzzetti (2000), an extensive literature search, augmented by archival research, was undertaken to document the circumstances of the greatest landslide disasters of the last millennium, A.D. 1000-1999 and to evaluate the role of massive rock slope failure in these disasters. Here a landslide disaster is defined as a single landslide event which resulted, either directly or indirectly, in the deaths of 1,000 or more people. To an extent this criteria is arbitrary, but it is suggested that this death toll is sufficiently high that it represents a scientific, administrative, and journalistic reporting threshold for the millennium. Data sources included a global data base of landslide disasters related to volcanoes contained in [64], a less than complete but important global survey of the world landslide problem edited by Brabb and Harrod [5], and detailed regional summaries (e.g., 46). For the period 1890 –1999 both the New York Times and the The Times (of London) Indexes were consulted as were other disaster listings published by the Munich Re-Insurance Company, Office of U.S. Foreign Disaster Assistance, and the United Nations Environment Program. 38 landslide disasters that met these criteria, many of them not widely known, form the global list in which a total about 260,500 deaths occurred during the millennium. This indicates an event frequency of about 1 in 26 years. The most recent was the rockslide-debris flow, which occurred on the flanks of the Casita Volcano, Nicaragua, triggered by heavy rains during Hurricane Mitch in 1998, which resulted in over 2,000 deaths (60). The completeness of the global list is subject to three important limitations; (a) despite what is considered to be a comprehensive review and an exhaustive literature search, it is recognized that an unknown number of catastrophes that meet these criteria have not been reported in historical records. (b) a related uncertainty surrounds a number of known landslide disasters in which the number of casualties is not definitively known events. (c) in some disasters, the actual catastrophic process has not been precisely definitely characterized (e.g., failure of natural dams leading to debris flows or floods). 3.2.


22 of the 38 events (58%) were directly or indirectly the result of massive rock slope failure (including volcano flank collapse). These events accounted for approximately 75% of the total number of deaths in the global record of landslide disasters. The five most disastrous single event landslides (in terms of life loss) are listed in Table 2; it is noted that 4 of these resulted from massive rock slope failure. 68% (26) of the events took place in Asia, 21 % (8) in South and Central America, and 11% (4) occurred in Europe. China has suffered at least 10 landslide disasters (26% of the total disaster events) in the millennium resulting in the deaths of

64 about 123,342 people. 47% of the total deaths in the record (31) took place in countries that are currently classified as emerging economies [cf. 2, 43]. Table 3. The major landslide disasters in the period 1000-1999 A.D. DATE









2 3 4 5

1985 1970 1792 1949

Colombia Peru Japan Tadjikistan

Nev del Ruiz Huascaran Ariake Sea Khait

Landslide Dam on Dadu River Lahar Rock avalanche Landslide-generated wave Rock avalanche

DEATHS 100,000 23,000 20,000 15,900 12,000

Analysis of Catastrophic Landslide Events; Identification of High Hazard Scenarios

The percentage of the total casualties that resulted from the different types of landslide processes was analysed in order to obtain an indication of the destructive impact of massive rock slope failure processes at a global scale. Surprisingly, a large percentage (68%) of the deaths due to massive rock slope failure resulted from indirect effects of landslides, i.e., the bursting of landslide dams (57%) and the occurrence of landslide-generated waves (11%). The greatest landslide disaster resulted from the bursting of a rockslide dam on the Dadu River, China, in 1786 in which as many as 100,000 people may have perished downstream (Li and Wang, 1992). Major landslide disasters are closely associated with earthquakes; over 79% of the deaths resulted directly or indirectly from earthquake-triggered massive rock slope failure. 12% of the total deaths were due to the direct or indirect effects of the flank collapse of volcanoes. 25% of the deaths resulted from the direct effects of non-volcanic rock avalanches. Some landslide disasters occurred at the locations of known previous landslide disasters (e.g., the 1985 event at Nevado del Ruiz (Colombia)) or where evidence of the occurrence of previous large-scale landsliding has since been documented (e.g., the events at Mayuyama (Japan) and Huascaran (Peru)) . Whilst some disasters have occurred without warning, in other disasters onset conditions have been observed but their significance was either misinterpreted or not recognised as such. Warnings that could have been given were not issued. Examples are the tragedies at Vajont (1963) and Nevado del Ruiz (1985). In the case of Plane D’Oisans, France, the 1219 A.D. disaster took place as a delayed response to the formation of a landslide dam and impoundment of a lake, 28 years after the damming landslide event. It is noted that the death toll from catastrophic landslides is independent of landslide magnitude; some high casualties have resulted from comparatively modest

65 initial rock slope failure (e.g. Casita Volcano, Nicaragua, 1998). This aspect of destructiveness is explored further.


The Destructiveness of Landslides

In attempt to link the magnitude of landslides and their destructiveness, the Landslide Destructiveness Index (LDI) has been proposed [24]. The LDI is defined in Equation 2 as the ratio of loss per unit volume of the damaging landslide in question. LDIvol = L/V


where L is loss and V is landslide volume measured in cubic metres. Loss can be measured in terms of mortality, monetary cost, damaged dwellings, etc. Where loss is 100 10

0.1 3




















1E-005 1E-006 1E-007 1E-008 1E-009 1E-010 1.0000E+000



Figure 7. Plot of Landslide Destructiveness Index (LDIvol) based on Equation 3 v. Landslide Volume (V). Black dots are data for Canadian landslide events in Evans, 2003. Other data points for some highly destructive landslides from other parts of the world (open squares) as follows (dates in brackets); A - Las Colinas, El Salvador (2001); B – Kelud, Indonesia (1919); C – Huascaran, Peru (1970); D – Nevado del Ruiz, Colombia (1985); E - Mont Granier, France (1248). Diagonal lines show loci of equal number of deaths from 1 to 105 fatalities.

66 measured in deaths, LDI is a function of the population density of the area struck by the landslide. The LDI tends to 1 in the case of small rockfalls. LDI can also be expressed in terms of landslide area [24]. In this case we examine the case of loss in terms of deaths. Evans (2003) showed that a plot of LDI v. landslide volume (Figure 7) shows a negative power - law relationship in which LDI is scaled to landslide volume (V) by Equation 3. LDI = aVb


where V is volume of deposit in m3, where a is a constant and b ~ -1 (Figure 7). The relationship is inverse showing that small magnitude-high frequency landslides are more destructive on a per unit volume basis than larger less frequent events. The landslide destructiveness space is now mapped out in Figure 7. A lower limit of 1 death per event establishes the lower boundary of the plot envelope. A series of parallel lines with a spacing of one log. cycle may be plotted to the right of the lower boundary (Figure 7). These correspond to 10, 102, deaths and so on to a maximum of 105 deaths. It is suggested that 105 is approaching the maximum credible death toll in a single event landslide. This maximum may be approximated by taking the highest recorded population density in the world (ca. 20,000 persons/km2 in a part of Tokyo) and the area of a huge landslide. In this case, we may take the debris area (45 km2) of the 1911 Usoi rockslide, Tajikistan, which is the largest landslide of the twentieth century [28]. If we assume that this landslide debris buries an area that has a population density corresponding to the maximum recorded on earth, a death toll as high as 200,000 could result. The upper boundary can thus be approximated as seen in Figure 7. The lower and upper boundary thus defines a landslide destructiveness space. Figure 7 is thus a plot of destructiveness for landslides and assumes that vulnerability is equal to susceptibility and that no resistance exists in the system. Some of the most destructive landslides in the global historical record plot near the upper boundary of the landslide destructiveness space (Figure 7) suggesting that this first approximation to an upper boundary is quite realistic.


Magnitude and Frequency of Fatalities from Landslides and Comparisons with Other Geological Hazards

The use of historical databases to obtain quantitative relations for the magnitude and frequency of fatalities was pioneered by L.F. Richardson [55] who studied the fatalities resulting from fatal quarrels. Historical databases of damaging landslides may be analysed by plotting the cumulative frequency per year of a disastrous consequence of a landslide (in this case deaths), commonly termed a F/N plot (see references in Cruden and Fell (Editors) 1997). This was first attempted for the Canadian fatal landslide record by Evans (1997). The method was also applied to Italian and other landslide data by Guzzetti (2000) and to landslides and flood fatalities in Italy by Salvati et al. (2003). The line formed by the F/N plot is termed a risk envelope [24]. Its use here is not to define a line of acceptable risk but rather to compare the destructiveness of

67 landslides with that of other major geohazards (earthquakes and volcanoes) during the millennium. Earlier work [44, 48] established scaling laws for the magnitude and frequency of natural disaster fatalities. They established that the distributions are self similar. Because of this, the probability of larger infrequent, catastrophic events can be directly estimated from the rate of occurrence of smaller, more frequent disasters providing, as discussed below, a lower threshold of fatalities is employed. In the global landslide risk envelope (Figure 8) for single-event landslides, the probability of fatalities above the threshold value of 1000 deaths is related to the number of fatalities in an event through a power law (Equation 4) F = aNb


where F is the annual probability of a disaster occurring with N or more deaths; a and b are constants. Evans (1997, 2003) has suggested that b in Equation 4 is ~ -1 indicating that the risk envelope represents one of constant risk. When the data for landslide disasters of the millennium, which resulted in a death toll of 1000, or more is plotted in Figure 8, a similar result is suggested, i.e., b ~ -1. Knopoff and Sornette (1995) found a similar result for historical earthquake death tolls. In Figure 8, the best fit power law, calculated without the 1786 Dadu River disaster data point, is as follows; F = 40.46N-1.00


In (5) the value of b is seen to be -1. This suggests that the number of fatalities in the 1786 Dadu River disaster may be overestimated or that the frequency of such a massive disaster is greater than the sampling period, which in this case is 1,000 years. The strong indications that disaster death toll data sets form robust power laws in which the exponent b is -1 holds some considerable promise for risk assessment at regional and national scales, since, to generate a risk envelope all that is required is the position of the events associated with the greatest losses. Figure 8 suggests that a landslide disaster resulting in 1,000 fatalities occurs with an approximate frequency of 1 in 25 years, a disaster resulting in 10,000 fatalitiesoccurs with a frequency of 1 in 250 years, and a landslide disaster of the scale of the 1786 Dadu River event occurs with a frequency of 1 in 2,500 years. The maximum credible landslide disaster has a possible frequency of 1 in 5,000 years. These statistics assume stationarity in the landslide disaster system, an assumption that requires further analysis. These estimates may be compared with the frequency of similar magnitude disasters involving volcanic eruptions and earthquakes (Figure 9). While the frequency of similar magnitude disasters involving volcanic eruptions is broadly similar to that of landslides, that of similar magnitude disasters involving earthquakes is approximately an order of magnitude more frequent (Figure 9).


Figure 8. Magnitude-frequency plot of deaths in single-event landslide disasters during the period 1000-1999. The power law fit is calculated excluding the 1786 Dadu River disaster (for discussion see text). A maximum credible limit for fatalities in a single-event landslide disaster is assumed to be 200,000.



The accumulating evidence that landslide magnitude-frequency relationships form orderly power law relationships is an important advance for hazard assessment in mountain areas. The developing view that massive rock slope failure is an important geomorphic process in the denudation of mountain terrain creates a promising framework for enhanced hazard assessment. Understanding initial failure mechanisms and the correct interpretation of precursor signs and identification of onset conditions initiates more site specific hazard assessment as does the prediction and modeling of post-failure behaviour. However, landslides resulting from massive bedrock failure are complex multiphase phenomena that may combine elements of fall, flow and sliding, dramatic volume change during movement, together with stop-start behaviour.


Figure 9. Fatal risk envelopes for landslides, volcanoes and earthquakes in the period 1000-2000 A.D. Regression lines with b = -1 have been superimposed on the data plots (Source of data available on request from the author) .

Massive rock slope failure is part of a landscape response system in which convergence of several factors may lead to a true geomorphic catastrophe. This is the case of excessive failure and/or entrainment volume, excessive mobility, water displacement (a landslide-generated wave), or a water release event. Social catastrophes are also seen to a result from the convergence of factors in the landscape system. The most hazardous scenarios are created by interaction between these processes and are not limited to the impact of landslide debris itself during initial movement, but also due to their secondary effects. The majority of landslide catastrophes are associated with other geophysical events (earthquakes, volcanic activity, and heavy rains) and therefore forms part of a spectrum of geological hazards in terms of vulnerability and risk. Power law scaling in the magnitude and frequency of fatalities landslide disasters provides a useful contribution to risk assessment. Finally, many of the most significant landslide catastrophes of the last thousand years are not well documented. This gap in our knowledge seriously restricts the usefulness of the historical record for hazard and risk assessment.

70 References 1. 2. 3. 4. 5. 6. 7. 8.

9. 10. 11.

12. 13. 14. 15. 16. 17. 18. 19.



22. 23. 24.

Abele, G. (1974) Bergstürze in den Alpen, ihre Verbeitung, Morphologie und Folgeerscheinungen; Wissenschafteliche Alpenvereinsheft 25, München, 230 p. Alcantara-Ayala, I. (2002) Geomorphology, natural hazards, vulnerability and prevention of natural disasters in developing countries. Geomorphology 47, 107-124. Alexander, D. (1989). Urban landslides. Progress in Physical Geography 13, 157-191. Benko, B. and Stead, D. (1998) The Frank slide: a reexamination of the failure mechanism; Canadian Geotechnical Journal, 35, 399-311. Brabb, E.E. and Harrod, B.L. (Editors) (1989) Landslides: extent and economic significance. A.A. Balkema, Rotterdam. Brabb, E.E. (1991) The world landslide problem. Episodes 14: 52-61. Blong, R.J. (1996) Volcanic hazards risk assessment, in R. Scarpa and R.I. Tilling (eds.), Monitoring and mitigation of volcano hazards, Springer-Verlag, Berlin, p. 675-698. Burbank, D.W., Leland, J., Fielding, E., Anderson, R.S., Brozovic, N., Reid, M.R., and Duncan, C. (1996) Bedrock incision, rock uplift and threshold hillslopes in the northwestern Himalayas, Nature 379, 505-510. Cassasa, G. and Marangunic, C. (1993) The 1987 Rio Colorado rockslide and debris flow, Central Andes, Chile. Bulletin of the Association of Engineering Geologists 30, 321-330. Chau, K.T., Wong, R.H.C., Liu, J. and Lee, C.F. (2003) Rockfall hazard analysis for Hong Kong based on rockfall inventory. Rock Mechanics and Rock Engineering, 36, 383-408. Chigira, M., Wang, W-N., Furuya, T., and Kamai, T. (2003) Geological causes and geomorphological precursors of the Tsaoling landslide triggered by the 1999 Chi-Chi earthquake, Taiwan. Engineering Geology 68, 259-273. Cornell, J. 1982. The great international disaster book. 3rd. Edition, Charles Scribner, New York, 472 pp. Cruden, D.M. and Fell, R. (Editors) (1997) Landslide risk assessment, A.A. Balkema, Rotterdam, 371 p. Dai, F.C. and Lee, C.F. (2001) Frequency-volume relation and prediction of rainfall-induced landslides, Engineering Geology 59, 253-266. Dai, F.C., Lee, C.F., and Ngai, Y.Y. (2002). Landslide risk assessment and management: an overview, Engineering Geology 64, 65-87. Densmore, A.L. and Hovius, N. (2000) Topographic fingerprints of bedrock landslides, Geology 28, 371-374. Densmore, A.L., Anderson, R.S., McAdoo, B.G. and Ellis, M.A. (1997) Hillslope evolution by bedrock landslides. Science 275, 369-372. Densmore, A.L.,Ellis, M.A., and Anderson, R.S. (1998) Landsliding and the evolution of normalfault-bounded mountains, Journal of Geophysical Research 103, B7, 15203-15219. Dussuage, C., Grasso, J-R., and Helmstetter, A. (2003) Statistical analysis of rockfall volume distributions: implications fro rockfall dynamics. Journal of Geophysical Research 108, B6, 2286, p. 1-11. Dussuage-Peisser, C., Helmstetter, A., Grasso, J-R., Hantz, D., Desvarreux, P., Jeannin, M., and Giraud, A. (2002) Probabilistic approach to rock fall hazard assessment:potential of historical data analysis. Natural Hazards and Earth System Sciences 2, 15-26. Eberhart-Phillips, D., Haeussler, P.J., Freymueller, J.T., Frankel, A.D., Rubin, C.M., Craw, P., Ratchovski, N.A., Anderson, G., Carver, G.A., Crone, A.J., Dawson, T.E., Fletcher, H. Hansen, R., Harp, E.L., Harris, R.A., Hill, D.P., Hreinsdottir, S. , Jibson, R.W., Jones, L.M., Kayen, R., Keefer, D.K., Larsen, C.F., Moran, S.C., Personius, S.F., Plafker, G., Sherrod, B., Sieh, K., Sitar, N., and Wallace, W.K. (2003) The 2002 Denali Fault Earthquake, Alaska: a large magnitude, slip-partitioned event. Science 300, 1113-1118. Eisbacher, G.H. and Clague, J.J. (1984) Destructive mass movements in high ountains:hazard and management, Geological Survey of Canada Paper 84-16, 230 pp. Evans, S.G. (1997) Fatal landslides and landslide risk in Canada, in D.M. Cruden and R. Fell (eds.), Landslide risk assessment, A.A. Balkema, Rotterdam, pp. 185-196. Evans, S.G. (2003) Characterizing landslide risk in Canada. Proceedings 3rd Canadian Conference on Geotechnique and Natural Hazards, Canadian Geotechnical Society, Edmonton Canada, pp. 1934.

71 25.

26. 27. 28. 29.

30. 31. 32. 33. 34.


36. 37.

38. 39. 40. 41. 42.


44. 45. 46. 47. 48.

Evans, S.G. and Clague, J.J. (1999) Rock avalanches on glaciers in the Coast and St. Elias Mountains, British Columbia. In Slope stability and landslides, Proceedings, 13th Annual Vancouver Geotechnical Society Symposium, p. 115-123. Evans, S.G. and DeGraff, J.V. (Editors) (2002) Catastrophic landslides: effects, occurrences and mechanisms. Geological Society of America Reviews in Engineering Geology, v. 15, 412p. Fuji, Y. (1969) Frequency distribution of the magnitude of the landslides caused by heavy rain-fall; Journal of the Seismological Society of Japan, 22, 244-247. Gaziev, E. (1984) Study of the Usoy Landslide in Pamir. Proceedings 4th International Symposium on Landslides, Toronto, 1 511-515. Govi, M., Gullà, G., and Nicoletti, P.G. (2002) Val Pola rock avalanche of July 28, 1987, in Valtellina (Central Italian Alps), in S.G. Evans and J.V. DeGraff (eds.), Catastrophic landslides: effects, occurrence, and mechanisms, Geological Society of America Reviews in Engineering Geology, v. 15, pp. 71-89. Gutenberg, B. and Richter, C.F. (1954) Seismicity of the Earth, Princeton University Press, Princeton, 440 p. Guzzetti, F. (2000) Landslide fatalities and the evaluation of landslide risk in Italy, Engineering Geology, 58, 89-107. Guzzetti et al. 2002. Guzzetti, F., Malamud, B.D., Turcotte, D.L., & Reichenbach, P. Power-law correlations of landslide areas in central Italy, Earth and Planetary Science Letters 195:169-183. Guzzetti, F., Reichenbach, P. & Wieczorek, G.F. (2003) Rockfall hazard and risk assessment in the Yosemite valley, California, USA, Natural Hazards and Earth System Sciences 3, 491-503. Hancox, G.T., and Perrin, N.D. (1994) Green Lake landslide: a very large ancient rock slide in Fiordland, New Zealand, Proceedings, 7th International Congress, Intaernational Association of Engineering Geology 3, 1677-1689. Harp, E.L., Jibson, R.W., Kayen, R.E., Keefer, D.K., Sherrod, B.L., Carver, G.A., Collins, B.D., Moss, R.E.S., and Sitar, N. (2003) Landslides and liquefaction triggered by the M 7.9 Denali Fault earthquake of 3 Novemeber 2002, GSA Today 13, 4-10. Harrison, J.V. and Falcon, N.L. (1938) An ancient landslip at Saidmarreh in southwestern Iran, Journal of Geology 66, 296-309. Hauser, A. (2002) Rock avalanche and resulting debris flow in Estero Parraguirer and Rio Colorado, Region Metropolitana, Chile, in S.G. Evans and J.V. DeGraff (eds.), Catastrophic landslides: effects, occurrence, and mechanisms, Geological Society of America Reviews in Engineering Geology, v. 15, pp. 135-148. Heim, A. (1932) Bergsturz und Menschenleben, Fretz and Wasmuth Verlag, Zurich, 218 p Hewitt, K. (1997) Risk and disasters in mountain lands, in B. Messerli and J.D. Ives (eds.), Mountains of the World; A Global Priority. Parthenon, New York, p. 371-408. Hovius, N., Stark, C.P., and Allen, P.A. (1997) Sediment flux from a mountain belt derived by landslide mapping, Geology 25, 231-234. Hungr, O. (1997) Some methods of landslide intensity mapping, in D.M. Cruden and R. Fell (eds.), Landslide risk assessment, A.A. Balkema, Rotterdam, pp. 215-226. Hungr, O., Evans, S.G., and Hazzard, J. (1999) Magnitude and frequency of rock falls and rock slides along the main transportation corridors of southwestern British Columbia, Canadian Geotechnical Journal 36, 224-238 Jones, D.K.C. (1992) Landslide hazard assessment in the context of development, in G.J.H. McCall, D.J.C. Laming, and S.C. Scott (eds.), Geohazards: Natural and Man-Made, Chapman Hall, London, 117-141. Knopoff, L. and Sornette, D. (1995) Earthquake death tolls, Journal de Physique I 5, 1681-1688. Kong, W.K. (2002) Risk assessment of slopes, Quart. Jo. Eng. Geolo. and Hydrogelogy, 35, 213-222. Li , T.and Wang, S. (1992) Landslide hazards and their mitigation in China, Science Press, Beijing, 84 pp. Newhall, C.G. and Hoblitt, R.P. 2002. Constructing event trees for volcanic crises. Bulletin of Volcanology, 64: 3-20. Nishenko, S.P. and Barton, C.C. (1996). Scaling laws for natural disaster fatalities, in J.B. Rundle, D.L. Turcotte, and W.Klein (eds.) Reduction and predictability of natural disasters, Proceedings Volume XXV in the Santa Fe Institute Studies of Complexity, Addison-Wesley, Reading, MA, p. 3-17.

72 49.



52. 53. 54. 55. 56. 57. 58. 59.



62. 63.

64. 65. 66.


68. 69. 70.

Ohmori, H. & Hirano, M. (1988) Magnitude, frequency and geomorphological significance of rocky mud flows, landcreep and the collapse of steep slopes, Zeitschrift fur Geomorphologie (Suppl. Bd.) 101, 149-164. Pedersen, S.A.S., Larsen, L.M., Dahl-Jensen, T., Jepsen, H.F., Pedersen, G.K., Nielsen, T., Pedersen, A.K., von Platen-Hallermud, F., and Weng, W. (2002) Tsunami-generating rock fall and landslide on the south coast of Nuussuaq, central West Greenland, Geology of Greenland Survey Bulletin, 191, 7383. Pelletier, J.D., Malamud, B.D., Blodgett, T., and Turcotte, D.L. (1997) Scale-invariance of soil moisture variability and its implications for the frequency-size distribution of landslides, Engineering Geology 48, 255-268. Philip, H. and Ritz, J-F. (1999) Gigantic paleolandslide associated with active faulting along the Bogd fault (Gobi-Altay, Mongolia), Geology 27, 211-214. Plafker, G. and Ericksen, G.E. (1978) Nevados Huascaran avalanches, Peru, in B. Voight (ed.), Rockslides and Avalanches, Vol. 1 Natural Phenomenon, Elsevier, pp. 277-314. Pyle, D.M. (1998) Forecasting sizes and repose times of future extreme volcanic events. Geology 26, 367-370. Richardson, L.F. (1948) Variation of the frequency of fatal quarrels with magnitude, Journal of the American Statistical Association 43, 523-546. Salvati, P., Guzzetti, F., Reichenbach, P., Cardinali, M., and Stark, C.P. (2003) Map of sites affected by landslides and floods with human consequences in Italy, CNR-IRPI Publication 2822. Sassa, K. (Editor), (1999) Landslides of the World. Kyoto University Press, 413 pp. Schuster, R.L. and Highland, L.M. (2001) Socioeconomic and environmental impacts of landslides in the western hemisphere, United States Open File Report 01-1276. Schwab, J.W., Geertsema, M. and Evans, S.G. (2003) Catastrophic rock avalanches, west-central British Columbia. Proceedings 3rd Canadian Conference on Geotechnique and Natural Hazards, Canadian Geotechnical Society, Edmonton Canada, pp. 291-298. Scott, K.M. (2000) Precipitation-triggered debris flow at Casita Volcano, Nicaragua: implications for mitigation strategies in volcanic and tectonically active steeplands, in G.F. Wieczorek and N.D. Naeser (eds.), Debris-flow hazards mitigation: mechanics, prediction and assessment, A.A. Balkema, Rotterdam, pp. 3-13 Sheridan, M.F., Bonnard, C., Carreno, R., Siebe, C., Strauch, W., Navarro, M., Calero, J.C., and Trujillo, N.B. (1999) Report on the 30 October 1998 Rock Fall/ Avalanche and breakout flow of Casita Volcano, Nicaragua, triggered by Hurricane Mitch, Landslide News 12, 2-4. Shou, K-H. and Wang, C-F. (2003) Analysis of the Chiufengershan landslide triggered by the 1999 Chi-Chi earthquake in Taiwan, Engineering Geology 68, 237-250. Siebert, L. (2002) Landslides resulting from the structural failure of volcanoes, in S.G. Evans and J.V. DeGraff (eds.), Catastrophic landslides: effects, occurrence, and mechanisms, Geological Society of America Reviews in Engineering Geology, v. 15, pp. 209-235. Simkin, T. and Siebert, L. (1994) Volcanoes of the World; 2nd. Edition, Geoscience Press, Tuscon, Arizona, 349 pp. United Nations (2002) Living with Risk; a global review of disaster reduction initiatives. Geneva, 382 p. Wang, W-N., Chigira, M., and Furuya, T. (2003) Geological and geomorphological precursors of the Chiu-fen-erh-shan landslide triggered by the Chi-chi earthquake in central Taiwan, Engineering Geology 69, 1-13. Watson, R.A. and Wright, H.E. (1969) The Saidmarreh Landslide, Iran, in S.A. Schumm and N.C. Bradley (eds.), United States Contributions to Quaternary Research, Geological Society of America, Special Paper 123, 115-139. Whitehouse, I.E., (1983) Distribution of large rock avalanche deposits in the central Southern Alps, New Zealand. New Zealand Journal of Geology and Geophysics 26, 271-279 Whitehouse, I.E. and Griffiths, G.A. (1983) Frequency and hazard of large rock avalanches in the central Southern Alps, New Zealand, Geology 11, 331-334. Woo, G. (1999) The Mathematics of Natural Catastrophes. Imperial College Press, London, 292 p.

73 71.


Worner, G., Uhlig, D. Kohler, I., and Seyfried, H. (2002) Evolution of the west Andean escarpment at 18 S (N. Chile) during the last 25 Ma: uplift, erosion and collapse through time, Tectonophysics 345, 183-198. Zhou, C.H., Yue, Z.Q., Lee, C.F., Zhu, B.Q., and Wang, Z.H. (2001) Satellite image analysis of a huge landslide at Yi Gong, Tibet, China. Quaterly Journal of Engineering Geology and Hydrogeology 34, 325-332.


ON THE INITIATION OF LARGE ROCKSLIDES: PERSPECTIVES FROM A NEW ANALYSIS OF THE VAIONT MOVEMENT RECORD D.N. PETLEY1 Department of Geography, University of Durham, Science Laboratories, Stockton Road, Durham, DH1 3LE, UK D.J. PETLEY School of Engineering, University of Warwick, Coventry, CV4 7AL, UK

Abstract The 250 million m3 Vaiont landslide of 1963 is the largest failure for which monitoring data has been collected in the period in which the failure was initiated. Despite this, and the availability of good geological and geotechnical datasets, considerable uncertainty remains regards the processes occurring within the landslide prior to and during the failure. In this paper, the monitoring data are re-examined to attempt to understand the triggering mechanisms and controls on movement. by plotting 1/velocity against time for the 1960 and 1962 movement events, and the 1963 final failure, it is demonstrated that the landslide appears to have undergone a change in deformation mechanism. The 1963 data suggest that the final failure was brittle in nature, but the lack of seismicity effectively rules out large-scale cracking in the limestone. The 1960 and 1962 failures suggest a ductile mechanism. This is used to support the previously proposed initiation mechanism in which deformation of the basal clays in the brittle-ductile transition regime allows movement that is characterised by micro-cracking on the small scale but apparent ductility on the large scale to be superseded by brittle failure when the microcracks coalesce.



The Vaiont landslide probably represents the most famous and widely discussed mass movement in history. Since its occurrence in October 1963 a wide range of studies have been conducted on the landslide, focussing on the creep prior to failure [10, 11, 12], the initiation of the failure itself [19], whether the landslide represented a first time failure or a reactivation [18, 9, 16], the high rates of movement of the landslide [5, 23, 20, 24], and the morphology of the landslide deposit [1].


E-mail of corresponding author: [email protected]

77 S.G. Evans et al. (eds.), Landslides from Massive Rock Slope Failure, 77–84. © 2006 Springer. Printed in the Netherlands.

78 Despite this remarkable concentration of effort, no consensus has emerged on many of these issues. In many ways this is remarkable as the landslide site is easy to access, there is good but spatially-limited monitoring data for the period leading up to the failure, and the geology of the site is well-established. In this paper we to concentrate on one of these issues, the mechanisms of the initial failure itself. As mentioned above, there is considerable uncertainty over the manner of the initiation of the failure, with on-going discussion about whether the failure occurred on pre-existing surfaces, over whether the controlling deformation occurred in the basal clays or in the limestones adjacent to the main fold axis, and whether the final failure could have been foreseen. Here, we address these issues by re-examining the movement record of the landslide for the three major phases of acceleration. These records hold important indicators of the processes operating within the landslide. 2.

The Vaiont Landslide

Detailed descriptions of the chronology of the Vaiont landslide are available elsewhere [10, 19], so we will provide only a brief review here. The landslide occurred in the deeply incised valley in the Dolomite mountains of northern Italy, approximately 100 km north of Venice. In 1965, construction was initiated of a tall (265.5 m) concrete arch dam to allow the generation of hydro-electric power by impounding 115 M m3 of water within the Vaiont Valley. The geological structure is a broad, asymmetric syncline formed from middle Jurassic limestone, overlain by upper Jurassic limestone with clay and Cretaceous limestones. As most of the formations dipped into the valley, the potential for dip-slope failures was clear, although such events were considered to be unlikely at Vaiont because the ‘chair’-shaped geological structure was thought to impede movement [10]. Impounding of the reservoir was initiated in February 1960. Nine months later in October, when the depth of the reservoir had reached 650 m asl (170 m water depth), a large crack almost 2 km long progressively opened on the flanks of Mt Toc on the south side of the lake, at 500-600 m above the valley floor. On 4th November 1960, a block of 700,000 m3 of material detached from the front of this mass and slid into the water. It was thus clear that a large detachment was occurring on the flanks of Mt Toc, and that this landslide represented a serious threat to the dam. As a result, a decision was taken to induce slow movement of the mass until it reached a more stable state by raising and lowering the lake level. As this might mean total blockage of the reservoir, a diversion tunnel was bored on the opposite bank to provide access to the turbines. Starting in October 1961 the lake level was slowly raised until, in November 1962, the rate of movement of the landslide increased to 11 mm/day, at which point the reservoir was drawn down and movement ceased. In April 1963 the lake level was again increased. However, in September 1963 the slope velocity increased to 35 mm/ day. Once again the lake level was drawn down in an unsuccessful attempt to reduce the rates of movement. At 22.39 local time on 9 October 1963, a catastrophic slope failure occurred, in which 260 M m3 of rock slid nearly 500 m northwards at up to 30 m/s. The reservoir contained 115 M m3 of water, from which a wave was pushed 245 m above the dam parapet and down onto the villages of Longarone, Pirago, Villanova, Rivalta and Fae in the valley below, with the loss of 2,500 lives.

79 3.

Analysing the Movement Records

Based upon key, pioneering work [2, 17], some attention has been paid to the use of 1/velocity relationships in the forecasting of the time of landslide occurrence [21, 22]. Some success has been noted in the analysis of catastrophic events, although there has been little theoretical or material-based research to justify these observations for landslides. Several authors have noted that the linear trend fits the final failure of the Vaiont landslide (Figure 1) [4, 6, 21]. Detailed analyses [4] have shown that there was some variation from the linearity of the trend, as is evident in Figure 1, which could lead to considerable uncertainty in the forecasting of the final time of the failure, but nonetheless the linearity is clearly illustrated for a period of some 60 days.

1/velocity (days/mm)

0.25 0.2

0.15 0.1

0.05 0 0





Tim e (days)

Figure 1. 1/velocity plot for the 1963 Vaiont failure, showing clear linear trend (data from [4]).

Given that the pattern of movement must be controlled by the stress – strain relationship of the landslide mass, and in particular the behaviour of the shear zone materials, it is interesting to review the mechanical causes of this linearity. In the case of Vaiont, it is now widely accepted that the basal zone of the landslide was located in clay layers within the limestone units [3, 20]. Investigations of the physics of linear time-to-failure analyses for seismic and volcanic phenomena have demonstrated that they can be explained in terms of subcritical crack growth in brittle materials [8]. This has since been proven mathematically [6, 15]. However, where ductile deformation or sliding on an pre-existing plane occurs, linearity in 1/velocity – time space would not be expected. Instead, the plot would have an asymptotic form, as a steady state velocity would be achieved that would be defined by an equilibrium condition with the stress state.

80 A recent study [15] has examined the movement records of a large number of landslides during which large movement events had occurred. The accelerating phase of each landslide was plotted in 1/velocity – time space. It was shown that in all cases in which the deformation occurred through the growth of fractures (i.e. through brittle failure mechanisms) the plot had a linear form, whereas all landslides that occurred in ductile/plastic materials, or which had a pre-existing shear surface, displayed an asymptotic form, although in some cases the rate of movement was very high. This presents an interesting perspective on the Vaiont case. The linearity suggests that a brittle mechanism was controlling the deformation at Vaiont. It is accepted that the main basal shear units were located in clay. Of course intact natural clays often deform through brittle mechanisms in which a shear surface is formed. However, if the landslide was a reactivation [19], it would be expected that the failure occurred on preexisting discrete plane. Thus, the linearity is difficult to explain. Several hypotheses can be proposed to explain this pattern: 1. Deformation occurred in clay layers that contained no pre-existing shear surfaces, suggesting that the landslide was a first time failure and not a reactivation. This contradicts the some evidence [19], and is problematic in terms of the deformation expected as a result of the flexural shearing associated with the synclinal form of the Vaiont valley. 2. Deformation occurred in clay layers that had been recemented by the deposition of calcium carbonate precipitated from ground water. 3. The controlling deformation occurred in the limestone layers adjacent to the fold that forms the ‘chair’-shape of the landslide. In this case, the kinematics of deformation to accommodate this shape would control the deformation. Whilst the deformation in the limestone layers is perhaps the most attractive explanation for this linearity, such brittle deformation in hard, brittle rocks should have a seismic signature. The seismicity of the Vaiont slide was recorded using single instrument located at the dam site. Recent studies [6] have noted that between May 1960 and October 1963, about fifty seismic events were recorded, with epicentres located throughout the unstable slope. Unfortunately, their depths are poorly constrained. However, during the final acceleration only seventeen seismic events were recorded in August and the first three weeks of September, but none were monitored between the last week of September and the collapse on 9 October. If the accelerating movement of the landslide was associated with cracking in the limestone mass then it is likely that seismic events would have been recorded. Thus, it can be concluded that the low level of seismicity recorded immediately prior to failure can probably be explained by the growth of microcracks within the clay, implying that the hypothesis that the failure occurred as a result of the cracking in a recemented clay is most likely to apply [6]. Examination of the 1/velocity – time plots for the accelerating phases in 1960 and 1962 are revealing in this respect (Figures 2 and 3). The 1960 movement shows a classic asymptotic form, with the rate of movement settling to a constant value of 1/velocity. The 1962 movement event also shows some of the signs of the asymptotic trend, but at about day 60 the rate of movement appears to increase once again. Note that the lowest 1/velocity rate is 0.89 days/mm, indicating that velocities were still low.

81 The 1960 plot therefore appears to indicate that the landslide was moving as a result of processes that were either ductile/plastic in origin, or were the result of sliding on pre-existing surfaces. The 1962 movement record appears to show that the landslide was undergoing a transition from this state to another, presumably the brittle state indicated by the 1963 movement record. Since the existence of pre-existing intact shear surfaces is ruled out by the 1963 movement record, it is clear that the shear zone in the Vaiont landslide deformed as a result of mechanisms that were initially ductile, but which transitioned to a brittle state as strain accumulated. This concurs with the earlier conclusions [11, 12], who used high mean effective stress triaxial tests to show that cemented/intact argillaceous materials have a brittle-ductile transition state that is characterised during the initial part of shear deformation by localised micro-cracking that gives an apparently ductile deformation signature on a larger scale. However, as strain is accumulated the material will eventually undergo brittle failure as the microcracks coalesce to form a shear surface [12, 13].


1/velocity (days/mm)

35 30 25 20 15 10 5 0 0




Tim e (days)

Figure 2. 1/velocity plot for the 1960 Vaiont movement event, showing asymptotic trend (data from [19]).

Thus, it appears that the most satisfactory explanation for the movement Vaiont landslide is the development of micro-cracks in the clay layers within the landslide. These micro-cracks allowed the initial creep-type movements to occur. The final failure was triggered by the development of a discrete shear plane as a result of the coalescence of the cracks. This explanation also perhaps provides the means to understanding one other perplexing aspect of the Vaiont failure. During the 1960 movement event the water level in the reservoir was at about 650 m a.s.l. datum. At this time the movement rate exceeded 30 mm/day. However, in late 1961 the lake level was slowly increased back to this level and beyond. Despite this, the rate of movement of the landslide remained slow, and indeed a significant acceleration of the landslide was not seen until mid 1962

82 when the water level had increased by a further 40 m. This delay in reactivation of movement was not seen when the lake level was increased once again in 1963. On this occasion, movement resumed as soon as the lake level returned to the late 1962 point.


1/velocity (days/mm)

70 60 50 40 30 20 10 0 0





Tim e (days) Figure 3. 1/velocity plot for the 1962 movement event, showing an intermediate state between the asymptotic and linear trends (data from [19]).

In simple slope stability terms this delay in the increase in movement rates in 1962 is difficult to explain. Rather complex explanations involving perched water tables in the landslide mass [3] or the precipitation input [10] have been proposed. However, the micro-cracking process may provide a simpler explanation. Between the 1960 movement event and the increase in lake level in late 1961, the reservoir was kept at a low level (approx. 600 m above sea level datum) for eleven months. During this time the water level in the slope would have reduced and there would have been flow of groundwater, perhaps aided by the opening of fractures within the landslide mass, as shown by the earlier seismic events and the large fracture at the landslide head. This ground water would have been saturated with CaCO3 from the limestone strata. It is possible that this water allowed the precipitation of small amounts of CaCO3 at the tips of the micro-cracks in the clay layers. Rapid precipitation of calcite from the through flow of enriched groundwater has been demonstrated experimentally [7]. Here, it was noted that calcite crystals of 50-60 P in average size could be deposited in approximately 50 days from a flow rate of less than 2 ml/h through a channel with a cross-section of 0.3 cm2. Notably, precipitation and crystal growth was noted to occur preferentially on dislocations, perhaps suggesting that the crack tip would be a likely source of precipitation. Thus, experimental evidence strongly suggests that the precipitation of CaCO3 might well have occurred as the flow of ground water proceeded during the long period of low lake level. This CaCO3 effectively strengthened the clay by removing the stress concentration at the crack tips. Of course the CaCO3 itself is relatively strong,

83 preventing failure. Thus, when the reservoir level was increased once again, deformation did not resume as this blunting of the crack tips prevented the growth of the now extensive micro-cracks through the clay. It was only when the normal stress was reduced still further by increased groundwater levels that cracking of the CaCO3 could occur, and the growth of the microcracks could resume. Due to the lower normal stress, it would be expected that crack growth would be more rapid than before, as is shown by the steeper gradient in 1/velocity – time space (compare Figure 2 with Figure 3, but note the different axis scales).



The datasets that were produced during the Vaiont landslide disaster provide researchers with an unmatched level of field data for the understanding of large, catastrophic landslides. Whilst considerable research has already been undertaken into the landslide mechanics, it is clear that we still have much to learn about this failure. However, in this paper we have shown that an analysis of the movement datasets can shed further light on the nature of the processes occurring within the landslide prior to an during the initial failure. The displacement records, analysed in the context of the recent work , suggest that the final failure occurred as a result of essentially brittle mechanisms [6, 15]. Given that the failure occurred in clays that had already been sheared through either tectonic deformation and/or an earlier movement of the mass, and given that the lack of seismicity suggests that the failure did not occur because of the controlling influence of the limestones close to the main fold, the most likely explanation is fracturing of precipitated CaCO3 cement within the clay layers. This occurred through the formation of pervasive micro-cracks that eventually coalesced to form a discrete shear surface. The growth of these micro-cracks meant that the earlier accelerating phases (in 1960 and 1962) had a ductile deformation signature.

Acknowledgements The authors would like to thank Dr Chris Kilburn (University College London), Dr Mark Bulmer (University of Maryland), Prof. Franco Mantovani (University of Ferrara) and Prof. John Hutchinson for the useful discussions and encouragement they have offered. Work on the Vaiont landslide has been partially undertaken under the LANDMOD research project, funded by a grant from the U.S. National Aeronautics and Space Administration Solid Earth and Natural Hazards Program (NAG5-9000).

References 1. 2. 3. 4.

Chowdhury, R.N. (1987) Aspects of the Vajont Slide. Eng. Geol., 24, 533-540. Fukozono, T. (1990), Recent studies on time prediction of slope failure. Landslide News, 4, 9–12. Hendron, A. J., and Patton, F. D. (1987) Vaiont Slide; a geotechnical analysis based on new geologic observations of the failure surface. Eng Geol, 24, 475-491. Hutchinson, J. (2001) Landslide risk – to know, to foresee, to prevent. Geologia Tecnica & Ambientale, 9, 3-24.

84 5. 6. 7. 8. 9. 10. 11. 12. 13. 14.

15. 16. 17. 18. 19. 20. 21. 22. 23. 24.

Jaeger, C. (1968) New considerations on the Vaiont slide; the dynamics of the slide. Felsmechanik. und Ingenieurgeologie Suplementum., 6, 243-247. Kilburn, C.R.J., and Petley, D.N. In press Forecasting giant, catastrophic slope collapse: Lessons from Vajont, northern Italy: Geomorphology. Lee, Y-J, and Morse, J.W. 1999. Calcite precipitation in synthetic veins: implications for the time and fluid volume necessary for vein filling, Chem Geo., 156, 151-170. Main, I.G. (1999) Applicability of time-to-failure analysis to accelerated strain before earthquakes and volcanic eruptions. Geophys J Int, 139, F1-F6. Mencl, V. (1966) Mechanics of landslides with non-circular slip surfaces with special reference to the Vaiont slide. Geotechnique, 16, 329-337. Muller, L (1964). The rock slide in the Vaiont valley. Felsmechanik Ingenieurgeologie, 2:, 148-212. Petley, D.N. (1996) The mechanics and landforms of deep-seated landslides. In: Brookes, S. and Anderson, M.G., Advances in hillslope processes, John Wiley & Sons, United Kingdom, 823-835. Petley, D.N. and Allison, R.J. (1997). The mechanics of deep-seated landslides. Earth Surf Proc Land, 22, 747-758. Petley, D.N. (1999a). Discussion of ‘A laboratory study of the strength of four stiff clays’. Geotechnique 49, 273 – 283. Petley, D.N. (1999b). Failure envelopes of mudrocks at high effective stresses. In A. C. Aplin, A. J. Fleet and J. H. S. Macquaker (eds) Physical Properties of Muds and Mudstones, Special Publication of the Geological Society of London 158, 61-71. Petley, D.N., Bulmer, M.H.K., and Murphy, W. (2002) Patterns of movement in rotational and translational landslides. Geology, 30, 719–722. Romero, U. S. and Molina, R. (1974) Kinematic aspects of Vaiont slide. Proceedings of the Congress of the International Society for Rock Mechanics, 2, 865-870. Saito, M. (1980). Semi-logarithmic representation for forecasting slope failure. Proceedings, International Symposium on Landslides, 1, 321-324. Semenza E. (1965) Sintesi degli studi geologici sulla frana del Vaiont dal 1959 al 1964. Memoir di Museo Tridentino di Scienze Naturali, 16: 1-51. Semenza, E. and Ghirotti, M. (2000). History of the 1963 Vaiont slide: the importance of geological factors. Bulletin of Engineering Geology and the Environment 59, 87–97 Tika, T. E. and Hutchinson, J. N. (1999) Ring shear tests on soil from the Vaiont landslide slip surface. Geotechnique, 49, 59-74 Voight, B. (1988) A relation to describe rate-dependent material failure: Science, 243, p. 200–203. Voight, B. (1989) A method for prediction of volcanic eruptions: Nature, 332, p. 125–130. Voight B. and Faust C. (1982) Frictional heat and strength loss in some rapid landslides. Geotechnique, 32, 43-54. Voight, B. and Faust, C. (1992) Frictional heat and strength loss in some rapid landslides; error correction and affirmation of mechanism for the Vaiont landslide. Geotechnique, 42, 641-643.


E. EBERHARDT Department of Earth and Ocean Sciences University of British Columbia 6330 Stores Road, Vancouver, British Columbia, Canada V6T 1Z 4

Abstract Despite improvements in recognition, prediction and mitigation, rock slope instabilities still exact a heavy social, economic and environmental toll in mountainous regions. This is largely due to the complexity of the processes driving slope failure and our inadequate knowledge of the underlying mechanisms. Ever increasingly, experts are called upon to analyse and predict the stability of a given slope - assessing its risk, potential mode of failure and possible preventive/remedial measures. To do so, it has become essential for the practitioner to be cognisant of the slope analysis tools that are available and to fully understand their strengths and limitations. This paper examines the use of numerical modelling and its role in aiding rock slope stability analyses by providing key insights into potential stability problems, failure mechanisms and mitigative solutions. Several examples will be presented to demonstrate the cause and effect relationships shaped by geological conditions (e.g. rock mass structure, strength degradation through weathering), coupled hydro-mechanical processes, interactions with engineered structures, and aspects of progressive failure as they apply to massive natural rock slopes. 1.

The Need to Understand Rock Slope Instability Mechanisms - Phenomenological vs. Mechanistic Approaches

The study of massive rock slope instability problems, in general, has largely been descriptive and qualitative. Studies that do focus on some quantitative aspect of largescale mass movements are often limited to individual processes or triggering mechanisms (e.g. correlating landslide activity with heavy precipitation events). As such, traditional treatments have primarily been directed towards phenomenological methodologies, both in terms of monitoring/prediction and stability analysis. Phenomenological-based approaches represent the interpretation of large-scale observations/measurements with the purpose of translating them into a useable form for engineering design (e.g. an empirical relationship, a failure criterion, etc.). One of the more common phenomenological techniques employed in rock slope stability 1

E-mail of corresponding author: [email protected]

85 S.G. Evans et al. (eds.), Landslides from Massive Rock Slope Failure, 85–101. © 2006 Springer. Printed in the Netherlands.

86 investigations involves the use of surface displacement measurements recorded over time, which are then extrapolated or analyzed for accelerations in order to predict catastrophic failure. Such measurements are commonly used as a form of early warning system and may cover a variety of scales from that of a crackmeter spanning an open tension crack to a system of geodetic measuring points covering an entire slope. Significant advancements have been made through these techniques [2], yet success is variable. Studies such as those by Kennedy and Niermeyer [11] have successfully extrapolated displacement versus time plots, based on open pit bench movements, to predict the correct date of catastrophic collapse - five weeks in advance. The collapse criterion used in this prediction was 6 m of displacement (Figure 1a), a value based on engineering judgement, displacement data, rock mass quality and lessons learned from two previous minor failures [18]. In this sense, the adopted methodology was purely phenomenological and relied heavily on experience gained over time. In contrast, similar predictions of catastrophic failure based on accelerations seen in displacement versus time measurements have led to incorrect forecasts, for example at Kilchenstock in the Swiss Alps - twice. Following the first incorrect prediction, and subsequent evacuation of an endangered town, Heim [9] reported that “lack of experience” was the reason for being misled. Howe ver, total displacements had reached magnitudes greater than 4 m at the time (Figure 1b), and it can be argued that Albert Heim’s experience was unparalleled. Less than two years later, the same methodology led to a second prediction of catastrophic failure followed by a second evacuation of the town, and again, a second unexpected cessation in slope movements [12].



Figure 1. Extrapolation of displacement vs. time records to predict catastrophic failure: a) successfully applied at the Chuquicamata mine in the Chilean Andes (after [11]); b) unsuccessfully applied at Kilchenstock in the Swiss Alps (after [12]).

The case of Kilchenstock may have occurred 70 years ago, but today, the same phenomenological-based “displacement versus time” analyses are still being employed and often with the same variability in terms of success. In summer 2001, near Innertkirchen in the Swiss Alps (Berner Oberland), accelerations observed through 3 surface displacement measurements of a 250,000 m unstable rock mass were interpreted as reaching a critical level (Figure 2a) thereby requiring the closure of the

87 only highway leading through the mountain pass (the Grimsel Pass). When the movements subsequently decreased, but the threat of failure still remained serious, a surface stream with a flow rate of 9000 litres/minute was diverted down the tension crack for a period of 18 days to accelerate/trigger failure [8]. Again, slope accelerations increased but failure did not occur (Figure 2b). A decision was then taken to use 19 3 tonnes of explosives to bring down approximately 150,000 m of unstable rock in what would be the largest controlled blast in Switzerland. Following the blast, official figures were not released but inspection of before and after photos suggested that maybe only 50% of the unstable mass was brought down. A second blasting campaign was subsequently carried out in Fall 2002. a).


Figure 2. Monitoring records near Innertkirchen showing: a) displacements vs. time and critical acceleration leading to forecasting of imminent failure; b) velocity vs. time during attempts to trigger failure (after [8]).

88 Such examples should not be interpreted as a criticism of the people involved in making these decisions. Instead, they demonstrate the inherent difficulty in relying on phenomenological-based analyses in which the underlying mechanics and mechanisms of the problem are largely ignored. The prevalent use of surface displacement measurements obviously addresses certain economic restraints in terms of what may be feasible for on-site monitoring of a given rock slope. Yet it must also be asserted that only so much can be inferred at surface when the problem itself transpires at depth. Thus lies the problem in relying solely on phenomenological methodologies.


Numerical Analysis of Rock Slope Instability Mechanisms

Phenomenological type approaches have also been widely adopted in rock slope analysis in the form of limit equilibrium analysis. These treatments traditionally involve the assumption or delineation of a two-dimensional slide surface along which a backanalysis is performed to determine the force/moment equilibrium conditions existing on the surface at failure. In other words, the analysis focuses on the back-analysis of stability along a fully developed failure plane, without considering how the failure plane evolved. All limit equilibrium techniques share a common approach based on a comparison of resisting forces/moments mobilized and the disturbing forces/moments (Figure 3), with methods varying in the assumptions adopted in order to achieve a determinate solution. Although limit equilibrium techniques are highly relevant where translational or rotational movements occur along discrete failure surfaces, analyses are limited to simple slope geometries and basic loading conditions, and provide little insight into rock slope failure mechanisms.

Figure 3. Limit equilibrium analysis (Morgenstern-Price solution) showing from left to right: the calculated critical slip surface and Factor of Safety; a free body diagram of the resulting forces acting on a slice; and the corresponding force polygon indicating that force equilibrium is satisfied in this solution.

89 Many rock slope stability problems on the other hand, involve complexities relating to geometry/topography, material anisotropy, non-linear behaviour, in situ stresses and the presence of coupled processes (e.g. hydro-mechanical). When examining the factors contributing towards massive rock slope instability, it becomes evident that a large number of physical processes are involved. Geological, geomorphological, hydrological, geomechanical and numerous other physical-based processes, all interact and contribute in one form or another to the destabilization of the slope. In effect, these processes may be viewed within a system as the result of a continuous series of events linked through cause and effect relationships (Figure 4).

Figure 4. Cause and effect relationship between rock mass processes and slope instability problems.

To properly examine these interactions, and their sensitivity to all relevant triggering mechanisms, numerical methods must be employed. Advances in computing power and the availability of relatively inexpensive commercial numerical modelling codes means that the simulation of potential rock slope failure mechanisms could, and in many cases should, form a standard component of a rock slope investigation [7,16]. 2.1.


Numerical methods provide approximate solutions to problems, which would otherwise not be solvable using conventional techniques (e.g. limit equilibrium). Numerical methods used for rock slope stability analysis may be divided into three approaches: continuum, discontinuum and hybrid modelling. Continuum modelling is best suited for the analysis of slopes that are comprised of massive, intact rock, weak rocks, and soillike or heavily jointed rock masses. Discontinuum modelling is appropriate for slopes controlled by discontinuity behaviour (Figure 5). Hybrid codes involve the coupling of these two techniques (i.e. continuum and discontinuum) to maximize their key advantages. Coggan et al. [5] and Stead et al. [16] summarize the advantages and limitations inherent in these different methodologies.


Figure 5. Overview of continuum and discontinuum numerical methods.

The technique chosen depends on both the site conditions and the potential mode of failure, with careful consideration being given to the varying strengths, weaknesses and limitations inherent in each methodology (finite-element, distinct-element, etc.). The quality of the input data made available for the analysis may also vary such that the objectives of the numerical analysis may focus on prediction when high quality instrumentation data is present (i.e. forward modelling of a potential instability or “Class A” prediction), or in cases where the data is limited, as providing a means to establish and understand the dominant mechanisms that may affect the behaviour of the system (Figure 6). As demonstrated by Coggan et al. [5], at all times good modelling practice must be exercised.

Figure 6. Spectrum of modelling situations and corresponding applicability (after [10]).

91 2.2.


Focussing on the use of numerical modelling as a means to establish and understand the dominant mechanisms contributing towards rock slope destabilization, modelling allows for the testing of several hypotheses with respect to the measured or expected behaviour of the rock slope in question. The following sub-sections provide several examples as to how numerical modelling can be used to better understand rock slope instability mechanisms due to complex geological conditions, coupled hydro-mechanical processes and interactions with engineered structures. 2.2.1. Geological Factors Stead and Eberhardt [15] examined the complex modes of failure commonly observed in thinly bedded, weak sedimentary rock slopes dipping parallel to topography. The mechanisms responsible for failure often involve some form of shear or tensile failure near the toe of the slope followed by planar sliding of the driving slab (Figure 7). The major failure mechanisms recognized include bilinear slab, ploughing, buckling, steppath and planar. In terms of geological controls, the influence of cross-cutting joints is key in generating the potential for the different complex modes of failure (Figure 8a).

Figure 7. Examples of complex failures in thinly bedded rock slopes (note that scales are approximate).

92 The effect of jointing in these problems, however, is often not fully appreciated given the difficulty in mapping the orientations of cross-cutting discontinuities normal to the slope face. In some cases, these joints can be extremely tight and go undetected due to the large driving forces acting across the joint surfaces (Figure 8b).


Figure 8. a) Cross-cutting joint with dip favouring development of a ploughing mode failure; b) slope face where cross-cutting joints are difficult to map due to large driving forces acting parallel to slope surface.

Distinct-element models generated to investigate these problems in thinly bedded slopes (Figure 9), show three different complex failure mechanisms – bilinear, threehinge buckling and ploughing, as the orientation of the dominant joint set varies with respect to the dip of the bedding planes. The complexity of these different failure modes involves both slip along the controlling discontinuities and plastic yielding of the intact rock material. Furthermore, their sensitivity to initial in situ stress states and different triggering mechanisms including pore pressures and seismic loading (e.g. [6]), can only be effectively treated using numerical techniques like the distinct-element method.


Figure 9. Distinct-element modelling of complex modes of slope failure in thinly bedded weak rock: a) bilinear; b) buckling; and c) ploughing modes of failure.

In such cases, the continuum behaviour of the intact rock contributes to the development of the instability in the discontinuous rock mass. Similar requirements are necessitated in cases involving weak rocks where the instability mechanism is related to progressive strength degradation, for example through physical and chemical weathering processes. Recent experiences from the 1999 Rufi slide [13], located in the sub-alpine Molasse of northern Switzerland and involving a series of interbedded

94 conglomerates and marls, suggest that failure did not occur along bedding plane contacts as had been assumed for similar slides in the region (e.g. the 1806 Rossberg slide at Goldau), but instead developed within the marl units and was controlled by the degree of weathering (Figure 10).

Figure 10. Bedding sequence of interbedded conglomerates and marls along side margin of the Rufi slide and conglomerate block overlying weakened weathered marl through which the slide surface passed (after [13]).

Numerical models were later used to better understand these processes through the utilization of discontinuum techniques (i.e. distinct-element method) that allow for the consideration of strength degradation of the marls (in this case the key process or “cause”) coupled together with increasing pore pressures between weak layers in the marls that served as the Rufi slide trigger (Figure 11). These findings provided key insights into the failure mechanisms responsible for the slide, especially with respect to the geological factors controlling the location and development of the shear surface. Ultimately, such information and understanding is essential to properly investigate and mitigate other problematic slopes in the region.

Figure 11. Slide surface of the 1999 Rufi slide and analysis showing numerical modelling of initiation of sliding due to yielding of underlying weathered layers.

95 2.2.2. Coupled Hydro-Mechanical Factors The coupling of complex interactions, e.g. hydro-mechanical behaviour, allows for numerical models to provide deeper insights into slope instability mechanisms but also, in cases where remedial measures are planned, to better understand how the slope will respond to different mitigative measures before and after implementation. This was demonstrated by Bonzanigo et al. [4] for a deep creeping landslide at Campo Vallemaggia in the southern Swiss Alps. The Campo Vallemaggia landslide is located in the crystalline penninic nappes of southern Switzerland. Two small villages, Campo Vallemaggia and Cimalmotto, are located on the toe of the slide mass where surface displacements have been geodetically measured for over 100 years. Surface and borehole investigations of the unstable mass revealed a 300 m deep, complex structure incorporating strongly weathered and broken metamorphic rocks divided into blocks along primary fault zones (Figure 12). Boreholes revealed the presence of deep artesian water pressures and data provided through in situ monitoring suggested that hydromechanical factors were controlling the unstable mass [3]. To stabilize the slope, a drainage tunnel was constructed but its effectiveness was questioned (despite surface measurements showing that the slope velocities had ceased). This was largely due to the smaller than expected flow rates into the drainage gallery, and competing arguments that erosion at the slope’s toe was the primary destabilizing factor.

Figure 12. Photo and block model of the Campo Vallemaggia deep creeping landslide (photo courtesy of Dr. Luca Bonzanigo). Note the toe scarp in the left-hand photo where erosional processes were counter-argued as being the primary factor driving the mass movements.

Coupled hydro-mechanical distinct-element modelling was therefore used by Bonzanigo et al. [4] to better understand the underlying mechanisms driving the unstable mass and those contributing to the apparently successful mitigative solution taken (Figure 13). Piezometer data was used to constrain the models, which in turn showed that high pore pressures were a major factor controlling slope movements and that very little drainage was required in the models (approximately 5 l/s) to significantly reduce pore pressures and to stabilize the slope (Figure 14).


250 m

location of drainage adit

Figure 13. Model geometry used for the coupled hydro-mechanical distinct-element analysis of the Campo Vallemaggia deep creeping landslide (after [4]).

50 m

50 m

drainage adit opened Figure 14. Coupled hydro-mechanical model showing relative slope velocities before and after introduction of drainage adit (after [4]).

The decision to use deep drainage to stabilize the slope at Campo Vallemaggia was primarily phenomenological in nature, i.e. correlations were made between pore pressure increases and slope accelerations [4]. In fact, drainage was recommended more than 100 years ago by Albert Heim, who interestingly met with the same political resistance due to counterarguments that erosional undercutting at the slide toe was the destabilizing factor [9]. Heim went on to lament that in the 30 years following his initial recommendation, the only mitigative measures taken at Campo Vallemaggia involved the planting of trees near the front of the scarp terrace to control erosion. He further added, “Understanding and courage are lacking in addressing the problem at the right places, and politics impedes progress. Lovely magnificent Campo, soon you will be merely a bygone dream ”! It took the better part of 100 years to overcome the scepticism and politics opposed to deep drainage, but current measurements suggest that Campo Vallemaggia may not be lost after all. Numerical modelling was able to reproduce the driving mechanism involved and explain the success of deep drainage where observations of tunnel drainage flow rates would otherwise suggest it failed. Accordingly, numerical

97 modelling revealed that since the groundwater system primarily involved fracture permeability, the resulting storativity would be low and therefore significant decreases in pore pressure could be achieved though the drainage of only small volumes of water. 2.2.3. Interaction with Engineered Structures The effects of underground openings on massive slope failures, and the use of numerical modelling to understand the failure processes involved, were demonstrated for the Frank slide by Benko and Stead [16]. Examining this cause and effect relationship from the opposite perspective, adverse tunnelling conditions may arise when active landslide processes are encountered at the excavation site of a tunnel or underground works. Such slope hazards act to increase the complexity of the geological conditions, induce tunnel instabilities, cause costly delays, interfere with construction logistics and shorten the life span of the final structure. Consideration must also be given to auxiliary structures, for example the influence of slope hazards on access roads, secondary adits and shafts. The Nathpa-Jhakri Hydroelectric Project (NJHP) in the Himalayas of northern India involves one of the largest ongoing civil works in India. The project centres around the construction of a 60.5 m high concrete gravity dam and a 27.3 km long headrace tunnel. The headrace tunnel was driven through quartzites, gneissic schists, quartz-mica-schists and amphibolites of low to medium grade metamorphism [17]. Foliation strikes parallel to the tunnel axis over long distances and dips towards the river valley (Figure 15).


Figure 15. Foliation dipping parallel to slope at surface and at depth at the Nathpa-Jhakri Hydroelectric Project (photos courtesy of Dr. Kurosch Thuro).

During tunnel construction, severe tunnel stability problems were encountered along certain sections of the headrace tunnel. Observations indicated that these sections passed through slope bodies in which signs of active sliding processes were present, e.g. visible open cracks within the body of the slope above the tunnel alignment [17]. In sections where the foliation was aligned parallel to the tunnel axis, a typical, almost symmetrical deformation pattern including shear deformation along the walls was observed (Figure 16). To better understand the mechanisms contributing to the tunnel stability problems, distinct-element modelling techniques were employed. Of particular concern was the role that foliation and fractures parallel to foliation play when acted upon by small-scale deformations derived from creeping slope mass movements.

98 Results from these models showed that if the slope was in a stable state and creep deformations were minimal, significant tunnel stability problems would not be expected. However, in the case of an unstable creeping slope, subsequent slope movements would begin to induce buckling failures in the tunnel roof once down-slope displacements exceeded 0.5 m. Figure 17 shows the progressive stages of tunnel failure modelled after down-slope displacements of 0.05, 0.5 and 1 m, respectively. These results correlated well with in situ observations and helped to establish the driving mechanism contributing to the destruction of the tunnel lining [17].

Figure 16. Cross section of the mountain valley and headrace tunnel showing typical deformation patterns encountered due to foliation orientation and deep-seated slope movements (after [17]).

Figure 17. Progressive stages of tunnel failure for down-slope displacements of 0.05, 0.5 and 1 m (after [17]).

99 4.

Conclusions and Future Developments

These examples demonstrate that when properly applied and constrained, numerical modelling can significantly assist in the design process by providing key insights into potential massive rock slope stability problems and failure mechanisms. Future work in this direction is focussing on the application of hybrid finite-/distinct-element analyses incorporating adaptive remeshing to model the progressive development of rock slope failure surfaces through brittle fracture processes ([18] this volume). It should be emphasized, though, that elements of field mapping, instrumentation monitoring, in situ measurements and laboratory testing must also be included if the overall state-of-the-art is to move towards the total assessment or prediction of the rock slope stability state. Currently, an integrated network of displacement, pore pressure and microseismic monitoring devices has been installed at a site in southern Switzerland (Randa), to help in quantifying the spatial and temporal evolution of such processes and to constrain complex numerical models [19]. Through such combined approaches, numerical modelling will be extended towards linking shear plane initiation and slope mass degradation to eventual catastrophic failure (Figure 18). Yet it must always be emphasized that numerical modelling is only a tool and not a substitute for critical thinking and judgement. Still, through the proper use of numerical modelling, key steps can be taken to build upon and transcend the phenomenological methodologies that dominate massive rock slope stability investigations today, thereby improving the visualization and comprehension of the coupled processes and complex mechanisms driving such instabilities.

Figure 18. Finite-/discrete-element analysis showing several stages of progressive brittle failure (after [8]).

100 Acknowledgements The author wishes to thank his collaborators in the various projects presented in this paper, especially Doug Stead, Luca Bonzanigo, Kurosch Thuro, Simon Loew, John Coggan, Mario Luginbuehl and Heike Willenberg.

References 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. 16.


Benko, B. and Stead, D. (1998) The Frank slide: A reexamination of the failure mechanism, Canadian Geotechnical Journal 35 (2), 299-311. Bhandari, R.K. (1988) Special lecture: Some practical lessons in the investigation and field monitoring of landslides, in C. Bonnard (ed.), Proc. of the Fifth Int. Symp. on Landslides, Lausanne, A.A. Balkema, Rotterdam, pp. 1435-1457. Bonzanigo, L., Eberhardt, E. and Loew, S. (2000) Measured response to a drainage adit in a deep creeping slide mass, in E. Bromhead et al (eds.), Landslides in Research, Theory and Practice: Proc. of the 8th Int. Symp. on Landslides, Cardiff, Thomas Telford, London, pp. 151-156. Bonzanigo, L., Eberhardt, E. and Loew, S. (2001) Hydromechanical factors controlling the creeping Campo Vallemaggia landslide, in M. Kühne et al. (eds.), United Engineering Foundation Int. Conf. on Landslides - Causes, Impacts and Countermeasures, Davos, Verlag Glückauf GmbH, Essen, pp. 13-22. Coggan, J.S., Stead, D. and Eyre, J.M. (1998) Evaluation of techniques for quarry slope stability assessment, Transactions of the Institution of Mining and Metallurgy, Section B 107, 139-147. Eberhardt, E. and Stead, D. (1998) Mechanisms of slope instability in thinly bedded surface mine slopes, in D.P. Moore and O. Hungr (eds.), Proc., 8th Int. Congress Int. Assoc. for Engineering Geology and the Environment, Vancouver, A.A. Balkema, Rotterdam, pp. 3011-3018. Eberhardt, E., Stead, D., Coggan, J. and Willenberg, H. (2002) An integrated numerical analysis approach to the Randa rockslide, in J. Rybár et al. (eds.), Proc. of the 1st European Conf. on Landslides, Prague, A.A. Balkema, Lisse, pp. 355-362. Gruner, U. (2001) Felssturzgefahr Chapf - Üssri Urweid (Gemeinde Innertkirchen): Angaben zur Geologie und zu den Bewegungen, Kellerhaus + Haefeli AG, Bern, Report 3661KB4336. Heim, A. (1932) Bergsturz und Menschenleben, Fretz and Wasmuth Verlag, Zurich. Itasca (2000) UDEC - Universal Distinct Element Code (version 3.1). Itasca Consulting Group, Inc., Minneapolis. Kennedy, B.A. and Niermeyer, K.E. (1970) Slope monitoring systems used in the prediction of a major slope failure at the Chuquicamata Mine, Chile, in P.W.J. Van Rensburg (ed.), Planning Open Pit Mines, Proceedings, Johannesburg, A.A. Balkema, Cape Town, pp. 215-225. Löw, S. (1997) Wie sicher sind geologische Prognosen?, Bulletin für Angewandte Geologie 2 (2), 8397. Luginbuehl, M., Eberhardt, E. and Thuro, K. (2002) Primary sliding mechanisms in dipping interbedded conglomerates and marls, in J. Rybár et al. (eds.), Proc. of the First European Conf. on Landslides, Prague, A.A. Balkema, Lisse, pp. 387-392. Stead, D. and Coggan, J. (2003) Numerical modelling of rock slopes using a total slope failure approach, in NATO Advanced Research Workshop - Massive Rock Slope Failure: New Models for Hazard Assessment, this volume. Stead, D. and Eberhardt, E. (1997) Developments in the analysis of footwall slopes in surface coal mining, Engineering Geology 46(1), 41-61. Stead, D., Eberhardt, E., Coggan, J. and Benko, B. (2001) Advanced numerical techniques in rock slope stability analysis - Applications and limitations, in M. Kühne et al. (eds.), United Engineering Foundation Int. Conf. on Landslides - Causes, Impacts and Countermeasures, Davos, Verlag Glückauf GmbH, Essen, pp. 615-624. Thuro, K., Eberhardt, E. and Gasparini, M. (2001) Adverse tunneling conditions arising from slope instabilities - A case history, in M. Kühne et al. (eds.), United Engineering Foundation Int. Conf. on Landslides - Causes, Impacts and Countermeasures, Davos, Verlag Glückauf GmbH, Essen, pp. 97107.

101 18. Voight, B. and Kennedy, B.A. (1979) Slope failure of 1967-1969, Chuquicamata Mine, Chile, in B. Voight (ed.), Rockslides and Avalanches, 2: Engineering Sites, Elsevier Scientific Publishing Company, Amsterdam, pp. 595-632. 19. Willenberg, H., Spillmann, T., Eberhardt, E., Evans, K., Loew, S. and Maurer, H. (2002) Multidisciplinary monitoring of progressive failure processes in brittle rock slopes - Concepts and system design, in J. Rybár et al. (eds.), Proc. of the 1st European Conf. on Landslides, Prague, A.A. Balkema, Lisse, pp. 477-483.


Abstract Pre-collapse creep is a widespread component of complex slope movements. Its study requires a preliminary kinematic classification, based on the relations between style and rate of gravitational dislocations and geological structure of rock massifs. Complexity and variability of slope deformations determine the shortcoming of the traditional deterministic approach. Thus, probabilistic analysis considering different scenarios and 'event trees' seems to be on the mainstream of rock slopes stability assessment. General aspects of geomechanical modelling and risk analysis of slope processes are discussed. 1.


Review on rock slopes stability demonstrates that creep on high slopes is typical and much more common than it was known before. Since term 'creep' is often applied to describe different slope and tectonic processes, it should be noted, that hereafter this definition corresponds to the pre-collapse gravitational deformation of rock slopes in the form of slow long-term movement with three-staged (primary, secondary and tertiary) rheological mechanism [8, 11]. It was found that the pre-collapse creep has close relationship with other gravitational processes like rock slumping, tension rupturing, etc. Such complex gravitational dislocations with large contribution of creep have nonuniform mechanism with various kinematics. This article bases on the results of detailed geological explorations at different dam sites. Particular attention is given to the main factors and geological conditions of creep formation and to the methods of their investigations, including modelling, kinematic and risk analyses. These slope deformations are of a special note because they occur under gravitational stress less than the rock mass strength, but this significant feature typical of creep often is not taken into account properly. It should be also mentioned that only the pre-collapse deformations are considered in this report without analysis of the collapse itself and of subsequent debris motion. 1

E-mail: [email protected]

103 S.G. Evans et al. (eds.), Landslides from Massive Rock Slope Failure, 103–110. © 2006 Springer. Printed in the Netherlands.

104 2.

General Features of Pre-Collapse Creep

Geological evolution of complex pre-collapse deformation and rock slope failure is closely connected with the erosion, valleys incision and corresponding increase of slope height and steepness, that cause slow growth of stress at the lower parts of slopes due to gravity loading. In turn, it results in different slope processes, among which two groups can be selected: relatively rapid (in a geological sense) with domination of sliding along initial surfaces of weakness (fractures, faults, schistosity and bedding planes), and relatively slow with large contribution of creep, which are considered hereafter. Secondary (stationary) creep starts when stress becomes equal to resistance to viscous flow. Conceptual equality of all types of the pre-collapse deformations (compression strain, shear, bending, and squeezing out) should be emphasised. Review of case studies of rock slopes' stability (Table 1) demonstrates wide distribution of creep as an ordinary element of complex slope processes, associated with reduction of rock mass strength under long-term gravitational loading. Similar active dislocations were discovered in the Italian Central Alps [5]. Close relationship of creep with other gravitational processes requires their joint investigation as complex systems named hereafter 'the gravitational (or active) dislocations'. It is important to note that complex creep mechanism in real rock masses characterised by complicated unmatched geological structure, discretion, non-uniformity, anisotropy, and 3D stresses, differs significantly from its simplified geomechanical interpretation based on standard laboratory tests. Meanwhile, problems of scaling for deformational and especially rheological properties and of the transition from laboratory tests to real slope processes are not yet solved. Gravitational dislocations usually occur on high (more than 100 m) steep (40q on the average) slopes, typical of neotectonic uplifts. Structural factors, especially bedding and fracture orientation relative to slope inclination strongly affect slope processes. For example, weak zones undercutting slope and dipping steeper than the angle of friction are unfavourable for creep since sliding of unstable block should occur earlier. Lithology is important for creep formation too. It occurs more often in schist and phyllites than in other types of sedimentary, metamorphic or igneous rocks. As a rule, it develops in highly weathered rocks [26] or in schistose and jointed ones with relatively low Q and RMR values. At some sites creep correlates with rate and amount of rainfall or with ground water level [1]. Raining often plays a role of triggering factor. Important manmade activity provoking creep includes slopes undercutting [3, 26] and rising of water table due to reservoir infilling [1, 10, 15, 16]. Diversity of factors and of the engineering-geological conditions of creep formation cause substantial scatters of volume and rate of the gravitational dislocations. Data on creep duration ere even more uncertain. It ranges from several months to tens of years, and is expected that can last even longer [20, 21]. The most important and noticeable feature of gravitational dislocations – their non-uniform mechanism with complicated variability in space and time. Variance in space is manifested by different creep-to-shear ratio along fractures in various parts of the same active zone. Rock mass deformation may be realised not by 'flow' only but also by bending with gravitational folding, tension, and compression and squeezing out. Creep usually concentrates at the lower portions of active zones, whilst in the upper and peripheral parts shears may dominate.


Table 1. Main types and sub-types of gravitational dislocations. Brief descriptions

Typical sub-types

I. Slow creeping of the whole active zone with concurrent shearing of some blocks along local sliding planes. Possibility of further transition of creep into rapid sliding is not excluded. This process sometimes is called 'deep-seated creep'.

1. Predominant non-stop creeping. Sliding along steep fractures is possible only in the upper part of active zone. At its' foot some squeezing out can be observed. 2. Movement as a result of coupling of creep and sliding of large blocks along fractures dipping towards the slope. Sometimes with rotation that can be caused by successive formation of several dislocations. 1. Buckling, i.e. slow sliding of some layers along bedding fractures parallel to slope surface and concurrent formation of recumbent gravitational folds with subsequent rapid rock slide along new sliding surface. 2. Slow movement of upper and middle parts of active zone along steep dipping faults with concurrent squeezing out in the slope toe. The latter is characterised by complex mechanism, including creep and small blocks motion along joints. 3. Slow displacement of active zone along some bedding planes being restrained by flattening of sliding surface due to folding that acts as an obstacle. 1. At first stage slow gravitational folding occurs with axial plane parallel to slope along which a shear plane for rapid sliding of the second stage is gradually formed. 2. At first stage there is very slow bending of some rock blocks, cut off by tension fractures. At second stage they overturn and fall down.

II. Two-stage dislocations. In the beginning, almost the entire active zone slides slowly along steep weak planes nearly parallel to the slope surface, being restrained by obstacle at the slope foot. In consequence visco-plastic deformation of obstacle occurs leading to the gravitational folding or squeezing out. At the second stage new single sliding surface forms causing rapid sliding of the whole active zone. III. Toppling, i.e. twostage dislocation in nearsurface zone of stress relaxation with the tension and creep deformations. At the first stage gravitational bending of layers and displacement or overturning of some blocks, cut off by extension fractures take place. 3. Stretching of gently dipping layers At the second stage rapid of hard rock into separate blocks sliding or rockfall occur. overlaying soft rocks. Slow movement and overturning of these blocks are connected with creep and deformation in basal soft rocks.

Examples Gepatsch dam site (Austria) Glunzerberg dam site (Austria) Wusheh dam site (Taiwan) Campo Vallemajia rockslide (Italy) Wahleach dam site (Canada) Tablachaka dam site (Peru)

References 16 20 26 6 14 1

Lettopalena rockslide (Italy) Lavini di Marco rockslide (Italy) Lujacia dam site (China)


Tehri dam site (India)


Vaiont rockslide (Italy)

15, 10

Clapiere rockslide (France) Revelstock dam site (Canada) Nevis Bluf dam site (New Zealand) Ginvaly dam site (Georgia) Chirkey dam site (Russia) Toctogul dam site (Russia) Boguchany dam site (Russia) Glennis Creek dam site (Ausralia) Boguchany dam site (Russia) Ust-Ilim dam site (Russia) Rufi Rockslide (Switzerland) Luchina rockslide (Yugoslavia)

2, 13 25

7 21

3 22 22 22 22 25 22 22 6 6

Multistage mechanism of slope dislocations expresses its variability in time. Usually there are two phases of slope destruction: the first one, with prevalence of long slow creep and the second one, characterised by rapid slide or rockfall. Mechanism of the first phase may be creep only [20] or combination of creep and sliding along weak planes [1, 14]. In general creep changes step by step from elasto-plastic to stationary visco-plastic (first and second stages), while its third stage (acceleration) occurs rarely.

106 Variability of dislocation mechanisms in time and space may be caused by the presence of an 'obstacle' at the foot of active zone restraining its motion. Such 'braking effect' generally is caused by absence of the pre-existing sliding planes undercutting rocky slope [21, 24] or by downslope flattening of rock beds as at the Vaiont slide [10, 15]. Stress caused by the weight of the upper part of an active zone can be insufficient to cut an obstacle but enough for slow creep, gravitational buckling or squeezing out with blocks displacements [24]. The pre-collapse deformation of an obstacle restrains displacement regardless of movement mechanisms along the entire boundary of an active zone. Variability may be also caused by lithological non-uniformity as in the case of very slow stress intensification in the rock mass composed of alternating rigid quartzite phyllites and weaker argillaceous phyllites. In such a case concurrent plastic yielding or transition to viscous flow through the whole active zone is hardly expected and traditional geomechanical modelling is insufficient. 3.

Kinematic Analysis and Classification of Gravitational Dislocations with Pre-Collapse Creep

Modelling of complex slope processes requires comprehensive understanding of causative mechanisms with due regard to time and space variability. It is important to investigate all possible kinematic types of particular dislocation mechanism. Review of case histories with large creep contribution allows general classification of such mechanisms based on the relation between gravitational dislocations and geological structure of rock massifs (see table 1). Dislocations of the first type are formed mainly in the absence of large structural anisotropy. Steep dipping of beds or strongly developed set of long fractures subparallel to high slope is favourable to development of buckling. Dislocations of the third type (toppling) are formed in the same longitudinal direction of beds as in buckling, but dipping into the slope or subvertical. Yet these multiple-factor and diverse mechanisms complicate selection of kinematic subtypes. Diversity of inclinometric data from single active zone demonstrates complexity of kinematic modelling (Figure 1) and emphasises significance of this method for kinematic analysis. Simultaneous presence of the creep and of the brittle failure leads to terminological difficulties. In particular, at a third subtype of toppling it is uncertain, if motion of an active zone as a whole occur before stress reach the ultimate value or not. In some cases there are transitional forms in absence of reliable criteria for kinematic subtypes attribution. Sometimes there is no single opinion on the origin of complex gravitational dislocations, like, for example, in cases of the sackungen [4]. It complicates kinematic systematisation, does not afford to standardise geomechanical modelling, and, thus, calls for further clarifying of kinematic types and subtypes. 4.

Key Problems of the Gravitational Dislocations Modelling and Slope Stability Assessment

Analysis of the engineering-geological data on gravitational dislocations indicates poor quality and lack of details in their geological conditions. Practical importance of geological investigations in comparison with geomechanical modelling is, generally, under-

107 estimated. Usually there is a shortage of boreholes, adits and of rock sampling (especially to measure rheological characteristics), inadequate geological maps and profiles, lack of geophysical and precise topographical monitoring, simplified description of joints and faults. It was noted by Hoek [12], Panet [17], Varga & Gorbushina [22]. 1





Figure 1. Kinematic analysis of dislocation mechanism based on inclinometer measurements: 1 – block displacement along sliding surface; 2 block and creep movements along the sliding surface; 3 – creep with different rates of movement; 4 – creep with linear decrease of rate in vertical direction; 5 – creep with non-linear decrease of its rate in vertical direction.

The reliable geomechanical analysis and modelling of slope processes require more comprehensive field explorations than it is carried out usually. Principal attention must be drawn to inclinometrical measurements allowing to delineate the limit of the active zone and to determine the rate of movement and changes of displacement mechanism with depth. Long-term monitoring of rainfall rates and of technogenous impacts should be performed. The results of these investigations have to be summarised in 3D engineering-geological model. Such model should be an obligatory stage of geomechanical analysis without which it seems that no further progress in geomechanical modelling of creep is practically possible. Such detailed investigations are carried out, for example, for slope stability assessment during dam design and construction. A current practice of geomechanical modelling with respect to ordinary investigations of slope processes is also unsatisfactory. Forecast of long-term slope stability based on the limit equilibrium method is especially poor for dislocations with pre-collapse deformation, characterised by complex kinematics. The shortcomings of traditional deterministic approach call for its replacement by risk analysis. However such replacement is accepted by many geologists and engineers with difficulty. Specific difficulties in modelling of the complex gravitational dislocations with creep usually include not only absence of exact boundaries of all elements with quaziuniform kinematics, but also insufficient elaboration of the appropriate modelling procedures of pre-collapse deformation and ignorance of scale effect in rock mass properties and slope processes. It is very difficult, if not impossible, to make exact kinematic analysis and geomechanical modelling of lithologically non-uniform rock mass where different rocks are characterised by different yield stresses. There is no clarity with the prediction of secondary creep transition into tertiary one. In practice secondary stationary creep can develop further in three different ways: 1) continue for a long time, 2) stop gradually and 3) be accelerated until the catastrophic rockslide or rock fall will occur. Since speeding is the main indication of such transition, traditionally speed-time plot is used to predict failure on the basis of different empirical criteria. However, use of this method is usually complicated by irregular form of graphs due to frequent temporary impacts such as heavy rainfalls or slope undercutting, and it is often impossible to determine the actual origin of this irregularity – is it a beginning of tertiary creep or a local deviation only.

108 In ordinary practice modelling of gravitational dislocations is carried out by the limit equilibrium methods (LEM). Geomechanical parameters obtained by the back analysis of monitoring data are used as the input data. Finite element (FEM) and distinct element (DEM) methods are usually applied for analysis of failure mechanism and for assessment of slope stability. They are also used for parametric analysis and for selection of stabilisation measures. It is important to examine boundary conditions and assumptions in relation to rock mass geomechanical characteristics, natural stresses, isotropy, and uniformity. It is believed that geomechanical models must be more co-ordinated with particular dislocation kinematics and with the degree of its complexity. New achievements in numerical modelling with due regard to gravitational dislocations with large creep contribution are described in the present book [6, 22] and new and perspective guidelines for creep analysis are suggested. 5.

Problems of Risk Analysis in Relation to Gravitational Dislocations

Traditional risk analysis is based mainly on the deterministic limit equilibrium methods, which are utilised to estimate factor of safety as a criterion of slope stability. However, these simplified and over-standardised methods of geomechanical modelling generally do not correspond to complex and kinematically variable slope processes. The shortcomings of traditional deterministic approach call for its replacement by risk analysis. Application of probabilistic analysis in engineering geology and geomechanics began 10-15 years ago, but was very much delayed in the field of slope processes. Limitations of traditional deterministic approach can be observed in slope stability assessment and in related hazard analysis. Deterministic estimation of limit equilibrium has to provide absolute stability to particular slope, but such 100% reliability is impossible in practice since it requires unrealistic complete engineering-geological investigations and elimination of human factors, which is impossible too. It is important to note that LEM reliability usually depends on the understanding of failure mechanism and on knowledge of geomechanical parameters and exact position of potential sliding surfaces, which are not always estimated sufficiently. Especially large scatter and low accuracy are typical of such parameters as cohesion of fractured rock massive or heterogeneity of natural stresses. There are cases of LEM complete unfitness when rock mass strength is reduced because of local and not-uniform development of different rheological mechanisms. It can be mentioned as well that risk probability analysis considers scatter of geological parameters and the errors of their measurement better. All the above indicates progressiveness of transition from the traditional conception of the absolute reliability to the conception of the acceptable risk [18]. Considerable shortcoming of traditional approach appears to be applied for simplified models of slope failure, aimed towards analysis of critical impact of single triggering factor. Meanwhile some actual engineering-geological conditions are characterised by development of multi-staged scenarios of potential hazard processes. In these cases it is necessary to use more appropriate models that take in mind interactions of different mechanisms and various factors, including secondary processes like, for example, the formation and failure of landslide dams. Considering probability as a degree of confidence in prognostic stability assessment, it is necessary to take in mind three important groups of data:

109 1) Probability and intensity of possible hazard; 2) Probability of the process development in a particular way at each bifurcation points, as can be seen in figure 2; 3) Recurrence of such factors as earthquakes or floods of definite intensity. With the help of the scenarios of process development it is possible to construct and analyse more complicated models like long-term sequence with several 'adaptational' periods, divided by bifurcation points. In each adaptational period process develops under the influence of some factors and the system is characterised by the quasi-uniform mechanism of slope process with limited change of parameters. The latter permits a deterministic approach in analysing and forecasting of slope process. However, when some parameters mount to critical values, the system reaches bifurcation threshold and further development of the hazardous process can turn sharp right up to catastrophic failure. Analysis of such complex process is possible only on the base of the 'event tree' probabilistic method (Figure 2). It provides a single criterion of economical, social and ecological risk that was impossible on the base of deterministic approach [18]. Trigger- Rockslide ing factor formation

Rockslide rea- Dam ches the river formation

Dam failure YES P5






1 - P5

YES Maximum rainfall P1

P2 NO 1- P2

NO 1 - P3

1 - P4

Da- Probability of mage each scenario

Risk of each scenario



P1P2P3P4P5 C1


P1P2P3P4(1-P5) P1P2P3P4(1-P5) C2



P1P2P3(1-P4) C3



P1P2(1-P3) C4



P1(1-P2) C5

Figure 2. Schematic 'event tree' with risk assessment for possible scenarios determined by maximum rainfall.



Creep is a typical component or phase of complex gravitational dislocations characterised by time and space variability of their mechanisms and by diversity of their kinematic types. Preliminary kinematic systematisation of such dislocations based on their kinematic and engineering-geological conditions is presented. Simplified and overstandardised methods of geomechanic modelling often used at present, generally do not correspond to the very complicated and not uniform gravitational dislocations, mainly due to shortage of engineering geological data. These methods do not allow either to obtain reliable assessment of slope stability or to forecast a transition from the secondary to tertiary creep. There is no doubt that rock slope analysis require approach, based on profound understanding of geological conditions and close collaboration of geologists and geotechnical engineers. Shortcomings of traditional deterministic approach require utilisation of probabilistic methods and analysis of scenarios for slope stability assessment. Further development of numerical modelling of slope dislocation mecha-

110 nisms with creep contribution is needed especially in the application of hybrid FEM/DEM analysis. References 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13.

14. 15. 16. 17. 18. 19.

20. 21. 22. 23. 24. 25. 26. 27.

Arnao, B.M., Garga, V.K., Wright, R.S., and Perez, I.Y. (1984) The Tablachaca slide 5, Peru, and its stabilization, in: IV Inter. Symposium on Landslides. Toronto, 1, 597-604. Blank, A., Durville, J.-L., Gandin, B., and Pincent, B. (1987) Methodes de surveillance d’un glissement de terrain de tres grande ampleur: le Clapiere, Alpes Maritimes, France, Bull. of IAEG, ʋ 35, 37-46. Brown, J.R., Henger, M.H., and Goodman, R.E. (1980) Finite Element Study of the Nevis Bluff Rock Slope Failure, New Zealand, Rock Mechanics, 12, 231-245. Clague, J.J., Evans, S. G. (2003) Gravitational origin of antislope scarps in British Columbia, (this volume). Crosta, G.B., Imposimato, S., and Roddeman, D.G. (2003) Continuum numerical modelling of flow-like landslides, (this volume). Eberhardt, E. (2003) From cause to effect – using numerical modelling to understand rock slope instability mechanisms, (this volume). Evans, R.S. (1981) An Analysis of secondary toppling rock failures – the stress redistribution method. Quart. Journ. of Eng. Geol., 14, 77-86. Gary, M., McAfee Jr., R., and Wolf, C.L (Ed.) (1972) Glossary of Geology. American Geol. Institute Genevois, R., and Tecca, P.R. (2002) Failure mechanisms and runout behaviour of three rockslidedebris avalanches in north-eastern Italian Alps, (this volume). Ghirotti, M. (2002) Eduardo Semenza: the importance of geological and geomorphological factors for the identification of the ancient Vaiont landslide, (this volume). Gladston, J., Fell, R., and Mostyn, G. (1999) Analysis and prediction of the pre-collapse deformation of cut rock slopes (Australia), in: Proc. of the IX ICRM, 1, Balkema, Rotterdam, 95-100. Hoek, E. (1993) When is a design in rock engineering acceptable? in: Proc. VII ICRM Congress, 3, Balkema, Rotterdam, 1485-1497. Merrien-Soukatchoff, V. (2002) Which models are available to understand a large landslide such as La Clapiere (Southern Alps, France). in: NATO ARW Massive rock slope failure: new models for hazard assessment. Abstract volume, 98-102. Moore, D.P., Ripley, B.D., and Groves, K.L. (1992) Evaluation of Mountain Slope Movements at Wahleach. in: Geotechnique and Natural Hazards, BiTech Publishers, Vancouver, 99-107. Müller, L. (1968) New considerations on the Vaiont slide, Rock Mech. Eng. Geol. 6, 1-91. Neuhauser, E., and Schober, W. (1970) Das Kriechen der Talhange und Elastische Hebungn beim Speicher Gepatsch. in: Proc. of the II ICRM, Beograd, 447-458. Panet, M. (1993) General Report: Rock Slopes, in: Proc. of the VII ICRM, Congress 3 Balkema, Rotterdam, 1577-1586. Risk Assessment as an Aid to Dam Safety Management (1999) ICOLD Bull. 9. Scarascia Mugnozza, G., Fasani, G.B., Esposito, C., Martino, S., Saroli, M., Di Luzio, E., and Evans, S.G. (2003) Rock avalanche and mountain slope deformation in a convex dip-slope; the case of the Maiella massif, Central Italy, (this volume). Scheidegger, A.E. (1976) Physical Aspects of Natural Catastrophes. Elsevier, Amsterdam. Shitian, W., Zhuoyuan, Z. (1984) Viscous-flow deformation of rock masses in near-surface conditions and the related rock slope failure, in: Proc. IV Intern. Symposium on Landslides, Toronto 1, 585-589. Stead, D., and Coggan, J. (2003) Numerical modelling of rock slopes using a total slope failure approach (this volume) Varga, A., and Gorbushina, V. (1998) Geostructural classification of unstable rock masses, in: Proc. of VIII IAEG Congress 8, Balkema, Rotterdam, 1477-1483. Varga, A.A. (2000) Engineering-geological analysis of gravitational creep in rock masses. Geoecology, Enggineering geology, Hydrogeology, Geocryology, ʋ 4, 291-306 (in Russian). Woodword, R.C. (1988) The investigation of toppling slope failures in welded ash flow tuff at Glennies Creek Dam, New South Wales. Quart. Journ. of Eng. Geol. 21, 289-298. Yu, C.W., and Chern, J.C. (1994) Creep modelling of an endangered slope adjacent to Wusheh Dam, Taiwan, in: Proc. of VII IAEG Congress 5, Balkema, Rotterdam, 3789-3795. Zaruba, Q., and Mencl, V. (1979) Engineering Geology. Academia Prague.


Abstract We examine the choices model-users have to face when applying models to better understand slope failure and prefailure movements of large rock slopes and illustrate how modelling can help understand rock slope movements. At first, the questions that the models can help to answer are listed. The different types of models available to understand the hydromechanical behaviour of rock slope movements are then specified. Different types of models have been applied to the La Clapière case. This movement is described and three different modelling analyses of this landslide are outlined. Each model allows an understanding of a part of the problem raised by rock slope movements, but none of them can actually represent totally the phenomena involved in the movement.



The use of models has now become commonplace in order to understand large unstable rock slopes. Many different types of models are indeed available. Among them however, which ones are truly helpful for comprehending major movements? What inputs are needed? What are they actually contributing to the understanding of the movement? What are the benefits from the outputs? The purpose of this paper is to summarize some of the knowledge existing about models, as well as the current thinking on modelling techniques, and then report the work undertaken on the La Clapière landslide. A model is a schematic, simplified and relatively abstract representation of an object or a process, which makes it possible to substitute a simpler system for the more complex natural system in order to describe, to explain or to forecast it [22]. This paper focuses on those models that explain the mechanical aspects of the initiation of rock slope movements. Others models may consider, for example, the explanation of the 1

E-mail of corresponding author; [email protected]

111 S.G. Evans et al. (eds.), Landslides from Massive Rock Slope Failure, 111–127. © 2006 Springer. Printed in the Netherlands.

112 hydrogeochemical processes in such landslides. The models studied herein are those that describe failure and pre-failure; it does not address models that reproduce post-failure motion and runout. Modelling needs to reconcile three fundamental features: 1. the particular query to be answered, 2. available information, and 3. available computation methods. It is important for the query to be clearly formulated and broken down into basic questions so as to easily determine what information is required to provide appropriate answers and what kind of model may be applicable. It is possible to assemble the typical questions associated with rock slope movements (where? when? how? and why?). The "when" question is slightly different in that a response generally proves impossible, except perhaps by means of estimating the probability of recurrence. The "where" question can be answered once the movement has been identified or once some predisposition factor has been recognized [8]; however, "where" potentially remains very difficult to localize with precision, unless the inherent complexity is handled by a stochastic approach [28]. This method will not allow locating the "where" precisely, but will make it possible to focus on the probability of event occurrence in a given area. The "how" and "why" questions are well suited to the field of mechanical modelling.


Different Types of Models

Models can be regarded as representations of objects and processes (essentially mechanical ones), which control the behaviour of slopes. We propose dividing the models that are helpful in the understanding of the mechanical behaviour of rock slopes into three categories, as depicted in Figure 1. The geomodels enable us to represent the geometrical and geological aspects of the system and object studied. For example a model-user has to decide how many different materials comprise the slope modelled and where are their boundaries. This choice is not always obvious especially when there is a continuous transition between different petrographic types of rocks. To represent the mechanical processes, the variables of the system (essentially displacement, stresses, pore pressure, water input and output) can be connected by: statistical relationships: these models are identified as statistical or correlative models, or mechanical and hydraulic principles. These are Mechanical Models. The Mechanical models can be themselves subdivided on those, which introduce these principles: into a physical (or analogical or small scale) model, in the form of equations, which will be solved either analytically or numerically (a theoretical solution of a mechanical problem must satisfy equilibrium equations, the behaviour of the constitutive material and boundary conditions).


Statistical models ("correlative")

Mechanical Models

Geomodel - Geometrical aspects - Geological aspects

Physical or analogical (small scale)



Failure and prefailure Stability Continuous Media

Postfailure motion and run out

Stress-Strain Discontinuous Media

Figure 1. Different types of models.

Statistical models serve to investigate and reveal the set of input parameters influencing displacements and stability. Mechanical models allow reproducing and testing mechanical behaviour. Physical modelling raises questions about similitude, the scale effect and the choice of equivalent materials. Numerical modelling addresses computing methods-related problems (finite elements, finite differences, boundary element, discrete element), as well as the choice of material behaviour: elastic (classical elasticity or Cosserat elasticity), elastoplastic (along with constitutive law choice) or elastofragile. Physical, numerical and analytical models all lead to making choices about the explicitness, or lack thereof, in the representation of discontinuities, should they exist. Model users must also make choices regarding the type of problem to be resolved: Are displacements analysed or the ground behaviour sought, or are the estimation of the distance to stability required? Are the displacements small enough to consider the stress and strain infinitesimal, or are large displacements required for consideration? In the latter case, the computer code must include the second-order development of the equilibrium equations. What equations are being solved: static equilibrium equations, for most of the finite element code, or the full dynamic motion equation (solved using the finite difference code FLAC or in UDEC). For the dynamic equation scenario, the equilibrium equation is formulated and solved iteratively until the contact laws and boundary conditions have been satisfied. A Lagrangian resolution method is applied, whereas for static equilibrium equations, a Eulerian resolution method is typically employed. Further details about the various computation methods can be found in Jing et al. [19]. In general, all types of models are useful in better understanding rock mass behaviour, yet they can only answer the part of the problem raised by rock slope movements. We now focus on the La Clapière site, where examples of the various modelling techniques have been implemented or are envisaged.

114 3.

The La Clapière Landslide

La Clapière is a well-known landslide located in southeastern France, within the Alps, some 80 Km north of the city of Nice. It is positioned on the left bank of the Tinée Valley and involves a slope, which culminates at 3,000 meters, over the stretch between 1,100 and 1,800 meters in altitude (Figure 2). The rock-mass is composed of gneissic rocks with a natural foliation (Figure 3). A large rupture was identified at the beginning of the 20th century. By 1936, the deformation at the top of the landslide was already quite visible. During the 1970's, movements became more continuous. The site has been monitored since 1982 ([9, 10]).


Ténibres valley

Rabuons valley

Dailoutre valley

Scree of Rabuons

Scree of Belloire

To Saint-Etiennede-Tinée (300 m)


Tinée Valley (altitude: 1100 m)

To Nice (90 km)

Figure 2. The landslide in 2001 photographed from across the Tinée Valley.

Despite the fact that this landslide has been receiving considerable attention from the engineering geology community in France, very little modelling work has actually been conducted on the site. A large amount of chronological information is available concerning this landslide including topographic, rain-fall, snow level, and hydrological measurement. Qualitative analyses reveal a correlation between these data [7], and an empirical quantitative relationship has been proposed. A more quantitative analysis, such as the Box and Jenkins methodology [2, 3, 21, 27], may be used to perform a time-series study, which could emphasize the temporal lag between rain-fall (or snow) impulse and displacement. This would yield information on the nature of the response to these inputs. It could also emphasize the portion of the movement explained by the variation in precipitation.


Figure 3. Cross-section along the Rabuons Valley (Figure 2), according to Laumonier & Gunzburger [15]. The top of the mountain is at 3,000 meters.

To complete the description of the state of knowledge at the site, table 1 presents laboratory mechanical characterizations carried out on the material to date. Table I. Characterization of the La Clapière rock matrix (according to Serratrice [18]). Loading orientation, compared to foliation Normal A Parallel // Oblique Anelle facies Compressive strength: Rc (MPa) 58,7 46,9 23,8 Tensile strength: Rt (MPa) 8,7 4,4 5,0 Iglière facies Compressive strength: Rc (MPa) 110,6 82,5 74,2 Tensile strength: Rt (MPa) 11,7 6,8

At the scale of 10- to 100-m2 outcrops, ground mapping and Schmidt hammer tests (at approximately 20 sites on the slope) have allowed generating a geotechnical classification of rock masses according to Bieniawski's Rock Mass Rating (RMR) system [1]. Using empirical the relationships (1), (2) and (3), these evaluations have been related to the mechanical strength and deformability of the rock mass ([14], [16], [17] and [24]):

Ceq (kPa) | 5 RMR

(Bieniawski, 1979, in [1])



) eq (°) | 0,5 RMR+8,3 r 7,2 E eq (GPa) | 10



(Trunck & Hönish, 1989, in [1])


(Serafim & Pereira, 1983, in [1])


The RMR values on the La Clapière slope varied from 37 to 59.25. The RMR value of the point nearest to the slip area (RMR = 42) was chosen as typical for later computations. The properties estimated from this value are: E = 8.9 MPa, C = 240 kPa, M = 32°, Vc = 4.8 MPa.


Numerical Mechanical Modelling

Different types of numerical mechanical modelling have been undertaken using finite element codes or the distinct element method. Various approaches were used in order to better understand the La Clapière landslide; this effort has led to use different modelling tools for testing assumptions, scenarios and in particular the role of: geometrical evolution, anisotropy, water, discontinuities. The presence of discontinuities implies the use of a discontinuous code, and the distinct element code UDEC, which models groundmass as an assembly of blocks separated by deformable joints was chosen. Three modelling tests will now be examined closer: the first considers the toppling scenario; the second is the evaluation of a particular view of the geometrical evolution of the slope (effect of glacier retreat, progressive failure from bottom to top); and the third focuses on the impact of anisotropy. The first two modelling tests have been carried out with the UDEC code, while the third used the CESAR-LCPC finite element code. 4.1.


The existence of foliation, the direction of which changes in the vicinity of the topographic surface, has often evoked a gravitational toppling scenario for explaining the origin of the La Clapière landslide [9, 10]. Let us note that the French term use to describe the La Clapière phenomenon is "fauchage" (or "mowing"), hence the French terminology is more closely related to the description of the instability itself and introduces the notion of a failure surface, whereas the term "toppling" refers to the movement style rather than to the presence or absence of a failure surface. Simple conceptual two-dimensional models, which represent the rock mass as an assembly of "rock columns" separated by discontinuities, were developed [26] in order to better understand the development of a flexural toppling phenomenon. In these models, vertical displacements are not allowed at the base and horizontal displacements


Figure 4. The toppling scenario of La Clapière, according to Follacci [9].

are not allowed on the right and left sides until a height of h (Figure 5). The H-h height is free and corresponds to an unloaded zone due to excavation by glaciers and the river. A parametric study was carried out while varying: the topographic slope D, the dip of discontinuities E, the mechanical characteristics of rock material (E, Q, Rt, C, M), joint characteristics (Kn, Ks, Rt, C, M), joint spacing (e), and the relative height of the unloaded zone H  h H .

Figure 5. Geometrical characteristics of the models attempting to reproduce flexural toppling (the symbol represents rollers: the movement is allowed along the roller and not authorized perpendicular to the rollers).

The analysis showed that the conditions required to initialize an inflection of the "columns" are the existence of an unloaded zone (Hzh), small spacing of the joints (less than 4 meters for a 100 m * 100 m model), and low joint characteristic values (C and

118 Rt < 10 kPa for the studied geometry). But modelling poorly reproduced significant toppling of the rock columns (Figure 6). The addition of vertical discontinuities, such as those observed on the La Clapière site or "median" discontinuities (as suggested by the plasticity zones in the model), makes it possible to simulate a more significant toppling: This new set-up, however, corresponds to a different geometrical representation and therefore a different conceptual model, which is no longer pure flexural toppling (Figure 7). To elaborate what was learned from these simple models, a model of the entire La Clapière slope (Figure 8), complete with a reconstituted topography (before the sliding began), was developed. In this model, foliation was represented as regularly-spaced discontinuities. The result affirmed the difficulty of simulating toppling using this geometry and did not allow reproducing of significant toppling. This result could have been anticipated from previous simple models because of the absence of a significant unloaded zone in the geometry of the entire slope.

Figure 6. Modelling of the toppling of continuous columns (geometry of Figure 5).


Figure 7. Modelling of the toppling in the presence of a "median" fracture (geometry of Figure 5).

Figure 8. Model of La Clapière slope.

The fact of not being able to numerically reproduce the tilting of the foliation observed at the site does not exclude the gravitational origin for the mechanism: This feature can be modeled either by adding geometric elements (discontinuities other than foliation) or by proposing other conceptual, hence numerical, models (e.g. a model that takes large deformations inside the continuous blocks into account). The tilting of the foliation could be of tectonic origin prior to glacial withdrawal. The questions raised by

120 this modelling approach necessitate a return to the site and lead to more site investigations and the conclusion that the tilting of the foliation is probably more likely to be of tectonic origin [15]. Given that modelling leads to reconsidering the toppling scenario, subsequent work has thus concentrated more heavily on acquisition of information on the ice age (glacial limits and evolution), determination of mechanical properties by means of the Equivalent Continuous Medium (ECM), analysis of the influence of water and anisotropy. 4.2.


Analysis of photographs of the slope, from 1938, reveals a regressive evolution of the landslide from bottom to top. Different areas can be identified in the slope, as seen in Figure 9. In 1938 signs of slope movement was essentially visible in area 1. Progressively observable indicators appear in areas 2, 3 and finally 4 (Figure 9). Modelling has been used to analyze the evolution of the slope from an initial state with estimated pre-ice age topography, through the current state [17]. The model dimension is shown in Figure 10. The topography is similar to that considered in Figure 8. The bottom of the valley was considered, as a plane of symmetry and a boundary with no horizontal displacement was positioned at this plane (Figure 10). This is in agreement with what was learned from the previous model. In order to take into account the large number of fractures of various scales and origins, which cannot be considered explicitly, the rock mass was homogenized (Figure 11) by applying the RMR methodology [16] to yield an Equivalent Continuous Medium (ECM). Four steps have been modeled (Figure 12) as follow: Step 1: The "consolidation step" ('ante-Würm' period): the model is consolidated under gravity to obtain an initial mechanical state. The valley has a pre-ice topography and mechanical stress-strain equilibrium is reached. For this calculation, material behaviour has been artificially restricted to the purely elastic range. While plastic deformations were allowed from the outset, some artifact plasticity zones could in fact appear because the initial stresses and strains are zero by default and thus very far from final equilibrium conditions. Since the effective initial state is (and will probably remain) unknown, it seems better not to consider plastic behaviour during this stage. Step 2: Glacial phase - Maximum Würm: the valley is completely filled by the glacier. Step 3: End of Würm: the glacier has completely melted and the valley is now empty. The period between Steps 2 and 3 serves to simulate deglaciation. To simulate a gradual melting of the glacier, three intermediate states (3a, 3b and 3c) of deglaciation have been introduced. Step 4: Current period: the alluvium has filled the bottom of the valley.


Rabuons Valley

SE Lobe



NW Lobe


to Nice


Dailoutre Valley

Tinee River

To SaintSaint-EtienneEtiennedede-Tinee Figure 9. Different areas of the La Clapière landslide.

3,000 m Crest line (Alt. 3,000 m)

Tinée valley Ice Bottom of the valley

C 3,000 m

Alluvial deposit


Rock mass


Figure 10. Geometry of the model, boundary conditions and constitutive materials.



Intact rock homogeneous anisotropic

« local » discontinuities

« extensive » discontinuities

Equivalent continuum (EC) anisotropic

Explicit joints

Figure 11. Mechanical equivalent characteristics (MEC).

Step 1 Consolidation

Step 2 Maximum glaciation

Step 3 Retreat of the glacier

Step 4 Deposit of the alluvium

Figure 12. The various modelling steps for the La Clapière slope (after Gunzburger [14]).


Since the computation was part elastic and part elasto-plastic, the evolution of the Mohr-Coulomb yield function,

f V V 1  K pV 3  V c (5) with K p

§S I · tan 2 ¨  ¸ © 4 2¹

1  sin I 1  sin I


was calculated at different points situated on the superficial part of the slope, at its bottom, middle and top (points A, B and C on Figure 10). The various slope components demonstrate radically different mechanical behaviour during deglaciation: whereas the highest part does not sustain major modifications to the yield function value, the lowest part of the slope is raised to nearly the yield surface. The final state is such that the toe turns out to be the most unfavorable area of the slope. Consequently, if it is assumed that the instability appearing after glacial melting, we could then presume that deglaciation-related unloading induced initiation of the landslide at the slope base followed by upslope expansion. This result is in good agreement with aerial photograph interpretations, which suggest that the La Clapière landslide could have begun at the level of the Tinée River Valley. Since this modelling set-up agrees with the scenario of a regressive evolution in the landslide from the bottom to the top of the slope, it is taken as encouragement to move ahead with this conceptual model and to improve it to take into account the following features: anisotropy, influence of valley width, 3D geometry, decrease in the mechanical characteristic due to weathering, water. The next section will present the test of anisotropy. 4.3.


Even though the toppling scenario was critiqued in Section 4.1, foliation does exist in the La Clapière slope and displays direction changes in the vicinity of the topographic surface. Further modelling was undertaken to test the influence of both anisotropy and the variation in anisotropy. Models using the CESAR-LCPC finite element code have been derived in order to investigate this consideration. The three configurations shown in Figure 13 have been studied in elasticity. The outputs of these 3 configurations are noticeably different even thought the dissimilarities are difficult to interpret. An influence of anisotropy on the V1 and V3 moduli and orientation can be observed, as well as on the value of the MohrCoulomb yield function f (Section 4.2). Yet it remains difficult to interpret in terms of stability improvement or worsening. The different results of V1 orientation are shown in Figure 14.







2 3


2 0 (isotropic case)








Figure 13. The various configurations of anisotropy studied (according to Dehail [6]).

It is necessary to define an anisotropic failure criterion to advance the interpretation of the influence of anisotropy. Due to the scarcity of laboratory data, no anisotropic criterion is available. Moreover, should one exist, it would be difficult to transpose it directly to rock mass behaviour. Even though the influence of anisotropy in terms of stability worsening is difficult to infer, the use of an anisotropic elastic model shows undeniably that this feature must be taken into account in future modelling and that it is better to represent foliation in the model by anisotropy than by series of joints.

0 (isotropic case)




Figure 14. Angle of the principal major stress with horizontal for the various configurations of anisotropy studied (according to Dehail [6]).

The use of anisotropy to represent foliation seems to be more satisfying than to view it as a set of discontinuities. If foliation is represented as a set of fractures, it is difficult to get field data on the spacing and the properties of fractures. At present, it is also difficult to obtain field rock mass properties, but it is possible to get more information from laboratory tests and to apply RMR methodology to an anisotropic rock.

125 5.


Various kinds of numerical mechanical models have been applied to better understand the La Clapière landslide. The three analyses presented herein show that each model has allowed a partial understanding of the problems raised by rock slope movements and has improved our knowledge of the site. The toppling scenario was discounted because of the questions raised by the first analysis. The model did not demonstrate the impossibility of this scenario but suggest a return to the field to obtain more information. From the field investigations another hypothesis on the origin of the toppling was put forward, i.e. the tectonic origin with field arguments detailed in [15]. The second model has shown that the regressive failure from the bottom to the top is mechanically plausible. The third analysis has confirmed the role of anisotropy is determining on stress state in the rock mass. Nevertheless, at present, it did not permit an assessment in terms of stability improvement or worsening. This third model has led to the conclusion that the foliation should rather be represented by anisotropy than by a set of discontinuities. The use of modelling has served as a complementary means of investigation for such sites. It has been introduced to help confirm or reject the geometrical and mechanical plausibility of qualitative movement scenarios. The model led to go back to the field with precise queries, instead of just having a look on the movement. The models described above are mechanical models. On this landslide, hydrogeochemical and hydrogeological models have also been developed in order to analyze hydraulic and hydrogeochemical information from a global conceptual standpoint.

Acknowledgments This work is part of a collective research effort sponsored by the French PNRN Program ("Programme National Risques Naturels"). The contribution by a number of colleagues working on the project is gratefully acknowledged, especially Y. Guglielmi.

References 1. 2. 3. 4.



Bieniawski, Z.T. (1989) Engineering Rock Mass Classifications, J. Wiley, New York. Box, G.E.P. and Jenkins, G.M., (1976) Time Series Analysis: Forecasting and Control, Holden-Day, San Francisco. Changnon, S.A., Huff, F.A. and Hsu C.F. (1988) Relations between precipitations and shallow groundwater in Illinois, Journal of Climate, Vol. 1, No. 12, pp. 1239-1250, Dec. 1988. Crosta, G.B. (2002) Rock Slope instability, the transition to catastrophic failure, Nato Advanced Research Workshop "Massive Rock Slope Failure – New Models for Hazard Assessment", Celano, Italy, June 16-21 2002. Cundall, P A. (1971) A computer model for simulating progressive, large-scale movements in blocky rock systems. Proceedings of the Symposium of the International Society of Rock Mechanics. Nancy, France. Vol. 1 Paper N° II-8. Dehail, V. (2002) Modélisation du glissement de terrain de la Clapière avec le logiciel CESAR, Projet de recherche de 3ème année Ecole des Mines de Nancy.

126 7. 8.

9. 10. 11.

12. 13.

14. 15.




19. 20. 21.






Durville, J.L. (1992) Mécanismes et modèles de comportement des grands mouvements de versants. Bulletin de l’Association Internationale de Géologie de l’Ingénieur, n°45, pp. 25-42. Dussauge-Peisser, C. (2002) Evaluation de l'aléa éboulement rocheux. Développement méthodologiques et approches expérimentales. Application au falaises calcaires du Y Grenoblois, Ph.D. thesis, LIRIGM, Université Joseph Fourier, Grenoble, France. Follacci, J.P. (1987) Les mouvements du versant de La Clapière à Saint-Etienne de Tinée (AlpesMaritimes). Bulletin de liaison du Laboratoire des Ponts et Chaussées. 150/151, pp. 107-109. Follacci, J.P. (1999) Seize ans de surveillance du glissement de La Clapière (Alpes Maritimes). Bulletin de liaison du Laboratoire des Ponts et Chaussées 220, pp. 35-51. Guglielmi, Y., Bertrand, C., Compagnon, F., Follacci, J.P. and Mudry J. (2000) Acquisition of water chemistry in a mobile fissured basement massif: Its role in the hydrogeological knowledge of the La Clapière Landslide (Mercantour Massif, Southern Alps, France). Journal of Hydrology, Elsevier. Vol. 229, no. 3-4, pp. 138-148. Guglielmi, Y., Seve, G. and David, E. (2002) Groundwater geochemistry for the surveillance of large moving rock masses: The example of La Clapière landslide (France). Geomorphology (paper accepted). Guglielmi, Y., Vengeon, J.M., Bertrand C., Mudry J., Folacci J.P. and Giraud A. (2002) Hydrogeochemistry. An investigation tool to evaluate infiltration into large moving rock masses (case study of La Clapière and Séchilienne Alpine landslides). Engineering Geology (paper accepted). Gunzburger, Y. (2001) Apport de l'analyse de la fracturation à l'étude du versant instable de la Clapière (Saint-Etienne-de-Tinée, Alpes-Maritimes), DEA PAE3S, LAEGO, INPL, July 18 (2001). 98 pages. Gunzburger, Y. and Laumonier, B. (2002) Origine tectonique du pli supportant le glissement de terrain de la Clapière (NW du massif de l’Argentera-Mercantour, Alpes du Sud, France) d’après l’analyse de la fracturation. Comptes Rendus Géoscience (Academie des Sciences). Vol. 334, no. 6, pp.415-422, May 2002. Gunzburger, Y. and Merrien-Soukatchoff V. (2002) Caractérisation mécanique d'un versant rocheux instable au moyen du système RMR - Cas de la Clapière (Alpes-Maritimes), Symposium International Param2002, Identification et détermination des paramètres des sols et des roches pour les calculs géotechniques, September 2-3, 2002, Paris, France. pp. 541-548. Gunzburger, Y., Merrien-Soukatchoff, V. and Guglielmi, Y. (2002) Mechanical influence of the last deglaciation on the initiation of the "La Clapière" slope instability (southern French Alps), 5th European Conference on Numerical Methods in Geotechnical Engineering (NUMGE) 2002, September 4-6, 2002, Paris, France. pp. 713-718. Interreg I. (1997) Risques générés par les grands mouvements de versant. Etude comparative de 4 sites des Alpes franco-italiennes. Edited by Pôle Grenoblois d’Etudes et de Recherches pour la prévention des Risques Naturels (Grenoble). Jing, L. and Hudson, J.A. (2002) Numerical methods in rock mechanics, International Journal of Rock Mechanic and Mining Sciences, Vol. 39, n° 4, August 2002, pp. 409-427. Laouafa, F., Darve, F. (2002) Modelling of Slope Failure by a material instability mechanism. Computers and Geotechnics 29, pp. 301-325. Lee, J.Y. and Lee, K.K. (2000) Use of hydrologic time series data for identification of recharge mechanism in a fractured bedrock aquifer system, School of Earth and Environmental Sciences, Seoul National University, Journal of Hydrology, 229, pp. 190-201. Merrien-Soukatchoff, V. (2002) Eléments de réflexion sur la qualité des modélisations en hydrogéotechnique, Mémoire d'Habilitation à Diriger des Recherches, LAEGO, Ecole des Mines de Nancy, INPL, Nancy, France. ( Merrien-Soukatchoff, V. and Gunzburger, Y. (2002) Modelling a tool of investigation for landslide: The case of La Clapière landslide (Southern Alps, France). International Symposium on Landslide, Risk Mitigation and Protection of Cultural and Natural Heritage, January 21-25, 2002, Kyoto University, Kyoto, Japan, 11 pages. Merrien-Soukatchoff, V. and Gunzburger, Y. (2002) Utilisation des classifications de massifs rocheux pour l'analyse de la stabilité de pentes. Présentation de deux cas d'application, Journées JNGG 2002, 89 octobre 2002, Nancy, France, 12 pages. Merrien-Soukatchoff, V., Quenot, X. and Guglielmi, Y. (2001) Apports de méthodes géomécaniques quantitatives à l’investigation de grands versants instables : application au glissement de la Clapière (Saint-Etienne-de-Tinée, Alpes Maritimes). XVème Congrès Français de Mécanique, Nancy, September 3-7, 2001. 6 pages. Merrien-Soukatchoff, V., Quenot, X. and Guglielmi, Y., Gunzburger, Y. (2001) Modélisation par éléments distincts du phénomène de fauchage gravitaire. Application au glissement de la Clapière




(Saint-Etienne-de-Tinée) Alpes-Maritimes. Revue Française de Géotechnique, Vol. 95-96, pp. 133142. Slini, Th., Karatzas, K. and Moussiopoulos, N. (2002) Statistical analysis of environmental data as the basis of forecasting: An air quality application, The Science of the Total Environment, 288, pp. 227237. Vengeon, J.M., Hantz, D. and Dussauge, C. (2001) Previsibilité des éboulements rocheux; approche probabiliste par combinaison d'études historiques et géomecaniques. Revue Française de Géotechnique, Vol. 95-96, pp. 143-154.


D. STEAD1 Simon Fraser University, Department of Earth Sciences, 8888 University Drive, Burnaby, B.C., V5A 1S6, Canada. J.S. COGGAN University of Exeter, Camborne School of Mines, Redruth, Cornwall, TR15 1SE, UK.

Abstract This paper illustrates the use of a combined finite-discrete element code with fracture propagation capabilities in the simulation of rock slope failure. Using this approach it is possible to investigate failure from initiation through transportation to deposition; in effect from a “total slope failure process” perspective. Selected failure cases and mechanisms are simulated, not as definitive analyses, but in order to show the future potential of this techniques in rock slope analysis.



Numerical modelling of rock slope failures has traditionally involved an emphasis on either the initiation or transport and depositional characteristics. Both continuum models, FLAC [7], Visage [15] and discontinuum models, UDEC [8] have been used increasingly in recent years to simulate the initiation of a rock slope failure. A range of constitutive models and methods have been applied with varying success. Coggan et al. [3] describe current methods used in rock slope analysis emphasising both the choice of appropriate models and good modelling practice. Recent developments have seen continuum codes developed which are specifically directed toward slope modelling, FLAC/Slope [9]. A major question that rock slope engineers will always need to answer is the risk posed by the rock slope failure. This requires an estimate of the spatial risk; something that is not readily available from routinely used numerical models. To date most analyses addressing the issue of spatial risk have involved analytical models based on rheological and frictional flow. These models allow an estimate of both the run-out distance and velocity of the slope failure and after calibration against field constraints 1

E-mail of corresponding author: [email protected]

129 S.G. Evans et al. (eds.), Landslides from Massive Rock Slope Failure, 129–138. © 2006 Springer. Printed in the Netherlands.

130 have been shown to be of significant practical application [6]; they, however, deal principally with the transport and depositional aspects of the failures. Couture et al. [4]. discussed the varied zones in a rock slope failure including the initiation/detachment zone, the transport zone and the zone of deposition. Attempts have been made to investigate the rock relationship of rock failure debris with the pre-existing rock mass structure using Stereoblock. The relatively recent introduction in rock engineering of particle flow codes responds to a need to simulate the complete failure process both in underground and surface environments. Calvetti et al. [1] show the potential for using the particle flow code PFC2D in the simulation of debris flow movements. In this method the failure mass is simulated using an arrangement of circular (2D) or spherical (3D) particles. These particles may be bonded together to form joint-bound blocks. The development of instability in the slope results in induced stresses that break the bonds between the rock blocks. In this way the initiation, fragmentation and flow of the failure debris can be simulated. The current paper presents an alternative technique for simulating the “total rock slope failure process” using a combined finite elementdiscrete element code, ELFEN [12].


The Combined Finite-Discrete Element Model

The combined finite–discrete element method, with fracture propagation, has been used successfully in rock engineering to simulate processes of rock bursting in deep underground mines, mining of tabular orebodies, and the blasting of rock slopes in open pit mining. Munjiza et al. [11] illustrated the use of a finite element mesh to represent the rock slope to be blasted, Figure 1. The effect of the detonation of successive blast holes is simulated by the progressive growth of fractures in the rock mass resulting in the formation of discrete elements. These deformable discrete elements are re-meshed and the analysis continued. The gradual breakdown of the rock mass in response to the blast is simulated by fracture growth, re-meshing and the formation of smaller discrete elements. The ELFEN code thus allowed the efficiency of the blast to be increased and in addition a comparison between the simulated and the observed blast muck pile size distribution.

Figure 1. Use of combined the finite-discrete element code , ELFEN, to simulate blasting, Munjiza et al. [11].


The combined finite element-discrete element code with fracture has also been used to investigate comminution during the grinding of materials in rock crushers. The ELFEN code, in addition, possesses a 2D/3D particle flow code module to allow simulation of the frictional flow of circular/spherical particles. The current researchers saw obvious parallels between the simulations of the processes involved in blast throw, rock pile movement, comminution/flow and those active in the “total rock slope failure process”. This paper provides a brief description of the combined finite-discrete element code, ELFEN, and preliminary results in its use to simulate varied rock slope failure mechanisms. ELFEN modelling of rock slopes presented in this paper and published in the engineering literature to date has been undertaken at Simon Fraser University, Canada and the Camborne School of Mines in the UK. It should be emphasised that the authors recognise that, as with any new method, a thorough and extensive process of back-analysis against major rock slope failure with available field/laboratory data is required. The prime purpose of this paper is to demonstrate the significant potential of this technique, recognising in their ongoing research the need to provide more constraint through the integrated use of alternative modelling methods and site observations. The combined finite-discrete element code utilises a variety of constitutive criteria including linear elastic (isotropic/orthotropic), non-linear (Mohr Coulomb, DrukerPrager, Von Mises, Rankine etc), visco-plastic and rigid. The tensile failure and crack propagation is modelled using a post initial yield rotating crack or Rankine formulation. Anisotropic damage evolution is simulated by degrading the elastic modulus,E, in the direction of the major principal stress invariant. The damage parameter, Z, is dependent on the fracture energy, Gf which is related to the critical stress intensity factor, KIC by Gf = K2IC/E. At some point in the analysis of a rock slope the adopted constitutive model predicts the formation of a failure band within a single element, Figure 2. The load carrying capacity across such localised bands decreases to zero as damage increases until eventually the energy needed to form a discrete fracture is released.

Existing Edges New Edges New Nodes

Figure 2. Formation of crack within finite element and use of ELFEN in simulating uniaxial compression test. [12].

132 At this point the topology of the mesh is updated, initially leading to fracture propagation within a continuum and eventually resulting in the formation of discrete elements as the rock fragments are formed [11]. Efficient algorithms have been developed for contact searching, however it is apparent that the continuous production of new smaller and smaller discrete elements for large problems may require considerable computing power, particularly for three dimensional problems. In response to this facilities exist for a parallel computing approach.


Preliminary Applications Of The Combined Finite Element-Discrete Element Code Elfen



The authors undertook a preliminary ELFEN modelling exercise on the April/May 1991Randa Rockslide, Switzerland. This was the first published use of the ELFEN code in rockslide simulation. [14]. The geology of the Randa rock slope is illustrated in Figure 3 and is seen to comprise predominantly gneiss with several well developed joint sets. This ELFEN model complemented continuum and discontinuum models previously undertaken by Eberhardt et al. [5]. Modelling analyses have emphasised the importance of progressive rock mass degradation and damage within the rock slope prior to eventual slope failure. This damage appears to be rooted as far back as the initial unloading of the rock slope during deglaciation, which has been shown to result in tensile damage at the slope toe. Figure 4 shows a series of snapshots of the failure process that occurred in two stages, a lower rockslide and a later upper rockslide. These analyses used as input the observed rock slide failure surfaces and were undertaken to illustrate the potential of the ELFEN code in simulating the progressive fragmentation of the rock slope mass both during initiation, transportation and deposition; that is “during the total slope failure process”.

Figure 3. Cross-section showing the geology and geometry of the 1991 Randa rockslides, after [5].


Please Wait..

Please Wait..


Please Wait..


Please Wait..



Please Wait..


Please Wait..


Figure 4. The use of the combined finite-discrete element code, ELFEN, to simulate the Randa Rockslide, Switzerland, 1991 [14].

Further ELFEN analyses have been undertaken at Simon Fraser University in order to simulate the Randa rock slope failure and in particular the actual development of the failure surface. Both Rankine rotating crack and Mohr constitutive criteria, incorporating crack propagation have been used and the influence of glacier unloading investigated. Preliminary interpretation of these models clearly indicates the importance of progressive rock mass damage in the development of major rock slides such as Randa, and will be the subject of a future paper.



The Elm rockslide occurred on September 11, 1881 and resulted in the loss of 115 lives. It is described in Heim’s classic work “Landslides and Human Lives” [13] as a clear example of what he termed “rubble stream” phenomenon in rock avalanche run out. The rockslide occurred within fine cleavage slates quarried at the slope toe and dipping into the mountain at 30o. The rock slope mass also comprised flysch sandstone, greensands and glauconitic limestones. A preliminary analysis of the Elm rockslide was undertaken based on the Heim’s section, Figure 5. In this analysis the development of fracturing due to the quarry at the toe is simulated along with the run out of a dry fragmented rock mass. Figure 6 illustrates stages in the rock slope failure and runout. Initial results were promising in terms of the comparative runout magnitude simulated assuming frictional transport. Further work is required to examine the development of failure with progression of the quarrying at the toe. Preliminary ELFEN analyses assumed properties for a weak highly anisotropic rock mass.


Figure 5. The Elm rockslide,1881 after Heim.[13], 1: 12500 scale.

Please Wait..


Please Wait..


Please Wait..

Please Wait..



Figure 6. Stages in the ELFEN simulation of the Elm slide.

Further sensitivity analyses are intended to examine the influence of varying rock mass strength on both the failure mechanism and transport process. 3.3.


Biplanar rock slope failures are a common feature in many UK opencast coal mines. These failures frequently comprise a rear failure surface along a fault and a lower basal failure along a plane of weakness, such as a seatearth. For such failure to occur a kinematic space constraint is assumed to require the formation of an inter wedge interface. A graben feature often develops in the post failure topography. Figure 7 illustrates a preliminary model of such a biplanar failure geometry and clearly shows the brittle internal fracturing that accompanies rock slope failure.






Figure 7. Preliminary model of biplanar rock slope failure, 50m high rock slope (1V:1H).

Complex rock slopes failures often involve kinematic considerations. This is illustrated in the literature by the frequent reference to internal deformation involving dilation, fracturing and the creation of voids within the slope. Such mechanisms were, for example, cited as important in the chair-shaped failure plane along which the Vaiont failure occurred and preliminary modelling of the Vaiont slide is underway in order to investigate the importance of this process. Similarly rock slope failures often involve complex lateral release mechanisms with plan rotational movements in order to provide kinematic release. The ELFEN code, at last provides a realistic model in which to study the kinematics of rock slope failure. The active-passive wedge failure mechanism described within the literature requires the presence of an interwedge interface – the development of this interface may be far more complex in rock slopes than previously envisaged. The importance of lateral release mechanisms has been neglected in rock slope deformation modelling and current work by the authors is directed towards assessing the progressive lateral release of rock slope failure blocks. 3.4.


Distinct Element modelling of a rock slope failure at Delabole Slate quarry rock of has been previously undertaken using the UDEC discontinuum code and is described in detail in Coggan and Pine [2]. Figure 8 shows the pre and post failure geometry of the failure. The mechanism of the failure has been described as due to the chisel shaped blocks (1 and 2) promoting toppling and sliding of the lower blocks. Toppling mechanisms are indicated by both UDEC modelling and a long period of surface slope monitoring using tension crack gauges and EDM. Selected snapshots, 1-4, from an ELFEN simulation of the Delabole failure are illustrated in Figure 9.


Approximate Pre-failure surface


74o 2 55o


Failure Surface 3

63o 4


45o Slide debris Claylodes

Figure 8. The Delabole slate quarry rock slope failure, [2].



Please Wait..


Please Wait..


Please Wait..

Please Wait..

Figure 9. Preliminary ELFEN model of rock slope failure shown in Figure 8 at the Delabole slate quarry, UK.


3.5. COASTAL EROSION AND ROCKFALL Coastal erosion and rockfall is a significant problem on the south east coastline of England, with recent major rockfalls in the Beachy Head area, [10] This problem involves chalk cliffs which, undertake significant fragmentation during slope failure. A series of ELFEN simulations are shown in Figure 10 which illustrate the potential for modelling toppling and rockfall problems. The ELFEN code had been used routinely for modelling the demolition of tall industrial chimneys and the parallel between this and failure of coastal sections is apparent.






Figure 10. Modelling of coastal erosion and rockfall in a chalk cliff with similar depositional character to the Beachy Head failures, [10].



The authors believe that combined finite–discrete element modelling with brittle fracture generation will form a major advance in rock slope analysis. Future work is required to apply this method to varied rock slope failure back-analyses in order to constrain input data and improve our understanding of rock slope failure mechanisms. A “toolbox approach” should be used in the analysis of major rock slope failures using appropriate methods for the varied stages of the total slope failure. It is important to adequately characterise the intact rock, discontinuities and rock mass within the initiation or trigger zone. Filed mapping should be targeted at the techniques of analysis that may be appropriate. It is critical to perform an engineering characterization of the rock mass with data on discontinuity persistence, spacing, roughness, block size etc, in addition to the routinely collected orientation data. Increasing use of remote sensing

138 techniques to study rock slides offers the potential of considerable improvements in constraint. Accurate and repeated surveys of the rock slope using techniques including 3D robotic ground based laser surveys, LIDAR, INSAR and DGPS all have considerable promise. The use of terrestrial photogrammetric techniques combined with image processing may provide valuable information on block size within the initiation zone. The majority of slope analysis techniques are confined to either evaluating factors of safety, or characterising continuum displacement/discontinuum displacements (stresses etc) within the initiation zone. Major movements and more importantly changes in kinematic restraints are not tackled. Rheology-based codes such as DAN [6] have proved extremely useful in modelling rockslide debris movement, velocity and runout, using experience of relevant field-based observations in a back-analysis manner to control future design. Such codes used in conjunction with ELFEN offer the chance to simulate rock slope failure from initiation though transportation to deposition. These varied approaches should be used as mutual constraints. Further advances in rock slope analysis require improved and relevant data collection, both surface and subsurface, in order to constrain the increasingly sophisticated codes that are now available.

References 1.


3. 4. 5.

6. 7. 8. 9. 10. 11. 12. 13. 14.


Calvetti, F., Crosta, G. and Tatarella,M. (2000) Numerical simulation of dry granular flows: from the reproduction of small scale experiments to the prediction of rock avalanches. Rivista Itaianal. Di Geotecnica , 2. p21-38. Coggan, J. S., and R. J. Pine. (1996) Application of Distinct-Element Modelling to Assess Slope Stability at Delabole Slate Quarry, Cornwall, England," Trans. Instn. Min. Metall., Sec. A, 105, A22A30. Coggan, J.S., Stead, D. & Eyre, J.M. (1998) Evaluation of techniques for quarry slope stability assessment. Trans. Instn Min and Metall - Section B 107: B139-B147. Minneapolis. Couture, R., Evans, S.G., Locat, J., Hadjigeorgiou and Antoine, P. (1999) A proposed methodology for rock avalanche analysis. Slope Stability Engineering. Eds. Yagi, Yamagami and Jiang. P1369-1378. Eberhardt, E., Stead, D., Coggan, J. & Willenberg, H. (2002). An integrated numerical analysis approach to the Randa rockslide. In Rybár et al. (eds.), Landslides in the Central Europe, Proceedings of the 1st European Conference on Landslides, Prague. Swets & Zeitinger, The Netherlands (In Press). Hungr, O. (1995). A model for the runout analysis of rapid flow slides, debris flows, and avalanches. Can. Geotech J. 32:610-23. Itasca 2002. FLAC 4.0. User’s Guide. Itasca Consulting Group, Inc. Minneapolis. Itasca 2002. UDEC Version 3.1. Universal Distinct Element Code. Itasca Consulting Group, Inc. Itasca 2002. FLAC/Slope. User’s Guide. A mini-version of FLAC to calculate factor of safety of slopes, 74pp, Itasca Consulting Group, Inc. Minneapolis. McWilliam, F. 1999. Just doing what comes naturally.Ground Engineering 32:4. Munjiza, A., Owen, D.R.J. and Bicanic, N. 1995. A combined finite-discrete element method in transient dynamics of fracturing solids. Eng Computations. 12:145-174. Rockflied Software. (2001). Elfen User Manual Version 3.04. Rockfield Software Limited, Swansea,UK. Skermer, N.A. (1989). Landslides and Human Lives, Heim, A. Translation. Bitech Publishers. Stead, D. Eberhardt, E. and Coggan, J.S. 2001. Advanced numerical techniques in rock slope stability\ analysis – Applications and limitations. In UEF International Conference on Landslides - Causes, Impacts and Countermeasures, Davos, Switzerland. Edited by M. Kühne, H.H. Einstein, E. Krauter, H. Klapperich and R. Pöttler, Verlag Glückauf GmbH, Essen, p615-624. VIPS 2001. Visage – Vectorial Implementation of Structural analysis and Geotechnical Engineering, Windsor, UK. Vector International Processing Systems.


Abstract Earthquake-triggered landslides are an important secondary effect of strong shaking. Such phenomena have occurred during a large number of earthquakes and have resulted in significant loss of life. Seismically induced slope failures have occurred in both rock and soil masses, and had sizes ranging from a few individual blocks up to several million cubic meters in volume. Since the mid 1960s numerous methods of analysis have been developed to assess the risk posed by these hazards, however, such investigations normally use a simplified form of ground motion input such as peak ground acceleration, or Arias Intensity (Ia). These take no account of potential amplification effects associated with steep topography. In this paper the mechanisms of topographic amplification are considered, along with the potential impacts on the stability of slopes during large earthquakes. The key uncertainties have been outlined along with some of the major research questions for the future. 1.


The initiation of landslides during earthquakes has been recognized for many years [16, 28, 37]. Such failures generally occur in a variety of geological and topographic conditions and may be related to complex phenomena such as liquefaction or thixotropic behaviour of materials. However, there are many landslides, such as those occurring in rock slopes, which do not involve such complex material behaviors. Additionally, there is a class of earthquake-related landslides that occur some time after shaking has occurred. A good example of such landslides would be the Calitri landslide, triggered by the 1980 Irpinia (Ms=6.7) earthquake [24]. It has been suggested [38] that such failures are related to dilation of dense soils during shaking, and the subsequent decrease in pore volume after vibration has ceased, resulting in a post seismic rise in pore water pressures. Earthquakes have traditionally triggered landslides in a variety of regions. Those dominated by problem soils such weak volcanic deposits [10] in a number of different earthquakes (e.g. the El Salvador earthquakes of 1986, 13, January 2001 and 13, 1

E-mail of corresponding author: [email protected]

139 S.G. Evans et al. (eds.), Landslides from Massive Rock Slope Failure, 139–154. © 2006 Springer. Printed in the Netherlands.

140 February 2001; [10, 40]). Additionally terrains marked by steep slopes show a large number of landslides triggered by seismic events. Examples of the latter group of phenomena include the Huscaran Rock Avalanche triggered by the 1970 Peru Earthquake [14] or more recently the Chi Chi earthquake, Taiwan [32]. This paper will concentrate on the latter group of phenomena, those occurring on steep terrains.


Earthquake Shaking and Topographic Amplification

It is perhaps unsurprising that earthquakes in mountain regions trigger large rock slope failures. The very existence of steep hillslopes is likely to result in landslide occurrence, which will be exacerbated during earthquakes. In addition to the ambient state of stresses that may cause landsliding in mountainous areas, the addition of dynamic, transient, stresses from earthquake shaking leads to additional instability. The effects of topography may result in an increase in the amplitude of ground motions [7, 17, 35, 36] known as topographic amplification. While hazard assessment for slopes may have made use of attenuation relationships to derive design ground motions [49], few if any will incorporate topographic amplification into the analysis. Furthermore, few strong ground motion records incorporate the effects of topography, as instruments tend to be deployed away from hillslopes. Therefore some attenuation relationships [41] may incorporate a term for soil conditions, but have no method for accommodating topographic effects. The limited number records that have been deployed on slopes record unexpectedly high ground accelerations [46]. 1996 report accelerations of 1.82g recorded during the MW=6.7 Northridge earthquake on a hillcrest [35] shows variation in ground motions along the length of a hillslope (Figure 1). How the interaction between topography and seismic waves have affected resultant ground motion has been a field of study for many years, often describing topography in terms of a simple shape ratio (Figure 2) [2, 12, 42, 50, 51]. Normally, however, such studies have been the results of 2-D modelling of canyons or buried valleys, and only limited work has been carried out to investigate the interactions found in 3 dimensions [43]. The effects of wave polarization have been suggested as a possible method of topographic amplification [1]. However data provided by the Central Weather Bureau recorded during the Chi Chi earthquake suggest that at least some amplification may be ascribed to the properties of the material. The peak accelerations shown in Fourier Amplitude Spectra in Figure 3, were clearly recorded in the range of frequencies that are characteristics of hard rock sites, which are normally in the region of 8-20 Hz. Based on the above observations, and on work carried out more recently [5,6] suggested a number of valuable observations about topographic amplification can be made. These include: 1. 2.

Amplification is greatest when polarized S waves have a direction of travel into the slope Amplification was noted to be greatest at steep incidence angles

141 3. 4. 5.


Amplification is greatest at the hill crest “De-amplification” occurs at valley bottoms as indicated by damage surveys [17, 36] The recognition of amplification is complicated by directional effects of strong vibration. Slope failures triggered by the Chi Chi earthquake for example in the Tachia River Gorge were found to be preferentially distributed on one side of the valley. This observation has been noted during other landslide-triggering earthquakes [47]. It was recognized in the studies cited above that analytical analyses of topographic amplification tended to identify correctly the frequency patterns and hill forms that would cause amplification, but the predicted size of the resultant vibrations was c. 50% below recorded ground motions. This observation suggests that there may also be a material component involved, a view supported after careful modelling of the amplification observed at Tarzana Hill during the 1994 Northridge earthquake [48].

Figure 1. Changes in ground motion characteristics with changes in geomorphology along a hillslope recorded at Mt. Ushibara, Japan shown as an amplification ratio relative to a base accelerometer [35]).


Figure 2. Variation between real hillslopes and models used in analytical studies.

Figure 3. Fourier amplitude spectra for E-W and N-S components of horizontal acceleration recorded at the Techi Dam.

143 Given the statements made above, it is useful to consider the impact of these observations on methods of assessing seismically triggered landslide phenomena.


The Newmark Sliding Block Model

The Newmark sliding block model [34] is a deterministic method of analysis that employs a bilinear approximation of the strength of soils and rock masses to assess the potential for instability to occur during earthquakes. Various authors have modified this basic methodology. The inputs include the Static Factor of Safety (F) calculated by any standard method; an estimate of ground motions normally expressed as an Arias Intensity (Ia) [4]; and some estimate of the amount of displacement required for the material to be reduced from peak to residual strength conditions. An excellent description of this method of analysis is given in the literature [38], so only a brief outline will be given here. Assume that a landslide can be considered as a rigid block. As this block is subjected to a forced vibration if the intrinsic strength of the mass is exceeded sliding will be initiated. This strength has normally been expressed as the critical acceleration (Acrit), which can be estimated as shown in equation 1: Acrit = (F-1)gsinD


where F is the static Factor of Safety; g is the acceleration due to gravity (specified in the units in which the final answer is required) and a is the angle of inclination of the shear surface (or the angle of thrust for a landslide with a rotational component; see Jibson & Keefer, 1993). In the case of earthquake shaking, such vibrations are normally recorded as ground accelerations. Convention dictates that these three components of motion are described as vertical; north-south and east-west, and may be expressed in terms of cm/s, m/s or g (a fraction of gravity). Traditional methods [44] have employed peak horizontal ground acceleration as an indicator of earthquake vibration. This leads to serious conservatism in hazard analysis as peak acceleration, by definition, occurs only once in the acceleration-time history (accelerogram). Two conceivable end possibilities exist. The first is that the ground motion never exceeds the critical acceleration. The second is that the vibrations always exceed this threshold. It is however more likely that for at least part of the earthquake, the ground motion will exceed the critical acceleration and some sliding will occur. This leads to the situation shown in Figure 4, i.e., when each acceleration pulse exceeds the critical acceleration some sliding occurs, which then stops as the acceleration pulse gets smaller. With each pulse, increasing strain potentially results in a reduction of strength, which in turn lowers the critical acceleration. The calculated Newmark Displacements shown in Figure 4 [15, 38], are the results of an analysis carried out for the lower part of the Calitri (Campania Region, Italy) landslide. This failure was re-activated by the 1980 Irpinia earthquake [24] showing displacements consistent with those calculated [15]. In addition to the use of actual input data, an attenuation relationship may be used that incorporates the magnitude of the causative earthquake (normally given as a moment magnitude MW, [19]) and the distance (d) between some measure of the source


Figure 4. Newmark’s approach to coseismic landslide displacement computation. Example from the Calitri accelerometric recording (PGA 0.156g) of the Irpinia 1980, Ms 6.8 earthquake with a critical acceleration K c =0.045g equal to that used [15] for modelling the lower part of the Calitri landslide.

(most commonly epicentre). There remains however some question about whether fault rupture distance is a more appropriate indicator of wave amplitude attenuation [29, 33]. While the work cited [15, 38] reported Newmark Displacements calculated from actual strong motion records, it is not always possible to employ accelerograms that are relevant to the geological or tectonic regime in question. In such a situation an attenuation expression is used to estimate the Arias Intensity, and ultimately the displacement of the sliding mass at any given couple of MW and d. Such a relationship, used in various forms [25, 26, 27] is shown in equation 2. Log(DN) = AMW – Bd - CAcrit +/- D


145 where DN is the Newmark Displacement (in cm), MW is the moment magnitude, d is the epicentral distance (in km) and Acrit is the critical acceleration. A, B, C and D are constants to be determined for different regions. Equation 2 has a number of limitations. Firstly, it has been satisfactorily compiled for only one region (California, USA) and therefore, should be treated with caution outside that tectonic regime. Secondly, since the measurements of Ia that equation 2 it is actually based on include few records where strong topographic effects have influenced motion, this is a relationship that is ultimately one for flat ground. Therefore, if the effects of amplification are to be considered in any seismically triggered landslide hazard analysis, it must be done outwith this stage of the assessment. 4.

An Example of Topographic Amplification – illustrating Complex Interactions Between Seismic Waves and the Ground

The Chi Chi earthquake occurred at 01:47 (local time) on 21, September 1999 [45]. It was caused by up to 7.5m of vertical (reverse) slip and 11.5m horizontal displacement on a 100 km length of the Chelungpu Fault, and the earthquake of MW=7.6 was immediately termed the ‘100 year event’ by the national press. The earthquake induced large numbers of landslides in rock slopes. The majority of these slides occurred in the hangingwall of the fault due to the nature of the terrain. The earthquake had a duration of c. 40-45s and the peak horizontal ground motion was recorded in the E-W direction was 0.98g [45]. Extensive landsliding occurred in rock slopes surrounding the Central Cross Island Highway near the Techi Dam. An investigation of twenty-three large rockslides that occurred along a 2km section of highway was carried out to investigate the role of topographic amplification in the initiation of these rock slope failures. 4.1.


A number of observations can be made about the spatial distribution of landslides triggered by the Chi Chi earthquake in this area. Firstly, there are a significantly larger number of failures on the south side of the river valley. This is contrary to existing models of topographic amplification, which suggest that the greatest increase in shaking should be in the direction of wave propagation, in this case, broadly west to east. Geological investigation excluded a major discontinuity set as a controlling influence. Failure occurred on a shear surface that formed due to the intersection of large numbers of smaller fracture sets. Secondly, there was a strong correlation between geomorphological break of slope and the crown of the landslide, suggesting some form of topographic control. If, as seems likely given the earlier discussion of amplification processes, slope angle and slope length has an influence on surface acceleration, then slope facets of different angles and lengths are likely to show differential responses to an input wave. It is also worth noting that the majority of landslides are shallow to medium thickness. Normal slope stability analysis has indicated that the depths observed in the field do not yield the lowest value for F. This however, could be a function of the mode of analysis, and should be treated with some caution.




The geology of each failure was identified, and a rock mass classification was carried out using Rock Mass Rating (RMR) [8] and Geological Strength Index (GSI) [20, 21, 22]. The rock mass was sampled and the materials tested in the laboratory under triaxial compression. The combination of laboratory and field observations was used to derive equivalent c and ɮ parameters for slope stability analysis. The parameters of shear surface angle and depth were based on field observations. The range of values used in the analysis is shown in Table 1 and further details are given elsewhere [31, 32]. Table 1. Range of parameters used in the analysis of rock slope failures near the Techi Dam triggered by the Chi Chi earthquake. Parameter Cohesion Friction angle Depth Shear surface angle F Acrit

Range 9-16 kPa 27-61o 5-15m 36-45 o 0.97-4.75 0-2.5g

The nearest available strong motion instrument was a freefield accelerometer operated by the Taiwanese Central Weather Bureau. This instrument recorded a Peak Ground Acceleration = 0.54g. Of the landslides back-analyzed, only three of these failures would have shown any displacement at this acceleration, and only one, which was marginally unstable under ambient stress conditions, would have shown any significant Newmark Displacement. Estimates of the wavelength of the recorded seismic waves from the Fourier amplitude spectra shown in Figure 3 indicated a length of seismic wave broadly consistent with the length of individual terrain facets identified through geomorphological mapping. The length of the slope facets in which the largest rockslides were observed, compared with slope angle is shown in Figure 5. Analysis of a larger dataset of landslides tends to support the original observations (S. Sepulveda, Personal Communication 2003). Figure 5 tends to suggest a geomorphological control, However, the observation made previously that the frequency content of the incident waves was in the range at which strong rock masses would resonate clearly complicates the interpretation of these results. Furthermore, the higher incidence of landslides on the south side of the Tachia River Gorge may also suggest a directional effect. Therefore, there were three possible phenomena that affected the ground accelerations, and the incidence of landsliding on the rock slopes in this area. 4.3. SUMMARY OF OBSERVATIONS ABOUT LANDSLIDING DURING THE CHI CHI EARTHQUAKE The mountainous nature of the terrain in the Central Mountains of Taiwan had a major role in the triggering of landslides during the 1999 earthquake. However, the


Figure 5. The influence of slope geometry on landslide activity. Those in which landsliding occurred had a slope facet length similar to the estimated wavelength of the seismic waves [32].

combination of field, laboratory and seismological investigations of the landslides triggered in the Tachia River Gorge near the Techi Dam perhaps raises more questions than are answered. The three principal questions to arise from the investigations of the Techi landslides are: i.




Landslides appear to have been triggered preferentially at breaks of slope. Many slope failures had their crowns at major geomorphic boundaries. This observation was not limited to the rockslides seen at Techi, but appear to be characteristic of landslides observed at other earthquakes (e.g. El Salvador, 2001). If the hypothesis of different slope facets, defined on the basis of length and angle, being accelerated at different rates is accepted, then the strongest effect would be expected on ridges. Various reports [e.g. 25] note such an observation. Within slope facets where failures were observed, back analysis of the landslides suggests that the accelerations required to induce sliding were up to 5 times higher than the recorded ground motions. It is interesting to note the similarity in depths of the landslides near the Techi dam. This is not consistent over the wider area. The El Salvador Earthquakes January – February 2001

The El Salvador earthquakes of January (MW=7.6) and February (MW=6.6) 2001 triggered extensive landslide activity. The incidence of landslides is somewhat complicated by the behavior of a major geological unit in the region, The Tierra Blanca, which displayed metastability [9]. However, it is clear that the occurrence of landslides on the Balsamo Ridge is significantly higher than debris slides in other areas. 5.1.


The geology of Balsamo Ridge is dominated by rocks of rhyolitic and dacitic composition. These lavas are frequently intercalated with weak layers of pyroclastic

148 rocks (GSI < 26) and palaeosols. The combination of these weak deposits and complex hydrogeological conditions results in a poor quality rock mass. Overlying these rocks are younger deposits of the Holocene San Salvador Formation, particularly, the Tierra Blanca. The rocks of the San Salvador formation, specifically the Tierra Blanca, are worth considering further. The Tierra Blanca is a highly heterogeneous pyroclastic ashfall deposit composed of a number of geological sub-units. The particle size distribution is best described as a sand or silty sand dominated by quartz and feldspar minerals. The structure of the rock is open with porosity measurements of up to 50%. Direct shear strength tests on undisturbed materials indicate values of c = 0-40kPa and I of between 34 and 38o. One-dimensional consolidation tests show that on saturation there is a substantial (up to 24%) decrease in pore volume. The presence of weak gel cements [11] partly accounts for the in situ strength of this material allowing road cuttings to stand up at high slope angles. It seems likely that there is a strong component of field strength derived from the effects of matric suction due to the high porosities. The type of landslides triggered by the earthquake of January 2001 was strongly linked to the geology. Slopes formed from the rocks of the Balsamo Formation mainly suffered rockfalls and debris falls. However, debris slides and debris flows were common in the Tierra Blanca. Such failures had been observed both during prior earthquakes (principally the much smaller 1986, ML=5.6 earthquake, [39]) and storm events.

Figure 6. SPOT monochromatic image (near infra red band) collected on the 28th January 2001 showing some of the landslides triggered by the 13, January 2001 El Salvador Earthquake.

149 5.2. LANDSLIDES TRIGGERED BY THE EARTHQUAKE Figure 6 shows landslides triggered on the main Santa Tecla to Comasagua highway. Extensive landsliding occurred along the ridge of the Balsamo Cordillera, giving rise to the classic shattered ridge (R. Jibson Personal Communication 2002). Examination of a SPOT image obtained on the 28th January shows that landslides were equally distributed both sides of the Balsamo Ridge (Figure 6). Such an observation suggests that no directional effects occurred as fault rupture propagated from east to west [9]. Furthermore, the distribution of landslides on numerous ridgelines of various orientations (Figure 7) reflects the distribution of slope orientations in the area. The landslides at Las Colinas and Las Barrioleras (c. 2km west of the Las Colinas slide) showed unusually large travel distances of c. 790 m and 1140 m respectively. These debris flows continued to move on slope angles as low as 6o. Back analysis of the landslide at Las Colinas [32] suggested that the failure could have been predicted using the Newmark Sliding Block model, although only by using the actual Arias Intensity recorded in San Salvador. Therefore, it is unnecessary to invoke the hypothesis of topographic amplification for this failure. The same is not necessarily true of landslides further along the ridge, where factors of safety for the observed landslides are somewhat higher (ca. 1-1.5).





Figure 7. Rose diagram indicating the orientation of 166 landslides triggered by the El Salvador earthquake. Although there appears to be an E-W trend, this is a reflection of the orientation of slopes associated with the Balsamo Ridge.

150 While the mechanics of this movement are intriguing, it is interesting to note that both of these landslides had, like the rockslides at Techi, their crowns at major geomorphic boundaries. This observation also applies to a large number of landslides. Published photographs [39] also show a major debris slide with an origin at the crest of a slope. In summary the landslides triggered by the El Salvador earthquake, fell into two broad categories: those resulting from the properties of the Tierra Blanca, and those that appeared to have a topographic control. The absence of the strong directional effects in the landslides triggered by the Chi Chi earthquake, which complicated the interpretation of the landslides from the latter event, is replaced by difficulties in understanding a possibly complex pore pressure response in the partially saturated Tierra Blanca.



The interaction of topography and seismic waves remains unclear. This uncertainty stems from a two possible reasons, these are: 1. 2.

The limited number of strong motion instruments deployed on ridgelines and hill crests; The absence of analytical studies that include more than one type of seismic wave, [5] for example assume that amplification processes stem only from horizontally polarized S waves.

It can be observed from damage patterns in buildings [17, 36] during historical earthquakes, and from instrumental records [46] that there are variations in ground motions on hillslopes and ridges during earthquakes. In addition to these ‘seismological’ problems, it is also difficult to rule out some major uncertainties in understanding the mechanics of a rock mass subject to shaking. Therefore, any form of hazard assessment that treats the amplitude of earthquake waves along hillslopes and ridgelines as homogenous will be undoubtedly simplistic. Given the observations made earlier regarding slope geometry and acceleration, there are a number of questions that cannot currently be answered regarding the interaction of seismic waves and topographic features (and subsequently on the generation of landsliding during earthquakes). These are: 1.



What is the interaction between waves of various incidence angles and the ground? The majority of existing research carried out to date has considered only polarized S waves, but P waves and Surface waves should be considered? Do differential stresses really occur at breaks of slope, or is the apparent clustering of seismically-triggered landslides around significant breaks of slopes (as indicated in Figure 8) a function of changing degrees of stability even in the absence of earthquake shaking? If there is a relationship between seismic wavelength, slope geometry and rock mass properties, what is the magnitude of amplification that may be


4. 5.



expected? Is such amplification systematically predictable based on current rock mass rating systems such as Q, RMR or GSI? If so, should this be considered using I(a) or should some alternative be proposed? What is the link between the resonant frequency of a rock mass, and the rock mass properties? What is the link between wave properties of the rock mass, and basic geomechanical parameters measurable in the field and the laboratory? Can rock mass classification schemes be used to predict topographic amplification as an input to hazard assessment for rock slope failures? What, if any changes in acceleration occur in strong rock masses with depth? Does this influence the depth of landsliding independently of the stability of the slope under ambient conditions? What is the role of progressive failure in the rock mass? How long are assessments of seismic slope stability valid? How much displacement in a sliding mass is required to change the seismic properties of the ground?

Figure 8. Schematic diagram showing a possible explanation of landslide activity at breaks of slope due to localized tension created by different accelerations on slope facets of different angles.

Until questions such as these can be answered it is likely to be difficult to be confident in any analysis of the stability of slopes during earthquakes. While the majority of assessments of landslide hazards are undertaken with imperfect knowledge of future events, in the case of earthquake-triggered landslides, the assumptions and uncertainties in estimating the stresses acting are extreme. Given the inability to predict the stresses acting on a rock mass during an earthquake, it should be no surprise that landslides continue to be triggered in places where we do not expect them to occur.



As a form of hazard assessment the Newmark Sliding Block Model gives some reasonable results for the evaluation of landslides during earthquakes. There are

152 however, a number of areas of uncertainty in the inputs to these models. Such difficulties are encountered in terms of defining material properties, but more importantly perhaps, understanding and characterizing the input ground motions for the analysis. Massive rock slope failures are, almost by definition, likely to occur in steep rock slopes, with possibly complex geomorphology. Therefore, these types of failure may occur in terrains where topographic amplification is likely to occur. While analytical studies [6] and summaries of the subject [18] demonstrate that the prediction where topographic amplification may occur is possible, under certain limited conditions, they do not describe well the magnitude of such amplifications, or give good assessments of ground motions for complex, potentially multilithologic, slopes. Therefore, one of the main inputs to the assessment of seismically triggered landslide hazards in rock slopes is only very poorly described. When the questions outlined in section 6 can be answered more fully, then a greatly improved means of assessing the hazard of massive rock slope failures during earthquakes will be possible.

Acknowledgements The observations presented in this paper were possible because of the financial support of the Natural Environment Research Council and the University of Portsmouth (UK). David Keefer, Randall Jibson and Ed Harp (all United States Geological Survey) are acknowledged for numerous discussions. David Petley of the University of Durham is gratefully acknowledged for allowing me to draw on his extensive knowledge of applied geomorphology.

References 1.

2. 3. 4. 5. 6. 7. 8. 9.


Aki K, (1988). Local site effects on strong ground motion. In: Van Thun, J.L., (Ed), Earthquake engineering and soil dynamics - recent advances in ground-motion evaluation. ASCE Geotechnical Special Publication, no. 20, 103-155. Aki, K. & Larner, K.L. (1970). Surface motion of a layered medium having an irregular interface due to incident plane SH-waves. Jl Geophys Res, 75, 1921-1941. Ambraseys, N.N. & Srbulov, M., (1995). Earthquake induced displacements of slopes. Soil Dyn Earthq Eng, 14, 59–71. Arias, A., (1970). A measure of earthquake intensity. In: Seismic Design for Nuclear Power Plants. Massachussetts Institute of Technology Press, Cambridge, MA,. 438–483. Ashford S. A. & Sitar N. (1997). Analysis of topographic amplification of inclined shear waves in a steep coastal bluff. Bull Seis Soc Amer, 87, (3), 692-700. Ashford S.A., Sitar, N., Lysmer J. & Deng N. (1997). Topographic effects on the seismic response of steep slopes. Bull Seis Soc Amer, 87, (3), 701-709. Athanasopoulos G.A., Pelekis, P.C. & Leonidou, E.A. (1999). Effects of surface topography on seismic ground response in the Egion (Greece) 15 June 1995 earthquake. Soil Dyn Earthq Engng. 18, 135-149. Bieniawski, Z. T. (1989). Engineering Rock Mass Classification. Wiley, New York. Bommer, J. J., Benito, M. B., Ciudad-Real, M., Lemoine, A., Lopez-Menjivar, M. A., Madariaga, R., Mankelow, J. M., Mendez de Hasbun, P., Murphy, W., Nieto-Lovo, M., Rodriguez-Pineda, C. E. & Rosa, H. (2002). The El Salvador earthquakes of January and February 2001: context, characteristics and implications for seismic risk. Soil Dyn Earthq Eng. 22, 389–418. Bommer, J. J. & Rodriguez, C. E. (2002). Earthquake-induced landslides in Central America, Engng Geol, 63, (3-4), 189-220.

153 11.

12. 13.

14. 15. 16. 17.

18. 19. 20. 21 . 22. 23.

24. 25. 26.

27. 28. 29. 30. 31.

32. 33. 34. 35. 36.

Bommer, J., Rolo, R. & Mendez, P. (1998). Propiedades mecanicas de la tierra blanca y la inestabilidad de taludes. Rev ASIA, 129, 15–21. San Salvador, El Salvador: Asociacion Salvadorena de Ingenieros y Arquitectos. Bouchon, M. (1973). Effect of topography on surface motions. Bull Seis Soc Amer, 63, (3), 615-632. Castaldini, D., Genevois, R., Panizza, M., Puccinelli, A., Berti, M. & Simoni, A. (1998). An integrated approach for analysing earthquake-induced surface effects: a case study from the Northern Apennines, Italy, Jl Geodyn, 26, (2-4), 413-441. Cluff, L. S. (1971). The Peru earthquake of May 31 1970: engineering geology observations. Bull Seis Soc Amer, 61, (3), 511-533. Crespellani, T., Madiai, C. & Maugeri, M., (1996). Analysis of the stability of a slope under seismic and post seismic conditions. Rivista Italiana di Geotecnica, XXX (1), 50–61. (In Italian). Crozier, M. J., Deimel, M. S. and Simon, J. S. (1995). Investigation of earthquake triggering for deepseated landslides, Taranaki, New Zealand, Quat Int, 25, 65-73. Faccioli, E., Vannini, M & Frassine, L. (2002). "Complex" Site Effects in Earthquake Ground Motions, Including Topography. 12th European Conference on Earthquake Engineering, Paper Reference 844. Elsevier Science. Geli, L., Bard, P. Y. & Beatrice, J. (1988). The effect of topography on earthquake ground motion: a review and new results. Bull Seis Soc Amer, 78, (1), 42-63. Hanks, T.C. & Kanamori, H., (1979). A moment magnitude scale. Jl Geophys Res, 84, 2348–2350. Hoek, E. (1990). Estimating Mohr-Coloumb friction and cohesion values from the Hoek-Brown failure criterion. Intl Jl Rock Mech, Min Sci Geomech Abs, 27, 227-229. Hoek, E. (1994). The strength of rock and rock masses. News Journal, International Society of Rock Mechanics, 2, 4-16. Hoek, E. & Brown, E. T. (1997). Estimation of rock mass strength. Intl Jl Rock Mech, Min Sci Geomech Abs, 34, 1165-1186. Huang, C. C., Lee, Y. H., Liu, H, P., Keefer, D. K. & Jibson, R. W. (2001). Influence of surface-normal ground acceleration on the initiation of the Jih-Feng-Erh-Shan Landslide during the 1999 Chi Chi, Taiwan, earthquake. Bull Seis Soc Amer. 91, (5), 953-958. Hutchinson, J.N., Del Prete, M., (1985). Landslides at Calitri, southern Apennines, reactivated by the earthquake of 23rd November 1980. Geol Appl e ldrogeologia, XX (1), 9–38. Jibson, R. W. (1987). Summary of research on the effects of topographic amplification of earthquake shaking on slope stability. United States Geological Survey Open File Report 87-268. Jibson, R.W., (1993). Predicting earthquake-induced landslide displacement using Newmark’s sliding block analysis. Transportation Research Board, National Research Council, Washington D.C., TR record 1411, 9–17. Jibson, R.W. & Keefer, D.K., (1993). Analysis of the seismic origin of landslides examples from the New Madrid seismic zone. Geol Soc Ame Bulletin, 105, 521–536. Keefer D. K. (1984). Landslides caused by earthquakes. Geo Soc Amer Bul, 95, 406–21. Mankelow, J. M. and Murphy, W. (1998). Using GIS in the probabilistic assessment of earthquaketriggered landslide hazards. Jl EarthqEnging. 2, (4), 593-623. Murphy, W. (1995). The geomorphological controls on seismically triggered landslides during the 1908 Straits of Messina earthquake, southern Italy. Q. Jl Engng Geol, 28 (1) 61-74. Murphy, W. Petley, D. N., Risdon, G. & Chen, M. C. (2000). Topographic Effects Influencing Large Scale Slope Failure During the 21st September Chi Chi Earthquake, Taiwan, 1999. Proceedings of the Second EuroConference on Global Change and Catastrophe Risk Management: Earthquake Risks in Europe, International Institute for Advanced Systems Analysis, Laxenburg, Austria, 6–9 July 2000. Murphy, W., Petley, D. N, Bommer, J. J. & Mankelow, J. M. (2002). Seismological uncertainty in the assessment of slope stability during earthquakes. Q Jl Engng Geol. 35, 71-78. Murphy, W. and Mankelow, J. M. (in press). Obtaining probabilistic estimates of displacement on a landslide during future earthquakes. Jl Earthq Engng, (in press). Newmark, N. M., (1965). Effects of earthquakes on dams and embankments. Geotechnique, 15, (2), 139–159. Paolucci, R., Faccioli, E. & Maggio, F. (1999). 3D response analysis of an instrumented hill at Matsuzaki, Japan, by a spectral method. Jl of Seism, 3, 191-209. Paolucci, R & Rimoldi, A. (2002). Seismic Amplification For 3-d Steep Topographic Irregularities. 12th European Conference on Earthquake Engineering, Paper Reference 087. Elsevier Science.

154 37. 38. 39. 40.

41. 42 . 43.

44. 45. 46.

47. 48. 49. 50. 51.

Rodriguez C. E., Bommer J. J. & Chandler R. J. (1999). Earthquake-induced landslides: 1980–1997. Soil Dyn Earthq Eng, 18, 325–46. Romeo, R. (2000). Seismically induced landslide displacements: a predictive model, Engng Geol, 58, (3-4), 337-351. Rymer, M. J. (1987). The San Salvador Earthquake of October 10, 1986 – Geologic Aspects. Earthquake Spectra, 3, 435-463. Rymer, M.J. & White, R. A. (1989). Hazards in El Salvador from earthquake-induced landslides. In: Brabb E.E. & Harrod B.L., (Eds). Landslides: extent and economic significance. 105-109. Rotterdam: Balkema. Sabetta, F. & Pugliese, A. (1987). Attenuation of peak horizontal ground acceleration and velocity from Italian strong motion records. Bull Seis Soc Amer, 77, 1491-1511. Sanchez-Sesma, F.J., Bravo, M.A. & Herrera, J. (1985). Surface motion of topographical irregularities for incident P, SV, Rayleigh waves. Bull Seis Soc Amer, 75, 263-269. Sanchez-Sesma, F.J., Chavez-Perez, S. & Aviles, J. (1994). Scattering of elastic waves by threedimensional topographies. Proceedings of the 8th World Conference on Earthquake Engineering, San Francisco, II, 639-646. Sarma, S. (1975). Seismic stability of earth dams and embankments. Geotechnique, 25, 743-761. Shin, T.C., Kuo, K.W., Lee, W.H.K., Teng, T.L., Tsai, Y.B., (2000). A preliminary report on the 1999 Chi-Chi (Taiwan) earthquake. Seis Res Let.. 71, 23–29. Spudich, P., Hellweg, M. & Lee, W.H.K. (1996). Directional Topographic Site Response at Tarzana Observed in Aftershocks of the 1994 Northridge, California, Earthquake: Implications or mainshock Motions. Bull Seis Soc Amer, 86, 1B Suppliment, S193-S208. Tibaldi, A, Ferrari, L. & Pasquarè, G. (1995). Landslides triggered by earthquakes and their relations with faults and mountain slope geometry: an example from Ecuador. Geomorph, 11, (3), 215-226 Vahdani, S. & Wikstrom, S. (2002). Response of the Tarzana strong motion site during the 1994 Northridge earthquake. Soil Dyn Earthq Eng, 22, 837–848. Wasowski, J. & Del Gaudio, V. (2000). Evaluating seismically induced mass movement hazard in Caramanico Terme (Italy). Engng Geol, 58, (3-4), 291-311. Wong, H.L. & Trifunac. M.D. (1974). Scattering of plane SH-wave by a semi-elliptical canyon. Jl Earthq Enging Soil Dyn, 3,157-169. Wong, H.L.(1982). Effect of surface topography on the diffraction of P, SV and Rayleigh waves. Bull Seis Soc Amer, 72, 1167-1183.


APPLICATION OF GROUND-BASED RADAR INTERFEROMETRY TO MONITOR AN ACTIVE ROCKSLIDE AND IMPLICATIONS FOR EMERGENCY MANAGEMENT N. CASAGLI*, P. FARINA University of Firenze, Italy, Earth Sciences Department I-50121, Via G. La Pira, n.4, Firenze (Italy) D. LEVA, D. TARCHI European Commission, Joint Research Centre, IPSC/HSU I-21020, Via E. Fermi n.1, Ispra (VA, Italy) Abstract A rockslide of about 106 m3 was reactivated in April 2002 on Monte Beni near the town of Firenzuola in Northern Tuscany (Italy). The rockslide caused the evacuation of 3 private houses and the interruption of the Regional Road n.65 “della Futa”, one of the main transportation routes connecting Firenze and Bologna. During the emergency, in order to rapidly acquire data on the state, distribution and style of activity of the moving rockslide, a monitoring campaign was carried out, using an innovative radar device capable of a remote sensing assessment of ground displacement fields with a high resolution and accuracy. The system is a ground-based radar interferometer, known as LISA (Linear Synthetic Aperture radar), which has been successfully tested in past experiences for landslide monitoring. The production of multi-temporal maps of ground displacements, over a time span of 5 days, has provided a clear picture of the rockslide mechanism and activity. These data were utilised by public authorities and decision makers to define temporary measures for risk reduction and risk management. 1. Introduction Rockslides are very hazardous slope movements which often evolve in catastrophic failures [21]. Sometimes they are activated as first-time failures and, in this case, the brittle nature of the shear strength is the main cause in the production of a sudden acceleration, leading to high velocities and large displacements. However, in the case of a reactivation of pre-existing shears, a rockslide can be catastrophic. Hutchinson [11] lists several of the factors producing large and rapid displacements in those slides which move along pre-existing shears. Many of these factors regard rockslides. The Vaiont case is probably the best known example of catastrophic reactivation of a pre-existing slide. For these reasons, when a rockslide initiates, it is extremely difficult to predict its *

E-mail of corresponding author: [email protected]

157 S.G. Evans et al. (eds.), Landslides from Massive Rock Slope Failure, 157–173. © 2006 Springer. Printed in the Netherlands.

158 consequences. The preparation of emergency plans is thus conditioned by the uncertainty of the short-term evolution of the failure. The monitoring of rockslide displacements provides key information for short-term and long-term predictions and for defining risk scenarios. Voight [20] and Fukuzono [9] proposed models and graphical methods to predict the timing of a catastrophic failure on the basis of the time-series of landslide displacements. Generally, the monitoring of active landslide movements is not easy, since most of the monitoring methods require the installation of instruments, sensors or benchmarks (control points: CPs) directly in the moving area. All the currently available techniques, both in the geotechnical and in the topographic realm, provide information about displacements only at selected points where the CPs are positioned. Therefore it is absolutely crucial that CPs are placed in locations where they can record movements that are significant in the landslide activity. Often the landslide displacement itself is the cause of the failure of the monitoring system. This is why in the majority of cases, the most active landslide sectors remain uncontrolled. Very often, large rockslides cannot be stabilised at a reasonable cost for the community. Therefore, the only measure to reduce the risk in settled areas, or in those sites of a high societal value, is to set-up an early-warning system associated with an emergency plan. Again, monitoring is the key element for any warning system. Current warning systems are based on costly and complex networks of CPs. For a variety of reasons, they do not work properly and are affected by several operational problems including: a) lack of maintenance; b) loss of CPs placed in sites with high landslide activity; c) CPs placed in location where they do not record significant information; d) the efficiency of the network and of the associated infrastructures (devices for data transmission, collection, storage and real-time interpretation of thresholds) is particularly vulnerable during emergency periods (e.g. during extreme meteorological events); e) warning systems are not very well understood by people, who are reluctant to accept the implementation of the emergency plan (e.g. evacuation) especially after a series of false alarms; moreover people and mass media prefer to see public money spent for structural measures and to trust “simple” engineering solutions rather than a warning system. Some of these limitations are technical and can be overcome through technological development. In this paper we present an example of how a novel technology can be usefully employed in a practical problem of landslide emergency management. 2. The Monte Beni Rockslide Monte Beni (elevation 1260.5 m a.s.l.) is located in Northern Tuscany (Italy), about 8 km on the NW of Firenzuola, a town well-known for the industrial production of stones from quarrying activity (Figure 1).


Road No.65



N ad Ro



Mt. Cimone 2165 m

MONTE BENI Firenzuola

Passo della Futa




Regional boundary

Regional Road

Figure 1. Location map of the investigated area.

The eastern slope of Monte Beni (slope direction 080°N), shown in Figure 2, was developed by quarrying activity, to produce construction material, from an ophilolitic sequence, which lies reversed over Cretaceous clay-shales. Figure 2 shows a complete view of the sequence which is composed, from the top to the bottom, of the following units: pc

mB pB

tf sd

bB L




Figure 2. General view of the eastern slope of Monte Beni from the relief of Pietramala (where the LISA was installed) showing the main geological and geomorphological features. mB = massive basalts; pB = pillow basalts; bB = basalt breccias; L = limestones; dc = debris cone; pc = perimetral crack; tf = thust fault with laminated cherts; sd = slope deposits (vegetated).

160 a) b) c) d) e)

jointed massive basalts; basalts with a pillow-lava structure; breccias of basalts; thin layer of laminated cherts (Diaspri Formation); limestones with marl inter-beds (Calcari a Calpionelle Formation) with a regular bedding dipping into the slope (238/32°). A thrust fault dipping 240/10° is located at the base of the breccias and caused the lamination of the cherts. The lower part of the relief is covered by slope-waste deposits with visible remains of ancient rockslides. The contours of discontinuity poles present in the sequence are shown in Figure 3. Four main joints sets are evident and dip respectively 065/76°, 120/80°, 019/78° and 159/82°. The quarry activity, developed in both the basalt and limestone layers, started on an industrial basis in the ‘40s, rose in intensity during the Post-War period and reached a maximum in the ‘80s. After a series of slope instability problems, associated with superficial rockfalls, the stone extraction was progressively reduced and limited to the removal of selected unstable masses, until it was definitively ceased in 1998.




Figure 3. Contours of pole density of discontinuities represented on the stereographic projection. (dd) = slope dip direction; (s)-(s) = slope strike.

A geological study [4], commissioned by the Firenzuola Municipality for the recovery of the quarry area, finalised in July 2000, confirmed the well-known instability problems associated with the fall of blocks of a few cubic metres but, for the first time, recognized the presence of a mass of heavily jointed rock in precarious stability conditions, located in the top part of the mountain and involving a volume estimated to be about 200,000 m3. A global stabilisation program was commissioned by the Municipality but, while this was still in preparation, in April 2002 after a moderate rainstorm (66.4 mm of rain between the 12nd and the 13rd of April), new dramatic evidences of instability appeared on the slope:

161 a) the activity of rockfalls, feeding two main debris cones, was markedly intensified; b) large fissures were opened in the upper part of the slope; c) underground noises were frequently heard. After a field survey carried out on the 16th of April, on behalf of the Civil Protection Survey of the Provincial Administration, it was possible to assess the actual dimension of the problem. The upper part of Monte Beni was affected by a rockslide, hanging over the quarry floor and involving an estimated volume of 1 - 2 × 106 m3. The upper boundary and the right flank of the rockslide were clearly defined by an opened perimetral crack, 320 m long. Two hypotheses were proposed for the lower and left boundaries (Figure 4): A) Best case: the basal sliding surface is placed within the basalt breccias whereas the left flank follows a joint. In this case the maximum depth of the sliding surface would be about 30 m leading to an estimation of the mobilised volume of about 1 × 106 m3; B) Worst case: the basal sliding surface is placed along the thrust fault on the top of the limestones and the left flank is controlled by a joint which is located outside the joint in A). In this case the maximum rockslide depth would be 60 m and the involved volume would be almost doubled, reaching 2 × 106 m3. In both cases, the estimation of the rockslide maximum depth was based on structural geology data and on the results of a seismic reflection survey carried out in the upper part of Monte Beni.

rim pe

ra et

ck ra c l

Hypothesis “A": best case

worst case Hypothesis “B":

Figure 4. Monte Beni showing the extension of the perimetral crack and the two hypotheses on the basal and left flank of the rockslide.

162 A planimetric view of the rockslide boundaries (based on hypothesis “A”) and of the main elements at risk is provided in Figure 5. Just after the first survey on the 16th of April it was decided to evacuate 3 private houses and to close the Regional Road n.65 “della Futa”, one of the main transportation routes between Firenze and Bologna.



Figure 5. Ortho-rectified aerial photograph of Monte Beni showing the presumed rockslide boundaries (according to hypothesis A), the abandoned quarry and the main elements at risk at the slope toe (the Regional road and three private houses).

It is very important to assess rapidly and precisely the volume of an incipient rockslide because volume is one of the main factors controlling the runout in the case of a catastrophic failure [16] and, therefore, the consequences in terms of expected loss. In particular the value of 1 million m3 seems to be a limit between low mobility and high mobility rock avalanches [16, 10]. Further all the empirical rules and the numerical models developed for the runout prediction of a rock avalanche, require volume as one of the main input data. For this reason it was very important to decide between the two hypotheses in order to provide decision makers with a reliable scenario for the preparation of the emergency plan and the implementation of the initial risk mitigation measures. Further data were necessary for assessing the actual rockslide boundary and these data could be obtained only by monitoring. Unfortunately, with the exception of the landslide head scarp, the rest of the rockslide body was not accessible for the installation of benchmarks or monitoring instrumentation due to the landslide movement.

163 For this reason it was decided to employ a new remote sensing technique, based on SAR (Synthetic Aperture Radar) interferometry, developed at the Joint Research Centre (JRC) of the European Commission, specifically for the field assessment of ground surface displacements. This novel system was used at Monte Beni in a monitoring campaign lasting 2 weeks, from the 5th to the 19th of May, with the following main purposes: a) assess the actual boundary of the moving zone and to decide between the two different hypotheses defined by the geological survey; b) assess the displacement field of the rockslide and validate the data acquired through a conventional monitoring network set up just after the first recognition of the hazard. 3. Conventional Monitoring In order to assess the evolution of the rockslide, a monitoring system was put in place directly on the day of the first survey (16th of April). This system is composed of 24 distometric bases installed across the perimetral crack which is accessible via a pathway running along most of the rockslide crown. Measurements of elongation of the bases have been carried out through a portable distometer with a daily frequency in the first phase and then, after the first two weeks, once per week. This simple system was set in place in order to obtain data during the first phase of emergency and to allow the installation of a permanent monitoring system by the local Civil Protection authorities. However for a series of reasons, it was not possible to complete this permanent system (composed of automatic extensometers) until November 2002 and so the manual distometers remained for 8 months the unique source of information on the rockslide activity (with the exception of the two weeks of radar measures described in the following sections). The data collected at the distometers over a period of one month are shown in Figure 6 as curves of cumulative displacements versus time. These data show quite a constant displacement rate with a maximum of 5mm/day. 4. Synthetic Aperture Radar Monitoring LISA (Linear Synthetic Aperture Radar) is a radar interferometer recently developed by the JRC for monitoring superficial displacements of slopes and engineering structures, directly in the field [15, 19]. The basic principles behind the system derive from the SAR interferometric technique developed in the last decades in the realm of satellite imagery [22, 13]. LISA is however a ground-based portable apparatus composed of a motorised sled, carrying a transmitting and a receiving antenna, which slides along a straight rail, about 3 m long (Figure 7). The synthetic aperture is realised by acquiring raw data at selected positions along the rail. The microwave source is provided by a network analyser operating in the frequency range between 30 kHz and 6 GHz.


Figure 6. Cumulated displacements assessed by the manual distometers. The time window covered by radar measures is shown shaded.

receiving antenna

transmitting antenna


linear rail Figure 7. Pictures showing the field arrangement of the LISA apparatus installed in Pietramala just in front of the Monte Beni rockslide.

165 Observations are usually carried out in the Ku-band (frequency between 12.5 and 18 GHz) by using a coherent conversion module. A complex radar image of the target area, containing both amplitude and phase information at discrete space intervals (pixels), is obtained through a single scan of the aperture length, which can be equal or less than the rail length. The system must be installed in a stable position in front of the area to be monitored, which must be completely visible from this position. The entire operational chain of data processing can be done directly in the field using portable instrumentation and a laptop PC, thus providing a real-time availability of information. The data processing procedure is described in detail by Tarchi et al. [17, 18] and includes: a) acquisition of raw radar data through a single scan at a given time b) calibration of the raw data by using information taken from periodic measurements on a metallic disc placed at a known distance and orientation from the LISA; c) focalization and geocoding of the radar image on a high resolution digital elevation model of the slope in a geographic co-ordinate system; This series of operations is repeated at selected time intervals in order to have a sequence of focalised radar images. The following further processing operations are carried out in this sequence: a) coherent averaging of a number of consecutive images in order to increase the signal-to-noise ratio; b) coherence analysis in order to distinguish the pixels where the phase measurement is significant (and representative of ground deformations) from those where the phase is disturbed by vegetation movements, by rapid changes in the target dielectric constants, or by atmospheric effects; c) application of a two-levels mask in order to neglect pixels where the phase information is not reliable or significant on the basis of the analysis of coherence; d) interferometric analysis carried out by comparing the phase information of consecutive averaged images. In particular an interferogram (I) is obtained from a pair of complex SAR images, where the former and the latter are referred to as master (m) and slave (s) images respectively, according to the following relationship: (1) I (k , l ) m(k , l ) ˜ s(k , l )* where * indicates the conjugate. The final product of the methodology is a sequence of interferograms showing, pixel by pixel, the phase differences between successive images. This phase difference is directly converted in ground displacements using the relationship:

O (2) M 4S where 's is the displacement along the line-of-sight, O is the wavelength and Mis the interferometric phase. The direct correspondence between phase difference and displacements is due to the following: a) images are taken exactly from the same position (zero baseline condition) thus avoiding the phase component related to the topography of the observed scene; b) successive images are taken at short time intervals (few minutes) and for this reason the derived interferograms are not affected by temporal decorrelation, which 's

166 is one of the main limitations to the use of satellite SAR data and which is due to the combined influence of atmospheric effects, noise produced by vegetation movements, and changes in the dielectric properties of the target. Interferograms, after the transformation of phase values into displacement, represent directly multitemporal deformation maps showing the field of displacements along the line-of-sight of the system over all the area where the radar backscattering is coherent enough (corresponding to the less vegetated sectors of the slope). The technique has been already validated for landslides differing in type, material and mechanism of movement [1, 2, 3, 5, 6, 7, 8, 14, 18, 17]. The comparison with independent monitoring data has showed that LISA is more accurate and robust than most of the current monitoring systems, such as GPS (global positioning system) or EDM (electronic distance meters). The main proven advantages of the LISA, which are also elements of innovation in the realm of landslide monitoring, are the following: a) remote sensing character of the technique which does not require the installation of benchmarks or reflectors in the moving area; b) capability of providing displacement fields over extensive areas rather than point information. 5. The SAR Monitoring Campaigns at Monte Beni From the 5th to the 19th of May, the LISA Prototype N.2 of the JRC, with a rail length of 2.85 m, was installed at Pietramala in front of the eastern slope of Monte Beni. The operational parameters employed are summarised in Table 1. Observations were carried out in the frequency band between 16.70 and 16.76 GHz, at frequency steps of 50 kHz, allowing the collection of 1601 frequency points, with VV polarization along the synthetic aperture. The transmitted power was of 300 mW (equal to about 25 dbm). The synthetic aperture was 2.85 m scanned at azimuth steps for 6 mm for a total of 476 azimuth points. The acquisition of the raw data for each image took an average of 40 minutes. The LISA was installed on a hill in front of the Monte Beni slope at an average distance of 1800 m (Figure 7). Table 1 – Operational parameters of the LISA system used in the radar campaign at Monte Beni. Frequency band: 16.70 – 16.76 GHz Frequency step: 50 kHz Frequency points: 1601 Polarization: VV Transmitted power: 0.3 W (25 dbm) Synthetic aperture: 2.85 m Azimuth step: 6 mm Azimuth points: 476 Measuring time per image: 40 min Number of collected images: ~1000 Target average distance: 1800 m Original spatial resolution: 5 u 2.5 m Expected accuracy: < 0.75 mm

167 According to these operational parameters the spatial resolution of the images is 5 m u 2.5 m on average, whereas the expected accuracy in the detection of displacement is lower than 0.75 mm. The resolution is slightly degraded to about 5 m u 5 m after the averaging procedures applied to increase the signal-to-noise ratio. More than 1000 radar images, arranged in a number of sequences were collected in two weeks. The longest continuous series of interferograms was obtained between 8/5/02 13:59 h and 13/05/02 18:12 h, spanning a total interval of 5 days, 4 hours and 13 minutes. Four interferograms of this sequence are shown in Figure 8. Figure 9 shows an enlargement of the last interferogram.


Figure 8. Three-dimensional representations of the sequence of cumulated displacements assessed by the LISA on Monte Beni from 8/5/02 13.59 and 13/05/02 18:12. Top left: 9/05/2002 14:05; Top right: 10/05/2002 16:45; Bottom left: 12/10/2002 20:10; Bottom right: 13/05/02 18:12 (end of the sequence).


DISPLACEMENT ALONG THE LISA LINE-OF-SIGHT (mm) Figure 9. Enlargement of the last image referred to the 13/05/02 18:12 after 5 days, 4 hours and 13 minutes of continuous monitoring.

The atmospheric conditions during this period were quite difficult since for most of the time Monte Beni was covered by dense fog. In these conditions it was not possible to use any optical topographic system whereas the radar observations were completely unaffected by fog. Each interferogram has been processed over the digital terrain model which was obtained by digitizing a 1:2,000 scale contour map. The scale in Figures 8 and 9 expresses the component of displacement, in millimetres, in the direction of the line-ofsight of the instrument. The pale grey zones correspond to those areas where the phase information was disturbed by vegetation and has been masked. The sequence allows us to follow in detail the evolution of the rockslide in space and in time. The high displacements concentrated in two zones at the scarp foot are related to rockfalls feeding the two debris cones on the quarry floor. The activity of the cone on the right flank of the rockslide (on the left side of the figures) is particularly

169 evident and it is confirmed by direct observations since this cone has doubled its initial volume in the span of 1 month. More coherent and extensive displacements affect the upper part of Monte Beni, within the basaltic layers, and these are related to the motion of a deep seated rockslide. The displacement pattern, also shown in planimetric projection in Figure 10, seems to confirm hypothesis “A” formulated in Section 2. Four points are spotted in Figure 10 and their history of cumulated displacements is shown in Figure 11. Point 1 is located on the debris cone on the right flank and shows a cumulated displacement of 57 mm over the investigated period (corresponding to a velocity of about 11 mm/day). Points 2, 3 and 4 are located on basalt bedrock and their displacements are representative of the rockslide motion. The highest displacement (37 mm, velocity of 7 mm/day) has been recorded at point 2. It is particularly significant that point 3, located near the perimetral crack , shows the same rate of displacement (5 mm/day) as the one observed on the distometer a-b which is placed at a very close distance across the crack (Figure 6).

Figure 10. Horizontal projection of the final map of cumulated displacements showing the position of 4 points where the history of displacement was assessed.

6. Implications for Emergency Management The data provided by the LISA define a clear picture of the rockslide mechanism and have allowed us to define, with high precision, the boundaries of the moving mass and the evolution of the movement in space and in time.


Figure 11. History of cumulative displacement in the 4 selected points shown in Fig.10.

As mentioned in the previous section, the results confirmed hypothesis “A” defined after the first field survey on the basis of a geomorphologic evidence only. After the LISA campaign it was possible to assess the effective magnitude of the rockslide confirming the estimate of 1 u 106 m3 of material involved. A series of risk scenarios were prepared for the Civil Protection Authorities based on this magnitude. For example a prediction of the runout area is represented in Figure 12 considering the failure of 1 u 106 m3 of rock and a travel angle of 30°. This value comes from an average of the observed travel distances of rock avalanches with comparable volume [16, 10, 12]. The scenario is obtained simply from the interception of the digital terrain model with a plane inclined 30° passing through the rockslide crown. The area affected by the rockslide runout would reach the Regional Road and two of the three houses located along the roadway. For this reason it was decided to maintain the closure of the road to traffic and the evacuation of the houses. 7. Conclusion Monte Beni represents the first case in which the new SAR technology of LISA has been applied in real operational conditions as a decision support to Civil Protection authorities.


Figure 12. Prediction of the most credible runout of the rockslide considering a volume of 1 u 106 m3 and a travel angle of 30°.

Other experimental campaigns carried out before, were aimed at the full validation of the technique and to its comparison with different traditional methods of landslide monitoring [1, 2, 3, 5, 6, 7, 8, 14, 18, 17]. After this long series of validation experiments, LISA is revealed as a powerful tool for the short-term monitoring of active movements. From this first operational test the following positive aspects arise: a) LISA can be used for monitoring movements during an emergency since its installation requires only a few hours and it is able to provide real-time data; b) LISA is capable of assessing the entire deformation field of a mass movement, at least in zones with scarce or moderate vegetation cover, providing high quality information on landslide boundaries, movement patterns and distribution of activity; c) LISA delivers multitemporal information at short time intervals (every 40 minutes in the investigated case, but it is possible to acquire up to an image every 10 minutes where the movement conditions require a higher observation frequency) and it represents a valid complement, or alternative, to traditional monitoring devices; d) LISA remotely assesses displacements, without the need of accessing the unstable area. This is a fundamental benefit for many cases where traditional techniques cannot be employed because of safety conditions.

172 Acknowledgements The authors wish to thank the Civil Protection Department of the Firenze Provincial Administration and particularly the director Dr. Luigi Brandi who was in charge of the definition of the emergency plan. The Mayor and the technical staff of the Municipality of Firenzuola are also acknowledged for the help provided to our study. The Parish of the Church of Pietramala gave us the permission to install the LISA inside the Church property. Prof. Canuti and his collaborators, Dr. Iotti and Dr. Tarchiani, are acknowledged for providing data on geology, geomechanical properties and on the monitoring system. Dr. Gigli and Dr. Nocentini, research assistants at the University of Firenze, are acknowledged for the help in the interpretation of geological and geostructural data. The research on new technologies for landslides monitoring is funded by the European Commission an by the National Research Council Group for Hydro-geological Disaster Prevention (CNR-GNDCI). References 1.



4. 5.




9. 10. 11. 12. 13.

Atzeni, C., Basso, M., Canuti, P., Casagli, N., Leva, D., Luzi, G., Moretti, S., Pieraccini, M., Sieber, A.J. and Tarchi, D. (2001) Ground-based SAR interferometry for landslide monitoring and control. Proc. ISSMGE Field Workshop on Landslides and Natural/Cultural Heritage. Trabzon (Turkey), 195209. Atzeni, C., Canuti, P. and Tarchi, D. (2001) Monitoring unstable cultural heritage sites with radar interferometry. In: K. Sassa (ed.), Proc. UNESCO/IGCP Symp. on Landslide Risk Mitigation and Protection of Cultural and Natural Heritage. Tokyo, Japan, 257-264. Atzeni, C., Canuti, P., Casagli, N., Leva, D., Luzi, G., Moretti, S., Pieraccini, M., Sieber, A.J. and Tarchi, D. (2002). Ground-based radar interferometry: a novel technique for monitoring unstable slopes and cliffs. In: R.G. McInnes and J. Jakeways (Editors), Instability planning and management: Seeking sustainable solutions to ground movement problems. Thomas Telford, London, 447-454. Canuti, P. (2000) Cava di Monte Beni: Relazione Geomeccanica. Commissione di studio per la messa in sicurezza di Monte Beni, Comune di Firenzuola. 36 pp. Canuti, P., Casagli, N., Leva, D., Moretti, S., Sieber, A.J. and Tarchi, D. (2002) Landslide monitoring by using ground-based radar interferometry. In: J. Rybar, J. Stemberk and P. Wagner (eds.), Landslides. Proceedings 1st European Conference on Landslides. Prague, Balkema, 523-528. Canuti, P., Casagli, N., Leva, D., Moretti, S., Sieber, A.J. and Tarchi, D. (2002) Some applications of ground-based radar interferometry to monitor slope movements. Proc. Int. Symp. Landslide Risk Mitigation and Protection of Cultural and Natural Heritage. UNESCO and Kyoto University, 357-374. Casagli, N., Farina, P., Leva, D., Nico, G. and Tarchi, D. (2002) Monitoring the Tessina landslide by a ground-based interferometer and assessment of the system accuracy. Proc. IGARSS 2002 International Geoscience and Remote Sensing Symp. Toronto, Canada. Casagli, N., Farina, P., Leva, D., Nico, G. and Tarchi, D. (in press) Landslide monitoring on a short and long time scale by using ground-based SAR interferometry. Proc. 9th Int. Symp. on Remote Sensing for Environmental Monitoring, GIS Applications, and Geology. Aghia Pelagia, Crete Greece,. International Society of Optical Engineering (SPIE). Fukuzono, T. (1985) A new method for predicting the failure time of a slope failure. Proc. 4th Int. Conf. and Field Workshop on Landslides, Tokyo (Japan), 145-150. Hungr, O. (2002) Analytical models for slides and flows. Proc. Int. Symp. Landslide Risk Mitigation and Protection of Cultural and Natural Heritage. UNESCO and Kyoto University, 559-586. Hutchinson, J.N. (1987) Mechanisms producing large displacements in landslides on pre-existing shears. 1st Sino-British Geol. Conf., Tapei, Memoir of the Geological Survey of China, 9, 175-200. Legros, F. (2002) The mobility of long-runout landslides. Eng. Geology, 63, 301-331. Massonnet D. and Feigl K.L. (1998) Radar interferometry and its applications to changes in the Earth’s surface. Reviews of Geophysics, 36(4), 441-500.

173 14.

15. 16. 17.


19. 20. 21. 22.

Pieraccini, M., Casagli, N., Luzi, G., Tarchi, D., Mecatti, D., Noferini, L. and Atzeni, C., in press. Landslide monitoring by ground-based radar interferometry: a field test in Valdarno (Italy). Int. J. Remote Sensing. Rudolf, H., Leva, D., Tarchi, D. and Sieber, A.J. (1999) A mobile and versatile SAR system, Proc. IGARSS’99 International Geoscience and Remote Sensing Symp., Hamburg, 592-594. Scheidegger, A. E. (1973) On the prediction of the reach and velocity of catastrophic landslides. Rock Mechanics, 5, 231-236. Tarchi, D. Casagli N., Leva, D., Moretti, S. and Sieber, A.J. (in press) Monitoring landslide displacements by using ground-based differential SAR interferometry: application to the Ruinon landslide in the Italian Alps. J. Geoph. Research. Tarchi, D., Casagli, N., Fanti, R., Leva, D., Luzi, G. Pasuto, A., Pieraccini, M. and Silvano, S. (in press) Landslide monitoring by using ground-based SAR interferometry: an example of application to the Tessina landslide in Italy. Eng. Geology. Tarchi, D., Ohlmer, E. and Sieber, A.J. (1997) Monitoring of Structural Changes by Radar Interferometry. Research in Nondestructive Evaluation, 9, 213-225. Voight, B. (1988) Material science law applies to time forecast of slope failure. Landslide News, 3, 811. Voight, B. (ed.) Rockslides and Avalanches Natural Phenomena. Developments in Geotechnical Engineering Elsevier Sc., Amsterdam, 14, 833 pp. Zebker, H.A. and Goldstein, R.M. (1986) Topographic mapping from interferometric Synthetic Aperture Radar observations. J. Geoph. Research, 91, 4993-4999.

MONITORING AND ASSESSING THE STATE OF ACTIVITY OF SLOPE INSTABILITIES BY THE PERMANENT SCATTERERS TECHNIQUE C. COLESANTI Politecnico di Milano Dip. di Ingegneria Elettronica Piazza L. da Vinci, 32 20133 Milano, Italy G.B. CROSTA1 Dip. Scienze Geologiche e Geotecnologie, Università degli Studi di Milano – Bicocca, Piazza della Scienza 4, 20126 Milano Italy A. FERRETTI Tele Rilevamento Europa, TRE S.r.l. Via V. Colonna 7, 20149 Milano, Italy C. AMBROSI Politecnico di Milano Dip. di Ingegneria Elettronica Piazza L. da Vinci, 32 20133 Milano, Italy Abstract The evaluation of the state of activity is an essential step for landslide hazard assessment. To attribute a state of activity class to a landslide a suitable set of monitoring data is needed. Different types of landslides present different characteristics and are subjected to different spatial and temporal evolution; these become relevant when working at a regional scale or when a catastrophic evolution of the movement is expected. The Permanent Scatterers (PS) technique is used to determine state of activity and long term behavior of rock slope instabilities. The PS technique, overcomes several limitations of conventional differential SAR interferometry (DInSAR) applications, and proves to be effective for high accuracy monitoring of gradual very slow slope deformation which may eventually transform into extremely rapid failures. The success depends on various factors including: available data, location and morphology of the area, PS density; and motion of the targets. We present the results of the application of the technique at a regional scale in the Central Italian Alps involving different types of instabilities in different materials. The utility of the technique in landslide hazard assessment is demonstrated by the quality of the results and by the integration of different datasets (inventory maps, site investigations, remote sensing). 1. Introduction The evaluation of the state of activity is a mandatory step for landslide hazard assessment both at regional and local scale. To correctly attribute a state of activity class 1

E-mail of corresponding author: [email protected]

175 S.G. Evans et al. (eds.), Landslides from Massive Rock Slope Failure, 175–194. © 2006 Springer. Printed in the Netherlands.

176 (i.e. active, quiescent, inactive) to a landslide, a suitable set of monitoring data is needed. Nevertheless, different types of landslides (e.g. rockslides, rock mass creep or deep-seated gravitational slope deformations, etc.) present different characteristics and different spatial and temporal evolution. This aspect becomes more relevant when working at a regional scale, or when a catastrophic evolution of the phenomenon is expected. Consequences of such high variability become more relevant when a new landslide is recognised for which no previous information is available. This can be thought of as a worst-case scenario because the discovery of the landslide could occur too late in the evolution of the phenomenon. Therefore, the lack of a displacement history for the landslide or of any sort of monitoring data, can hamper both the interpretation of the process and the forecast of future development. In this paper we tackle this problem by using Earth Observation Synthetic Aperture Radar (SAR) datasets from currently operating SAR sensors mounted on the ERS-1 and ERS-2 satellites of the European Space Agency (ESA). The results of this approach are then integrated with geological, geomorphological, geomechanical and topographical data to reach a more complete and substantiated conclusion about activity of slope instabilities. The increasing interest in this remote sensing technique is connected to a series of factors, namely, the greater data availability due to the launches of new satellite systems (e.g. European ERS-1 and ERS-2 satellites, ENVISAT, SIR-C/X – Synthetic Aperture Radar flown on-board of NASA’s Space Shuttle, Japanese JERS-1 and ALOS, Canadian RADARSAT and RADARSAT2), the improved capabilities of the space sensors, and the development of more advanced data processing techniques. On the other hand, the most important factor limiting the usefulness of many available earth observation satellite datasets is the coarse ground resolution of space imagery. To increase the awareness of the utility and limitations of radar-based remote sensing we give some basic principles of SAR interferometry and relevant processing techniques, with special reference to the Permanent Scatterers (PS) differential interferometry technique >8, 9, 3@.

2. Basic Principles of Space-Borne SAR Interferometry Satellite radar interferometry involves phase comparison of synthetic aperture radar (SAR) images, gathered at different times with slightly different looking angles >18, 14, 4, 2, 12, 10@. Synthetic Aperture Radar (SAR) is an active microwave device capable of recording the electromagnetic echo back-scattered from the Earth surface. This echo is arranged into a two-dimensional image map, whose dimensions are the sensor-target distance (the slant range or Line of Sight (LOS) direction, which can be projected on the ground to obtain the ground range) and the platform flight direction (along track or azimuth direction). SAR systems work independently of Sun illumination and microwaves can penetrate clouds, and depending on the operating frequency (longer wavelength for deeper penetration), even soil, vegetated canopies, and to some extent snow (up to several cm).


Look angle (J) J

S = G*cos(90-J) = G*cos(\)

Slant range (S)


\ Ground range (G)

N I = 8.5°

Descending orbit

LOS T= 23° Ground range


Azimuth Figure 1. ERS acquisition geometry: images cover a 100 x 100 km area with a ground resolution of 80 m2. LOS = Line of Sight.

The ESA ERS-1 and ERS-2 satellite data, acquired in C-band (carrier frequency 5.3 GHz) according to a specific geometry (Figure1) have been used in this study. ERS-1/2 are characterised by a 35 day revisiting time (i.e. time span between successive overpasses on the same area with the same acquisition geometry). ERS-1 interferometric data cover the time interval from July 1991 to March 2000, whereas

178 ERS-2 was launched in April 1995 and it is still operating. The ERS side-looking SAR sensor images the Earth surface from an orbit at a height of about 780 km illuminating a 100 km wide strip, called a swath, with a constant off-nadir angle (strip map mode) of around 21q (at mid-range). The ERS orbit is nearly polar and data are acquired both along ascending and descending orbits, respectively moving approximately from South towards North and vice versa. The angle between the directions the antenna is pointing to and the nadir is called look angle. The angle between the radar beam centre and the normal to the local topography is referred to as incidence angle, T. This angle, ranges from 19q (near range) to 26q (far range) for flat terrain conditions, and it is slightly larger than the off-nadir angle due to the Earth surface curvature. A SAR image (covering an approximately 100x100 km2 area) is a matrix of complex numbers. Amplitude values are related to local ground reflectivity (i.e. the amount of backscattered energy) whereas phase values are the sum of two contributions: local reflectivity and a quantity proportional to the sensor-target distance. The sensitivity of phase data to the sensor-target distance is extremely high. A twoway path difference of O (i.e. a single way path difference of 0.5O=2.83 cm for ERS and RADARSAT) translates in a full phase cycle (2S). ERS image resolution amounts to about 5m in azimuth direction and 9.5m in slant range. Slant range can be transformed into a ground range by considering the local topography (local incidence angle, T and local terrain slope, D). For flat terrain at mid range (T=23q), this corresponds to a ground range resolution of about 25 m. The concept of resolution should not be confused with the one of sampling step (i.e. the actual size of a single image pixel). The sampling step is 4 m in azimuth and 8 m in slant range (corresponding to 20 m in ground range, for flat terrain at the centre of the swath). Furthermore, the radar ranging mechanism induces a slope dependent resolution (and pixel size) along ground range. This is responsible for geometric distortion effects (Figure 2) which are very important when studying slope instabilities. Foreshortening is the effect occurring when slopes facing the sensor (011@ and the application in suitable areas (slope instabilities above the tree line in alpine areas, >16, 17, 15@, DInSAR cannot be currently considered as an operational tool for landslide monitoring. The probability of success of a DInSAR analysis in detecting slope instability is controlled by different factors, namely, geometric and temporal decorrelation, atmospheric artefacts, scale constraints (the unstable area should include at least a few hundreds of resolution cells), phase ambiguity, aspect and local inclination of the slope. 3. Permanent Scatterers (PS) Technique The PS approach, recently developed and patented for commercial exploitation at Politecnico di Milano (Italy, Italian Patent Office, 1999), is based on a few observations >8, 9, 3@. Atmospheric artifacts show a strong spatial correlation within every single SAR acquisition, but they are uncorrelated in time. Conversely, target motion is usually strongly correlated in time and can exhibit different degrees of spatial correlation depending on the phenomenon at hand (e.g. subsidence due to water pumping, fault displacements, localized sliding areas, collapsing buildings, etc.). Atmospheric effects can then be estimated and removed by combining data from a long time series of SAR images, such as those available in the ESA-ERS archive. In order to exploit all the available images, and then improve the precision of the estimate, only scatterers slightly affected by both temporal and geometrical decorrelation are selected. Therefore, one of the basic principles of the PS techniques consists of the fact that the scattering mechanism of a certain amount of image pixels is dominated by a single point-wise element. The contribution of this element overpowers the coherent sum of all other scattering elements within the same sampling cell. As long as these dominant scatterers correspond to objects whose reflectivity does not vary in time (e.g. man-made structures and rock exposures), temporal decorrelation is also negligible. Possible stable and point-targets, so-called Permanent Scatterers (PS), are then detected on the grounds of the stability of their amplitude returns. This allows pixel-bypixel selection with no spatial averaging. Due to high spatial correlation of the atmospheric contribution, even a sparse grid of measurements may allow proper sampling of the atmospheric components, provided that the PS density is larger than 3-4 PS/km2. Of course, a sufficient number of images should be available (usually more than 20), in order to identify PS and separate the different phase contributions. Even though precise state vectors are available for ERS satellites, the orbit indeterminations and their impact on the interferograms cannot be neglected. Estimated APS (Atmospheric Phase Screen) is actually the sum of two contributions: atmospheric effects and orbital fringes due to baseline errors. However, the latter correspond to loworder phase polynomials and do not change the low-wave number character of the signal to be estimated on the sparse PS grid. At the PS, sub-metre elevation accuracy (due to the wide dispersion of the incidence angles available, usually r 70 millidegrees with respect to the reference orbit) and millimetric terrain motion detection (due to the high phase coherence of these scatterers) can be achieved, once atmospheric contributions are estimated and removed. In particular, the relative target LOS velocity can be estimated with unprecedented accuracy (sometimes even better than 0.1 mm/yr, due to the long time span). The higher

183 the accuracy of the measurements, the more reliable the differentiations between models of the deformation process under study. This is a key issue for landslide hazard and risk assessment. The results of this multi-interferogram approach are: (1) a map of the PS identified in the image and their coordinates - latitude, longitude and precise elevation; (2) the average LOS velocity (deformation rates determined with submillimetric precision, from 0.1 to 1.0 mm/yr of each PS); (3) the estimated LOS motion component of each PS as a function of time (time histories or time series). As in all differential interferometric applications, the results are computed with respect to a Ground Control Point (GCP) of known elevation and motion. Permanent Scatterers can be thought of as benchmarks of a high density geodetic network, forming a sparse grid whose spatial density varies locally and can exceed 500 PS/km2 in urban areas. Because the number of “natural” PS (corresponding to exposed rocks) seems to vary strongly in relation to local lithology and morphology >5@ it is not possible to provide a generally valid figure for the PS density in rural or low urbanisation areas. Tests performed during this study have shown that natural PS density varies from 0-10, 20-50 and up to 200 PS/km2 . It must be stressed that the actual PS density is also a function of the threshold value set on phase stability to identify PS. Usually reliability and precision of the measurements carried out at individual PS are summarised in the so-called single pixel multi-interferogram coherence J, which is a normalised index: 0 d |J| d 1 (thresholding is actually carried out on |J|). The PS approach and, in general DInSAR, presents a major advantage with respect to in situ measurements (e.g. optical levelling and GPS surveys) consisting of the availability of time histories for displacements. This is allowed by the time span covered by the ESA ERS archive gathering data since 1991/92. Some drawbacks of DInSAR techniques still limit the performance of a PS analysis for unlimited use in slope instability studies. The intrinsic ambiguity of phase measurements implies serious difficulties in detecting single PS deformation effects with average LOS rates exceeding 8-10 cm/yr. The theoretical limit, in the absence of noise amounts to O/4 =1.4 cm/35 days, i.e. 14.5 cm/yr. Of course if a priori information (e.g. spatial correlation) of the phenomenon at hand this limit can be overcome. Further limits, in common with conventional DInSAR, are: the fact that the position of "natural" PS benchmarks cannot be chosen (as can coherent areas in conventional interferograms), and the capability of measuring 1D deformation (2D involving both ascending and descending data). Moreover, it is still difficult to anticipate the PS density in rural areas without carrying out at least several processing steps on a significant number (>20) of SAR images. 4. Example of the Integration of Multiple Datasets With PS Data for Landslide Monitoring Landslide hazard assessment requires the definition of the state and style of activity for the specifically analysed phenomenon. Nevertheless, the state of activity is quite difficult to define on the basis of field observations only, especially for relatively slow and continuous movements. For the same type of phenomena, the identification of

184 unstable or moving slope sectors is quite difficult in the absence of existing structures infrastructures or visible damage. This becomes more important when a regional landslide hazard assessment has to be performed. Furthermore, slope movements can become manifested only after a specific or total displacement has occurred and at that moment no monitoring data is usually already available for any interpretation of the phenomenon itself. DInSAR and Permanent Scatterers can help in these cases. The usefulness of PS measurements can be particularly relevant due to limited impact of the main factors limiting the performances of DInSAR. In the following we present some case studies from the Lombardy Region (Northern Italy). These case studies have been extracted from a large set of deformation data obtained by performing a regional scale PS study covering 4900 km2 in an alpine and prealpine area (Figure 5). This study has been the first one covering a very large mountainous area interested by different morphological and local meteo-climatic conditions. The average elevation ranges between 200 m a.s.l. and about 4000 m a.s.l., involving different amounts and types of precipitation. Aims of the project were the application of the PS technique at regional and local scale, in rough alpine areas, to recognize slow slope movements (mm - cm/y) and their state of activity, in areas with limited urbanization. Furthermore, we were interested in the identification of the type of natural and artificial scatterers, in the interpretation of landslide mechanisms, and in the understanding of the major limitations and possibilities of the technique when working with low scatterers densities. The total number of scatterers recognised in the different areas varies according to the type (descending and ascending) and number of ERS passes (29 ascending and 51 descending in the 1993-2000 time interval), and the threshold value accepted for the coherence. In particular, from 8,300 up to 36,800 PSs have been found in zone 1; from 13,100 up to 18,500 PS in zone 2, from 11,300 up to 55,600 PS in zone 3 and from 68,300 to-160,500 in zone 4. The average spatial PS spatial density amounts to 77/km2 and they were generally located at the valley bottom because of the higher degree of urbanization. Three local case studies have been extracted from the regional dataset, namely, a large deep-seated slope gravitational deformation, a rockslide included in a large deep seated slope gravitational deformation, and an unstable scree slope.

Figure 5. Location of the 4 study areas in the alpine and prealpine sectors of the Lombardy Region (northern Italy, Central Italian Alps). Different areas have been studied through different sets of SAR images (descending and ascending coverages) and a different number of scatterers has been extracted for each zone.

185 These and other examples have been analysed according to the availability of historical data, of field evidence of deformation and diffused (or localised) damage, of morphologic evidences from aerial photo-interpretation, and of topographic measurements. 4.1. THE VARADEGA DEEP-SEATED SLOPE GRAVITATIONAL DEFORMATION Very large deep seated slope gravitational deformations (DSSGD) are a typical feature of alpine and prealpine areas. They are usually considered as relict forms because of their size and the relative difficulty in monitoring displacements. Toe bulging is frequently observed and this is often the sector with evidence of major localized activity. Rockfalls, rock slides and debris flows have been frequently mapped in these sectors and interact with existing structures and infrastructures. The Varadega DSSGD is located within zone 3 (Figure 5) and it is part of a larger phenomenon covering about 30 km2 (Figure 6). Main features within this area have been mapped during the development of a regional landslide inventory map at a 1:10,000 scale. Limits of unstable areas, crown areas and accumulation of deep landslides, scarps and counterscarps have been mapped (Figure 6) within the limits of the DSSGD. The next step consisted of the blind application of the PS technique without considering the results of the landslide inventory. The analysis of descending ERS data led to a large number (about 300, see Figure 7) of PSs well distributed along the middle and lower slope sectors with a consistent trend of rate of displacement. PS correspond both to natural and artificial features, from bedrock outcrops to corrugated iron roofs and metallic pylons or structures.

Hydroelectric power plant Figure 6. The Varadega deep seated slope gravitational deformation, covers about 9 km2 within a larger phenomenon along the left hand flank of Middle Valtellina, Lombardy, Central Italian Alps. Toe bulging is recognizable both in plan and lateral view, together with major scarps in the upper slope sectors. Total relief is about 1600 m. The hydroelectric power plant, built in 1908, is visible on the alluvial plain of the Adda River.



Figure 7. Varadega deep seated slope gravitational deformation: landslide inventory map with limits of main slope instabilities, scarps and counterscarps, and structural lineaments. Dots represents the PS position and the relative rate of displacement (mm/y) as derived from the time series.

Figure 8. Time series for different Permanent Scatterers positioned along the Varadega slope. The rate of displacement ranges between -0.2 mm/y and -6.5 mm/y and it is obtained by performing a linear regression of the data measured with respect to a reference master image. A) data from a longitudinal alignment of PSs, showing the change in the rate of vertical displacement moving from the upper to the lower slope sectors, B) data from a group of PSs located in the middle sector of the Varadega slope.


Figure 9. Sketch summarizing the possible interpretation for the observed PS rate of displacement according to the position along the slope and the expected geometry of the failure surface. A rotational movement results in a progressive downslope decrease in vertical displacement rate. A planar failure surface implies a homogeneous distribution of vertical and horizontal velocity vectors along the profile.

We could observe a progressive decrease in the vertical displacement rate when moving downslope as well as a certain lateral variability of the LOS velocity values relative to individual PS. When comparing the distribution of the PS with the one of the mapped landslides, we noticed a very good qualitative agreement. In fact, PS with similar average rates of displacement were grouped within specific landslide bodies or within different areas of the same landslide body (Figure 7, 8). An initial kinematic interpretation of the PS displacement data has been attempted. We tried to find a relationship between the distribution of the measured rate of displacement, the location of the PS along the slope, and the supposed failure surface (Figure 9). In fact, Figure 9 shows two possible distributions of vertical displacement components according to two different limit conditions: translational and rotational failure surfaces. We believe that the progressive decrease in the vertical displacement rate downslope originates in the subcircular failure surface, steeper in the upper slope sector and almost horizontal or dipping within the slope, at the toe. In this last position, very small negative up to positive (upward vertical component) values where measured. The absence of ground control points prevents us from the possibility to perform a quantitative assessment of the quality of our remote measurements. We did access however, some topographic measurements carried out in the 19701985 period along a penstock and on the main building of a hydroelectric power plant (Figure 6). The time intervals investigated by means of the two different techniques are not coincident. This could represent a problem for relatively fast moving landslides. Fortunately, we are studying a very slow continuous deformation phenomenon which probably maintains a similar behaviour over long time spans. The different measurements are compared in Figures 10a and b. Figure 10a shows a plan view of the power station, the penstock up to the small loading reservoir with the location of the PS and the topographic benchmarks. The PS are more numerous than the topographic benchmarks. They are located on the structures and some are almost coincident with topographic benchmarks. Down-slope movements are recorded at many points with a rapid decrease in the rate of displacement moving down-slope along the penstock. In correspondence of the power station building both the PS technique and the topographic levelling read an upward movement with comparables values (0.75 and


Figure 10. Varadega deep seated slope gravitational deformation. Plan view of the hydroelectric power station, the penstock and the loading reservoir with position of the PSs (number in italics, circles) and of the topographic benchmarks (black triangles). Movements recorded by two topographic benchmarks on the building and on the penstock, during the observation interval, are reported in the plot.

1.0 mm/y; Figure 10b). These results suggest a subcircular geometry of the failure surface and the unexpected temporal continuity and spatial extension of the movements. The state of activity of such a large phenomenon has been determined more exactly with a jump from a relict state (from aerial photo interpretation) to an active state. Finally, this analysis shows how such large instabilities are subjected to a continuous movement also in the actual climatic condition. 4.2. THE RUINON ROCKSLIDE AND THE CONFINALE-CIMA DI SALINE DSSGD Large deep-seated slope gravitational deformations are characterised by abundant morpho-structures (e.g. scarps, counterscarps, trenches, etc.) and often by an obvious toe bulging. Rock falls, rock slides and debris flows can occur in correspondence of these bulging toe sectors because of steep slope angles and pervasive jointing of the rock mass. The Ruinon rockslide is a large slope instability, located in Upper Valtellina (Lombardy Region). It was reactivated in 1987, and is presently moving at a relatively high velocity, up to some metres per year. This rockslide is part of a major deep seated slope gravitational deformation >1@ involving the entire slope from 1400 m a.s.l. up to 3000 m a.s.l.


Figure 11. A) Location of the PSs in the Upper Valtellina area (zone 1). B) Simplified landslide inventory map for the Ruinon rockslide – Confinale-Cima di Saline DSSGD area. Major lineaments and slope instabilities are represented together with PSs position. Maximum down-slope displacement rates up to 20 mm/y are reported.


Figure 12. Ruinon rockslide, Cima di Saline Confinale DSSGD. A) Example of cumulative displacements recorded in the crown area of the Ruinon rockslide since 1997; B) displacement rate of the DSSGD as determined through the PS technique.

-20 y -10 -10 y -8


-8 y -6.25 -6.25 y -4.5 -4.5 y -2.75

500 m

-2.75 y -1 -1 y 0

Figure 13. PSs distribution along a scree slope in the Premadio area (Lombardy Region, zone 1) by adopting a 0.6 value for the coherence threshold. A progressive increase in the displacement rate is visible moving upslope along the talus.

191 This case study represents an example of the potential integration of different systems, namely, topographic surveys, GPS fixed stations, extensometers and distometers, the PS technique. The monitoring activities are localized within the Ruinon rockslide area, whereas the PS technique has been applied to the entire area (Figure 11a). Figure 11b shows the entire slope sector with the relative PSs. We observe that PS are located only outside of the rockslide area. This observation is the consequence of two possible factors: the very high displacement rate characterizing the rockslide, the continuous acceleration recorded during the entire observation period (Figure 12a). On the other hand, Permanent Scatterers located within the slope show that the deep seated slope gravitational deformation is presently moving in different sectors at different velocities (Figure 12b). Measured LOS displacement rates range between 7 mm/y and 25 mm/y in the DSSGD area (Figure 11b).

4.3. MONITORING OF ACTIVE SCREE SLOPES As mentioned above, the number of PS for a specific area can be increased through a high-detail analysis or by adopting a lower coherence threshold value. In this last case the number of PS increases rapidly but the reliability of the corresponding deformation measurements decreases. The effects of the acceptance of a lower coherence value on the number of PSs for the different study area are summarised in Table 1. Nevertheless, we maintain that in some cases this approach can give very interesting results especially in the understanding of the phenomena occurring in the study areas. As an example of this type of application we present a case extracted from the Upper Valtellina area (Zone 1, in Figure 5). The case study area is characterised by abundant scree slope deposits located at the foot of high rocky cliffs in limestones and dolostones. Scree slopes are both vegetated and active. Granular debris flows occur along parts of the slopes and include large blocks. Lowering the coherence threshold values we observed an important increase in the number of PS and their progressive spreading toward the upper slope sectors where a lower number of PSs is usually observed. In the Premadio area (Figure 13) we observed an interesting pattern for the displacement rate values. A progressive decrease in the displacement rate was recognized moving upslope along the talus up to the limit where main debris flow lobes occur. We believe that this is an example of phenomenological use of the PS technique. Scree slopes are known to creep because of the limit equilibrium conditions at which the material occurs and of the presence of geomorphologic processes (thaw and freezing, avalanching, debris flows, rockfalls, etc.). These processes cause both the reworking and the growth of the scree slopes. Nevertheless, it is difficult to monitor scree slope evolution because of their relative homogeneity and the lack of evident features. In this example, the upslope increase of the displacement rate, recognised over a relatively wide area, is the result of the relative degree of activity along the slope. Activity increases in the upslope direction and the PSs stops exactly at the limit coincident with debris flow lobes which occur with a yearly frequency.


Table 1. Effects of the decreased coherence threshold for the selection of PSs in the study areas. 'PS represents the increase in the number of PS by adopting different coherence values. Zo n e


Th re s h ol d






36 800

328 00


Ascendin g

0.7 75

23 850

500 0


Ascendin g

0.7 75



0.6 25

71 800

458 00


Ascendin g

0.7 75



0.6 25

249 700

1 387 00

5. Discussion and Conclusions Landslide hazard assessment requires different steps including landslide recognition (location, geometry, type) involved materials, triggering and controlling factors, state and style of activity. The first step includes the production of a landslide inventory map and this is usually performed through aerial photo interpretation that can be acquired in different ways, including high resolution stereo satellite images. The automatic recognition of landslides from different images is still an unfeasible goal. A possible way to overcome this difficulty consists of the identification of slope areas subjected to instabilities through the measurement of their movement. This can be done for some specific landslides already recognised in the field through different techniques (e.g. topographic levelling, on site monitoring, aerial photogrammetry) but it can hardly be done at a regional scale or for areas where movements have not been noticed or are too slow. SAR interferometry could be a suitable way to perform such a recognition but several factors (e.g. decorrelation noise, atmospheric artifacts, local movements) limit the practical applicability of conventional DInSAR techniques to landslide recognition and monitoring. The Permanent Scatterrers technique >7, 17, 2, 12@ solves some of these problems linked to classic DInSAR and can be used to integrate other information levels (e.g. field survey and monitoring, aerial photointerpretation). In this paper we present a series of examples concerning the application of the PS technique to landslide monitoring and recognition in mountainous areas. Furthermore, we demonstrate that this technique is capable of improving the level of knowledge of unstable slopes, especially those affected by movements ranging from extremely slow (less than 1 mm/y) to slow (some tens of millimetres per year). The capabilities of this technique are supported by a series of in-situ measurements through topographic levelling techniques and GPS measurements. The PS approach is a fully operational tool for mapping slow evolving (up to ~10 cm/yr) ground deformation phenomena in areas where man-made structures and/or exposed rocks are available. In particular, the technique was shown to be useful in evaluating the state of activity of landslide phenomena due to the millimetric precision and the spatial density of PS benchmarks.

193 The main limitations are related to the intrinsic ambiguity of phase measurements and to the fact that, especially in rural areas, it is still very difficult to evaluate in advance the expected density of PS benchmarks. In rural areas it is, therefore, difficult to quantify the success probability of a PS analysis. With the aim of reducing this drawback we are convinced that the correlation between the spatial density of Permanent Scatterers and the local morphology and lithology deserve being investigated in depth. Presently, PS techniques investigates linear type of motions. This can be a suitable way to verify the state of activity of large to very large slope instabilities that are often also considered as relict ones. Nevertheless, recent experiences show the possibility of recognizing and investigating non-linear motions. This opens an important field of research and applications concerning slope motions that evolve from a slow constant rate of deformation phase to a rapid accelerating phase. The other important advantage of the technique consists of the availability of a tenyear dataset of SAR images (ESA ERS1/2 since 1991) and the relatively low cost of some ground based techniques (e.g. topographic levelling, GPS). This last point becomes quite important when looking at the number of measuring points (PS and benchmarks) that can be investigated with the different techniques. Historical images allow the reconstruction of the “recent” targets and landslide motion. This is not available when a monitoring network is installed on a recently recognised unstable slope or structure. Major limitations of the PS technique consist of the relatively long revisiting time (presently limited to 35 days for ERS2 but to be overcome by the availability of new satellites), the difficulty in recognizing high displacement rates, and the limited number of PS in vegetated and non-urbanized areas.

Acknowledgements The authors wish to thank: the Lombardy Region Geological Survey for allowing the publication of the data and the staff of the T.R.E.- Telerilevamento Europa company for their work in the construction of the PS dataset. References 1. 2. 3. 4. 5.


Agliardi, F., Crosta, G.B., Zanchi, A.: Structural constraints on deeps seated slope deformations kinematics. Engineering Geology, 59(1-2), 83-102, 2001. Bamler, R., Schättler, B., Karlsruhe, 1993. SAR Data Acquisition and Image Formation, in Schreier, G., Editior, SAR Geocoding: Data and Systems, Wichmann. Colesanti, C., Ferretti, A., Prati, C., Rocca, F., 2003. Monitoring Landslides and Tectonic Motion with the Permanent Scatterers Technique, Engineering Geology, 68, 3-14. Curlander, J. C., McDonough, R. N., New York, 1991. Synthetic Aperture Radar: Systems and Signal Processing, John Wiley & Sons, Inc. Dehls, J. F., Basilico, M., Colesanti, C., 2002. Ground deformation monitoring in the Ranafjord area of Norway by means of the Permanent Scatterers technique, Proceedings of the IEEE International Geoscience and Remote Sensing Symposium (IGARSS 2002), Toronto, Canada. Ferretti, A., Prati, C., Rocca, F., 1999. Multibaseline InSAR DEM Reconstruction: The Wavelet Approach, IEEE Transactions on Geoscience and Remote Sensing, Vol. 37, No. 2.

194 7. 8.

9. 10. 11. 12. 13. 14. 15. 16.


18. 19.

Ferretti, A., Prati, C., Rocca, F., 2001b. Multibaseline Phase Unwrapping for InSAR Topography Estimation, Il Nuovo Cimento, Vol. 124, No. 1. Ferretti, A., Prati, C., Rocca, F., 2000a. Nonlinear Subsidence Rate Estimation Using Permanent Scatterers in Differential SAR Interferometry, IEEE Transactions on Geoscience and Remote Sensing, Vol. 38, No. 5. Ferretti, A., Prati, C., Rocca, F., 2001a, Permanent Scatterers in SAR Interferometry, IEEE Transactions on Geoscience and Remote Sensing, Vol. 39, No. 1. Franceschetti, G., Lanari, R., New York, 1999. Synthetic Aperture Radar Processing, CRC Press. Fruneau, B., Achache, J., Delacourt, C., 1995. Observation and Modeling of the Saint-Etienne-de-Tinée Landslide Using SAR Interferometry, Tectonophysics, 265. Henderson, F. M., Lewis, A. J., Editors, New York, 1998. Principles and Applications of Imaging Radar, Vol. 2 of Maunal of Remote Sensing, 3rd Edition, John Wiley & Sons, Inc. Massonnet, D., Rossi M., Carmona, C., Adragna, F., Peltzer, G., Feigl, K., Rabaute, T., (1993) “The Displacement Field of the Landers Earthquake Mapped by Radar Interferometry”, Nature, 364. Mensa, D. L., Norwood, 1991. High Resolution Radar Cross-Section Imaging, 2rd Edition, Artech House. Nagler T., Rott, H., Kamelger, A., 2002. Analysis of Landslides in Alpine Areas by Means of SAR Interferometry, Proceedings of IEEE IGARSS 2002. Rott, H., Scheuchl, B., Siegel, A., Grasemann, B., 1999. Monitoring very slow slope movements by means of SAR interferometry: a case study from a mass waste above a reservoir in the Ötztal Alps, Austria, Geophysical Res. Letters, 26. Rott, H., Mayer, C., Siegel, A, 2000. On the operational potential of SAR interferometry for monitoring mass movements in Alpine areas, Proceedings of the 3rd European Conference on Synthetic Aperture Radar (EUSAR 2000), Munich, Germany. Ulaby, F. T., Moore, R. K., Fung, A. K., Norwood, 1981. Microwave Remote Sensing: Active and Passive, Vol. 2, Artech House. Zebker, H. A., Rosen, P. A., Goldstein, R. M., Gabriel, A., Werner, C. L., (1994) “On the Derivation of Coseismic Displacement Fields Using Differential Radar Interferometry: The Landers Earthquake”, Journal of Geophysical Research, 99.


FORECASTING RUNOUT OF ROCK AND DEBRIS AVALANCHES R.M. IVERSON U.S. Geological Survey 1300 SE Cardinal Ct. #100, Vancouver, WA 98683 USA

Abstract Physically based mathematical models and statistically based empirical equations each may provide useful means of forecasting runout of rock and debris avalanches. This paper compares the foundations, strengths, and limitations of a physically based model and a statistically based forecasting method, both of which were developed to predict runout across three-dimensional topography. The chief advantage of the physically based model results from its ties to physical conservation laws and well-tested axioms of soil and rock mechanics, such as the Coulomb friction rule and effective-stress principle. The output of this model provides detailed information about the dynamics of avalanche runout, at the expense of high demands for accurate input data, numerical computation, and experimental testing. In comparison, the statistical method requires relatively modest computation and no input data except identification of prospective avalanche source areas and a range of postulated avalanche volumes. Like the physically based model, the statistical method yields maps of predicted runout, but it provides no information on runout dynamics. Although the two methods differ significantly in their structure and objectives, insights gained from one method can aid refinement of the other. 1.


Forecasts of hazards from rock and debris avalanches must address two kinds of questions: (1) where and when will slope failure occur, and (2) how far and how fast will down-valley runout occur? Although runout of rock avalanches is commonly regarded as a enigmatic phenomenon, forecasting runout may be a more tractable problem than forecasting the location and timing of rock slope failure that occurs in the absence of observed precursory deformation. Whereas slope failure is governed by a balance of quasistatic forces that can be exceedingly delicate (e.g., a factor of safety ~ 1.001 implies that a slope is stable but precariously poised), rock avalanche motion is governed by an imbalance of dynamic forces that can be immense. The speeds and masses of moving rock avalanches dictate that bulk inertial effects commonly dominate motion, and the laws of classical mechanics dictate that the effects of bulk inertia are relatively predictable. Thus, there is cause for optimism about forecasting avalanche runouts.

E-mail of author:

[email protected]

197 S.G. Evans et al. (eds.), Landslides from Massive Rock Slope Failure, 197–209. © 2006 Springer. Printed in the Netherlands.

198 As might be expected for a phenomenon driven by gravity and dominated by inertia, runout of rock avalanches depends strongly on avalanche volume (or mass) and runout-path topography [15]. The topography of prospective avalanche paths is generally well known, whereas the volumes of prospective avalanches are typically poorly constrained because the size of slope failures commonly depends on subtle geological structures and heterogeneities, and on transient forcing due to rainfall or earthquakes. Therefore, methods for forecasting runout should take advantage of reliable knowledge of topography and take account of limited knowledge of avalanche volume. These considerations argue strongly for use of forecasting methods that represent three-dimensional topographic effects rigorously and treat avalanche volumes as independent variables that have inherent uncertainty. This paper discusses two methods of forecasting rock avalanche runouts across threedimensional terrain. One method aims mainly at enhancing scientific understanding by devising a physically and mathematically rigorous theory that yields testable predictions of runout dynamics. The other method aims mainly at expediting practical hazard assessment by using statistical analysis of runout trends to forecast the probable extent of future runouts. The physically based model yields greater returns of information, at a cost of greater demands for input data and computation. The statistically based approach has a more limited scope, but it requires relatively modest computation and no input data other than path topography and a postulated distribution of avalanche volumes. The choice of a particular method of runout forecasting depends principally on the objectives of the forecast. Some practical objectives (e.g., hazard-zone assessment) may be met most expediently with statistically based methods, whereas scientific objectives (e.g., improved understanding) can be met most rigorously with physically based models. Diverse objectives and methodologies can be synergistic, however. Improved physical understanding can lead to improved hazard forecasts, and information gained in hazard assessments can lead to improved physical understanding. 2.

Physically Based Modeling

This section describes formulation and testing of a physically based model that uses universal principles (i.e., mass and momentum conservation) and well-tested formulas (i.e., the Coulomb friction rule and Terzaghi effective-stress principle) to compute avalanche motion from initiation to deposition. The model, developed by Iverson and Denlinger [3, 11], predicts the behavior of granular avalanches under a wide variety of soil and rock states, which may range dry to water-saturated and from rigid to fully fluidized. Conditions in which the model applies include static limiting equilibrium (which exists at the onset of slope failure), dynamic states dominated by bulk inertia, and subsequent static states that result from deposition. Furthermore, the model accounts for the effects of evolving porefluid pressure and three-dimensional path topography. 2.1.


The guiding philosophy of the model of Iverson and Denlinger [3, 11] is to represent the well-constrained aspects of avalanche dynamics as thoroughly as practicably possible, and to minimize assumptions about the more puzzling aspects of avalanche dynamics. For example, aspects of avalanche dynamics dictated by momentum conservation are

199 completely constrained by physical law. Therefore, bulk inertia terms (which express momentum transport without energy dissipation) in the avalanche dynamics model involve no assumptions, and involve only mathematical approximations that are rigorously justifiable in view of the pertinent physics. The more puzzling aspects of avalanche dynamics result from dissipative (i.e., resisting) forces, and the model assumes that these forces obey well-tested formulas of classical soil and rock mechanics (i.e., the rules for Coulomb friction and effective stress mediated by pore-fluid pressure). This parsimonious approach to physically based modeling provides the surest route to rigorous understanding of avalanche runout, because it employs no coefficients that are adjusted to fit model predictions to data, and it thereby facilitates conclusive hypothesis tests. Use of adjustable resisting forces in avalanche dynamics models is unwarranted unless a model that rigorously conserves momentum and employs parsimonious assumptions about resisting forces is demonstrably inadequate [10]. 2.2.


As noted above, the central postulate in the avalanche dynamics model of Iverson and Denlinger [3, 11] is the well-known Coulomb-Terzaghi equation for resistance to basal sliding:

W bed = ( V bed - pbed ) tan ϕbed Here V bed is basal normal stress,


pbed is basal pore-fluid pressure, and ϕ bed is the basal

Coulomb friction angle, which is constrained by experiments to range from about 30 to 40 degrees for most fragmented rocks and granular soils. Application of the CoulombTerzaghi equation to avalanche dynamics involves a number of subtleties, however. First, a resistance equation consistent with Coulomb-Terzaghi behavior must be employed to describe not only basal sliding but also shear and normal stresses within deforming avalanches. Second, basal pore-fluid pressure in an avalanche mass can change as a function of position and time, and evolving pore-fluid pressure must therefore be evaluated simultaneously with evolving avalanche motion. Third, mass and momentum must be conserved in four dimensions (space plus time) within the moving avalanche. A detailed mathematical derivation of equations with the properties described above is beyond the scope of this summary but has been provided by Iverson and Denlinger [11]. To simplify the four-dimensional equations, they are integrated through the avalanche thickness to eliminate explicit dependence on the velocity component normal to the bed. This simplification is typically justifiable because tabular avalanche geometries dictate that bed-normal velocities are much smaller than bed-parallel velocities in most instances [18, 19]. The resulting depth-integrated equations governing evolution of mass, momentum and pore-pressure distributions are referenced to a coordinate system fitted to local bed topography (Figure 1) and are summarized as follows:

200 ∂U ∂F ∂G + =S + ∂x ∂y ∂t




h νx

h νy


    hν ν  h ν 2 + 1 h c2  hν  y x  Sx      x 2  x        G =  2 1 2 S =  F= U=  Sy  (2b,c,d,e)   c h ν + h ν ν h ν h x y     y 2    y λh ν y λh ν x λh Sλ   3 υf µ ν x 2 ∂ θx g g h tan ϕbed − Sx = x h − sgn(ν x )(1 − λ) z + ν x (2f) ρ h ∂x   2 2 υf µh ∂ ν x υf µh ∂ ν x ∂ ∂ νx h kact/pass − sgn [gz h(1 − λ)] sin ϕint + + ∂y ρ ∂ x2 ρ ∂ y2 ∂y   3 υf µ ν y ∂ θy h tan ϕbed − Sy = gy h − sgn(ν y )(1 − λ) gz + ν 2y ∂y ρ h   2 υf µh ∂ 2 ν y υf µh ∂ ν y ∂ νy ∂ g − sgn [ h(1 − λ)] sin ϕ + h + k z act/pass int ρ ∂ x2 ∂x ∂x ρ ∂ y2

D 2 Sλ = g ρ z h λ0

∂p | + ρ gz ∂z h f



kact/pass = 2


pbed ρ gz h

1 ∓ [1 − cos2 ϕint (1 + tan2 ϕbed ) ] cos2 ϕint c=


[(1 − λ) kact/pass +λ] gz h




Although this set of equations is mathematically complex, the physical concepts entailed are simple and few: mass and momentum conservation, Coulomb friction, and effective stress mediated by evolving pore-fluid pressure. The independent variables in the equations are the orthogonal planimetric coordinates x and y (which are rotated to fit local


Figure 1. Schematic illustrating variables in the avalanche model summarized by equations (2 a-k). Vertical exaggeration is about 10 to 100+. Labels of the front, body, and tail characterize an avalanche or debris flow that is partially liquefied by high pore-water pressure.

topography) and time t. The dependent variables are the depth-integrated velocity components v x (x, y, t) and v y (x, y, t) , the avalanche thickness h(x, y, t) , and the basal pore-pressure ratio,

O (x, y, t) .

For granular avalanches without pore-fluid effects (i.e.,

P = 0, O = 0 ), the only relevant parameters are the basal and internal friction angles of the granular debris, I bed and I int . If pore-fluid effects are present, additional relevant parameters are the bulk density of the avalanche debris, ȡ, bulk density of the pore fluid, U f , volume fraction of the pore fluid (i.e., porosity), X f , viscosity of the pore fluid, µ, and the pore-pressure diffusivity (i.e., consolidation coefficient), D. The x and y components of gravitational acceleration, g, local slope angle, ș, and local bed curvature, 1 / r x = w T x /wx , are determined by the local terrain (Figure 1). An important feature of all of these quantities is that they are independently measurable on maps or in standard laboratory tests; none of the quantities is an adjustable tuning coefficient. The equations defining k act/pass , O , and c are mathematically derived and have straightforward physical interpretations [3, 9, 11]:

k act/pass is a Rankine earth-pressure

coefficient that applies in cases with simultaneous internal deformation and slip along the bed; Ȝ is the ratio of basal pore pressure to basal lithostatic stress; and c is a gravity-wave speed that governs the maximum rate at which disturbances propagate longitudinally

202 through the deforming avalanche material. In general this gravity-wave speed is influenced by intergranular friction and includes as a special case the analogous speed used in shallow-

c = ( g z h )1/2 . Here, c = ( g z h )1/2 applies only under conditions of full avalanche fluidization ( O = 1 ).

water wave theory,

Initial conditions used to begin model calculations specify zero avalanche velocity ( v x = v y = 0 ), an initial pore-pressure distribution, O 0 (x, y) , and an initial thickness distribution h0 (x, y) . The thickness distribution defines the avalanche volume, which is assumed to remain constant throughout motion and deposition. Terms that allow for variable avalanche volume due to progressive erosion or sedimentation can easily be added to the mass- and momentum-conservation equations. To date, however, no rigorous experiments or calculations have been performed to constrain the magnitude of such terms, and they consequently are omitted. To assess behavior of prospective avalanches with differing initial volumes or distributions of mass as specified by h0 (x, y) , multiple model runs are required. The system of equations (2a-k) listed above includes some well-known equations as special cases: the standard shallow-water equations [22], the dry granular avalanche equations of Savage and Hutter [18, 19], the multidimensional Savage-Hutter equations of Gray et al. [4], the simplified Savage-Hutter equations of Hungr [7], and the slidingconsolidation equations of Hutchinson [8]. Also, despite their generality, equations (2a-k) contain at their core the simple one-dimensional equation of motion for a rigid Coulomb slide block,

d v x /dt = g ( sin θ - cos θ tan ϕ bed )


The mathematical complexity that distinguishes (2a-k) from (3) is necessary to account for the distributions of velocity, thickness and pore pressure in three-dimensional rock and debris avalanches. 2.3.


One of the biggest obstacles to developing a robust, physically based model of rock and debris avalanches is the difficulty of conclusive testing. Such models can seldom, if ever, be tested against field data, because field data generally leave many factors (such as initial and boundary conditions) poorly constrained. Therefore, models are generally fitted to field data rather than tested against field data, and model veracity remains equivocal. As an alternative to fitting field data, models can be tested against data from controlled experiments in which all parameter values, boundary conditions, and initial conditions are independently constrained. However, a potential difficulty with such experiments is scaling, because controlled experiments generally cannot be conducted at the scale of large geological events. This difficulty can be addressed in several ways. For example, normalization of the equations of motion (2a-k) identifies relevant scaling parameters, which imply that purely frictional avalanches without pore-fluid effects will behave in a manner that is independent of scale [11]. On the other hand, the same scaling parameters indicate that avalanches with pore-fluid effects can be expected to behave in a scaledependent manner, whereby increasing mobility occurs with increasing scale if all other

203 factors are constant. These scaling considerations motivated Denlinger and Iverson [3] to test model predictions against data from two kinds of experiments: (1) bench-top experiments with miniature avalanches of about 0.001 m3 of dry, well-sorted sand in which pore-fluid effects were negligible, and (2) large-scale, outdoor experiments with avalanches of about 10 m3 of poorly sorted, water-saturated sand and gravel with significant pore-fluid effects. Both types of experiment highlighted the importance of multidimensional momentum transport in avalanches, and both demonstrated good agreement between data and model predictions. However, owing to the complexity of the numerical method used to compute solutions to the model equations, additional testing under more realistic scenarios with complex topography and boundary conditions is necessary before runout predictions of natural rock avalanches can be made with confidence. Experiments like those reported by Denlinger and Iverson [3] also reveal phenomena that provoke new kinds of questions and motivate further model refinements. For example, experiments show that grain-size segregation can be an extremely efficient process in poorly sorted avalanches. This segregation influences pore-pressure generation and dissipation (because fine sediments sustain high pore pressures more readily than do coarse aggregates,) and it therefore holds large implications for macroscopic dynamics. Indeed, feedbacks between the micro-dynamics of grain-scale processes and the macro-dynamics of avalanche motion may be crucial in some circumstances [13]. Such complexities demonstrate the need for continuing efforts to refine physically based models. 3.

Statistically Based Forecasting

The complexity of rigorous, physically based modeling of rock avalanche runout indicates a need for simpler forecasting methods that can be readily employed in hazard assessments. Commonly such assessments must be performed in situations where limitations of time, money, or information preclude detailed modeling. The summary below describes a statistical forecasting method developed by Iverson et al. [12] to delineate areas likely to be inundated by lahars, which are water-saturated debris flows that originate on volcanoes. With modification of the pertinent data sets and statistics, as summarized below, this method can be adapted to forecasting runout of rock and debris avalanches. 3.1.


The framework of this forecasting method rests on three observations: (1) the primary factor determining runout of a prospective rock avalanche is its volume, but this volume is unknown a priori; (2) a secondary factor determining runout is the three-dimensional topography of potential runout paths, which may be known with good accuracy as a result of standard topographic mapping; (3) runout patterns of large avalanches appear geometrically similar to those of small avalanches viewed at a larger scale. That is, runout patterns appear to exhibit fractal scaling. In accord with these observations, the forecasting methodology described here entails formulation and statistical testing of empirical, scale-independent runout equations that employ avalanche volume as the independent variable. In this methodology a range of avalanche volumes is postulated and used to compute runout zones that are constrained by

204 path topography. The computed runout zones reflect both the statistical uncertainty of the runout equations and geological uncertainty about prospective avalanche volumes. The possibility of valid, scale-independent runout equations has been suggested previously. Many investigators have noted that runout of rock avalanches is characterized better by the planimetric area inundated than by Heim=s [5] famous fahrböschung or H/L ratio [1, 2, 6, 12, 14, 15, 16, 21]. Whereas H/L ratios decline markedly and nonlinearly with increasing avalanche volume (V), planimetric areas of inundation ( A1 ) scale quite consistently with V 2/3 , as might be expected on the basis of the dimensional equivalence of 2/3 A1 and V . Thus, the power-law hypothesis 2/3 A1 = D 1V


where D 1 is a constant, can be assessed statistically to evaluate whether it provides a useful runout equation that is independent of scale.

A2 = D 2 V



Although equation (4) may be useful for assessing runout, by itself it is insufficient for delineating runout zones, because it provides no information about the spatial distribution of A1 . An additional equation is needed to constrain the lateral limits of inundation and thereby place planimetric bounds on the area defined by A1 . On the basis of physical scaling arguments detailed by Iverson et al. [12], an additional power-law equation is postulated to relate the maximum vertical cross-sectional area of valley inundation ( A2 ) to the avalanche volume. This equation constrains the lateral limits of inundation if the runout path topography is known. If initiation areas and avalanche volumes are specified, use of equation (5) in conjunction with equation (4) defines the extent of prospective avalanche runout zones. Of course, the utility of this approach depends on the statistical validity of the hypotheses represented by equations (4) and (5). 3.2.


A complete description of the statistical rationale and methods used to test and calibrate runout equations (4) and (5) was provided by Iverson et al. [12], who applied these equations to lahar runouts. For the rock avalanche runouts addressed here, a detailed report is in preparation; the results summarized below aim only to demonstrate the potential viability of the method. As an initial step in statistical testing and calibration, equations (4) and (5) are logarithmically transformed to the linear forms,

log A1 = log D 1 + 23 log V



log A2 = log D 2 + 23 log V


The log transformation facilitates use of least-squares linear regression as a tool for testing and calibration, and also implies that scatter of observed values of A1 and A2 as functions of V is expected to scale with the magnitude V. Table 1 summarizes results of statistical testing and calibration of (6a) and (6b) using published data for rock avalanches worldwide. Published data on planimetric innundation area A1 are common, whereas published data on valley cross-sectional inundation area A2 are rare. Therefore, different sets of avalanches were used to compile the data for testing (6a) and (6b), and using these disparate data sets it is not feasible to assess the possibility of cross-correlation between the A1 and A2 values. Instead, the data sets for A1 and A2 are treated as completely independent. The most important information in Table 1 is provided by the F statistics, which compare alternative linear models of log A1 and log A2 as functions of log V. Very large values of the F statistic for the Aspecified zero slope@ models indicate that such models can be rejected with a high degree of confidence (exceeding 99.5 %, as inferred from tabulated values of the F distribution). In other words, regression models that assume linear dependencies of log A1 and log A2 on log V are clearly superior to models that assume log A2 and log A1 lack dependence on log V. In contrast, the small values of the F statistic for the Aspecified 2/3 slope@ models indicate that such models cannot be rejected with even a 90% degree of confidence. In other words, the 2/3 power laws represented by equations (6a) and (6b) are not clearly distinguishable from the best-fit linear regressions representing log A1 and log A2 as functions of log V. On this basis the 2/3 power laws are adopted as acceptable models of the data. Calibration of the 2/3 power laws entails determining optimal values of the coefficients D 1 and D 2 . To accomplish this, best-fit values of D 1 and D 2 are obtained by minimizing the mean square error in (6a) and (6b) (as in a linear regression procedure), and then rounding these values to two significant digits. In this way D 1 = log


(1.3617) = 23 and


D 2 = log (-0.699) = 0.20 are obtained, yielding the predictive equations A1 = 23 V 2/3

A2 = 0.20 V 2/3


The forms of these equations are similar to those of equations used by Iverson et al. [12] to forecast lahar inundation, but the coefficients that multiply V 2/3 are unique to rock avalanches. For rock avalanches, the planimetric runout coefficient 23 replaces the lahar coefficient 200, and the cross-sectional inundation coefficient 0.20 replaces the lahar coefficient 0.05. Thus, on the basis of these coefficients, rock avalanches can be expected to inundate planimetrtic areas about nine times smaller than those inundated by lahars of similar volume and can be expected to inundate valley cross-sectional areas about four times larger than those inundated by lahars of similar volume. These contrasts reflect the fact that rock avalanches generally undergo less liquefaction and exhibit more flow resistance than lahars.

206 Table 1. Parameters and analysis-of-variance statistics for alternative linear models of log-transformed runout data for rock avalanches.

______________________________________________________________________ Models for prediction of planimetric inundation area, A1

Parameter Best-fit regression slope of line 0.7207 intercept of line at log V =0 0.9296 number of data pairs 136 degrees of freedom 134 sum of squares 30.5634 mean square 0.2281 standard error 0.4776 r2 statistic 0.7602 F statistic not applicable

Specified 2/3 slope 0.6667 1.3617 136 135 31.1179 0.2305 0.4801 0.7559 2.4311

Specified zero slope 0.0000 6.8146 136 135 127.4689 0.9442 0.9717 0.0000 424.87

Models for prediction of cross-sectional inundation area, A2

Parameter Best-fit regression slope of line 0.6304 intercept of line at log V =0 -0.3971 number of data pairs 12 degrees of freedom 10 sum of squares 1.9930 mean sqaure 0.1993 standard error 0.4464 r2 statistic 0.7867 F statistic not applicable

Specified 2/3 slope 0.6667 -0.699 12 11 2.0182 0.1835 0.4283 0.7840 0.1262

Specified zero slope 0.0000 4.7077 12 11 9.3434 0.8494 0.9216 0.0000 36.88

______________________________________________________________________ The uncertainty of the runout forecasts obtained from equations (7a) and (7b) can be estimated from the associated standard errors, which fall between 0.4 and 0.5 (Table 1). Because these values apply to log-transformed data, they imply standard errors of about 100.4 to 100.5 ( 3) for constraining either A1 or A2 as a function of V. At first glance the three-fold standard error associated with (7a) and (7b) might seem unacceptably large for runout forecasting, but the error appears less severe if uncertainty associated with postulating prospective avalanche volumes V is considered concurrently. In most hazard assessments, only broad bounds can be placed on V, and a population of possible avalanche volumes must be postulated to provide a useful forecast. Thus, the standard error associated with using (7a) and (7b) to forecast inundation by an avalanche with a particular V must be gaged relative to the uncertainty about V itself. Viewed in this context, a threefold standard error associated with (7a) and (7b) appears tolerable. Of course, any hazard assessment that employs (7a) and (7b) superposes the uncertainty of the equations and the uncertainty in postulated ranges of V to produce a runout forecast that is probabilistic, not deterministic. 3.3. EXAMPLE OF APPLICATION The implications of the uncertainties associated with (7a) and (7b) are illustrated by example in Figure 2, which depicts a map that forecasts rock avalanche runout patterns on


Figure 2. Shaded relief map of the western flank of Mt. Rainier, Washington, USA, with runout hazard zones for rock avalanches originating in the Sunset Amphitheater. Nested hazard zones were computed for avalanches with hypothetical volumes of 0.1 km3 (light tone), 0.316 km 3 (intermediate tone), and 1 km3 (dark tone) that might descend one or more of three valleys.

the western flank of Mount Rainier, Washington, USA. The algorithm used to compute these avalanche inundation patterns is the same as that described by Iverson et al. [12] and Schilling [20] for computing lahar inundation patterns; only the coefficients in (7a) and (7b) are novel. The spiny appearance of the hazard zones depicted in Figure 2 is a consequence of the relatively coarse resolution (62.5 m) of the digital elevation model (DEM) used to represent Mount Rainier=s topography. Spines tend to disappear if DEMs with finer resolution are employed (S.P. Schilling, personal communication, 2002). To generate the hazard zones depicted in Figure 2, a prospective avalanche source area in and around the steeply sloping Sunset Amphitheater at about 4000 m elevation on the western flank of Mount Rainier was chosen. On the basis of its geometry, lithology, and geological history, the Sunset Amphitheater has been identified as an area particularly susceptible to slope failure [17]. To compute the three nested hazard zones in Figure 2, three values of V were postulated: 0.1 km3, 0.316 km3, and 1 km3, corresponding to log V = -1, -0.5, and 0, respectively. This ten-fold range in V results in a 4.64-fold variation in the predicted inundation areas A1 and A2 (because 102/3  4.64). This variation is about 50%

208 larger than the three-fold variation of inundation areas expected on the basis of the standard errors of equations (7a) and (7b). Thus, the inner hazard zone in Figure 2 would almost certainly be inundated by an avalanche with V ~ 0.3 km3, and an avalanche of this volume would be very unlikely to inundate an area larger than that of the outer hazard zone in Figure 2. A similar rationale can be extended to smaller and larger avalanches by computing additional inundation areas (for additional postulated V) and depicting them on a map like Figure 2. Iverson et al. [12] provide a more thorough discussion and detailed example of this methodology, albeit in the context of lahars rather than rock avalanches. The hazard zones of Figure 2 illustrate both advantages and limitations of the statistical forecasting method. Depiction of nested hazard zones with specified ranges of uncertainty is a clear advantage in hazard assessments. On the other hand, the depicted hazard zones take no account of dynamic effects (such as runup and superelevation) that occur as avalanches interact with topography, and they provide no information on avalanche speeds and impact forces. 4.


Forecasting runout of rock and debris avalanches is a longstanding problem with both scientific and practical importance. Scientific questions about the mechanics of the runout process can be best addressed with a physically based runout model that rigorously conserves momentum and avoids used of adjustable coefficients. On the other hand, practical questions about the likelihood of down-valley inundation can be addressed most expediently with a statistical model developed specifically for hazard assessment. The two methods can be synergistic, however. Elementary physical reasoning leads to scaling relationships that provide some constraints on statistical forecasting by guiding formulation of pertinent empirical equations, and statistically verified empiricisms summarize data trends that can be targets for prediction by rigorous, physically based models. Continued progress in scientific understanding and hazard assessment will likely entail melding of information obtained from physically based and statistically based investigations. Acknowledgments Julia Griswold compiled the data and performed the calculations for application of the statistical forecasting method to rock avalanche runout, and she computed the hazard map for the western flank of Mount Rainier (Figure 2) . Steve Schilling wrote the GIS program LAHARZ, which performs the computations. References 1. 2. 3.

Dade, W.B., and Huppert, H.E. (1998) Long-runout rockfalls, Geology 26, 803-806. Davies, T.R.H. (1982) Spreading of rock avalanche debris by mechanical fluidization, Rock Mech. 15, 9-24. Denlinger, R.P., and Iverson, R.M. (2001) Flow of variably fluidized granular masses across threedimensional terrain: 2. Numerical predictions and experimental tests, J. Geophys. Res. 106 B, 553-566.

209 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. 16. 17. 18. 19. 20. 21. 22.

Gray, J.M.N.T.,Wieland, M., and Hutter, K. (1999) Gravity driven free surface flow of granular avalanches over complex basal topography, Proc. Roy. Soc. London, Ser. A. 455, 1841-1874. Heim, A. (1932) Bergsturz und Menschenleben, Fretz and Wasmuth, Zürich. Hungr, O. (1990) Mobility of rock avalanches, Report of the National Research Center for Disaster Prevention (Japan) 46, 11-19. Hungr, O. (1995) A model for the runout analysis of rapid flow slides, debris flows, and avalanches, Can. Geotech. J. 32, 610-623. Hutchinson, J.N. (1986) A sliding-consolidation model for flow slides, Can. Geotech. J. 23, 115-126. Iverson, R.M. (1997) The physics of debris flows, Rev. Geophys. 35, 245-296. Iverson, R.M. (2003) How should mathematical models of geomorphic processes be judged? in P.R. Wilcock and R.M. Iverson (eds.) Prediction in Geomorphology, Geophys. Monograph 135, American Geophysical Union, Washington, D.C. Iverson, R.M., and Denlinger, R.P. (2001) Flow of variably fluidized granular masses across threedimensional terrain: 1. Coulomb mixture theory, J. Geophys. Res. 106 B, 537-552. Iverson, R.M. Schilling, S.P., and Vallance, J.W. (1998) Objective delineation of lahar-inundation hazard zones, Geol. Soc. Amer. Bull. 110, 972-984. Iverson, R.M., and Vallance, J.W. (2001) New views of granular mass flows, Geology 29, 115-118. Kilburn, C.R.J., and Sørensen, S-A. (1998) Runout lengths of struzstroms: the control of initial conditions and of fragment dynamics, J. Geophys. Res. 103 B, 17877-17884. Legros, F. (2002) The mobility of long-runout landslides, Eng. Geol. 63, 301-331. Li Tianchi (1983) A mathematical model for predicting the extent of a major rockfall, Ziets. Geomorph. 27, 473-482. Reid, M.E., Sisson, T.W., and Brien, D.L. (2001) Volcano collapse promoted by hydrothermal alteration and edifice shape, Mount Rainier, Washington, Geology 29, 779-782. Savage, S.B., and Hutter, K. (1989) The motion of a finite mass of granular material down a rough incline, J. Fluid Mech. 199, 177-215. Savage, S.B., and Hutter, K. (1991) The dynamics of avalanches of granular materials from initiation to runout, Part I. analysis, Acta Mechanica 86, 201-223. Schilling, S.P. (1998) LAHARZ: GIS programs for automated delineation of lahar hazard zones, U.S. Geol. Sur. Open-file Rep. 98-638. Vallance, J.W., and Scott, K.M. (1997) The Osceola mudflow from Mount Rainier: sedimentology and hazards implications of a huge clay-rich debris flow, Geol. Soc. Amer. Bull. 109, 143-163. Vreugdenhil, C.B. (1994) Numerical Methods for Shallow-Water Flow, Kluwer, Dordrecht.

CONTINUUM NUMERICAL MODELLING OF FLOW-LIKE LANDSLIDES G.B. CROSTA1 1 Dip. Scienze Geologiche e Geotecnologie Università degli Studi di Milano – Bicocca Piazza della Scienza 4 20126 Milano, Italy S. IMPOSIMATO Dip. Ingegneria Strutturale Politecnico di Milano Piazza Leonardo da Vinci 32, 20133 Milano, Italy


D.G. RODDEMAN 3 FEAT – Finite Element Application Technology The Netherlands Abstract Continuum modelling of flow-like landslides is a possible approach that can be adopted to simulate landslide instability, and the transition to catastrophic failure up to flow development. Models based on continuum mechanics and associated with different rheological models are usually preferred to predict landslide runout and relevant parameters. A finite element method approach is here presented and contrasts previous research where depth-averaged equivalent-fluid approaches were adopted. We developed a 2D/3D finite element code to analyse slope stability and to model runout of mass movements characterised by very large displacements. Different material laws already known, tested and verified for granular materials have been implemented. Materials laws include classical elasto-plasticity, with a linear elastic part and different applicable yield surfaces with associated and non-associated flow rules. A series of simulations has been performed. The effects of different morphological conditions have been simulated to understand processes close to obstacles, both deformable and perfectly rigid, or of trough shaped profiles where accumulation is favoured. MohrCoulomb, Drucker-Prager, von Mises models with or without strain softening have been adopted for the granular materials. Results for Mohr-Coulomb material are presented and demonstrate the capability of this approach and the relevance of internal deformation within the flowing material. 1.


Slope stability and runout analysis of very large landslides are usually difficult to perform. Large fast moving landslides often assume a complex behavior showing a continuum passing from sliding to flowing. These phenomena are defined as complex 1

E-mail of corresponding author: [email protected]

211 S.G. Evans et al. (eds.), Landslides from Massive Rock Slope Failure, 211–232. © 2006 Springer. Printed in the Netherlands.

212 and composite flows [8] and are characterized by a complicated set of behaviors and a pattern of deformation: an upper layer slightly deformed and a lower-basal layer with maximum shear deformation. Both translation or sliding and internal deformation participate in the motion and deposition of flow-like landslides. At the same time the term flow-like landslides implies that the material behaves as a complex fluid. Translation or sliding implies the resisting action at the base of the flowing material along the infinitesimal thickness of the failure/sliding surface, but the internal deformation or the interaction with a rough basal – topographic surface imply the resisting action of the flowing material. Flow like landslides are able to entrain or deposit material while moving, allowing for important changes in mass and volume. Among the natural processes showing such characteristics we can list rock and debris avalanches, rockfalls, debris flows, earth flows, dry grain flows, etc.. They can involve different types of material, namely: rock, soil, a mixture of the two, water and ice, in different proportions, both in volcanic and non volcanic environments. The comprehension and modelling of these phenomena are made difficult by the incompleteness of their description which often forces us to assume non-consistent or incomplete behavior and models. Pre-, syn- and post-failure observations and data are fundamental constraints to develop, run, calibrate and validate a model of such events. This difficulty is witnessed by the number of theories and models suggested in the literature [24, 51, 22, 25, 42, 30, 48, 3, 36, 12, 15, 16]. Rock and debris avalanches are a major hazard in mountainous areas and are characterized by an extreme mobility and extremely high velocities. Extensive damages and casualties have been caused by these phenomena through the centuries both in non volcanic [24, 1, 14, 16, 17, 18, 19], volcanic [52, 55, 56, 54, 21] and marine environments [43] as well as in waste mining dump materials [35]. The wide spectrum of observed behaviors and the occurrence of long runout distances and large invasion areas are among the consequences of the natural variability of these phenomena. The definition of hazard and possible terrain zonation is therefore complicated and requires the determination of the size or volume of the unstable mass, the triggering probability of the initial failure, the geometry of the invasion area and of the deposit, the intensity of the phenomenon along the path, the probability to reach a specific point in space (reaching probability) and the time needed to reach any specific point along the path. Relationships between volume and mobility of rock avalanches have been presented and discussed in the past [24, 50, 1, 25, 11, 39] but they do not give a complete description of the phenomenon. In the following we will propose a new continuum finite element modelling approach to describe the transient movement of flow-like landslides. A Lagrangian finite element model is presented in its present two-dimensional version and is used to simulate: the applicability of classical material laws for granular material, the effects of morphological features on flow behavior and those of deposition and entrainment of material along the flow path. The modelled phenomena are mainly of the rock and debris avalanche type where large volumes of dry rock and/or debris are involved.

213 2.

Mechanics and Modelling

Large flow like landslides are characterized by exceptional runout and runup distances and unique morphological features. Numerous theoretical models to explain such behaviors and to describe the motion and dynamics of rock avalanches have been presented. Rock avalanche deposits are described as formed by extensively fractured/shattered material [51, 23, 34, 41, 55, 9, 10]. Large blocks are found at the upper free surface and a progressive and irregular decrease in deposit grain size, from gravel to clay, is observed with depth. Large fractured rock blocks immersed in a finer matrix are common and they can result from the high longitudinal stresses acting within the material in motion or from the rock mass disintegration occurring in correspondence with the toe of the failure surface. At the same time, it has been demonstrated that shear strength is controlled by matrix properties when this is present in a certain percentage. Eventually, the progressive decrease in grain size with depth can be explained with the higher acting stresses and with the increase in accumulated shear strain with depth. This is partially in contrast with the assumption that shearing is concentrated to a thin basal boundary layer. This assumption is often accepted but relatively few field observations support it. Both empirical and numerical models have been presented in the literature. Empirical models [24, 50, 1, 25, 11, 39] are subjected to a high degree of approximation arising from the difficulties to find a comprehensive description of: the actual processes, the initial volume and geometry and detachment position of the unstable mass, the conditions of the slope where movement, scouring and deposition occurred, and the total duration. As a consequence, statistical analyses and models lack robustness. Numerical models [30, 48, 53, 4, 45, 26, 3, 32, 13] can simulate the flow behavior of flow-like landslides and can be used to predict runout and to perform hazard zonation. They generally use finite difference schemes in Eulerian coordinates [53, 4, 45]. Nevertheless, a Lagrangian coordinates frame is more suitable for long runout phenomena and in FEM analyses it avoids excessive distortion of the computational grid. Suggested models include sliding friction and velocity-dependent resistance models for point mass motion [37, 38, 46] coupled to digital elevation models [43], sliding block models that include friction [24, 16] and pore-water pressure parameters [29] [48], modified flood prediction models [33, 20, 48, 45], two dimensional depthaveraged Lagrangian frictional models [30] or using a wider range of rheological models for the basal highly-sheared layer [26], using a fully two dimensional discrete element approach [2] or a finite element model [6, 7]. Highly refined mathematical solutions and a Coulomb-like behavior have been successfully used for a threedimensional flow description [32, 13]. Laboratory flume experiments over non-smooth beds [30, 31] and high speed ring shear tests [44, 27, 49], at shear rates comparable to those in rock avalanches, demonstrates that the internal friction angles show no systematic dependence on shear strain rate when the material is let free to expand. In these cases, the granular materials have been usually treated as an incompressible Coulomb continuum characterized by a constant friction angle independent of shear rate and solids concentration. Iverson and

214 Denlinger [32] showed that Coulomb behavior can be adopted to simulate flowing materials and they developed a Coulomb mixture theory. Numerical models are capable of simulating both runout and velocity distribution along the path under a broad spectrum of capabilities, limitations, and degrees of sophistication. All these methods depend on the suitable choice of model parameters, requiring proper calibration. This calibration is hardly definitive when only geometrical information (e.g. deposit thickness, maximum or leading-edge runout distance, trim-line tilting derived velocities, runup distance or relief) is available and boundary conditions are complex or partially known (e.g. basal and lateral containment, free surface drag, basal scouring and entrainment and/or deposition during motion, water absorption and material mixing, liquefaction, conditions of the material along the flow-path, etc.). Additional constraints on model parameters can be provided by other field data such as flow duration, velocity estimates or measurements, or debris distribution to achieve a unique solution [6, 7]. A final remark must be made on the fundamental character of flow-like landslides. These phenomena usually entrain (e.g. basal scouring, liquefaction and entrainment, etc.) or deposit (e.g. at the tail along the failure surface, branching, lateral levees and lobes, etc.) material during their motion with major consequences on their dynamic behavior. Nevertheless, very few analyses and models consider this peculiar aspect and try to understand their effects on motion. Furthermore, material entrainment and deposition as well as field observations of a pervasive block shattering and grain size decrease with depth suggest that the use of depth averaged continuum models oversimplify the occurring processes. We performed some numerical models to simulate material entrainment and deposition during motion of granular materials. We tried to analyse cases with simple boundary conditions by using constitutive models for granular materials. 3.

Finite Element Analyses

Continuum models invoke the classical conservation equations of mass, momentum and energy. They have been frequently applied through a depth-averaged approach, where properties and volume remain constant through material columns and no exchange or internal flow of the material is allowed. In general, the adoption of a continuum approach presents the advantage of representing the complex geometry of the avalanche (i.e. flow height, deposit size and geometry, runout length, velocity distribution). In this section we present and discuss the Tochnog Professional Finite Element code [47] an approach that has been used for the modelling of flow-like landslides (e.g. rock and debris avalanches) in this study. More information about this code can be found at 3.1.


Sliding and flowing rock and soil masses show very large displacements and deformations. If a traditional Lagrangian finite element method would be used, the finite element mesh would be subjected to these large displacements and deformations. This would lead quite rapidly to a highly distorted mesh and, as a consequence, the

215 calculated results become inaccurate. For this reason we decided to use a particular type of combined Eulerian-Lagrangian method. For such method, material displacements do not distort the FE mesh, such that accurate calculation results can be obtained. As mentioned above, the displacements and strains in slides of rock and soil masses can be very large, especially for flow-like movements. Besides discretization issues, this also influences the material law that should be used. It should be such that any rigid body rotation should only lead to stress rotation, but not to additional stresses within the flowing mass. The problem, however, consists of the fact that the rigid body rotation component is not uniquely defined as part of an arbitrary deformation pattern. Numerous definitions are possible. For sliding and flowing masses, as in the case of large rock and debris avalanches, an updated Lagrange model is the more suitable. We applied an incrementally objective Lagrangian model, based on a polar decomposition of the incremental deformation tensor. The discretization in space is done with isoparametric finite elements. Several types of elements can be used in the Tochnog FE code. For the present calculations we used triangular three-node elements in 2D, and hexahedral eight-node elements in 3D. For discretization in time, Euler backward time stepping is applied, because of its high numerical stability. On top of this Euler scheme, we apply automatic time stepping and control of the number of equilibrium equations, such that a guaranteed bound is obtained of the unbalance error at the end of each time step. Since we disconnect material displacements from the finite element mesh, state variables need to be transported through the mesh. This is done by a Streamline Upwind Petrov Galerkin method (SUPG). For the present research we applied classical elasto-plasticity to model the nonlinear path-dependent behavior of soils. This is the most complete and widely accepted approach to model both rock and soil-like materials. At the same time, Coulomb like behavior has been suggested as valid by different researchers([30, 26, 32, 13]. The parameters adopted for the linear elastic part are the traditional Young modulus and Poisson ratio. As a yield rule, the Mohr-Coulomb, Drucker-Prager and von Mises surfaces have been used and can be coupled. Their characteristic parameters are, namely: a friction angle, a cohesion, a dilatancy angle. Since we cope with granular materials, a non-associated flow rule has been applied. The values of the material properties can be evaluated and attributed in different ways, namely by a simple trial and error calibration procedure, or by selecting representative values according to the type of involved rock and/or soil. This second approach is clearly the most general one because it does not require an antecedent event for calibration to have already occurred. Different constitutive laws are available or under implementation within the code. They include plastic and hypoplastic laws, hardening anisotropic laws and softening laws. To start the numerical calculations, we must reach the initial equilibrium stress state. This section of the computation is performed through quasi-static time stepping at which all inertial effects were left out. The assumption is that this part of the computation models the very long time nature took to establish the initial gravity state. A pre-defined slip or failure surface has been used during the performed simulations. Such pre-defined failure surface can either be determined, in real case

216 studies, from preliminary finite elements stability calculations, from in-situ evidence like major tension or shear cracks, or from post-event descriptions of the main failure surface. The initial movement or occurrence of the landslide can be triggered by either lowering cohesion in time (for example, to simulate heavy rainfall effects), or imposition of a base acceleration diagram (to simulate seismic triggering). After the landslide is triggered and the mass starts moving, time steps are taken until a solution at rest is obtained. Profile geometry can be controlled with high detail and different types of material can be contemporaneously used. Furthermore, layers of material can be placed along the profile to simulate natural conditions, introduce a factor of variability and simulate the effect of material entrainment during the flow. Several post-processing options have been made available in the code and help in the visualization of the results. The material flow, the velocity pattern in time, and other variables, can be shown as avi-true-color movies and mpg files. Pictures of the initial state, intermediate and ultimate at rest state can be obtained in different digital formats. 4.

Numerical Simulations

To understand the capability of the new modelling code, to verify the applicability of this numerical approach to flow-like landslides and to understand the mechanisms acting during their occurrence we ran a set of different models with different boundary conditions and a general curvilinear profile. We decided to use a simple geometry, both for the landslide mass (triangular section) and slope profile, to make easier the comparison of the results and the understanding of the processes. These models include a curvilinear profile with no geometrical irregularity (all cases in Table 1 and case NO in Table 2), a curvilinear profile with a trapezoidal obstacle (O1, O2, O3 in Table 2) or trough (ditch case D1 in Table 2) of different dimensions. A successive set of simulations have been run by placing an elongated trough shape, or a layer of erodible material, to simulate deposition or entrapment of material as well as entrainment of material from the “sliding” surface. Material laws have also been varied. We used, different constitutive laws and yield rules, namely: plastic and strain softening, Mohr-Coulomb, Drucker Prager, von Mises. The choice is linked to their validity for granular materials, to the straightforward evaluation of the property values and to the possibility of simulating their change during flow and strain accumulation. In the following we will show the results from simulation performed using a Mohr-Coulomb yield rule. The softening constitutive law has been applied both to the landslide material and to the interface. Softening is represented in Figure 1 where angle of friction is represented versus the plastic invariant, t


³ kdt .



The plastic invariant k can be written in its incremental form as:


0.5Hijpl Hijpl





is the tensor of the increment of the plastic strain. We applied a logarithmic

law for this relationship, starting from a friction angle of 40° for k = 0 down to 20° for k = 100, and a constant value was adopted beyond this limit. 4.1.


4.1.1. Flow along a smooth profile The first set of simulations has been carried out to check the influence of changes in friction angles, both of the landslide material and of the interface. Then, we decreased the interface friction angle maintaining constant that of the landslide (cases A1 and A2 in Table 1) and viceversa (cases A2 and A3 in Table 1). Table 1 - List of models run with simple geometry, no obstacle, ditch and entrainment: Slope of the inclined plane is 31°. For all the reported simulations we adopted a Mohr-Coulomb yield rule and a linear-elastic law and the following parameters, namely: E = 100.000 kPa, Q = 0.23, c = 0.05 kPa. Size of triangular mass in initial position used to express volume of landslide material: 2900 m2 (V=1).

Mod el A1 B1 A2 A3 S1


Interface properties, Ii 27.8° 27.8° 18.5° 18.5° 18.5°

Landslide properties Friction angle, Ir 37° 37° 37° 32° 37°

1 ½ 1 1 1

Runout beyond curve sector (m) 12 4 72 82 220

friction angle [deg]


40 35

friction angle [deg]




25 0 1 2 3 4 5 6 7 8 9 10 kap

25 20 15 10 5 0 0





50 60 kap



90 100

Figure 1. Strain softening law expressed in terms of decrease in friction angle with increasing value of the plastic invariant, K. Logarithmic decrease is applied up to k = 100 and beyond this threshold value a constant friction angle of 20° is assumed.


Figure 2. Geometrical conditions adopted for the numerical simulations. The landslide mass has triangular section and has been divided in 6 different layers of equal thickness (different grey tones).

The ratio between the friction angles of the two materials is maintained constant to verify the influence of the friction angle of landslide material under constant reduction factor conditions. The initial geometry is represented in figure 2 with the initial sloping sector (31°), the curved joining part and the horizontal sector. The landslide material is coloured in six different horizontal layers to follow the redistribution of material during flow and deposition. The same representation has been chosen to compare the results from the different models listed in Table 1 (Figure 2a, 2b, 2c, 2d). Looking at the final distribution of the material for cases A1 and A2 (Figure 3a and b) we note that the runout for case A2 is longer but that the distribution of the different layers is quite similar, with upper layers of the sequence moving at the front and passing the lower layers which occupy the middle and tail sectors. Similar but more evident conditions are recognizable for cases A2 and A3. In particular, decreasing the angle of friction of the landslide material (case

a) L =12 m

c) L =82 m

b) L =72 m

d) L =220 m

Figure 3. Landslide mass redistribution after the end of the movement (t = 100 s) for cases a) A1, b) A2, c) A3 and d) S1 listed in Table 1. The different material distribution is visible and the runout length (L) beyond the curvilinear limit is reported . The strain softening case (S1) has been run with a longer horizontal sector and the arrow in d) show the profile limit as in case a), b) and c).


Table 2. List of models run to study the influence of the constitutive law and the friction angle on initial motion. Slope of the inclined plane is 31°. Maximum velocity computed at different time steps and total distance covered by the front are reported. Model



C1 S1 C3 S3

40° 40°-20° 40° 40°-36°

20° 40-20° 36° 40-36°

Maximum Velocity (m/s) T=1s 8.5 8.8 8.5 8.7

T=2s 15.0 15.3 15.4 15.8

T=3s 18.6 20.1 18.5 19.9

Distance run by the front (m) T=4s 20.3 22.1 19.6 21.2

49 55 44 49

A3), by maintaining the same value for the friction angle along the interface, causes a more continuous deposition of the material from the lower layers along the sloping sector of the profile. In the case of strain softening the upper layer covers the entire a)












Figure 4. Comparison of geometry, mass ridistribution, failure mechanism, and plastic strain for the initial stages of motion. 4 different models (S1, C1, S3 and C3) are represented with properties described in Table 3. Figures a) to d) show the geometry and failure mechanisms, whereas Figures e) and f) show plastic strain for two cases (S3 and C3).

220 length of the deposit, including part of the sloping sector of the profile, and forms also the thin and elongated front of the flow. This is due both to the larger size of the upper layer (as visible in figure 1) but it also demonstrates, when compared with the other examples, that yielding occurs initially in the bottom layers to diffuse progressively through to the upper ones. The consequences of the strain softening law can be investigated also by looking at the initial stages of motion. We present in the following the results for the initial 4 seconds after motion onset (Table 2). A couple of tests (C1, C3) have been performed adopting a plastic law with no strain softening, a constant friction angle for the landslide material (40°) and with two different friction angles for the interface, 20° and 36° respectively. For the second couple of tests (S1 and S3) we introduced the same softening law as presented above (Figure 4). The consequences of the choice of these sets of values are also represented in Figure 4 in terms of mass distribution (layers of different grey tone) and contour lines for the plastic invariant k. Again the adoption of the softening law increase the distance run by the front within the same time interval and its maximum velocity. A small decrease of the friction angle (from 40° to 36°) with deformation due to the softening law allows the front to reach (case S3), within the initial 4 s, the same distance for the case with an interface friction angle of 20° (case C1). This suggests that it is unnecessary to have a very low friction angle along the interface or basal surface since the beginning of the motion. In the strain softening cases we computed a decrease of about 13°, during the initial 4 s, at the point of maximum accumulated strain. Furthermore, the interface friction angle seems to play an important role in the typology of the failure mechanism. In fact, a global failure, involving the entire landslide mass, can be recognised in the two cases with low basal friction angle (20° for case S1 and C1) after 4 s. In these two examples, the upper layers do not appear at the landslide front. On the contrary, a partial failure can be observed if the basal friction angle is higher (36°, case S3 and C3), as suggested by the preservation of the horizontal layering in the rear part of the landslide mass. In these two cases the upper layers are an important part of the landslide front. Nevertheless, the two different failure mechanisms produce the same final values of maximum front velocity in contrast with the idea that a lower basal friction angle can give a higher mobility. After the analysis of the initial stages of motion, we examine the effects of the choice of the constituve laws and parameter values on the flow and total runout. The friction angle of the interface clearly results in more influence on the flow behavior (runout length and velocity) than the angle of internal friction of the landslide material (Figure 5a and b). The strain softening model presents higher values both for the runout length and the computed maximum velocity. In fact, the strain softening constitutive law implies a self-feeding mechanism progressively decreasing the angle of friction. Lower the initial friction angle, larger is the displacement and consequently greater the plastic strains which will induce a further decrease in the friction angle. The velocity distribution within the flowing mass for the strain softening model is reported in Figure 6. It is interesting to note that the maximum velocity occurs at the flow front up to 28-30 s into motion. After this instant a low velocity sector developes





500 450



I if.  soil I



350 300 250 200

I  if. soil I


I if.  soil I


I  soil I if.

vel. max [m/s]


x front [m]

I if.  soil I I soil I if. 







50 0

0 0




20 25 time [s]








20 25 time [s]





Figure 5. Influence of landslide material and interface friction angles on flow behavior. A) Position of the flow front vs time for different models. The two horizontal lines (xfront = 275 and 325 m) represent the beginning and the end of the curved sector along the path. The numbers for each curve show the arrest time for each model. B) Maximum velocity computed within the flowing mass at different time steps as a function of material and interface properties.

along the horizontal zone of the profile behind the frontal part of the mass and front velocity becomes comparable with, or lower than, the velocity computed within the landslide mass still flowing along the main slope sector above the curved connection. Dependence of flow behavior (in terms of front position, runout length and maximum computed velocity) on the volume of the landslide has been analysed through 30


time [s]

16 2









velocity [m/s]








30 32 34


36 38 40


0 0



300 x [m]




Figure 6. Isochrones showing velocity distribution within the flowing mass, in proximity of the interface, at different time steps (2 s interval). The material is characterised by a Mohr Coulomb strain softening constitutive law. The two vertical lines (xfront = 275 and 325 m) represent the beginning and the end of the curved sector along the path.

222 a couple of models. The two adopted volumes are identified through the cross section areas of 1450 m2 (V=1/2 in Table 1) and 2900m2 (V=1 in Table 1). The results are summarized in Figure 7 and in Table 1. It can be observed that the runout lengths have higher values for larger volumes at a given time. More relevant, is the different position of the flow front at different time steps during the motion (Figure 7a). Larger masses seem to flow more rapidly but stop more rapidly than smaller ones. The same trend is followed by the maximum velocity. We also observe (Figure 7b) that, after the flow front passes the curved sector of the profile (between 22 and 30 s ca.), the maximum velocity is higher for the small volume landslide. 4.1.2. Deposition and entraiment As mentioned above, deposition of landslide material and entrainment of material available along the slope can strongly affect both depositional and flow features of flow-like landslides [28].To study this problem we decided to run a set of simulations with a trapezoidal trough or ditch geometry, located along the horizontal sector of the profile, infilled or not with an erodible material. We will start looking at the case of infilled troughs which have been thought to simulate the case of valley sediments at the toe of unstable slopes. Three major cases have been considered (Figure 8 and Table 3), namely: 1) Trapezoidal layer deposited above the horizontal part of the profile with a high length vs. height ratio (case E1, Figure 8a and b) 2) Double trapezoidal layer, with a lower trapezoid forming the infilling of a trough and an upper trapezoidal layer symmetric to and laying above the previous one (Case E2, Figure 8c) and emerging from the horizontal profile line 3) Trapezoidal layer forming the infilling of a trough perfectly aligned with the horizontal part of the profile (Case E3, Figure 8d). Each trapezoid has a maximum and minimum base length of 97 m and 50 m, respectively with a height of 5 m. The main results are summarised in Table 4, and Figure 9 shows the final geometries of the deposit for the three cases. We attributed a low friction angle (15°) to the erodible material and to the interface (7.5°) to simulate a very weak (e.g. alluvial and saturated) basal layer. 30

36 s


40 s




vel max [m/s]

x front [m]

250 200 150


10 100 2 I if.  A = 2900 m soil I


2 I if.  A = 2900 m soil I



2 I if.  A = 1450 m soil I

A = 1450 m I if.  soil I


0 0




20 25 time [s]








20 25 time [s]




Figure 7. Influence of volume on flow behavior. The two curves in each plot are relative to two different volumes expressed in terms of the cross section area (1450 m2 and 2900 m2). A) front position versus time. The two horizontal lines (xfront = 275 and 325 m) represent the beginning and the end of the curved sector along the path. B) Maximum computed velocity versus time.

223 a) case E1

b) case E1 b)


c) case E2 c)

d) case E3 d)

Figure 8. Geometric conditions examined to simulate material entrainment during flow of a granular mass. a) and b) Trapezoidal deposit with landslide layers in the initial position; c) double trapezoidal deposit emerging 5 m above the horizontal profile line; d) trapezoidal deposit, 5 m deep, perfectly matching the horizontal profile.




Figure 9. Final geometries of the landslide mass for the three cases a) case E1, b) case E2, and c) case E3 (see Table 4) used to simulate effects of material entrainment along the path.. A longer horizontal sector has been adopted for Case E3 in figure c).

Table 3. List of models run to simulate material entrainment during motion. Slope of the inclined plane is 31°. For all the reported simulations we adopted a Mohr-Coulomb yield rule and elastic linear law, and the following parameters: E = 100.000 kPa, Q = 0.23, c = 0.05 kPa. Model



Trapezoid above profile Trapezoid below and above profile Trapezoid below the profile

E2 E3

Landslide (LS) and Deposit (DP) properties Friction angle, Ir LS =37°, DP = 15° LS =37°, DP = 15° LS =37°, DP = 15°

Interface properties, Ii

Runout beyond curve sector (m)

Arrest time (s)

LS =18.5°, DP = 7.5° LS =18.5°, DP = 7.5° LS =18.5°, DP = 7.5°







224 Erosion and material entrainment can occur more easily in case E1 where the layer is rising above the profile and the basal friction angle is lower than the internal friction angle of the landsliding material. In this case, as for the following one, the three major steps are: the impact between the mass in motion and the one resting along the profile, the front acceleration of the trapezoid, and the arrest. In Figure 10 (Case E1) we observe the impact and the slightly upward direction of the velocity vectors (after 12 s) followed by their re-orientation parallel to the horizontal layer. A progressive deceleration behind the front is visible whereas the front accelerates. The flow front is formed in small part by the landslide front overcoming the weak sediment and in a larger part by the weak sediment itself. A similar pattern for velocities is observed also in cases E2 and E3 with different geometries of the weak sediment layer. These two cases are quite similar but for the displaced volume (larger for case E2) and the total runout length. In fact, the impact and thrust action of the landslide mass in case E2 (Figure 11) induce the entrainment and mobilization of a larger mass of weak sediment, and a more rapid arrest of the motion. The entrained material stays at the front of the moving mass and it is pushed, and transported forward without being passed by the landslide material. Only a thin layer of weak sediment remains close to its original position and below the landslide mass. The original landslide material runs in case E2 the shortest distance with respect to cases E1 and E3 and the same occurs for the maximum runout length (inclusive of the mobilized and entrained sediment: 181.6 m). The conditions for case E3 (Figure 12) are the ones most frequently occurring at flow-landslide sites. In fact, the model simulates the condition of the flow-slide riding on a weak sediment layer filling the valley bottom.

T = 12 s

T = 18 s

T = 14 s

T = 20 s

T = 16 s

T = 22 s

T = 24 s

Figure 10. Case E1 – Trapezoidal sediment layer laying above the horizontal sector of the profile. Velocity vectors are reported at different time steps between 12 s and 26 s.


T =10 s

T =12 s

T =14 s

T =16 s

T =18 s

T =20 s

T =22 s

Figure 11. Case E2 - Mass distribution and velocity vectors for the case of a weak sediment layer filling a trough and rising above the horizontal sector of the profile. The impact and thrust action of the landslide mass cause entrainment and mobilization of a mass of weak sediment.

In this case there is no direct impact between the landslide front and the weak layer. The landslide material moves along the curved connecting sector and at its exit it conserves a small vertical component of velocity. The layer of weak material is almost completely eroded and mobilized by the landslide. The original landslide material runs the longest distance with respect to the other case studies. The same is true for the material pushed at the flow front (runout length = 226.7 m).


T =10 s

T =12 s

T =14 s

T =16 s

T =18 s

T =20 s

T =22 s

Figure 12. Case E3 - Mass distribution and velocity vectors for the case with a weak sediment layer filling a trough up to the same elevation of the horizontal sector of the profile. The impact and thrust action of the landslide mass cause entrainment and mobilization of the entire mass of weak sediment.

4.1.3. Flow against obstacles Flow-like landslides have the ability to flow around and beyond obstacles of different shapes located in different positions along the path. Furthermore, carefully designed artificial obstacles could be used to contain flow-like landslides of relatively small volume. We carried out simulations on models with the same profile geometry as before but inserting obstacles of different shape (embankment and ditch), position along the

227 profile, and mechanical characteristics (Figure 13, Table 4). Results of these simulations have been compared with those along an identical profile with no obstacle (Table 4) and the final geometry of the material for the examined cases is reported in Figure 14. Case O1 simulates the presence of a deformable obstacle, with better properties compared to the landslide mass, at the curved connection along the profile. This is the point where flow-slides generally show their maximum velocity. The landslide mass hits the steep upslope side of the obstacle, deforms and mobilizes it while surmounting it, and rapidly stops (Figure 14b). The runout length is about the 33% of the case with no obstacle (case NO, Figure 14a). The distribution of velocity vectors for case O1 is also shown in figure 15. We observe that obstacle mobilization starts immediately after the impact of the landslide material on the uphill flank. Then, the obstacle moves but with lower velocity with respect to the landslide material which surmounts it and stops with a characteristic steep snout. Case O2 considers a rigid obstacle (Figure 14c) with properties equal to the rest of the profile whereas an elastic obstacle is introduced in case O3 (Figure 14d). In these two cases the obstacle remains in its position greatly increasing energy dissipation. The maximum runout reaches the 57.6% and 31%, of the no obstacle case (NO), respectively.

a) case O1

b) case O2,

c) case O4

d) case O4

Figure 13. Model geometries adopted to study the behavior of flow-like landslides in presence of obstacles of different shape. See Table 4. Inclination of embankment and ditch slopes is 33°. Maximum height of the embankment is 23 m for case O1 and 6 m in all the other cases. Table 4. List of models run with obstacle, or ditch and/or entrainment, located along the horizontal part of the profile with the exception for case O1. For all the cases, we adopted the following parameters: E = 100.000 kPa, Q = 0.23, c = 0.05 kPa. Volume equal to 1 represents the model with cross section area of the initial landslide mass of 2900 m2. Model


Obstacle, Ditch, Entrainment


1 1

O2 O3 O4

1 1 1

No obstacle Elasto-plastic obst. On the curved connection Rigid obst. Elastic obst. Elasto-plastic obst.




Landslide (LS) and Obstacle (OB) properties Friction angle, Ir 37° 40°

Interface properties, II

Runout beyond curve sector (m)

18.5° 34°

122.3 40.9

37° 37° LS = 37° OB = 37°

18.5° 18.5° LS = 18.5° OB = 18.5° 18.5°

70.5 38.8 105.0




a) Case NO

b) Case O1

c) Case O2

d) Case O3

e) Case O4

f) Case D1

Figure 14. Flow of landslide material against obstacles of different shapes compared to the case with no obstacles (NO). Contours of plastic invariant, k, showing the behavior of the material are reported.

The runout reduction is minimum for case O2 because energy dissipation occurs only through the friction force acting along the interface with the same angle of friction characteristic of the profile. For an elasto-plastic obstacle, characterized by the same properties as the landslide mass (case O4, Figure 14e), we simulate a condition similar to erosion and material entrainment. The maximum runout is 86% of the no obstacle case and the obstacle is stretched below the landslide material and forms the front of the flow. The obstacle material is tapered with minimum thickness upslope and maximum thickness at the front. Obstacles can coincide with a concavity in the profile. This is the case of a ditch like geometry located along the horizontal sector of the profile. In this condition (Case D1, Figure 14f) the runout reduces to the 52% of the NO case. When the material reaches and fills the ditch a shear zone develops within the landslide material. Part of the landslide mass is lost in the ditch and the shear of the landslide material is controlled by its angle of friction. Comparing this case with the symmetric one including a rigid embankment (case O2), the ditch is less effective in slowing down and arresting the flow. This could be considered in designing passive countermeasures to flow-like landslides. 5.

Discussion and Conclusions

Numerical methods for flow-like landslide modelling depend heavily on different factors, namely, the type of natural phenomenon, the type of modelling approach, the appropriate selection of yield and constitutive models, and model parameters, the understanding of the assumptions and limitations of the adopted modelling procedure, the suitability of the model to simulate or include both morphological and physical-


T =6 s

T =8 s

T =10 s

T =12 s

T =14 s

Figure 15. Flow of landslide material against a deformable obstacle placed in correspondence of the curved sector along the profile (O1). Vector velocity at different time steps are shown.

mechanical constrains and boundary conditions. In this paper we presented the application of a new numerical modelling approach based on a highly sophisticated finite element technique. We applied this approach to different geometrical conditions and to material with different sets of properties. This modelling activity was aimed at the understanding of the behavior of flow-like landslides and of the suitability of numerical modelling approaches proposed in the literature. We back-analysed [6, 7] large rockslides and rock avalanches (Vajont and Val Pola) for which initial and final geometries, runup on the opposite valley flank, total duration, material characterization and released seismic energy records were available. This evidenced the capabilities of the adopted numerical approach and the possibility to perform a sound calibration of two-dimensional numerical models. The material involved in large flow-like landslides is often characterised by a wide range of grain sizes. The deposits include massive blocks as well as very fine particles (up to clay size). Fragments less than 20 cm are the most frequent, but the heterogeneity often precludes the performance of meaningful laboratory and field measurements [51, 34, 41]. Behaviour of the finer fraction is not generally representative of the entire flow but different researchers have shown that fine materials can control the shear behavior

230 of granular materials when they amount to the 50 to 60% of the flowing mass. Furthermore, deformation mechanisms and abnormal pore fluid pressure conditions typical of large scale flows can occur and are difficult to replicate in small scale testing. Sedimentological observations suggest that a continuous and progressive disaggregation and fragmentation continues all along the flow path. Thus, an important model limitation is the assumption of constant mechanical properties throughout flow occurrence. For long-runout flows modelled with constant mechanical parameters and calibrated according to the total runout length it seems more correct to adopt values of the material properties typical of the conditions existing during the intermediate to terminal stages of flow [53]. This assumption can be at the origin of high early-stage velocities. To overcome this limitation, we have implemented in the numerical code a strain softening behavior for the landslide material as well as for the surface on which movement takes place. As shown in the paper this assumption is able to increase considerably the runout potential of granular flows without an excessive decrease in the friction angle value since the beginning of motion. In fact, a resistant interface does not guarantee a slow movement of the mass in the initial phases but influences mainly the runout stage. This observation is quite interesting because this result cannot be observed using a depth-averaged model. Most of the continuum models presented in the literature to simulate flow-like landslides apply a depth-averaged approach. In our study we tested the possibility to consider and analyse internal material strains checking for their influence on flow development. As presented in the simulations, the depth-averaged assumption can be quite limiting especially in presence of such morphologic features as sharp changes in flow direction, knee points, obstacles of different nature and deformability, etc. These conditions induce relative displacements, distorsion and plastic deformations within the flowing mass which cause energy dissipation and which cannot be simulated through a depth-averaged approach. Similar problems in the simulation of flow-like landslides arise when trying to include the effects of material entrainment and deposition, In fact, it is known that entrainment for some rock avalanches and especially for small flow-like landslides, might become relevant, implying an increase in volume up to 3 orders of magnitude. Eventually, debris cover (talus, colluvial and alluvial deposits) could play a determinant role in increasing flow mobility by different mechanisms (undrained loading and liquefaction, strain localization in soft materials, weak wet sediment and river water entrainment, etc.). A fully two dimensional or three dimensional model must be able to simulate material entrainment and deposition. We have shown that entrainment has an important influence on flow behavior and that eroded and entrained material slow down the landslide by mass increase and increased friction action. Momentum transfer occurs between the landslide material and the eroded material placing the latter in front of the flow. Furthermore, mass removal and entrainment require energy dissipation but the increase in volume induces an increase in length of the flowing mass and then of the runout length. Acknowledgements All finite elements plots are courtesy of Finite Element Application Technology, FEAT. The research has been funded by the EC Project Damocles EVG1-CT-1999-00007.

231 References 1. 2.

3. 4. 5. 6.

7. 8. 9. 10. 11. 12. 13.

14. 15. 16. 17.



20. 21. 22. 23. 24. 25. 26. 27.

Abele, G. (1974) Bergusturze in den Alpen. Wissenschaftliche Alpenvereinshefte 25, Munchen. Calvetti F., Crosta G. and Tatarella M. (2000) Numerical simulation of dry granular flows: from the reproduction of small-scale experiments to the prediction of rock avalanches. Rivista Italiana di Geotecnica, A.G.I., 21, 2/2000, 21-38. Campbell, C.S. (1989) Self-lubrication for long runout landslides. J. Geol., 97, 653-665. Chen, H. and Lee, C.F. (2000) Numerical simulation of debris flows. Canadian Geotechnical Journal 37, 146-160. Crosta, G. (1991) Studio di movimenti in massa. Modellizzazione teorica e sperimentale con osservazioni e rilievi di campagna. PhD Thesis, Univ. Studi di Milano, 137 pp. Crosta, G., Imposimato, S., and Roddeman, D. (2002) Numerical modelling of large landslide stability and runout, EGS 27th General Assembly, Geophysical Research Abstracts, European Geophysical Society, vol. 4, 2002 EGS02-A-02241. Crosta, G., Imposimato, S., and Roddeman, D. (2003) Numerical modelling of large landslide stability and runout,. Natural Hazard and Earth System Sciences, in press. Cruden, D.M., Varnes D.J. (1996) Landslide types and processes. In: A.K. Turner and R.L. Schuster (eds.) Landslides: Investigations and Mitigation, Transportation Res. Board, Special Report 247: 36-71. Cruden, D.M., Hungr, O. (1986) The debris of the Frank Slide and theories of rock-avalanche mobility. Canadian Journal of Earth Sciences, 23, 425-432. Cruden, D.M., Lu, Z.Y. (1992) The rockslide and debris flow from Mount Cayley, B.C., in June 1984. Canadian Geotechnical Journal, 29: 614-626. Davies, T.R.H. (1982) Spreading of rock avalanche debris by mechanical fluidization. Rock Mechanics, 15, 9-24. Davies T.R., McSaveney M.J., Hodgson K.A. (1999) A fragmentation-spreading model for long-runout rock avalanches. Canadian Geotechnical Journal, 36 1096-1110. Denlinger, R. and Iverson, R.M. (2001) Flow of variably fluidized granular masses across threedimensional terrain: 2. Numerical predictions and experimental tests Journal of Geophysical Research, v.106, no.B1, p.553-566, 2001. Eisbacher, G.H., Clague, J.J. (1984) Destructive mass movements in high mountains: hazard and management. Geol. Survey of Canada, Paper 84-16, 230 pp. Erismann, T.H. (1986) Flowing, rolling, bouncing, sliding: Synopsis of basic mechanisms. Acta Mech., 64, 101-110. Erismann, T.H. and Abele G. (2001) Dynamics of rockslides and rockfalls. Springer, Berlin, 316 p. Evans, S.G., and Clague, J.J. (1988) Catastrophic rock avalanches in glacial environments. In Proceedings of the 5th International Symposium on Landslides. Edited by C. Bonnard, 10–15 July, Lausanne, Switzerland. A.A. Balkema, Rotterdam, The Netherlands, Vol. 2, 1153–1158. Evans, S.G., Aitken, J.D., Wetmiller, R.J., and Horner, R.B. (1987) A rock avalanche triggered by the October 1985 North Nahanni earthquake, District of Mackenzie, NWT. Canadian Journal of Earth Sciences, 24: 176–184, 1987. Evans, S.G., Hungr, O., and Enegren, E.G. (1994) The Avalanche Lake rock avalanche, Mackenzie Mountains, Northwest Territories, Canada: dating, description, and dynamics. Canadian Geotechnical Journal, 31: 749–768. Fread, D.L. (1988) The NWS DAMBRKmodel: theoretical background and user documentation. National Weather Service, Silver Spring, Md, Hydrologic Research Laboratory, HRL-256, 315 pp. Glicken, H.X. (1998) Rockslide–debris avalanche of May 18, 1980, Mount St. Helens volcano, Washington. Bulletin of the Geological Society of Japan, 49: 55–105. Habib, P. (1976) Production of gaseous pore pressure during rock slides. Rock Mechanics, 7, 193-197. Hadley, J.B. (1978) Madison Canyon rockslide, Montana, USA. In Rockslides and avalanches. Vol. 1. Edited by B. Voight. Elsevier, Amsterdam, The Netherlands, 1, 170-183. Heim A. (1932) Bergsturz und Menschenleben. Fretz und Wasmuth, Zurich, Switzerland, 218 pp. Hsu, K. (1975) Catastrophic debris streams (sturzstroms) generated by rockfalls. Geol. Soc. of America Bull., 86, 129-140. Hungr O. (1995) A model for the runout analysis of rapid flow slides, debris flows and avalanches. Canadian Geotechnical Journal, 32: 610-623. Hungr, O. and Morgenstern N.R. (1984) Experiments on flow behavior of granular materials at high velocity in a n open channel flow. Geotechnique, 34: 405-413.

232 28. 29. 30.

31. 32. 33. 34. 35.

36. 37. 38. 39. 40. 41. 42. 43.

44. 45. 46.

47. 48. 49. 50. 51. 52. 53. 54.

55. 56.

Hungr, O. and Evans, S.G. (1996) Rock avalanche runout prediction using a dynamic model. Proceeding of the 7th Inter. Symposium on Landslides, Throndeim, Balkema, Rotterdam, 233-238. Hutchinson, J.N. (1986) A sliding-consolidation model for flow slides. Canadian Geotechnical Journal, 23: 115-126. Hutter, K., Savage, S.B. (1988) Avalanche dynamics: the motion of a finite mass of gravel down a mountain side. In Proceedings of the 5th International Symposium on Landslides. Edited by C. Bonnard, 10–15 July, Lausanne, Switzerland. A.A. Balkema, Rotterdam, The Netherlands, 1, 691-697. Hutter K., Koch T., Plüss C., and Savage S.B.(1995) The dynamics of avalanches of granular materials from initiation to runout. Part II. Experiments. Acta Mechanica, 109, 127-165. Iverson, R.M. and Denlinger, R. (2001) Flow of variably fluidized granular masses across threedimensional terrain: 1. Coulomb mixture theory. Journal of Geophysical Research, 106, B1, 537-552. Jeyapalan, J.K., Duncan, J.M., Seed, H.B. (1983) Investigation of flow failures of tailing dams. J. Geotech. Eng., ASCE, 109(2), 172-189. Johnson B. (1978) Blackhawk landslide, California, USA. In Rockslides and avalanches. Vol. 1. Edited by B. Voight. Elsevier, Amsterdam, The Netherlands, 1, 481-504. Kent, A. and Hungr, O. (1995) Runout characteristics of debris from dump failures in mountainous terrain: stage 2: analysis, modelling and prediction. British Columbia Mine Waste Rock Pile Research Committee and CANMET-Western Research Centre, Edmonton, Alta. Kilburn C.R.J., Soresen, S.A. (1998) Runout lengths of sturzstroms: the control of initial conditions and of fragment dynamics. Journal of Geophysical Research, 103: 17,877- 17, 884. Koerner, H.J. (1976) Reichweite und Geschwindigkeit von Bergsturzen und FlieEschneelawinen. Rock Mechanics 18, 225-256. Koerner, H.J. (1977) Flow mechanisms and resistances in the debris streams of rock slides. Bull. Int. Assoc. Eng. Geol., 16, 101-114. Legros, F. (2001) The mobility of long runout landslides. Eng. Geol., 63, 3-4, 301-331. McEwen, A.S. and Malin, M.C. (1989) Dynamics of Mount St. Helens’ 1980 pyroclastic flows, rockslide-avalanche, lahars, and blast. Jour. Volcanology and Geothermal Research, 37: 205-231. McSaveney, M.J. (1978) Sherman Glacier rock avalanche. In Rockslides and avalanches. Vol. 1. Edited by B. Voight. Elsevier, Amsterdam, The Netherlands, 71–96. Melosh, H.J.: (1979) The physics of very large landslides. Acta Mech., 64, 89-99. Moore, J.G., Clague, D.A., Holcombe, R.T., Lipman, P.W., Normark, W.R., and Torresan, M.E.: (1989) Prodigious submarine landslides on the Hawaiian ridge. Journal of Geophysical Research, 94B: 17,465 – 17,484. Novosad J. (1964) Studies on granular materials II. Colln. Czech. Chem. Commun., 29, 2697. O’Brien, J.S., Julien, P.Y., Fullerton, W.T. (1993) Two dimensional water flood and mudflow simulation. Jour. Hydraulics Div., ASCE, 119(HY2), 244-261. Pariseau, W.G., and Voight, B. (1979) Rockslides and avalanches: basic principles, and perspectives in the realm of civil and mining operations. In Rockslides & Avalanches. Edited by B. Voight. Developments in Geotechnical Engineering, 14b, 1-92. Roddeman, D.G. (2001) TOCHNOG user’s manual a free explicit/implicit FE program. FEAT, 177 pp. Sassa, K. (1988) Geotechnical model for the motion of landslides (Special lecture). Proceedings of the 5th International Symposium on Landslides, 37-56. Savage S.B., Sayed M. (1984) Stresses developed in dry cohesionless granular materials sheared in an annular shear cell. J. Fluid Mech., 142, 391-430. Scheidegger, A.E. (1973) On the prediction of the reach and velocity of catastrophic landslides. Rock Mechanics, 5, 231-236. Shreve, R.L. (1968) Leakage and fluidization in air-layer lubricated avalanches. Geol. Soc. of America Bull., 79, 653-658. Siebert, L. (1984) Large volcanic debris avalanches: characteristics of source areas, deposits and associated eruptions. Jour.Volcanology and Geothermal Research, 22:163-197. Sousa J., Voight, B. (1991) Continuum simulation of flow failures. Geotechnique, 41, 515-538. Sousa J., Voight, B. (1995) Multiple-pulsed debris avalanche emplacement at Mount St. Helens in 1980: Evidence from numerical continuum flow simulations. Journal of Volcanology and Geothermal Research, 66: 227-250. Voight, B., Glicken, H., Janda, R.J., Douglass, P.M. (1981) Catastrophic rockslide-avalanche of May 18. U.S. Geol.Surv., Prof. Pap., 1250: 347-377. Voight, B. and Sousa, J. (1994) Lessons from Ontake-san: a comparative analysis of debris avalanche dynamics. Eng. Geol., 38, 261-297.

LANDSLIDE MOBILITY AND THE ROLE OF WATER F. LEGROS1 Instituto Geofísico del Perú Urb. La Marina B19, Cayma, Arequipa, Peru

Abstract Landslides are known to travel further than expected from the coefficient of friction of their material. In some cases, this is just because the ratio of the height lost to the horizontal distance travelled (H/L), which is compared to the coefficient of friction, is not computed from the centre of mass of the deposit, as it should be, but from the distal end. Simple spreading of the landslide mass can then explain the excess runout. However, spreading alone is not able to explain the spectacular runout of most landslides, for which the centre of mass does travel further than predicted for a frictionally-controlled slide. The long travel distance of the centre of mass cannot be explained by dry granular models. As it is well known that water reduces solid friction in debris flows, and that significant amounts of water are present in many landslides, it is proposed here that water is the main cause for the unexpectedly high mobility of landslides. Water in the debris also introduces a viscous dissipative stress which can account for the relatively channelled behaviour of landslides over topography. The difference between landslides and debris flows is wholly gradational and related to the water content.



Landslides are gravity-driven, mass movements which occur in a wide variety of terrestrial and extraterrestrial environments. In mountainous areas, they are an important agent of erosion and affect profoundly the morphology. They also have a significant impact on human activities, being able to destroy and bury large areas in an extremely short time, or to generate destructive tsunamis where they enter a piece of water. An issue of particular interest for human safety is the distance that a landslide is able to reach and the area that it can sweep. On Earth, subaerial landslides commonly travel several kilometres, and runout distances of several tens of kilometres are not uncommon, even in the historical record. The runout distance may depend on several factors including the volume of the landslide, its lithology, the topography of the path and the 1

E-mail: [email protected]

233 S.G. Evans et al. (eds.), Landslides from Massive Rock Slope Failure, 233–242. © 2006 Springer. Printed in the Netherlands.

234 nature of the substrate, the amount of water and other fluids in the landslide, the value of the gravity acceleration and the subaerial or submarine environment. The volume can vary from a few cubic metres to tens of cubic kilometres in subaerial landslides [9, 36, 37], and up to thousands of cubic kilometres for submarine landslides [8, 23] and for landslides on Mars [27]. The topography can be steep or gentle, and the landslide can be channelled in a narrow valley or spread in an open landscape. The lithology can be strong or weak, the substrate can be variably permeable, saturated with water or not. In order to improve assessment of the hazards associated with landslides, the effect of each of these factors on the landslide runout should be better understood. A main point discussed in this paper is the influence of water present in the landslide mass on the runout distance. With regard to this issue, it is necessary to define more closely what we understand by landslide, and more particularly the difference between landslide and debris flow. Most of what is known about landslides comes from the study of their deposits. Given the variety of conditions in which landslides occur, it is not surprising that the characteristics of their deposits are also varied. However, for most authors, megablocks with jig-saw fractures and a hummocky topography seem to be two important elements to distinguish landslide from debris-flow deposits. Regarding the mechanisms, there is agreement that debris flows generally contain more water than landslides, and that this is a main reason for their smoother deposits and longer runouts. Clearly, landslides and debris flows are part of a continuum. There is no particular reason to think that mass flows should be either totally dry or fully saturated with water. To the contrary, there is good evidence that many landslides do contain a substantial amount of water. Evidence for a continuum of water content is given by landslides which laterally transform into debris flows [17, 29, 31, 41, 42, 45]. Landslides involve the failure of large masses of rock, sometimes to great depths, which, in the rainy environment where they usually occur, are likely to contain a water table. The presence of a water table is indeed often necessary for failure to occur [17]. Water can later be added to the landslide by mixing with river or sea, or by entrainment of watery sediment from a valley or the seafloor. Direct and indirect evidence for the presence of water in specific landslides have also been presented, including for several landslides in New Zealand [30], the Ontake-san landslide [44], the Blackhawk landslide [19], and the Arequipa landslide [22], to quote a few. The amount of water beyond which a landslide transforms into a debris flow (i. e., beyond which the megablocks with jig-saw fractures and hummocks tend to disappear) is not well known, but it is likely that flows in which the pores are fully saturated with water tend to a liquefied state. In contrast, the megablocks in landslides likely represent dry or undersaturated, unliquefied portions of terrain. Landslides are therefore mostly undersaturated mass flows, while debris flows would be mass flows saturated with water.


Mobility of Landslides

More than 20 models have been proposed to explain the great mobility of landslides [33]. The reason for such a productive debate probably comes from the fact that, contrary to the intuitive feeling, landslides do not obey simple frictional laws. According

235 to frictional law, the horizontal distance (L) travelled by a solid block or a granular material sliding on a slope and coming to rest on a flat should be proportional to the elevation lost (H) and inversely proportional to the coefficient of friction, so that the H/L ratio is equal to the coefficient of friction. In contrast, the H/L ratios computed for many landslides are generally much lower than the normal coefficient of friction of rocks (about 0.6), which means that landslides travel further than expected from frictional arguments. Furthermore, H/L ratios are smaller for large landslides than for small ones, while, in the frictional model, there is no effect of the mass on the distance travelled. It should be noted that the same occurs for debris flows, which usually travel much further than landslides. However, the abundance of water is so obvious in debris flows that no special model has been needed to explain their long runout. The common assumption that landslides are essentially dry has been the reason for the need of an alternative model for their mobility. The runout distance of a landslide is the parameter most frequently reported in the literature, because it is both easy to determine and of basic relevance for hazard assessment. Total elevation lost, measured from the top of the scar to the distal end of the deposit, is also often presented because of the implicit idea that it somehow determines, or at least has a strong influence on, the runout distance. As mentioned above, the ratio of these two parameters is generally much lower than the normal coefficient of friction of rocks (Figure 1a). However, in a frictional model, for the H/L ratio to be equal to the coefficient of friction of the material, it must be measured from the centre of mass. It has therefore been argued that the low H/L ratios of landslides could just be due to the fact that they are estimated from the distal end of the deposit. Davies [4] pointed out that landslides spread much during emplacement, and proposed that, just by spreading, landslides with normal coefficient of frictions could achieve the H/L ratios observed in nature. To verify this hypothesis, it is therefore essential to estimate H/L ratios from the centre of mass. Although data on the travel distance of the centre of mass are generally not available in the literature, for well documented landslides deposits, they can be estimated with some accuracy. This was done by Legros [21], who showed that the H/L ratios measured from the centre of mass of seven landslides were much lower than the coefficient of friction of rocks, and were as low as 0.1 for two very large landslides. This demonstrates that, although spreading of landslides does occur and does increase their runout, a reduction of the solid friction is still necessary to account for the high mobility of many of them, especially of the largest ones. Another interesting feature of Figure 1 is the fairly good correlation between runout and volume on the one hand (Figure 1b), and area and volume on the other hand (Figure 1c). Runout is actually a roughly linear function of the cubic root of the volume, while area is a roughly linear function of the volume at the two-third power. This suggests that the geometry of landslide deposits is scale-invariant, just as it has been observed for debris flows [18].


Figure 1. Relationships between total fall height (H ; Figure 1a), runout distance (L ; Figure 1b), area covered by the deposit (A ; Figure 1c) and volume (V) and for landslides in various environments and debris flows. Modified after Legros [21].


Velocity of Landslides

In order to constrain the dynamics of landslides, data on their velocity are extremely useful. These are not as straightforward to obtain as the runout distances, but peak velocity can be estimated where the landslide has run up against topographic obstacles. For three well documented landslides (Mount St. Helens [45], Ontake-san [44], and Nevado de Colima [38]), the velocities estimated from the run up against transversal obstacles are compared with the velocities predicted by a model that assumes a constant coefficient of friction, or any other model in which energy linearly decreases with distance (Figure 2). It is seen that the kinetic energy of the three landslides is always much lower than that predicted by a straight energy-line model. Energy dissipation is therefore not constant, being greater on the initial, steep slopes, where velocity is high, and lower on the distal, gentle slopes, where velocity is low, thus suggesting that energy dissipation is positively correlated with velocity. By attempting to model the 1980 Mount St. Helens landslide numerically, McEwen and Malin [28] also found that a frictional model was not appropriate and that it was necessary to add a velocitydependent, viscous term of friction to reproduce the channelled path of the landslide. The evidence for a velocity-dependent term of energy dissipation must be taken into account in any model that attempts to explain landslide mobility.

237 4.

Fluid-absent Models

Several models have been proposed in which landslide mobility was explained in the absence of any fluid [1-6, 39,40]. Most of them are somehow based on the concept of the collisional regime of grain flows, or some variation of it. A grain flow that moves slowly dissipates energy through enduring, frictional contacts between grains, and its runout is controlled by the internal coefficient of friction of the material. In contrast, in a fast-moving grain flow, rather than rubbing against each other, grains collide and provoke an expansion of the flow. Energy is dissipated through the short-lived, inelastic collisions, while friction is reduced or suppressed. Some authors assumed that, by this mechanism, the total dissipation of energy might be reduced and that landslides might travel further than in the frictional regime. However, both experiments and theory show that this is not the case, as grain flows in the collisional regime are self-regulated in such a way that they dissipate the same amount of mechanical energy as in the frictional regime [13, 14, 20, 32]. Numerical experiments by Straub [39, 40] confirm that grain flow in the collisional regime have straight energy lines. Surprisingly, Straub concludes that this argues for the importance of collisional grain flow as a major flow mechanism in a wide variety of geological flows. In contrast, here, it is argued that the wide range of H/L values displayed by geological flows, their inverse correlation with the mass of the flow, the fact that they are often lower than the coefficient of friction of rocks, and the tendency of geological flows to follow the topography (thus reflecting relatively low velocities) all indicate that a collisional grain flow is not an appropriate model for most geological flows, and that other mechanisms must operate. A way to increase the runout of a flow, and even the travel distance of its centre of mass, without reducing the frictional stress, was proposed by Van Gassen and Cruden [43]. They suggested that, in a flow which progressively loses mass by deposition along the path, the moving part of the flow would decelerate less and so could achieve a longer runout. This means that, instead of producing a uniform deceleration, friction concentrates on some part of the flow, which slows down and comes to rest more rapidly, while the rest of the flow is not (or less) affected by friction and can travel further. Hungr [15] and Erlichson [7] criticised the analysis and showed that, for the law of momentum conservation used by Van Gassen and Cruden [43] to be valid, an additional source of energy (other than the release of potential energy) was required for the flow. Legros [21] later argued that progressive deposition along the path could actually increase the runout of a flow, but that, in the absence of a source of additional energy, the conservation of mechanical energy should be used instead of the conservation of momentum. A direct consequence is that the distance travelled by the


Figure 2. Energy lines for models that assume a constant coefficient of friction, compared with data of natural landslides (solid diamonds). The velocities inferred for the three landslides are small (diamonds close to the topography line). Vertical and horizontal coordinates, x and z, are normalised by H and L, respectively. Shaded area represents the failing mass and CM is the centre of mass. Modified after Legros [21].

239 centre of mass cannot be increased. Therefore, transfer of mechanical energy from the rear of landslides to their front, as envisaged by Heim [10], may partly explain their spreading, but not the long runout of all of them, particularly of the largest ones, unless a source of additional energy is found. It has been recently proposed that such a source of energy might be provided by the elastic energy stored in the compressed rock mass (J.N. Hutchinson, Personal Communication, 2002). When the mass fails and the landslide spreads, rock would decompress and release this energy. The amount of energy that can be stored and then released by this way is easy to estimate, as it is equal to the potential energy that the rock pile loses by compaction under its own weight. Given the low compressibility of rocks, this should generally represent only a very small part of the total potential energy released by the fall of the rock mass. Furthermore, water, which may be present in the pores of the rock, is also very imcompressible, and it is difficult to imagine how a large volume of air could be compressed and then maintained at high pressure inside the rock. The decompression of rock during emplacement of landslides can therefore not cause a significant increase of their runout. If there is no convincing evidence that the centre of mass of a dry grain flow can travel further than expected from its coefficient of friction, it is well established that dry masses of grains can spread during transportation and reach relatively large runout distances. An experimental evidence of this is given by the laboratory sand avalanches of Davies and McSaveney [6], which achieve H/L ratios (measured from the distal end of the deposit) as low as 0.15. Therefore, some landslides might be dry and still have relatively long runouts. This is more likely to occur for small landslides that involve superficial failure. However, for such landslides, the H/L ratio measured from the centre of mass should still be equal to the coefficient of friction of rocks.


Models Involving Fluids

Three types of fluids have been invoked for the partial lubrication or fluidisation of landslides : air, water and volcanic gases. In addition, Hsü [12] proposed that fine particles in the vacuum could form a fluidising suspension. This hypothesis must be discarded, as there can be no pore pressure in the vacuum. Even very fine particles have ballistic trajectories and they cannot form a suspension. Volcanic gases may have played a role in some specific, volcanic landslides, such as the one at Mount St. Helens [45] but cannot be invoked for non-volcanic landslides. Fluidisation by atmospheric air requires the entrainment of a large volume of air in order to fill the pores at lithostatic pressure. For large landslides, this volume becomes huge and probably impossible to achieve [21]. On the other hand, the hypothesis that landslides could glide over a cushion of entrapped, compressed air that would gently leak up through the narrow pores of the debris [34, 35] relies on the unlikely assumption that the loose debris would behave as a coherent block. More likely, the debris would collapse by batches through the air layer, that would be passed through the debris as large bubbles, a process called aggregative fluidisation and observed in experiments [46].

240 Owing to its incompressibility, water in the pores of a debris can be pressurised to lithostatic levels with only a very small reduction of pore volume. Thus, there is no need to incorporate large volumes of water to liquefy a landslide; water initially present in the failing mass is enough. Moreover, aggregative fluidisation does not occur with water [46]. Water is therefore adequate as a fluidising medium. Indeed, it is now well understood how water acts to reduce solid friction in mudflows and debris flows [16]. Due to the load of the debris, pore pressure is raised to lithostatic value, so that water supports the debris and solid friction is strongly reduced. Water is progressively expelled out of the debris during a process called consolidation, which is slow enough to allow the debris to travel significant distances [25, 26]. The presence of water also introduces a viscous, velocity-dependent stress. This allows the debris to be easily channelled by the topography and to flow over gentle slopes. As the consolidation process means that the flow progressively loses material, the runout distance is largely controlled by the total volume of debris mobililised, which can explain why debris flow deposits have a roughly scale-invariant shape [18]. Now, if a substantial amount of water is present in landslides, which, as previously discussed, is the case for many of them, it seems logical that it should play the same role as in debris flows. Therefore, the long runouts of landslides, their responsivity to topography, and the scale-invariant geometry of their deposits can all be explained by the presence of water. A body of evidence which has sometimes been presented against the importance of water for the mobility of landslides is the existence of long-runout landslides on Mars [27] and on the Moon [11]. On Mars, huge landslides (up to 1013 m3) have travelled distances of up to 100 km, with H/L ratios (taken from the distal end of the deposit to the top of the scar) as low as 0.1. H/L ratios for the centre of mass are unknown, but, anyway, the probable presence of ground ice at a depth of a few hundreds of metres [27] suggests that many of these large landslides contained a substantial amount of water. On the Moon, where there is no water, long-runout landslides are in fact extremely rare. Actually, only one has been described [11], but its location near an impact crater suggests that it may be the result of a meteor impact, and so may have been partly emplaced as ejecta [24]. The Moon might therefore provide evidence for, rather than against, the essential role of water for the generation of long-runout landslides.



Landslides are driven downslope by gravity, while dissipative stresses progressively consume their mechanical energy until they come to rest. If the stress is caused by solid friction against the ground or between particles, or by collisions between particles, the horizontal distance travelled by the centre of mass of the landlside is entirely controlled by the coefficient of friction of the material and the height lost. As the landslide can spread, however, the distal end of the deposit can reach a somewhat greater distance. This scenario could apply to dry, typically small, landslides. Many landslides, especially the large ones, contain substantial amounts of water, either already present before failure occurs or incorporated during transport. Water can

241 support part of the load of the debris, thus reducing total solid friction. At the same time it introduces a viscous term of friction, which is velocity-dependent. The result is that these landslides are channelled by the topography and that the distance travelled by their centre of mass exceeds that predicted by the coefficient of solid friction of their material. Such landslides are therefore very similar to debris flows, the difference being that debris flows are essentially saturated with water, while landslides conserve substantial portions of dry or undersaturated rocks, which are responsible for the hummocks and shattered megablocks typical of their deposits. The volume of water exerts an important control on the distance that a landslide can travel in excess of that predicted by solid friction alone. As the volume of water must be positively correlated with the volume of the landslide, the runout distance increases with the volume.

References 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. 16. 17. 18. 19. 20.

Campbell, C.S. (1989) Self-lubrication for long runout landslides, J. Geol. 97, 653-665. Campbell, C.S., Cleary, P.W., and Hopkins, M. (1995) Large-scale landslide simulations: Global deformation, velocities and basal friction, J. Geophys. Res. 100, 8267-8273. Cleary, P.W. and Campbell, C.S. (1993) Self-lubrication for long-runout landslides: examination by computer simulations, J. Geophys. Res. 98, 21911-21924. Davies, T.R.H. (1982) Spreading of rock avalanche debris by mechanical fluidization, Rock Mech. 15, 9-24. Davies, T.R.H. and McSaveney, M.J. (1999) Runout of dry granular avalanches, Can. Geotech. J. 36, 313-320. Davies, T.R.H., McSaveney, M.J., and Hodgson, K.A. (1999) A fragmentation-spreading model for long-runout rock avalanches, Can. Geotech. J. 36, 1096-1110. Erlichson, H. (1991) A mass-change model for the estimation of debris-flow runout, a second discussion: conditions for the application of the rocket equation, J. Geol. 99, 633-634. Hampton, M.A, Lee, H.J., and Locat, J. (1996) Submarine landslides, Rev. Geophys. 34, 33-59. Hayashi, J.N. and Self, S. (1992) A comparison of pyroclastic flow and landslide mobility, J. Geophys. Res. 97, 9063-9071. Heim, A. (1932) Bergsturz und Menschenleben, Fretz und Wasmuth, Zürich. Howard, K.E. (1973) Avalanche mode of motion: implications from lunar examples, Science 180, 1052-1055. Hsü, K.J. (1975) Catastrophic debris streams (Sturzstroms) generated by rockfalls, Geol. Soc. Am. Bull. 86, 129-140. Hungr, O. and Morgenstern, N.R. (1984) Experiments on the flow behaviour of granular materials at high velocity in an open channel, Géotechnique 34, 405-413. Hungr, O. and Morgenstern, N.R. (1984) High-velocity ring shear test on sand, Géotechnique 34, 415421. Hungr, O. (1990) A mass-change model for the estimation of debris-flow runout: a discussion, J. Geol. 98, 791. Iverson, R.M. (1997) The physics of debris flows, Rev. Geophys. 35, 245-296. Iverson, R.M., Reid, M.E., and Lahusen, R.G. (1997) Debris-flow mobilization from landslides, Ann. Rev. Earth Planet. Sci. 25, 85-138. Iverson, R.M., Schilling, S.P., and Vallance, J.W. (1998) Objective delineation of lahar-inundation hazard zones, Geol. Soc. Am. Bull. 110, 972-984. Johnson, B. (1978) Blackhawk landslide, California, U.S.A, in B. Voight (ed), Rockslides and avalanches. 1. Natural phenomena, Elsevier, Amsterdam, pp. 481-504. Legros, F. (2002) Can dispersive pressure cause inverse grading in grain flows ?, J. Sedim. Res. 72, 171-175.

242 21. 22.

23. 24. 25. 26. 27. 28. 29. 30.

31. 32. 33.

34. 35. 36. 37. 38.

39. 40. 41. 42. 43. 44. 45. 46.

Legros, F. (2002) The mobility of long-runout landslides, Eng. Geol. 63, 301-331. Legros, F., Cantagrel, J.-M., and Devouard, B. (2000) Pseudotachylyte (frictionite) at the base of the Arequipa volcanic landslide deposit (Peru) and implications for emplacement mechanisms, J. Geol. 108, 601-611. Lipman, P.W., Normark, W.R., Moore, J.G., Wilson, J.B., and Gutmacher, C.E. (1988) The giant submarine Alika debris slide, Mauna Loa, Hawaii, J. Geophys. Res. 93, 4279-4299. Lucchitta, B.K. (1977) Crater clusters and light mantle at the Apollo 17 site; A result of secondary impact from Tycho, Icarus 30, 80-96. Major, J.J. (2000) Gravity-driven consolidation of granular slurries: implications for debris-flow deposition and deposit characteristics, J. Sedim. Res. 70, 64-83. Major, J.J. and Iverson, R.M. (1999) Debris-flow deposition: effects of pore-fluid pressure and friction concentrated at flow margins, Geol. Soc. Am. Bull. 111, 1424-1434. McEwen, A.S. (1989) Mobility of large rock avalanches: evidence from Valles Marineris, Mars, Geology 17, 1111-1114. McEwen, A.S. and Malin, M.C. (1989) Dynamics of Mount St. Helens 1980 pyroclastic flows, rockslide-avalanche, lahars and blast, J. Volcanol. Geotherm. Res. 37, 205-231. Mothes, P.A., Hall, M.L., and Janda, R.J. (1998) The enormous Chillos Valley Lahar: an ash-flowgenerated debris flow from Cotopaxi Volcano, Ecuador, Bull. Volcanol. 59, 233-244. Palmer, B.A., Alloway, B.V., and Neall, V.E. (1991) Volcanic-debris-avalanche deposits in NewZealand: lithofacies organisation in unconfined, wet-avalanche flows, Sedimentation in volcanic settings, SEPM Special Publication 45, 89-98. Plafker, G. and Ericksen, G.E. (1978) Nevados Huascarán avalanches, Peru, in B. Voight (ed), Rockslides and avalanches. 1. Natural Phenomena, Elsevier, Amsterdam, pp. 277-314. Savage, S.B.and Hutter, K (1989) The motion of a finite mass of granular material down a rough incline, J. Fluid Mech. 199, 177-215. Shaller, P.J. and Smith-Shaller, A (1996) Review of proposed mechanisms for Sturzstroms (longrunout landslides), in P.L. Abott and D.C Semour (eds), Sturzstroms and detachment faults, AnbzaBoreego Desert State Park, California. South Coast, Geological Society, Santa Ana, pp. 185-202. Shreve, R.L. (1968) The Blackhawk landslide, Geol. Soc. Am. Special Paper 108, 1-47. Shreve, R.L. (1968) Leakage and fluidisation in air-layer lubricated avalanches, Geol. Soc. Am. Bull. 79, 653-658. Siebe, C., Komorowski, J.-C., and Sheridan, M.F. (1992) Morphology and emplacement of an unusual debris-avalanche deposit at Jocotitlán volcano, Central Mexico, Bull. Volcanol. 54, 573-589. Siebert, L. (1984) Large volcanic debris avalanches: charcateristics of source areas, deposits, and assiociated eruptions, J. Volcanol. Geotherm. Res. 22, 163-197. Stoopes, G.R. and Sheridan, M.F. (1992) Giant debris avalanches from the Colima Volcanic Complex, Mexico: Implications for long-runout landslides (>100 km) and hazard assessment, Geology 20, 299302. Straub, S. (1996) Self-organisation in the rapid flow of granular material: evidence for a major flow mechanism, Geol. Rundsch. 85, 85-91. Straub, S. (1997) Predictability of long runout landslide motion: implications from granular flow mechanics, Geol. Rundsch. 86, 415-425. Takarada, S., Ui, T., and Yamamoto, Y. (1999) Depositional features and transportation mechanism of valley-filling Iwasegawa and Kaida debris avalanches, Japan, Bull. Volcanol. 60, 508-522. Vallance, J.W. and Scott, K.M. (1997) The Osceola Mudflow from Mount Rainier: Sedimentology and hazard implications of a huge clay-rich debris flow, Geol. Soc. Am. Bull. 109, 143-163. Van Gassen, W. and Cruden, D.M. (1989) Momentum transfer and the friction in the debris of rock landslides, Can. Geotech. J. 26, 623-628. Voight, B. and Sousa J. (1994) Lessons from Ontake-san: A comparative analysis of debris avalanche dynamics, Eng. Geol. 38, 261-297. Voight, B., Janda, R.J., Glicken, H., and Douglass P.M. (1983) Nature and mechanics of the Mount St. Helens rockslide-avalanche of May 1980, Géotechnique 33, 243-273. Wilson, C.J.N. (1984) The role of fluidisation in the emplacement of pyroclastic flows, 2: Experimental results and their interpretation, J. Volcanol. Geotherm. Res. 20, 55-84.

ROCK AVALANCHE OCCURRENCE, PROCESS AND MODELLING O. HUNGR1 Department of Earth and Ocean Sciences,University of British Columbia 6330 Stores Road, Vancouver, B.C., Canada V6T 1Z4

Abstract Rock avalanches are relatively infrequent, but highly destructive. From the point of view of societal costs, direct damage due to rock avalanches is of similar magnitude as loss due to defensive actions, esp. slope stabilization and alienation of land. Runout estimates are needed in order to minimize hazard areas that need to be set aside to protect population against perceived rock avalanche hazards. Empirical methods reviewed include the travel angle approach and area-volume correlation. Theories attempting to explain high mobility of rock avalanches are reviewed. In particular, the hypothesis of lubrication by liquefied saturated soil entrained from the landslide path is discussed. In a discussion of analytical modelling approaches, preference is given to an empirical approach, calibrating simple rheological models by means of back-analysis of existing cases.



Increasing population density and development of mountainous terrain bring human settlements within reach of landslide hazards. Perhaps the most serious threat arises from small, high frequency landslides such as debris flows and debris avalanches. On the other hand, large and relatively rare rock avalanches also constitute a significant hazard, due to their prodigious capacity for destruction. Such landslides involve the spontaneous failure of entire mountain slopes, with volumes measured in tens or hundreds million m3 and travel distances of several km. Individual rock avalanches have caused as many as 15,000 fatalities [38]. Stability analysis of entire mountain slopes is exceedingly difficult. Thus, concern about the possible occurrence of a rock avalanche usually arises only once certain precursory signs of impending failure appear. Once such signs are identified, monitoring of displacements, strains, piezometric pressures or rock noise can be used to gauge deterioration in stability and signal the onset of failure. Stabilization of potential rock avalanche sources is possible only in exceptional cases, where extreme consequences could result. As an example, certain mountain 1

E-mail: [email protected]

243 S.G. Evans et al. (eds.), Landslides from Massive Rock Slope Failure, 243–266. © 2006 Springer. Printed in the Netherlands.

244 slopes forming the shores to artificial lakes have been stabilized by drainage, at costs in the order of tens of millions of dollars per site. Under normal circumstances, however, such stabilization projects are not economically feasible. Hazard mitigation can then be achieved only by removing vulnerable developments from the path of the potential landslide. For this, a reliable runout analysis is required. The purpose of this article is to review the process of rock avalanche motion and to suggest a means of predicting runout distance and velocity. The term "rock avalanche" has developed naturally in the literature, as a simplification of the complex "rock slide-debris avalanche", proposed by Varnes [45]. Hungr et al. [24] suggested that the term "rock avalanche" be reserved for flow-like movements of fragmented rock resulting from major, extremely rapid rock slides. This contrasts with the term "debris avalanche", which should be reserved for landslides originating in unconsolidated material. It is true that many rock avalanches entrain unconsolidated debris along their long travel paths. Hungr et al. [24] recommended that the term "rock slide-debris avalanche" be used when a rock slide mobilizes a large quantity of debris by entrainment of liquefied substrate from the path.


Rock Avalanche Occurrence

Large landslides do not, in general, occur cyclically at any given location. But their occurrence can be considered cyclic in a large area of mountainous terrain. Unpublished data collected by the author indicates that major rock avalanches occur with an annual frequency of 1/500 to 1/5,000 per 10,000 km2 of mountainous terrain, being somewhat more frequent in sedimentary and metamorphic, than intrusive rocks. Somewhat higher average rock avalanche frequencies are observed on volcanoes. Given these relatively low frequencies, potential damage is perhaps more significant in case of rock avalanches than reflected in the historical record. A rock avalanche in Canada could easily cause a national disaster. A certain extreme rock avalanche scenario, such as the destruction of a major dam, propagated down a cascade of other installations, could possibly reach the level of a catastrophe of global significance. Thus, rock avalanche risk management presents a difficult dilemma. On one hand, rock avalanches are potentially very destructive and we must be careful not to underestimate this. On the other, major rock avalanches are quite rare and overestimating their risks could easily lead to excessive and unwise expenditures. The ability to provide realistic predictions of rock avalanche failure behaviour, particularly their runout, is therefore important.


Empirical Methods of Runout Prediction

The sole established means to predict rock avalanche runout relies on simple empirical correlations [33]. For example, a correlation between rock avalanche volume and the area of deposits, derived from data by Abele [1] and supplemented by additional points from the author's files appears in Figure 1. The logarithmic correlation plot shows a linear relationship with an approximate slope of 2/3. indicating geometrical similarity

245 between deposits [18]. As pointed out by Davies [9] and implemented by Iverson et al. [29], it is possible to use such a correlation to predict the length of the deposit and thus the runout distance from the toe of a steep valley slope. Unfortunately, the correlation contains large scatter and uncertain decisions must be made regarding the position of the slope toe and the plan shape of the deposit. Another correlation, first noted by Heim [16] and developed by Abele [1] and Scheidegger [43], exists between the volume and the tangent of the travel angle (fahrböschung), shown in Figure 2. As defined by Heim [16], the travel angle is the slope angle of a line drawn between the crown of the source area and the toe of the deposit, following the centerline of the path. It is often equated with the slope of the energy line of the landslide which, assuming frictional motion, equals the average apparent friction angle prevalent during motion. Evidently, many rock avalanches move with friction angles far lower than the dry friction angle of soil or crushed rock. This is known as the problem of excessive mobility and it has been the subject of considerable research for at least 30 years. The fahrböschung plot again contains strong scatter.



AREA (m2)





1x104 1x105







VOLUME (m3) Figure 1. A correlation between rock avalanche magnitude and the plan area of deposits.

246 1

0.9 0.8 0.7 0.6 0.5 0.4


tan fahr.

0.3 0.2




















Figure 2. A correlation between rock avalanche volume and the tangent of the fahrböschung angle (vertical angle between the crown of the landslide source and the toe of the deposit). From [43].

One could argue that the assumption behind equating fahrböschung and the slope of the energy line is inaccurate, as the later would be better represented by the slope of a line connecting the centres of gravity of the slide mass, before and after movement [7]. Hungr [17] reconstructed the slope between centres of gravity (') approximately for several well-known rock avalanches (Figure 3). Both the trend and scatter of this diagram are similar to those of the fahrböschung plot.


0.9 0.8 0.7 0.6 0.5 0.4

tan α′




















Figure 3. A correlation between rock avalanche volume and the tangent of the vertical angle ' between the centres of gravity before and after the slide (After [17]).

Several authors attempted to reduce the scatter of the fahrböschung plot by accounting for the shape and confinement of the runout path and other morphological factors [1], [37], [6]. However, the improved correlations still contain very considerable error. The error may partly be due to inaccuracies in volume estimates, especially where the rock avalanches entrained large quantities of material from their

247 path. Some authors justifiably argue that the fahrböschung plot is meaningless, if the motion of the rock avalanche is not, in fact, frictional [26], [11], [32].


Reasons for Excessive Mobility of Rock Avalanches

A number of hypotheses have been advanced to explain the excessive mobility of rock avalanches: a) Shreve [44] proposed that the sole of the rock avalanche is lubricated by a cushion of trapped air. If this were the case and air pressure was sufficient to provide significant uplift of the average column of debris, then parts of the debris sheet which are thinner than average would be completely fluidized. This would result in normal grading with the coarsest particles falling to the base, elutriation of fines, massive air jetting and formation of craters and fallback cones or fans on and surrounding the debris sheet. No such features have been observed [8]. In fact, reverse grading, with the largest particles resting on the top surface of the debris is usual (Figures 4 and 5). The same argument can be advanced against hypotheses postulating fluidization by escaping steam, formed by vaporization of ground water [15].

Figure 4. Reverse grading and normal grading.

b) The mechanical fluidization theory anticipates that the friction angle of a rapidly-sheared dry granular material will decrease with increasing velocity, as dynamic interactions (bouncing) force particles away from each other. A similar effect was shown in a computer model based on particle dynamics [4]. However, no one has so far been successful in demonstrating this effect in the laboratory using rapid ring shear tests, flume tests or similar. For example, flume tests reported by Hungr and Morgenstern [20] reached velocities of up to 6 m/s and displayed measurable decrease in the density of the flowing mass. Yet, the measured dynamic friction angle increased with speed, contrary to the mechanical fluidization hypothesis. The same effect was shown for materials of varying shape, grading and particle properties (including varying restitution coefficients for inter-particle contacts), with materials ranging from graded angular sand to uniform, spherical polystyrene beads. c) The acoustic fluidization of Melosh [35] relies on energy input from external vibrations, set up by the boundary of the moving mass. This effect is difficult to

248 examine, as it will occur only in a full-scale event. We do not know whether such vibrations are sufficiently regular to cause a loosening effect. It is also not clear why should such an effect lead to a scale dependency and why it should occur at all when the rock avalanche masses travel over soft subgrades, liable to dampen out any vibration. d) Davies and McSavenney [10] proposed that mobility is the result of gradual fragmentation of the moving mass. However, fragmentation is an energy-consuming process. Thus, while it may promote lateral and longitudinal spreading, it cannot increase the mobility of the centre of the mass (cf. Figure 3). Also, some extremely mobile granular flow-slides initiating in broken rock, such as those occurring on coal mine waste dumps, do not involve fragmentation (Figure 6).

Figure 5. Reverse grading observed in the Frank Slide debris, Alberta, Canada.


Figure 6. A flow slide from a coal waste dump in south-western British Columbia, Canada. A part of the source scar is shown in the background. The flow slide traveled over 2 km at velocities of up to 20 m/s (see also [25]).

e) Lubrication by liquefied saturated soil entrained from the path of the rock avalanche was first proposed by Buss and Heim [3], following their examination of the Elm Slide in Switzerland. The hypothesis was further advanced by Abele [1], Sassa [41], Legros [32] and others. Evidence of displacement and liquefaction of saturated substrate can be found in and around most rock avalanche deposits. For example, the debris of the 1903 Frank Slide displaced so much of the alluvial deposits in the channel of the Old Man River, that the present debris surface in the proximal part of the deposit lies below the original stream bed (elongated lake in Figure 7). The margins of the deposit were found surrounded by a "splash" of liquefied mud [8]. Lateral displacement of mud expelled from beneath the debris sheet was responsible for much of the damage in the town of Frank [34]. A schematic representation of the entrainment / liquefaction process is shown in Figure 8. An initially dry mass of fragmented rock (a) partly displaces (b) and partly over-rides (c) a layer of saturated soil near the toe of the valley slope. The movement of both the rock fragments and the mud resulting from impact liquefaction is enhanced by the low strength of the liquid soil (d).


Figure 7. A part of the deposit of Frank Slide, viewed from the crown scarp facing in the direction of movement.

The process illustrated in Figure 8 can be clearly observed in a landslide which occurred in May, 1999 at Nomash River, Vancouver Island, Canada (Figure 9). A block of marble and andesite, over 300 000 m3 in volume, slid from the side of a Ushaped glacial valley. The rock mass deeply scoured a till-derived colluvial apron at the toe of the source slope and entrained another 300 000 of liquefied soil, before turning a


Figure 8. A schematic representation of the simultaneous overriding, liquefaction and displacement of a layer of saturated substrate at the front of a rock avalanche mass.

90º bend and continuing to flow for about 1 km along the thalweg of the valley. The fahrböschung of the movement was 13º, much less than could be derived from the Scheidegger plot (Figure 2). Such a mass movement, originating in rock but involving a substantial volume of liquefied soil, may be called a rock slide-debris avalanche, in contrast to a rock avalanche, dominated by fragmented rock, and debris avalanche, dominated by soil. The role of saturated soil is unambiguous in such cases. There is a gradational boundary between rock slide-debris avalanches and "pure" rock avalanches, in which the lubricating mud layer is largely covered by rock fragments and therefore invisible. One of the objections to the mud lubrication hypothesis was that highly mobile rock avalanches have been observed on the Moon and on Mars. However, while the lunar examples may well be related to cratering processes rather than simple gravitational sliding, the Martian example may represent fossil phenomena which were once facilitated by liquid water [32]. Thus, mud lubrication is the most likely explanation for the great mobility of many, if not all, rock avalanches. Some of the other mechanisms, such as fragmentation spreading and acoustic fluidization may also simultaneously play a role.


Lumped Mass Models

These models, also referred to as "block" or "sled" models, represent the rock avalanche as a dimensionless point of mass sliding along a prescribed path. The technique was developed and practically used in the field of snow avalanche science, where it was


Figure 9. The Nomash River rock slide - debris avalanche of May, 1999. A rock slide originating from the left side of the valley entrained large quantities of saturated colluvium and alluvial soil and travelled for about 1 km along the thalweg. The light-coloured part of the debris consists of fragmented rock, the darker areas represent mud. Note that a small independent tongue of rock debris is situated in the forefront; the main slide is behind it. There is a light snow cover on the ground. (Photo courtesy of R. Guthrie, B.C. Ministry of Environment, Nanaimo, Canada).

meant to simulate the motion of the avalanche front. Of course, such a representation is oversimplified, as it neglects the important effects of internal deformation of the flowing mass. Since more complete models are now readily available, lumped mass modeling is no longer useful. However, it is instructive to use this model to introduce the concept of energy line, as proposed by Körner [31]. The energy line is a line raised above the movement path by an elevation equal to the kinetic energy head, He=v2/2g, where v is the velocity. For a block sliding on a dry frictional surface, where the resisting force is proportional to the normal force, the energy line is a straight line, inclined at the friction angle, ij (Figure 10). Should pore pressure be present, the energy line will remain straight, provided that we can assume the pore pressure, u, to be proportional to the normal stress, ın:

u ruV n (1) where ru is a proportionality constant, similar to the "pore pressure ratio" used in Soil Mechanics. The magnitude of ru is about 0.5 for full saturation and more if excess pore pressure exists. The slope of the energy line will equal to the "bulk friction angle":



atn[(1  ru ) tan I )


Energy Lines

He v Figure 10. Energy lines corresponding to various types of rheology (see text). He is th enenrgy head, v velocity.

In case of non-frictional rheology, where either the friction angle or the pore pressure ratio is not constant, the energy line will be curved. Figure 10 illustrates two types of behaviour. If ij increases or ru decreases as a function of velocity, a concave line will result. Thus, lower velocities will correspond to a given overall displacement of the block. Körner [31] showed that the frictional model over-estimates the velocity in most cases of rock avalanches and snow avalanches. He adopted Voellmy's [46] frictional/turbulent rheology, where the resisting forces are assumed to depend on the effective normal stress and the square of velocity. Presented in the same form as Equation [2], the Voellmy relationship can be written as:


atn[(1  ru ) tan I 

v2 ] H[ cos E


where H is the average flow depth,  the local slope angle and ȟ a coefficient with dimensions of LT-2, similar (although not exactly equal) to the Manning's coefficient of turbulent flow. Körner [31], McLellan and Kaiser [36] and others showed that the Voellmy lumped mass model gives good representation of the velocity profiles of rock avalanches. Hungr [19] applied the same equation in the framework of an unsteady flow model, as described below. A concave shape of the energy line (Figure 10) can also be obtained by allowing ijb to decrease with displacement. For example, Sassa [41] introduced a model in which the ru coefficient increases in the downslope direction, as the landslide over-rides materials of increasing water content and decreasing mean grain size. Sassa [41] advocated the use of undrained ring shear tests to derive the appropriate ru for each of a variety of geological materials found along the landslide path. He also proposed that such tests would account for possible changes in ij and pore pressure resulting from

254 grain crushing in a thin shear zone. Further discussion of this approach is given in Section 7. The long-dashed line in Figure 10 shows the opposite effect where the energy line is convex, predicting higher velocities for a given overall displacement. Such an effect could result from decrease in ijb with increasing velocity, as would be predicted by the various "fluidization" theories, discussed earlier. For example, production of steam by means of frictional heating, or the controversial "mechanical fluidization" phenomenon would both generate a convex energy line. Alternatively, a convex shape could result from resisting forces increasing with displacement (or time), as would be the case if gradual consolidation and dissipation of pore pressure were coupled with movement [28], [11], [30] and [12]. All such approaches are likely to lead to an overestimation of rock avalanche velocities. The main types of rheological relationship and their influence on model velocities are summarized in Table 4. Table 4. Rheological relationships and their influence on model velocities. Energy Line straight concave

Velocities high moderate


very high


Models and Reference frictional [16] turbulent, viscous rate effects [31] resistance decreases with displacement [41] fluidization, pore pressure increase [15] sliding consolidation [28], [11], [30],[12]

Dynamic Runout Modelling

The simplest forms of dynamic analysis of rock avalanches are based on the St. Venant Equation of unsteady flow. With reference to a fixed coordinate system shown in Figure 11 and a depiction of a typical normal column in the flow (Figure 12), a onedimensional "Eulerian" form of the equation can be written as:


dv dv  dx dt

g sin D 

T dH  gk cos D UH dx


Here,  is the slope angle, v is velocity, H is flow thickness, T is the total resisting stress on the column base (normally a function of velocity and depth),  a velocity correction coefficient and  the fluid density. k is a lateral pressure coefficient, discussed below. The first term in Equation [4] represents acceleration, consisting of a local and convective part. The second term is the downslope component of gravity, the third the resisting force on the column base and the forth a pressure thrust resulting from the longitudinal gradient of the upper surface. When the problem is re-stated so that the longitudinal coordinate x is a moving coordinate attached to each column, a simpler "Lagrangian" form of the equation results:


dv dt

g sin D 

T dH  gk cos D UH dx


Here, the convective acceleration term, with its correction coefficient disappears, allowing direct explicit numerical integration of the equation. A continuity condition must also be satisfied. As shown by Hungr [19], it is possible to account for gradual widening and narrowing of the path, although the momentum changes in directions perpendicular to the direction of movement are neglected.

 Figure 11. A coordinate framework for the derivation of a one-dimensional equation of unsteady flow (See text).

Flow surface

Figure 12. The dynamic equilibrium of a column (after [19]).

The pressure term requires consideration. The formulation of the last term in Equation [5] is based on Figure 12 and contains an assumption that the flow lines are parallel with the bed of the flow. In a fluid, under hydrostatic conditions, the coefficient k equals 1.0. If the landslide mass is composed of a granular material, however, k could range between an "active" value (less then 1) and "passive" value in excess of 1,

256 depending on whether the material is expanding or contracting [39]. Hungr [19] used a numerical procedure to estimate the development of longitudinal strain in the sliding body. The lateral pressure coefficient, k, varies between the active and passive values depending on strain. The importance of considering an anisotropic stress distribution in the flowing mass is illustrated by the analysis of a rock avalanche that produced the largest run-up elevation ever recorded [14]. As can be seen on Figure 13, the Avalanche Lake rock avalanche began as a slab slide from a 1200 m high isoclinal slope in dolomite. After disintegrating and scouring the river valley of alluvial gravels, the leading edge of the rock avalanche mounted the opposite slope to a height of 600 m. Most of the slide mass then returned back to the valley and even ran-up again against the original slope. However, the leading edge, containing entrained alluvial gravel, continued flowing in a curved path over the surface of the 600 m high bench and eventually returned down the run-up slope back to the valley (Figure 14). An attempt to analyse this example using Hungr's [19] dynamic model "DAN", using the properties of a fluid (i.e. k=1.0) is shown in Figure 15a. The leading edge begins to mount the run-up slope, but soon builds into a gigantic standing wave and collapses backwards, failing to reach more than about 2/3 of the height. The model with anisotropic stress state correctly simulates the projection of the leading edge to the shelf (Figure 15b). True three-dimensional models can be derived by the generalization of Equations [4] and [5] ([5], [30]). In three-dimensional models, the earth pressure coefficient should have different values in directions the longitudinal and transverse to the direction of motion, according to the orientation of the strain tensor.


Rheology and Calibration

Various rheological relationships can be introduced into the model, by formulating a suitable expression for the resisting stress, T, as a function of velocity, depth and other variables and constants. Hungr [19] proposed seven different rheologies, including frictional, perfectly plastic, viscous, turbulent, Bingham [40], power law, Coulombviscous and Voellmy. In the frictional case, the resisting stress can be expressed as:


JH tan Ib cos E


where Ȗ is the unit weight of the material. Pore water pressure can be accounted for by using ijb from Equation [2] and the Voellmy rheology is obtained by substituting for ijb from Equation [3]. A key question with regard to the frictional and Voellmy equations is whether it is reasonable to assume that the pore pressure coefficient, ru can remain constant during motion. As the mass of broken rock flows down, overriding and incorporating substrate material and shearing primarily in a thin basal layer of liquefied soil, the pore pressure within the shear zone may be influenced by the following phenomena (Figure 16):

257 1) Stress level changes: as the flowing sheet becomes thinner, the pore pressure should decrease (ru will remain constant, as long as the pore pressure change is proportional). 2) Velocity changes: as velocity increases, the soil in the shear zone will tend to expand through dispersion. If there is insufficient time for water to flow into the zone, the pore pressure will decrease and ru will be reduced. Such an effect could be captured by the "turbulent" term in the Voellmy equation. 3) Consolidation drainage: any excess pore pressure may be reduced with time, if water has time to escape from the shearing zone.

Figure 13. The Avalanche Lake rock avalanche [14]. A planar slide from a 1200 m high slope resulted in a run-up of 600 m.


Figure 14. Avalanche Lake, flow of debris on the bench, 600 m above the original valley floor. The avalanche arrived from the left margin of the photo.

Figure 15. Two attempts to analyse the run-up of the Avalanche Lake rock avalanche, using the model DAN [19].

259 4) Mixing: in contrast to the previous point, vigorous mixing of the shear zone may prevent drainage and may give rise to fluid drag forces affecting the soil skeleton. Bagnold [2] referred to this process as "autosuspension". 5) Comminution: Sassa [42] showed that grain crushing within the shear zone may lead to change in gradation of the material and consequent tendency for densification, hence an undrained increase in ru.

Figure 16. Schematic representation of a flowing rock avalanche. A liquefied basal layer supports coarse, granular body of the flow. Velocity gradient is concentrated in the basal layer. Arrows indicate possible drainage paths.

6) Material entrainment: as substrate material of varying gradation and water content is liquefied and incorporated into the shear zone, indeterminate changes in pore pressure will result. 7) Surface water entrainment: water from streams or lakes found in the landslide path may be incorporated into the shear zone, changing its water content. It is this writer's opinion that all these processes cannot possibly be simulated by a detailed mechanistic model. The processes will be operative in varying degrees at various points of the landslide footprint. The character and thickness of the shear zone will vary in time and space in a complex manner, determined mainly by geological factors. The alternative approach is to largely ignore the detailed mechanics of the shear zone and to model the entire body of the landslide as a homogeneous domain, composed of an "equivalent fluid". As defined by Hungr [19], the equivalent fluid has properties such, that it will produce approximately the correct external dimensions (velocity, depth distribution) when flowing along the landslide path. Of course, the properties, or even the type of constitutive relationship for the fluid cannot be determined by laboratory tests, but must be derived by backanalysis of full scale observed events.

260 Using the shareware program "DAN" [19], a calibration procedure has been established for several types of mobile landslides: rock avalanches [21], flow slides in coal mine waste [25] and debris flows and avalanches in pyroclastics (Revellino et al., in prep.). The procedure is organized as follows: 1) Case histories for calibration are selected so that they all belong into a fairly uniform typological class. 2) Using several alternative rheologies, analyses of the case histories are completed, choosing input parameters such that the requisite runout occurs. 3) All of the rheologies used in DAN have only one or two adjustable parameters. Where there are two, it is necessary to impose additional conditions. This is done by attempting (by trial and error) to simulate flow velocity and depth data at any locations where observations exist. 4) In some cases, it is possible to generalize and to select a best-fit set of parameters for an entire group of case histories. An example of the procedure is shown in Figure 17 [13]. A small rock avalanche from a volcanic source (Figure 18) has been analysed using four alternative rheologies. The resulting velocity profiles are compared with a group of field velocity estimates, derived from superelevation of the flow in bends of the path. The best fit is obtained using the Voellmy model, with a friction coefficient of 0.1 and a turbulence coefficient of 500 m/sec2. A collection of similar "opportunistic" observations from a number of rock avalanches is shown in Figure 19. Again, the Voellmy model provides the best estimates of velocities, consistent with the observations by Körner [31]. The influence of rheology on the longitudinal distribution of depth of the deposits is shown in Figure 20. The frictional rheology tends to produce relatively short deposits, thin in the proximal part and tapering forward. The Voellmy result produces longer deposits and the opposite efect. The Bingham model produces relatively uniform ("critical") depth, leaving a substantial amount of deposits on the steep proximal part of the flow path. In the case illustrated by Figure 20 (the Frank Slide), the Voellmy result is judged the most realistic. Hungr and Evans [21] also showed that in about 70% of rock avalanche cases, good first order predictions of the total runout distance can be obtained using the Voellmy model, with a fixed pair of resistance parameters (friction coefficient of 0.1 and a turbulence coefficient of 500 m/sec2). The comparison is shown in Figure 21. The model underpredicted total runout in four cases. One of these (Sherman) was a rock avalanche travelling over a glacier. The remaining three were cases which incorporated large amounts of substrate material. Runout was underpredicted in four cases of rock avalanches depositing on relatively dry slopes of upper mountain sides, lacking substrate prone to liquefaction. This shows that future calibration efforts must concentrate on classifying the substrate material found in the path of the potential rock avalanche. Ideally, sets of rheological parameters will be found corresponding to different type of substrate.

261 8.

Material Entrainment

Since it is based on a Lagrangian formulation of the movement equations, the DAN model also permits the simulation of large scale material entrainment, such as occurs in rock slide-debris avalanches [22]. An example is shown in Figure 22, an analysis of the Nomash River case described briefly in Section 4. The initial rock slide phase Voellmy, 0.1,500 Voellmy, 0.2, 1500 Frictional Bingham

100 90


80 70 60 50 40 30 20 10 0








Frictional: fi=30, ru=0.45 Bingham: tau=18kPa, viscosity=1 kPa.s

Figure 17. Calibration procedure: selection of an optimal rheological relationship by matching the velocity profile [13]. Frictional: ij=30º, ru=0.45; Bingham, IJ=18kPa, µ=1 kPa..s.

Figure 18. The 1984 Mt, Cayley rock avalanche, analysed in Figure 17. View downslope from the source (note superelevation in first bend [13].




80 Voellmy 60

Frictional Bingham


20 Figure 19. Comparison of calculated and actual rock avalanche velocities [21].

0 0

20 40 60 80 FIELD VELOCITY (m/s)





1800 1600 1400 1200 1000 1000








1800 1600 1400 1200 1000 1000






2000 1800


1600 1400 1200 1000 1000





Figure 20. Frank Slide: predicted deposit depth profiles resulting from various rheologies.




Figure 21. A comparison of predicted and actual rock avalanche runout distances, obtained using the Voellmy model, with a friction coefficient of 0.1 and a turbulence coefficient of 500 m/sec2 [21].







900 800 700 600 500 400 300

5000 10000 ACTUAL RUNOUT (m)


300 200 WIDTH




0 1000











20 10 0 0







Figure 22. An analysis of the Nomash River rock slide-debris avalanche, depicted in Figure 9. a) slope profiles plotted at 10 sec. intervals (all depths of the landslide mass and erosion depths are exaggerated 10X). Entrainment ("erosion") begins at a distance of 350 m. The Voellmy reheology replaces the frictional rheology at the same point. b) velocity profiles for the front and tail of the flow, compared with field observations (crosses).

was modelled using frictional rheology, with no pore pressure and a friction angle of 30º. Entrainment was specified according to yield rates (m3/m) estimated in the field. The Voellmy rheology (friction coefficient of 0.05 and a turbulence coefficient of 400 m/sec2)was introduced at the same point where entrainment of the first batch of colluvial soil took place. The model produced a very satisfactory representation of both

264 velocity and depth distribution for this case. However, the magnitude of the resistance coefficients cannot yet be generalized, until additional similar cases can be backanalysed.



Rock avalanches are complex, but tools now exist that make it possible to simulate many of their motion characteristics. This author favours an approach based on observation of external characteristics of flow behaviour, without attempting to quantify every detail of the actual movement mechanism. Such an approach is strongly dependent on calibration, using real events. The adopted rheological models should have only a few adjustable parameters, so that they can be full constrained. Ideally, each calibration should involve several events and one should attempt to generalize sets of rheological parameters to groups of similar events. Future calibration work should also discern and classify types of substrate over which the rock avalanches travel. Material entrainment should be included in the analysis, where it is significant. We should also look for means of predicting entrainment yield rates.

Acknowledgements The Celano workshop organizers have prepared a highly stimulating and interesting meeting.

References 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12.

Abele, G., 1974, Bergstürze in den Alpen. Wissenschaftliche Alpenvereinshefte, Number 25, Munich, (in German), 230 p. Bagnold, R.A., 1961. Auto-suspension of transported sediment; turbidity currents. Procs., Royal Society of London, Ser. A, 265: 315-319. Buss, E. and Heim, A., 1881, Der Bergsturz von Elm. Worster, Zurich, 133 p. Campbell, C.S., 1989, Self-lubrication for long runout landslides. Journal of Geology, 97: 653-665. Chen, H. and Lee, C.F., 2000. Numerical simulation of debris flows. Canadian Geotechnical Journal, 37:146-160. Corominas, J., 1996. The angle of reach as a mobility index for small and large landslides. Canadian Geotechnical Journal, 33:260-271. Cruden, D.M., 1980. A large landslide on Mars: Discussion. Bulletin, Geol. Society of America, 1,91, p.64. Cruden, D.M. and Hungr, O., 1986, The debris of the Frank Slide and theories of rockslide-avalanche mobility. Canadian Journal of Earth Sciences, 23: 425-432. Davies, T.R.H., 1982. Spreading of rock avalanches by mechanical fluidization. Rock Mechanics, 15: 9-24. Davies, T.R.H., and McSavenney, M.J., 1999, Runout of dry granular avalanches. Canadian Geotechnical Journal, 36: 313-320. DeMatos, M., 1987. A sliding consolidation model for rock avalanches. Ph.D. Thesis, Univ. of Alberta. Denlinger, R.P. and Iverson, R.M., 2001. Flow of variably fluidized granular masses across threedimensional terrain. 1. Numerical predictions and experimental tests. Journal of Geophysical Research, 106:553-566.

265 13.

14. 15.

16. 17. 18. 19. 20. 21. 22.


24. 25.

26. 27. 28. 29. 30. 31. 32. 33. 34. 35. 36. 37. 38. 39.

Evans, S.G., Hungr, O., and Clague, 2001. The 1984 rock avalanche from the western flank of Mt. Cayley, Garibaldi Volcanic Belt, British Columbia: description and dynamic analysis. Engineering Geology, 61:29-51. Evans, S.G., Hungr, O. and Enegren, E.G., 1994, The Avalanche Lake rock avalanche, Northwest Territories, Canada: description, dating and dynamics. Canadian Geotechnical Journal, 31: 749-768. Goguel, J., Pachoud, A., 1972, Geology and dynamics of the rockfall of the Granier Range which occurred in November 1248. Bulletin, Bureau de Récherches Geologiques et Miniéres, Hydrogeologie, Lyon, 1: 29-38. Heim, A., 1932, Landslides and human lives (Bergsturz und Menschenleben), N. Skermer Ed., Bi-Tech Publishers, Vancouver, B.C., 196 p. Hungr, O., 1981, Dynamics of rock avalanches and other types of mass movements, Ph.D. Thesis, University of Alberta, Edmonton, 500 p. Hungr, O., 1990, Mobility of rock avalanches. Reports of the National Research Institute for Earth Science and Disaster Prevention, Tsukuba, Japan, 46:11-20. Hungr, O., 1995, A model for the runout analysis of rapid flow slides, debris flows and avalanches. Canadian Geotechnical Journal: 32: 610-623. Hungr, O. and Morgenstern, N.R., 1984, Experiments in high velocity open channel flow of granular materials. Géotechnique, 34: 405-413. Discussion and Reply, 35: 383-385. Hungr, O., and Evans, S.G., 1996, "Rock avalanche runout prediction using a dynamic model." Proceedings, 7th International Symposium on Landslides, Trondheim, Norway, 1: 233-238. Hungr, O., and Evans, S.G., 1997, A dynamic model for landslides with changing mass. In Marinos, P.G., Koukis, G.C., Tsiambaos, G.C. and Stournaras, G.C., Eds., Procs., IAEG International symposium on engineering geology and the environment, Athens, June, 1997, 1: 719-724. Hungr, O., Evans, S.G., and Hazzard, J. (1998): Magnitude and frequency of rock falls and rock slides along the main transportation corridors of south-western British Columbia. Canadian Geotechnical Journal, 36: 224-268. Hungr, O., Evans, S.G., Bovis, M., and Hutchinson, J.N., 2001, Review of the classification of landslides of the flow type. Environmental and Engineering Geoscience, VII: 221-238. Hungr, O., Dawson, R., Kent, A., Campbell, D. and Morgenstern, N.R., 2002, Rapid flow slides of coal mine waste in British Columbia, Canada. In "Catastrophic Landslides" Geological Society of America Reviews in Engineering Geology No. 14 (in press). Hsu, K.J., 1975. Catastrophic debris streams (sturzstroms) generated by rock falls. Bull., Geological Society of America, 86:129-140. Hutchinson, J.N. and Bhandari, R.K., 1971. Undrained loading, a fundamental mechanism of mudflow and other mass movements. Géotechnique, 21:353-358. Hutchinson, J.N., 1986. A sliding-consolidation model for flow slides. Canadian Geotechnical Journal, 23: 115-126. Iverson, R.M., Schilling, S.P. and Vallance, J.W., 1998. Objective delineation of lahar-inundation hazard zones. Geological Society of America Bulletin, 110:972-984. Iverson, R.M. and Denlinger, R.P., 2001. Flow of variably fluidized granular masses across threedimensional terrain. 1. Coulomb mixture theory. Journal of Geophysical Research, 106:537-552. Körner, H.J., 1976, Reichweite und Geschwindigkeit von Bergsturzen und fleisschnee-lawinen. Rock Mechanics, 8: 225-256. Legros, F., 2002, The mobility of long runout landslides. Engineering Geology, 62: 100-130. Li-Tianchi, 1983. A mathematical model for predicting the extent of a major rockfall. Zeitschrift fur Geomorphologie, 27: 473-482. McConnell, R.G. and Brock, R.W., 1904, The great landslide at Frank, Alberta. Canadian Parliament Sessional Paper, No.25, Department of the Interior, Government of Canada. Melosh, H.J., 1979, Acoustic fluidization: a new geologic process? Journal of Geophysical Research, 84:7513-7520. McLellan, P.J. and Kaiser, P.K., 1984. Application of a two- parameter model to rock avalanches in the Mackenzie Mountains. Procs., 4th. Internat. Symposium on Landslides, Toronto, 1: 135-140. Nicoletti, G. and Sorriso-Valvo, M., 1991. Geomorphic controls of the shape and mobility of rock avalanches. Geological Society of America Bulletin, 103:1365-1373. Plafker, G., Ericksen, G.E., 1978, Nevados Huascarán avalanches, Peru. Rockslides and avalanches, Voight, B., Ed., Elsevier, Amsterdam, 1: 277-314. Savage, S.B. and Hutter, K., 1989. The motion of a finite mass of granular material down a rough incline. Journ. Fluid Mechanics, 199: 177-215.

266 40. 41. 42.

43. 44. 45.


Sousa, J. and Voight, B., 1991. Continuum simulation of flow failures. Geotechnique, 41: 515-538. Sassa, K., 1988. Geotechnical model for the motion of landslides (Special lecture). Landslides, C. Bonnard, Editor, Proceedings, 5th International Symposium on Landslides, 1: 37-56. Sassa, K., 2002. Mechanism of rapid and long traveling flow phenomena in granular soils. Proceedings, International Symposium on Landslide Mitigation and Protection of Cultural and Natural Heritage. Kyoto University, Kyoto, Japan, 11-30. Scheidegger, A.E., 1973, On the prediction of the reach and velocity of catastrophic landslides. Rock Mechanics, 5:231-236. Shreve, R.L., 1968, The Blackhawk Landslide. Geological Society of America, Special Paper No. 108. Varnes, D.J., 1978, Slope movement types and processes. In R.J. Schuster and R.J. Krizek (editors), Landslides, Analysis and Control: Transportation Research Board, National Academy of Sciences, Washington, DC., Special Report 176: 11-33. Voellmy, A., 1955, Über die Zerstorungskraft von Lawinen. Schweizerische Bauzeitung, 73: 212-285.

MOBILITY OF ROCK AVALANCHES TRIGGERED BY UNDERGROUND NUCLEAR EXPLOSIONS V.V. ADUSHKIN1 Institute of Geospheres Dynamic, Russian Academy of Sciences Moscow, Russia Leninskiy Avenue 38/6, Moscow 117334, Russia Abstract Several large rockslides and rock avalanches ranging in volume from 105 m3 up to 108 m3 were triggered by underground nuclear explosions at the Novaya Zemlia test site. Rapid filming of rock avalanche formation allowed direct measuring of the velocities of debris spreading. Dynamics of two case studies derived from the real time observations and from the analysis of debris morphology and grain size composition is discussed in details. Factors determining runout of artificial rock avalanches such as variability of debris grain size composition and topography of the transition and deposition zones are examined. Relationships of rock avalanche runout and their volume are determined and compared with those of the natural events of different origin. Critical conditions of slope failure occurrence depending on intensity of seismic effects of the explosions and slope angles are examined as well. 1. Introduction Major rockslides, the volumes of which exceed few millions m3 are characterised by flow-like debris motion when dry material moves as a liquid. Their runout can be much larger than fall height, sometimes exceeding ten kilometres, and deposits can cover tens and even hundreds square kilometres, and cause severe disasters in populated regions. The phenomenon is known as rock avalanche. Such catastrophic events occurred in the Alps [1, 12], Mackenzie Mountains [8], Tien Shan [11] and other mountain systems [7, 9, 11, 13, 23]. That is why study of rock avalanche phenomenon should be considered as an important and actual task of geomechanics and engineering geology. Numerous models have been proposed to explain mechanism of their extra-mobility. Some of these models explain reduction of apparent friction by influence of air [17], water [19], dust [14] and saturated soil [22]. Reliability of these models was examined, in particular, by Erismann [10] and Hungr [15]. Other models explain long runout of large dry rock avalanches, without any lubricants. Campbell [5] and Campbell et al. [6] proposed that low friction may be explained by granular mechanics. Melosh [20] developed theory of acoustic fluidisation that explain this phenomenon as the reduction of friction coefficient due to elastic acoustic vibrations in the rock avalanche body during its high-speed motion. In this model debris has a power-law shear-stress/strain-rate dependence similar to that of 1

E-mail: [email protected]

267 S.G. Evans et al. (eds.), Landslides from Massive Rock Slope Failure, 267–284. © 2006 Springer. Printed in the Netherlands.

268 a vibrating sand. A systematic, although limited, attempt to derive equivalent fluid parameters was made by Hungr and Evans [16]. Using a dynamic model, configured with Bingham friction and Voellmy rheologies, they obtain the best possible simulation of the deposits length as well as of their velocities and thickness. Elaboration of the reliable mechanical models of rock avalanche formation and motion and their numerical simulation require objective input data. Such data can be obtained in the course of the detail field measurement of the deposits’ geometrical parameters, study of the geological structure of the rock massifs and of the mineralogical and grain size composition of the debris. Very informative data can be obtained by the real time observations of the process, especially by filming of a rock avalanche during its motion. Systematic observations of rock slope failures and of the evolution of ‘secondary’ effects of underground nuclear explosions in rock massifs have been carried out at the Novaya Zemlya nuclear test site (Figure 1). As far as both time and place of rock slope failure were known in advance, it gave a chance to record whole process and to determine geometric parameters of the source zones and resultant deposits with high accuracy, rarely attainable in the study of the similar natural phenomena.

Figure 1. Schematic map of the Novaya Zemlia test site and location of rock avalanches triggered by underground nuclear explosions. Contour lines interval is 100 m.

2. Rock Avalanches Triggered by Underground Nuclear Explosions In a number of cases underground nuclear explosions were accompanied by large-scale rockslides, some of which transformed into rock avalanches. It happened that such slope

269 failures caused significant material damage and destroyed registering equipment. To exclude rock avalanche formation in future testing, it had been necessary to determine the conditions under which rockslides occur and convert into rock avalanches and to predict their runout. Processes of rock slope failure and avalanche motion were fixed by rapid filming from helicopters and ground observational points. Intensity of seismic shaking was measured by accelerometers and velocimeters. For better registration of surface motion after the camouflet explosions, special lights were burned at several points on the slope just before blasting and velocities of these lights' motion were measured. Volumes and areas of source zones and resultant deposits were calculated by surveying and aerial photography before and after the event with an accuracy of about 10-20%. Grain size composition of the resultant debris was studied by profile measuring of fragment dimensions on the surface of the rock avalanche deposits. Mineralogical composition of rocks was studied too. Rockslides ranging in volume from tens of thousands up to nearly one hundred million cubic meters [2] were recorded and studied. Parameters of largest events, which location is shown on Figure 1, are presented in Table 1. Table 1. Dimensions of rock avalanches on the Novaya Zemlya test site. Explosion V (m3) S (m2) h (m) H* (m) L* (m) l (m) L*/H*

B-1 8˜107 3.5˜106 23 400 1900 1600 4.75

A-8 2˜107 7.5˜106 26.7 350 900 700 2.57

A-6 8˜106 4˜105 20 450 1200 750 2.67

A-10 5˜106 2.6˜105 19.2 350 800-950 600 2.3-2.7

A-2 2˜106 1.5˜105 13.3 300 700 450 2.33

A-9 5˜105 6.5˜104 8 350 750 400 2.14

A-3 105 3˜104 3.3 430 900 350 2.1

A-12 4˜104 2˜104 2 300 550 200 1.83

Here and below, ‘V’ and ‘S’ are the volume and the area of rock avalanche deposits, respectively; ‘H*’ is the height of the centre of gravity of the source zone; ‘h’ is the average thickness of the rock avalanche deposits; ‘L*’ is the maximum horizontal distance of the avalanche front from the centre of gravity of the source zone, and ‘l’ is the length of rock avalanche deposits (Figure 2).

Figure 2. Relationships between parameters, traditionally used for rock avalanche description, and parameters, used in the present paper.

The site topography was similar for almost all explosions: the falling rock was able to move free down the slope and spread without significant confinements over the wide

270 valley bottoms, which are nearly horizontal or are inclined at an angle from 2º to 10º. It was found out that avalanche fronts moved much further than might be expected for a rockslide according to the law of dry friction. I must note that above definitions of ‘H*’ and ‘L*’ are traditionally used in our studies, though they differ from ‘H’ and ‘L’ parameters proposed by Heim [12] (see Figure 2). It should be taken into account in comparison of our data with estimates of rock avalanches' runout based on ‘H’ and ‘L’ values. However, our analysis show that in most of cases difference between H*/L* and H/L ratios does not exceed their scatters. 2.1. ROCK AVALANCHE TRIGGERED BY THE EXPLOSION B-1 The largest rock avalanche, 8˜107 m3 in volume (Figures 3-5), was created by the camouflet explosion in the tunnel B-1. The ridge at the test site was 850-900 m high and the crown of the scar rose up to 800 m. The steepest part of the slope, where its angle increased up to 40º-45º, was at the elevation from 300 to 500 m. Below it the slope angle gradually decreased to 5º-10º and to 2º-3º at the foot. The massif is composed of carbonaceous clayey shale with dolomite limestone interbeds, striking 110º-160º with dip angle 20º-30º. Rocks are intensively fractured: fractures' density varies from 5-10 to 3050 per meter. Main fracture system coincides with bedding planes. Along the tunnel fault zones from 10-20 cm up to 1-2 m and, rarely 5-10 m thick were observed. The thickness of scree varies from 0.1-1 m on the slope to 5-6 m at its foot. Permafrost spreads inside the massive up to 500-600 m with constant temperature –4ºC. The subsurface zone 10-15 m thick is exposed to annual temperature fluctuation, decreasing inside the massif.

Figure 3. Schematic map of the source zone and of the deposition area of the B-1 rock avalanche. Contour lines correspond to the pre-failure relief.

Figure 4. Photographs of the rock avalanche 8˜107 m3 in volume formed after the underground nuclear explosion in the tunnel B-1. A and B – views from helicopter; C – view from the downstream part of the Zhuravlevka River valley.



Figure 5. Aerial photo of the rock avalanche 8˜107 m3 in volume created by explosion in the tunnel B-1. White rectangle marks area shown on Figure 3; O -2,3,8 – control points used for survey.

Topographic sketch on Figure 3 shows the position of the source zone which occupied almost the whole mountainside and the final position of the rock avalanche deposits. Average depth of the scar that originated on the slope was about 80 m. Rock avalanche formed deposits up to 1.6 km long while L* was 1.9 km. Its width along the slope foot was 2.2 km, and thickness of the deposits varies from 10-20 m to 30-50 m. Rock avalanche covered 3.5˜106 m2, and its average thickness was calculated as 23 m. This rock avalanche spread over the depositional area rather uniformly, so that the ratios of its area and width and those of source zone were 2.5 and 1.3 respectively. Rock avalanche debris blocked a rather wide valley of the Zhuravlevka River (the valley width at that place is 2 km approximately) and formed an artificial Nalivnoe (Im-

273 pounded) Lake, approximately 1×2 km in size, which still exists several tens of years after its formation. Filtration rate through the dam’s body varies so that the spring flood does not overtop dams crest and, on the other hand, lake exists during low water period. Main morphological features of the rock avalanche can be seen on Figure 5. Two morphological zones can be distinguished in the deposits. Larger frontal part of rock avalanche is characterised by radial alternating ridges and furrows. In contrast, its proximal part is formed by transverse ridges. At the lower part of the proximal zone there is a depression striking along the slope, that is marked by the deep bay of the Nalivnoe Lake. Basing on the above morphology, I assume that debris came to a halt first at its proximal part. Sequential stop of the tailing portions of debris, probably due to momentum transfer from them to distal portions of debris (similar to the mechanism proposed by Van Gassen and Cruden [25]) finally caused the stop of the entire mass of rock avalanche. Along the distal rim of rock avalanche debris, where it started to ascend the opposite slope, thickness of the deposits significantly increase. This frontal zone is characterised by more 'chaotic' micro-morphology. It can be also seen that rock avalanche expanded downstream the river valley more than upstream, and that the chaotic zone is much wider at the downstream limit of the rock avalanche (see Figure 5). Since in the time of explosion was known with high accuracy, the whole process of slope failure and rock avalanche motion was filmed and motion parameters were derived. Figure 6 presents the graph of the velocity of the rock avalanche front motion measured from filming data.

Figure 6. Front velocity of the B-1 rock avalanche as a function of time. Dashed line corresponds to that period of the rock avalanche motion when its front was masked by dust cloud.

Apparently, the debris velocity and, thus, its kinetic energy increased during first 1520 seconds of motion when rock mass moved downslope. During this period it was seen that material that formed the slope surface settled down faster then deeper units. After 20-25 seconds, at a distance of about 1.0-km from the slope foot, the rock avalanche

274 front was formed, which velocity reached maximum value of 60 m/s (about 220 km/s). On the surface of the frontal part of moving rock avalanche we recognised debris that originated from rocks, which rested initially at the uppermost part of the slope. It was found out due to snow spots, because before the explosion snow covered only the very top of the mountain. High, almost maximal velocity of front motion remained for about 20 seconds. Assuming that the entire moving debris had the velocity close to the maximal, we found that kinetic energy of the avalanche (Ek) was about half of the potential energy of the rock mass involved in slope failure (Ep): Ek|0.5 Ep. It means that other half of potential energy was already spent on friction and rock destruction during rock avalanche motion. Avalanche stopped 50 seconds after the explosion. Therefore rock avalanche front came to a halt rather abruptly, approximately in 10 seconds (see Figure 6). Such abrupt decrease of the velocity and termination of motion should be considered as characteristic feature of rock avalanche motion. 2.2. ROCK AVALANCHE TRIGGERED BY THE EXPLOSION A-10 Rather unusual rock avalanche was triggered by seismic shaking after the powerful underground camouflet explosion in the tunnel A-10. Failure took place at the steepest part of the slope due to seismic effect of the explosion. Rock massif in the source zone is composed of terrigenous metasediments of the Silurian age, mainly mica schist, crystalline schist and mica-crystalline schist. Schistosity dips at an angle of 40º-60º in the same direction as the slopes' inclination. The total volume of the rock avalanche was about 5˜106 m3. The pre-failure site topography can be seen on Figure 7. On the film that was made during the explosion and subsequent slope failure it was seen that in the beginning of motion the detached rock mass moved as a single unit and later subdivided into two parts with volumes ratio of approximately 1:4: Avalanche-1, 106 m3 in volume, and Avalanche-2, 4˜106 m3 in volume (Figures 8 and 9). Their main geometrical characteristics are presented in Table 2. Table 2. Geometrical parameters of the Avalanches 1 and 2 triggered by the A-10 explosion. Parameters Avalanche-1 Avalanche-2

V (m3) 106 4˜106

S (m2) 9.7˜104 1.6˜105

h (m) 10 25

H* (m) 350 350

L* (m) 950 800

l (m) 600 500

L*/H* 2.71 2.28

Gravity centres of both parts of rock avalanche were nearly at the same elevation of 350 m, though the crown of the scar above Avalanche-1 was 45-50 m higher. The average slope angle (30º-35º) and seismic intensity characterised by the maximum mass velocity (10-25 m/s), were generally the same for both avalanches. Both avalanches moved along unconfined surface and stopped on the valley bottom dipping 7º-9º. The main difference between the two parts of this rock avalanche is their potential energy corresponding to the volume of each part. One could expect larger runout for Avalanche-2 due to its larger potential energy. But in the case we can see reverse situation: runout of Avalanche-1, with smaller volume, is approximately 1.2 times larger than the runout of Avalanche-2 that is 4 times larger in volume. Similarly, the L*/H* value for Avalanche-1 is bigger that that of Avalanche-2 also by a factor of 1.2. Although the Avalanche-2 is 4 time bigger than the Avalanche-1, it covered the area only 1.6 times larger. It is caused by essentially smaller thickness of the Avalanche-1

275 deposits, which is 2.5 times less than that of the Avalanche-2 on an average. I should also note that thickness of the Avalanche-1 deposits is generally the same along its entire depositional area, while the Avalanche-2 is much thicker at its proximal part then at the distal one. Partially it can be explained by influence of relief: at the end of its path Avalanche-2 moved along small gully (see Figure 7). According to our experience channelling should lead to bigger runout of debris (the same was mentioned by Nicoletti and Sorriso-Valvo [21] for natural events). However, in our case situation is opposite, perhaps due to small extent of channelling at this site, as far as gullies' depth (10-15 m) was less than debris thickness and, thus, could not affect its motion significantly.

Figure 7. Topographic map showing pre-failure relief and configuration of the source zone and deposition area of the rock avalanche formed after the underground nuclear explosion in the tunnel A-10. Contour lines interval is 5 m. 1 and 2 – Avalanches-1 and 2, respectively.


Figure 8. Oblique aerial view of the A-10 rock avalanche. The Avalanche-1 is at the foreground (1); part of the Avalanche-2 is at the background (2); the foot of the source zone on the upper left (S), and the secondary scar of the Avalanche-2 is marked by (SS).

It was assumed that difference in runout might depend on the grain size distribution in the rock avalanche deposits, which is significantly different for these two parts of rock avalanche [3]. Rock fragments on the surface of the Avalanche-2 body are less than 1 m in size, and grain size distribution is characterised by rather small scatter with most abundant fraction of about 15-30 cm. On the other hand, deposits of Avalanche-1 consist of debris with more variable grain size, varying from a few centimetres to 5-10 metres (Figure 10). Difference between debris composition of Avalanches-1 and 2 can be attributed to the difference in the mineral compositions and rock structure due to local peculiarities of metamorphism. However, other explanation of smaller runout of Avalanche-2 in comparison with Avalanche-1 can be proposed (A.L. Strom, Personal Communication, 2002). As could be seen on Figure 7, transition from the slope at the foot of source zone to the deposition area is more smoothed above the Avalanche-1, where it coincide with a small gully, rather then above the Avalanche-2, with prominent steep massif between two gullies.


Figure 9. Combination of vertical aerial photographs of the rock avalanche formed after explosion A-10. The Avalanche-1 is on the left, the Avalanche-2 - on the right. Secondary scar above the tongue of the Avalanche-2 is marked by (SS).

278 It caused different stile of motion of corresponding parts of the rock avalanche. The Avalanche-1 was completely involved in the flow-like motion while the Avalanche-2 can be divided by secondary scar (marked by 'SS' on Figures 9 and 10) into proximal unit that accumulated at the slope's foot and avalanche unit. Similar debris distribution is typical of numerous natural events [24]. Thus, partial involvement of the Avalanche-2 debris in the flow-like motion could lead to its smaller runout.

Figure 10. Grain-size distribution of debris on the surface of the Avalanches-1and 2 of the A-10 event.

Processes of slope failure and avalanche motion at the A-10 test site were fixed by rapid filming. Figure 11 shows successive profiles of the Avalanche-1, as it moved downslope, derived from the film. In the beginning, rock avalanche was obscured by a cloud of dust, which followed the slide. Contour labelled (1) on Figure 11 is the first one when rock avalanche front became visible as it passed through the front of the dust cloud, and successively numbered contours represent later stages in rock avalanche evolution. Based on these data the velocity of the avalanche front was determined.

Figure 11. Dynamics of the Avalanche-1 derived from the rapid film. Numbers correspond to the position of rock avalanche front at the specified time points (in seconds) after the explosion.

279 This part of sliding mass accelerated during its motion along the first 500 m of the runout that started on the slope of 35º and continued until the slope angle decreased up to 15º. Then rock avalanche decelerated. Final stage of motion occurred on the slope, which angle varies from 7º to 10º. The maximum velocity of the Avalanche-1 front reached nearly 40 m/s and lasted for about 10-15 seconds (Figure 12). Assuming, that the whole mass of debris had maximum velocity at this period, its kinetic energy (Ek) could be estimated as about 0.4 of the potential energy of the descended rock massif (EP). Thus, at this stage of rock avalanche motion more then half of the entire energy dissipated due to friction and rock destruction. Total duration of slope failure and rock avalanche motion was ~30 seconds.

Figure 12. Front velocity of the Avalanche-1 as function of time. Dashed line corresponds to that part of the avalanche motion when its front was masked by dust cloud.

2.3. CRITICAL CONDITIONS OF LARGE-SCALE ROCK SLOPE FAILURE Parameters of seismic motion on the slopes caused by powerful underground explosions. were registered by accelerometers and velocimeters. Maximum particle (mass) velocity on the slopes dipping 30º-50º, where large-scale failures occurred, ranged from 8 m/s up to 25 m/s and average acceleration value, measured during such explosions as A-10 or B-1, varied from 10 to 30 g. On the basis of these data critical conditions of large-scale rock slope failure, depending on the intensity of seismic effects and on the slope angle, were determined (Figure 13). Three dashed curves on Figure 13 reflect data obtained in the course of field measurements at the test sites with different slope steepness, geological structure and rock strength and firm line corresponds to the empirical relation of the threshold combination of mass velocity and slope angel described by equation 1.

U cr

5.2 (m/s); D>25º tg (D  25 q )



Figure 13. Critical conditions of the large-scale rock slope failure. 1-3 – empirical data on mass velocity of seismic waves measured on the slopes of different steepness at the test sites: 1 – Lazarev Mountain, 2 – Moiseev Mountain, 3 – Chrnay Mountain. 4 – graph of equation (1); hatched zone corresponds to failure conditions.

Function (1) can be applied for large-scale events only, when linear dimensions of the area on the slope affected by seismic wave with mass velocity Ucr and higher, is not less than 102-103 m. 2.4. RELATIONSHIPS OF ARTIFICIAL ROCK AVALANCHES PARAMETERS Data obtained by rapid filming of rock avalanches, triggered by the B-1 and the A-10 explosions show that their fronts moved with velocities up to 40-60 m/s. Under the same conditions, the larger is the rock avalanche mass, the higher is its front velocity. Distance of rock avalanche front motion is considered as the main characteristic of this phenomenon. It is determined by the kinetic energy value, which rock mass gains while it descend moving downslope under the influence of gravity force. It is evident that L*/H* ratio of rock avalanches, triggered by seismic effect of underground explosions, that have roughly similar H* values (300-450 m), distinctly increase with increase of avalanche volume. Rock avalanches with volumes of 104-105 m3 have the L*/H* ratio about 2. Rock avalanches, which volume is of the order of 108 m3 have the L*/H* ratio up to 5 (Figure 14-A). It implies that with increase of slope failure volume, style of debris motion changes and becomes similar to the motion of the viscous fluid. Fluid-like motion of artificial rock avalanches that spread over unconfined surface becomes apparent from the analysis of the relationship between average thickness of the deposits and their volume (Figure 14-B). While volume grows from 104 up to 106 m3,

281 the average thickness of rock avalanche deposits gradually increase. For volumes exceeding 5×106 m3, average thickness of the deposits reach 20-25 m and further increase of rock avalanche volume is not accompanied by proportional thickening of deposits. It means, that in the case of large rockslide, debris maintain avalanche-like motion until its thickness decrease up to some critical value. After that, resistance to motion increases abruptly being governed by dry friction, and rock avalanche stops.

Figure 14. Length of rock avalanche fronts runout (A) and average thickness of deposit (B) as function of their volume.

3. Comparison With Natural Events To analyse motion of rock avalanches of larger volumes we used data on several wellknown natural events, which volumes range from 107 up to n×1010 m3 [23, 26]. Some of them are comparable in size with the largest artificial avalanches at the Novaya Zemlya test site and have L/H ratio of the same order ranging from 3 to 5 [23]. As noted above, for general qualitative comparison it is acceptable to use both L and H, and L* and H* values (see Figure 2) and corresponding ratios. Length of runout of larger natural rock avalanches involving cubic kilometres of debris, is much bigger – up to 13-16 km [23] and corresponding L/H values exceed 10. Several rock avalanches associated with volcanic eruptions, such as the Shiveluch, Bezymyannyi and Kamen in Kamchatca and the St. Helens event in the U.S.A., were analysed too. One more group of slope failures, resulting in the avalanche-like motion, which was utilised for the comparison of the artificial and natural rock avalanches, is failure of the rock waste dumps of the Central mine in the Khibini Mountains [18]. Dumps are stacked on the steep slopes of the Rosvumchorr Plateau at elevation of 900-1000 m a.s.l. Slope angles of the plateau reach 30-50º. As the waste accumulates sudden collapses of debris mixed with ice and snow occur, creating rock-ice avalanches. Parameters of the largest rock-ice avalanches are presented in Table 3. It should be noted that rock-ice avalanches are extremely mobile comparatively to their volumes and, thus, pose a significant threat. Correlation of L/H as well as L*/H* and l/H* ratios of artificial and natural rock avalanches of different origin versus their volumes are shown on Figure 15. It represents general relationships of these parameters that fit to the same equation both for natural and artificial events.

282 Table 3. Parameters of the rock-ice avalanches. V (m3) L* (m) H* (m) L*/H*

2˜106 1100 400 2.7

7˜105 400 250 1.6

1.5˜106 1000 400 2.5

1.3˜106 500 250 2.0

2.4˜106 1100 400 2.7

2.6˜106 900 370 2.4

6˜106 3000 650 4.6

3˜105 240 170 1.4

4˜106 1600 500 3.2

Figure 15. Dependence of the rock avalanche length to fall height ratio, versus their volume (after [2], modified). Key to legend: A – rock avalanches triggered by underground nuclear explosions (see table 1). Upper circle – L*/H* value, lower circle – l/H* value. B – rock-ice avalanches at the Khibini Mountains (L*/H* values). C – volcanic rock avalanches (L/H values with their scatter). D – natural non-volcanic rock avalanches (L*/H* values). The following natural events are enumerated: 1- Saidmarreh, 2 – Flims, 3 – St. Helens volcano, 4 – Shiveluch volcano , 5 – Kamen volcano, 6 – Bezymyannyi volcano, 7 – Khait, 8 – Aini, 9 – Goldau, 10 – Frank, 11 – Madison, 12 – Elm. Curves 1-3 are explained in the text.

Curve 1 on Figure 15 is drawn through a set of points that represent rock avalanches at the Novaya Zemlya test site and natural rock avalanches that moved in the unconfined environment. For rather small volumes (104-106 m3), the L*/H* ratio is practically

283 constant and equal to |2. In the range of volumes V=107–108 m3, there is a good agreement between data on rock avalanches of explosive and natural origins. For bigger volumes, curve 1 also fits well with data on both volcanic and non-volcanic events. For Vt106 m3 this curve can be described by the empirical function: L(L*)/H(H*) = 0,13V0,2


Curve 2 on Figure 15 reflects growth of l/H* ratio versus rock avalanche volume for the artificial events. For volumes more than 108, when length of unconfined debris apron significantly exceeds dimensions of the source zones, it becomes very close to the curve 1. Shape of these curves show that massive rock slope failures in which more than one million of cubic meters of rocks are involved, are accompanied by a peculiar scale effect that leads to anomalous mobility of debris and a to corresponding increase of rock avalanche runout. Lastly, the rock-ice avalanches formed on waste dumps in the Khibini Mountains have much higher mobility, demonstrated by curve 3 on Figure 15. It can be assumed that presence of the ice and snow reduces the resistance force of such avalanches and their mobility starts growing at lower volumes. 4.


Real-time observations of the processes of massive rock slopes failure and rock avalanches formation at the Novaya Zemlia nuclear test site provided data for better understanding of this hazardous phenomena. Numerous artificial rock avalanches ranging in volume from 105 m3 up to 108 m3 have been studied in details. Critical conditions at which rock slopes' failure occur, depending on mass velocity caused by seismic waves and on slope angle, have been established for seismic effects of underground nuclear explosions. They can be applied for natural conditions too, especially in the areas, where strong motions, corresponding to 9 or more points of the MM or MSK scales are expected. Direct measurement of rock avalanches front velocities show that they reach 40-60 m/s and that rapidly moving debris came to a halt in few seconds when its velocity decreases from more than 90 km/h up to zero. It proves the indirect observations, indicating the abrupt stop of natural rock avalanches. It was also found out that when rock avalanche front velocities were maximal, more then half of the initial energy of descending rock mass already dissipated due to friction and debris fragmentation. Differences in geometrical parameters of two parts of rock avalanche triggered by A-10 explosion allow proposing that grain size distribution of rock avalanche debris can affect their mobility. I hypothesise that rock avalanches composed of uniform material are less mobile than those composed of fragments more variable in size. However, it can not be excluded that the observed differences were due to peculiarities of the site topography, which caused different mass distribution in both parts of this rock avalanche. Comparison of rock avalanches triggered by powerful underground explosions and both volcanic and non-volcanic large-scale natural events demonstrates that their high mobility obeys the same relationships and, thus, should be governed by the same me-

284 chanical processes. The empirical formula (2) and curves 1-3 on Figure 15 allow predicting roughly geometrical parameters of unconfined rock avalanches for the given volume and the height of collapse. References 1. 2. 3.

4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. 16. 17. 18. 19. 20. 21. 22.

23. 24. 25. 26.

Abele, G. (1974) Bergsturze in den Alpen, Wissensch, Alpenvereinshefte, Munchen H. 25. Adushkin, V.V. (2000) Explosive initiation of creative processes in nature. Combustion, Explosion, and Shock Waves 36, No 6, 21-30. Adushkin, V.V., and Spungin, V.G. (1998) The influence of granular structure of rockfalls on their speading along mountain slopes. In: Proceedings of the third international Conference on Mechanics of Jointed and Faulted Rock. Vienna, Austria, 6-9 April 1998, 541-546. Adushkin, V.V., Rodionov, V.N., and Shcherbakov, S.G. (2000) Mechanism of the spontaneous generation of rock avalanches on mountain slopes. Doklady Earth Sciences, 373 A, 958-959. Campbell, C.S. (1989) Self-lubrication for long runout landslides, J. Geology 97, 653-665. Campbell, C.S., Cleary, P.W., and Hopkins, M. (1995) Large-scale landslide simulations: Global deformation, velocities and basal friction, J. Geophys. Res. B100, 8267-8283. Costa, J.E., and Schuster, R.L. (1988) The formation and failure of natural dams, Geol. Soc. Am. Bull. 100, 1054-1068. Eisbacher, G.H. (1979) Cliff collapse and rock avalanches (sturzstroms) in the Mackenzie Mountains, northwestern Canada, Can. Geotechnical J. 16, 309–334. Eisbacher G.H., Clague J.J. (1984) Destructive mass movements in high mountains: hazard and management. Ottawa, Geological Survey of Canada Paper 84-16. Erismann, T.H. (1986) Flowing, rolling, bouncing, sliding: synopsis of basic mechanisms, Acta Mechanica 64, 101-110. Fedorenko, V.S. (1988) Mauntainous Rockslides and Rock Falls, their Prediction, Moscow State University (in Russian). Heim, A. (1932) Der Bergsturz und Menschenleben, Fretz und Wasmuth, Zurich. Hewitt, K. (2001) Catastrophic rockslides and the geomorphology of the Hunza and Gilgit river valleys, Karakoram Himalaya, Erdkunde 55, 72-93. Hsu, K.J. (1975) Catastrophic debris streams generated by rockfalls, Geol. Soc. Am. Bull. 86,129-140. Hungr, O. (1990) Mobility of rock avalanches. Reports of the National Research Institute for Earth Science and Disaster Prevention, Tsukuba, Japan, 46, 11-20. Hungr, O., and Evans, S.G. (1996) Rock avalanche runout prediction using a dynamic model, in: Proceeding 7 th International Symposium on Landslides, Trondheim, Norway, 1996, Balkema, 233-238. Kent, P.E. (1966) The transport mechanism in catastrophic rockfalls, J. Geology 74, 79-83. Krasnoselsky, E.V., and Kalabin, G.V. (1975) Rock dumps on mountain slopes, Nauka, Leningrad, (in Russian). Legros, F. (2002) The mobility of long-runout landslides, Engineering Geology 63, 301-331. Melosh, H.J. (1979) Acoustic fluidization: a new geological process?, J. Geophys. Res., 84, 7513-20. Nicoletti, P.G. and Sorriso-Valvo M. (1991) Geomorphic controls of the shape and mobility of rock avalanches, Geol. Soc. Am. Bull. 103, 1365-1373. Sassa, K., Fukuoka H., Lee, J-H., Shoaei Z., Zhang, D., Xie, Z., Zeng, S, Cao, B. (1994) Prediction of landslide motion based on the measurement of geotechnical parameters, in: Developement of a new Cyclic Loading Ring Shear Apparatus to study earthquake-induced-landslides. Report for Grant-in-Aid for Developmental Scientific Research by the Ministry of Education, Science and Culture, Japan (Project No 03556021), DPRI, Kyoto, 72-106. Shaller, P.J. (1991) Analysis and implications of large Martian and Terrestrial landslides, Ph.D. Thesis, California Institute of Technology. Strom, A.L. (2003) Morphology and internal structure of rockslides and rock avalanches; grounds and constraints for their modelling. This volume. Van Gassen, W., and Cruden, D.M. (1989) Momentum transfer and friction in the debris of rock avalanches, Can. Geotechnical J. 26, 623 - 628. Voight, B., Landa, R.G., Glicken H., Douglass, P.M. (1983) Nature and mechanics of the Mount St.Helens rockslide-avalanche of the 18 May 1980. Geotechnique 33, 243-273.


M.J. McSAVENEY* Institute of Geological & Nuclear Sciences Ltd, PO Box 30 368, Lower Hutt, New Zealand

T.R.H. DAVIES Department of Geological Sciences University of Canterbury Christchurch New Zealand

Abstract Dynamic fragmentation is a new hypothesis for the mechanism of rock-avalanche long runout. Low-strain-rate fragmentation is dominated by growth of a few flaws. It is the regime leading to initial failure of many landslides. Static rock strength is largely independent of loading rate. The dynamic regime is entered when growth of a few flaws does not relieve elastic strain fast enough, and stresses rise adjacent to the flaws, forcing many new ones to nucleate and grow. Strengths of dynamically fragmenting materials increase at about the 4th root of strain rate. Elastic strain energy, W, per unit volume, released at failure is given by W=Q2/(2E), where Q is strength and E is elastic modulus. Its explosive release as kinetic energy provides a large, isotropic, clast-dispersing stress, every time any clast is stressed to failure. Fragmentation-induced dilation is a positive granular “pressure”, but also causes low pore-fluid pressure, and is incompatible with saturation of voids by liquids and therefore is incompatible with high pore pressure and undrained loading. Driven entirely by internal deformation within the avalanching mass, dynamic fragmentation propels the distal margins of large avalanches of brittle rock further than they could travel had they just collapsed to joint-bounded clasts. 1. Introduction The puzzle of the unexpectedly long runout of large rock avalanches (Figure 1) has been “solved” many times since Heim [15] first drew attention to it. Here, we elaborate on our presently favoured solution for what, in the progress of 120 years, has now become the expected, if not fully explained, long runout of large rock avalanches – this solution is dynamic fragmentation [14]. It is necessary to invoke the process of dynamic fragmentation in large rock avalanches to explain the abundance of finely comminuted * Email of corresponding author: [email protected]

285 S.G. Evans et al. (eds.), Landslides from Massive Rock Slope Failure, 285–304. © 2006 Springer. Printed in the Netherlands.


Figure 1. Oblique aerial view to the southeast of the long-runout, earthquake-triggered Falling Mountain rock avalanche of 1929 in New Zealand’s Southern Alps (42˚54´ S, 171° 41´ E). Some 55 x 106 m3 of rock fell from Falling Mountain and ran out to 4.54 km in a fall of 1200 m [28]. At the close of the avalanche, the deposit had bulked up to ~66 x 106 m3 and its interior had become finely comminuted (Figures 3 and 4), consisting of a matrix-supported deposit with 50% of clasts smaller than 10 mm (Figure 10). Figures 3 and 4 are of the wall of the canyon eroded in the avalanche deposit at lower left (locations marked on Figure 5). The deep amphitheatre scar on the flank of Falling Mountain appears to be diagnostic of an earthquake trigger, although not all earthquake-triggered landslides exhibit this. Photograph by Lloyd Homer.

rock in their deposits. Our hypothesis [8–10, 28] is that, in driving the fragmentation, the portion of rock-avalanche kinetic energy consumed in straining innumerable clasts to breaking point is released explosively (Figure 2), creating a positive isotropic clast pressure within the avalanche interior, as breaking clasts interact with surrounding clasts. In this continuous explosion of kinetic energy, work is done in dilating the rock mass, spreading it further than it could have spread without the additional pressure. Thus, the positive internal pressure from fragmentation drives longer runout, without a need to invoke lowered basal or internal friction coefficients, or special pore fluids, or pore conditions. In this paper, we further develop and illustrate our fragmentation concepts [9, 10] and explore the effects of comminution on rock avalanches, drawing from concepts developed in petroleum extraction [14], Rock mechanics [3, 4, 6, 34] mining [16] and astrogeology [29].


Figure 2. Violent mode of failure of a 30 cm cube of coal described by Bieniawski [4] as occurring “in a manner tantamount to an explosion.” The fragmentation hypothesis [9, 10, this paper] holds that the very high stress environment in the straining interior of large rock avalanches creates a myriad of such explosions, which may be likened to rock bursts in deep mines. The analogy with rock bursts is highly relevant because in both rock avalanches and rock bursts, the rock is strained to the point at which it ruptures catastrophically (as illustrated), and so, both hold similar amounts of stored elastic energy, despite the widely differing external stress environments. It is immaterial how long the energy has been stored, for millions or billions of years in some rock bursts, or for a few minutes as in the illustrated experiment, the outcome of its sudden release is the same—dynamic fragmentation (see section 5).


Collapse versus Fragmentation

We recognised two “separate phenomena” in the disintegration of source masses [9, 10]. The first is the falling apart of the rock mass along pre-existing joints and other large defects that give the in-situ rock a relatively low strength. We called this process collapse. We distinguish it from the subsequent fracturing during which a large part of the rock mass is comminuted to particles well below the sizes dictated by the spacing of original rock-mass discontinuities such as joints. We call this latter process fragmentation. Specialists in rock mechanics will recognise these processes as part of a continuum in which rock-mass strength is determined by the probability-densty distribution of defects and the size of the rock mass, as investigated by Bieniawski [4]. Collapse is the failure of lower frequency (more widely spaced), weaker defects, and fragmentation exploits the stronger and smaller defects that have a much higher spatial frequency (are more closely spaced). Collapse occurs in all landslides of jointed rocks. Fragmentation occurs to a limited extent in all of them too, but it becomes dramatically dominant in the larger rock avalanches. A markedly contrasting display of released kinetic energy between the extremes in this continuum was noted by Bieniawski [4].


Figure 3. View of interior of the Falling Mountain rock avalanche deposit 3.5 km from source. Alternating light and dark bands are from fragmentation of light grey sandstone and dark grey mudstone beds from the original source block. Note coarse, clast-supported deposit texture near the upper surface, and fine, matrixsupported texture below. Fifty percent of the interior is finer than 10 mm (Figure 10) [28]. Side stream has exhumed the former valley wall. Basal avalanche debris includes wood fragments, eroded soil and waterworn cobbles eroded from the upper valley floor. Locally, severely abraded, in situ tree stumps are rooted in soil and protrude into the base of the avalanche. See Figure 5 for site location.

3. Comminution and Runout To demonstrate the ubiquitous distribution of finely comminuted rock in long-runout landslides, we need hardly look beyond the presentations at the Celano workshop [1, 17, 24, 27, 31, 32], but see also [9, 10, 28] and references cited therein. The degree of comminution can be deceptive in outcrop. At New Zealand’s Falling Mountain (Figure 1)[28] and Round Top rock avalanches [35], and at Flims, Switzerland, [30] some outcrops of apparently intact rock that display continuity of structural elements across the outcrop, collapse to coarse sand when sampled because of the pervasive fracturing. The comminution, however, is extremely variable from outcrop to outcrop, precluding characterisation of the complete particle-size distribution for a rock avalanche. The greatest variation, however, is from the bouldery, clast-supported exterior to the sandy, matrix-supported interior (Figure 3). The failing rock mass must fracture at the onset of deformation for the mass to be released from its source to begin granular flow. Fracturing must continue to the last moments of significant deformation because it is driven by the deformation-induced stresses of grain interactions [10]. Across a single outcrop (Figure 4), we find broken rock in all stages of separation, from tightly packed “sibling” clasts to chaotically disordered masses.


Figure 4. Basal interior of the Falling Mountain rock avalanche [28] beneath 60 m of overburden, 4 km from source. Width of photograph is ~5 m, but the illustrated features are somewhat scale invariant from macro to micro scales. Note the regions on the lower left and middle right with clusters of fractured but only partially disaggregated clasts, and a central zone of disordered fragments in a fine matrix. Within the clusters, there is progressive disorder from a central zone of order to the zone of disordered clasts. The progressive disorder is accompanied by isotropic volume dilation. The dilation is zero for unfractured rock and ~42% for fully crushed, and disordered rock [12]. Fracturing is progressive throughout the runout or the ordered clasts would be destroyed. It represents a volumetric tensile strain against a compressive load of ~1.2 MPa. In order for this to occur, the fracturing of the rock (fragmentation) must have been accompanied by a dispersive stress ≥1.2 MPa. Our hypothesis is that this is provided by the positive clast pressure exerted by explosively fragmenting clasts.

Many cobble-sized and smaller clasts from New Zealand’s Mount Cook and Mount Fletcher rock avalanches (Figure 12) [26] are bounded by a mix of hackly, highly irregular surfaces (Figure 6) and sections of conchoidal fracture surfaces, often with clear evidence of an impact origin (Figure 7). Jointing in the source mass of the Mount Fletcher rock avalanches, however, locally was so closely spaced that some clasts as small as 1 cm3 are bounded by planar joints on all surfaces. Fragmentation can, however, involve the creation and collapse of new closely spaced joint sets, as in the Flims landslide [30, 32, 33] (Figure 11), suggesting that some fragmentation there was caused by sound waves resonating to very high amplitude within clasts. Rock comminution plays an undeniable role in the spreading of landslide masses (Figures 1 and 12). Without any comminution, the landslide mass would be locked in the source, or would translate as a single block without spreading. Despite the outward appearance of a coarse blocky, clast-supported deposit (Figure 12 lower), the interior of a typical rock avalanche deposit is finely comminuted, to the point of being matrixsupported, with a finely powdered matrix of sand and silt [10] (Figure 10).


Figure 5. Portion of vertical aerial photograph of the Falling Mountain rock avalanche showing outline of the source scar and deposit (dashed line) and the lines of the profile (dotted A’-B’) used for the diagram in Figure 16. Sample locations refer to sites of Falling Mountain grain-size distributions in Figure 10. Fig. 3 and Fig. 4 refer to sites where Figures 3 and 4 were taken. The upper reach of the west branch of the Otehake River has exhumed a pre-avalanche bedrock gorge, exposing a longitudinal section to the base of the distal third of the avalanche deposit.


Figure 6. Hackly fracture surface on pebble of highly indurated greywacke siltstone form the Mount Cook rock avalanche of 1991 [26]. Width of view 50 mm.

Fragmentation is a necessary and sufficient condition for “long runout” [9, 10], but it is not the only process contributing to runout. Granular flow is another necessary condition, but it alone is not sufficient to explain long runout [10]: it also is necessary in short-runout landslides (Figure 13). Davies and McSaveney [9] show that a low coefficient of friction is a sufficient, but not a necessary condition for long runout. Lowered friction through some lubrication mechanism is not likely to be a “universal” solution to the long-runout puzzle. Although pore-fluid saturation and undrained loading [21] may be sufficient conditions for long runout [19, 22], for most lithologies, they are incompatible with rock-mass dilation through the growth of intergranular void space that is a necessary consequence of fragmentation. Saturation and undrained loading cannot occur in a shearing rock mass while fragmentation continues, unless the rock mass has very specific and unusual properties. The required conditions are a saturated, porous rock


Figure 7. Cobble from the distal surface of the Mt Cook rock avalanche of 1991 [26] showing surface texture resulting from high velocity impacts with other clasts. Scale is in millimetres. Figure 8 is close view of impact scar adjacent to digital callipers. Clast lithology is very highly indurated greywacke siltstone, which displays fine detail much better than the dominant coarse sandstone clasts.

mass whose initial void space can collapse with increasing strain [20] – such as chalk, quick clay, and pumaceous tephra.

4. Isotropic Dispersive Stress of Fragmentation – a Positive Granular Pressure A simple energy-balance analysis demonstrates that if low coefficients of friction are not universal in long-runout landslides, then another force besides gravity must be acting directly to achieve the additional runout. We [8–10, 28] put forward the hypothesis that this force is an internal dispersive stress originating from the continual break up of the rock mass to grain sizes well below that dictated by the spacing of joints in the source mass before break up. This force arises indirectly from gravity. No energy other than that acquired from the loss of potential energy in the fall is involved. The process of energy use is more complicated than the simple conversion of potential energy to kinetic energy and loss of kinetic energy to heat energy through friction. As clasts are stressed to their breaking point, they convert kinetic energy to potential energy as they deform elastically [3, 4]. This then is released as kinetic energy as the clasts break [3, 4, 6]. The release is expressed as an isotropic dispersive strain on the clast fragments [3, 4, 6, 34] (Figures 2 and 15), with no net gain or loss of momentum


Figure 8. Close view of impact scar preserved on highly indurated greywacke siltstone cobble from the Mt Cook rock avalanche of 1991. Width of view is 70 mm.

from that of the clast prior to fracturing. The elastic strain energy (W) released by failure of a unit volume of uniaxially stressed rock is: W=Q2/(2E)


where Q is the strength of the rock and E is the elastic modulus [16]. The potential energy is released explosively as kinetic energy at failure [3, 4, 6] (Figure 2, the analogous phenomenon of rock bursts in mines is a serious danger to life, and even the most modern rock-strength testing equipment still requires a safety shield around the tested sample). It provides a powerful internal dispersive stress within the fragmenting mass so that fragmenting clasts exert a strong positive pressure against their adjacent clasts (Figure 15 right). Equation 1 is valid for either tensile or compressive strength, depending on the mode of stress causing clast failure. The evidence of conchoidal surfaces (Figure 8) and impact scars (Figures 7–9) on clasts indicates that compressive loading in impacts is a widespread cause of fragmentation in rock avalanches. There is much to be learned of the pressure generated by the fragmentation process in rock avalanches from the behaviour of test samples in laboratory rock-strength testing machines. In early machines, stress was used as the independent variable, and brittle rocks often responded with exceptional violence because the resulting strain on fracture was unrestrained (as in Figure 2) [3, 4, 34]. More recent servo-controlled testing machines can use strain as the independent variable [18] and so can eliminate the explosive release of energy, but this is achieved by strong mechanical restraint on


Figure 9. Impact scar on highly indurated greywacke siltstone cobble (rear side of cobble in Figure 7) from the Mount Cook rock avalanche of 1991 [26]. Width of view is 20 mm.

the motion [18]. No such restriction on strain is present in a rock avalanche, and the release of energy is unrestrained in most fragmentation events in nature. Unweathered greywacke rock of New Zealand’s Southern Alps typically has a compressive strength Q = ~280 MPa and E = ~70 GPa. Strength in impulse loading appropriate to rapid landslides may be 5–10 times the static strength [14], and so W =~14–56 kPa/m3. At failure, about half of this potential energy is manifest as kinetic energy (the remainder is manifest as heat) [2], implying average root-mean-square unconfined fragment velocities of ~50-100 m/s, with some particle velocities of up to ~100-200 m/s. Similar particle velocities are calculated for rock bursts [25]. Wawersik and Fairhurst [34] recognise two classes of brittle rock failure behaviour (Classes I and II, Figure 15 right) that are independent of the ‘stiffness” of the testing machine. Class II brittle rocks are able to dilate axially against stresses equal to the rock strength (Figure 15), indicating the magnitude of the dispersive pressure involved in failure of a clast of Class II brittle rock. It is clear, however, from the behaviour of rock avalanches that they are not powered by average internal pressures of the order of 1–2 GPa that might be encountered during the impulse-loaded fragmentation of strong brittle rock such as fresh New Zealand greywacke. We can deduce from this that only a small proportion of the interior clasts can be fragmenting at any instant, a further argument supporting our contention that fragmentation is a significant process through most of the duration of motion. Although there is a positive granular pressure from the fragmenting interior for the dispersal of the fragments, there must be reduced pressure in the material occupying

295 100% Alpine fault Acheron


Casey Coleridge

Proportion finer


Falling Mtn 1


Falling Mtn 2 Falling Mtn 3

Mt Cook Round Top


South Ashburton Frank Slide (base) Mt St Helens (average)


0% 0.001







Grain size (mm)

Figure 10. Some rock-avalanche grain-size distributions. Data are mostly from single samples collected by the authors. Casey landslide data from [5], Frank Slide from [7] and Mt St Helens from [13]. Alpine fault data are from a sample of fault gouge from within an exposure of a pure thrust portion of the New Zealand Alpine fault. Locations of the three Falling Mountain samples are shown in Figure 5.

the growing void-space volume. Hence fragmentation is incompatible with mechanisms of long runout requiring water saturation and undrained loading. 5.

Dynamic Fragmentation

We can now add the concept of dynamic fragmentation [14] to our earlier fragmentation hypothesis [9, 10]. Grady and Kipp [14] recognise dynamic fragmentation as distinct from static fragmentation. Static fragmentation arises at low strain rates though gradual crack initiation and growth, although the behaviour of the mass at failure is far from static. In the quasi-static failure regime, fragmentation is dominated by the growth of a single, weak flaw. This is the failure regime leading to the initial failure of many landslides, and it must be the failure process involved in the break down of rock in creeping landslides. The strength of the rock mass in this regime is largely independent of the loading rate. The dynamic regime is entered when the growth of this flaw does not relieve the applied elastic strain, and stresses rise in the material adjacent to the flaw, forcing new flaws to nucleate and grow. Some landslide failures under the dynamic loading of earthquake shaking may occur in this regime, but here we are concerned more with the behaviour of innumerable individual clasts as they interact with adjacent clasts during their high-speed travel across the landscape.


Figure 11. Closely spaced “playing card” joints formed in the fragmentation of fine-grained limestone in the prehistoric Flims rock avalanche, Rhine Valley, Switzerland [30, 33] (58 mm lens cap provides scale). If the joints were formed by impact-induced internal pressure waves reflecting from the surface within a previously intact clast, they would have formed at a spacing of one quarter wavelength. For a seismic velocity of ~4 km/s, this suggests that a resonating acoustic frequency of ~100 kHz caused this spacing of joints.

Fragmentation at high strain rates is by the rapid growth of all available crack nuclei in the stressed region, with bifurcation of propagating cracks, as must be occurring in the fragmentation event illustrated in Figure 2. Fragmentation in this regime is controlled by the dynamic as opposed to quasi-static propagation of cracks hence the name dynamic fragmentation. There is a transition between static and dynamic fracture at some critical strain rate. The strength of dynamically fragmenting material increases with approximately the 4th root of the strain rate [14]. Melosh et al. [29] show through Grady-Kipp fragmentation theory that the transition strain rate defining the boundary between static and dynamic fracture is a


Figure 12. Contrast in style of distal lateral margin between the Mount Cook rock avalanche of 1991 (upper) and the first Mount Fletcher rock avalanche of 1992 (lower) [26]. The two were of similar volumes of highly indurated greywacke sandstone and siltstone, and both ran out on glaciers. These two long runout landslides differed in height of fall (2.7 vs 1.2 km) and degree of comminution, resulting in greater fragmentation-driven spreading, and hence a thinner deposit, at Mount Cook.


Figure 13. An informative path to a cause of long runout leads past landslides that lack it. The “unexpectedly short” runout South Ashburton landslide [28] fell onto a saturated substrate of gravel overlain by loess. It eroded the loess, but shows no reduced basal friction in its runout (volume 300,000 m3; fall height >740 m; runout 300 m; apparent friction 0.4-0.5). Excavations into the deposit (Figure 14) suggest that the source mass collapsed to the limit imposed by joint spacing, producing a clast-supported deposit. It did not fragment to the matrix-supported deposit we find universally in long-runout rock avalanches.

function of the size of the fracturing mass, and decreases exponentially as the volume of the mass increases. Several lines of evidence attest to the occurrence of dynamic fragmentation in rock avalanches. Events such as are illustrated in Figure 2 must occur repeatedly during the course of a rock avalanche. Rough, hackly surfaces on clasts (Figure 6) indicate multiple bifurcation of fracture surfaces, as do apparently structurally intact clasts in which no fragment is larger than coarse sand. Impact marks (Figures 7-9) and conchoidal fracture surfaces (Figures 7, 8) resemble those made with a hammer blow. Also, a large rock avalanche typically completes its runout in the order of 100 seconds, during which about half of its mass is reduced to particles smaller than 10 mm in diameter (Figure 10). The dynamic regime is required if this task is to be accomplished in such a short time.


Figure 14. The interior of the South Ashburton landslide is clast supported, with interior grain sizes similar to those of the exterior cobbles (but for a matrix of inwashed sand and silt).

In rock avalanches, there is increasing basal shear stress and increasing maximum strain rates in increasingly larger events. At the same time, the transition to dynamic fragmentation occurs at lower strain rates, and dynamic strength increases at higher strain rates. There also are more clasts. Hence, as avalanche mass increases, more and more clasts undergo dynamic fragmentation, while at the same time being subject to an increasing amount of energy being recycled through elastic energy to be manifest as an internal dispersive stress within the dilating, fragmenting avalanche mass. This results in an increasing internal granular pressure during runout with increasing avalanche mass. We see the outcome of this dynamic-fragmentation theory in greater spreading of a more finely comminuted, fragmented rock mass in larger rock avalanches. The hypothesised sequence of events is illustrated in Figure 16. 6.

Fracture-surface Energy and Fragmentation

In earlier work [9], we described fragmentation as a net energy sink, with energy absorbed in fracturing bonds within a rock’s constituent mineral crystals. We calculated that the increase in clast surface area in rock avalanching might absorb 4% of the total energy of the fall [9]. Here we note the spurious basis of our earlier work. Fracture-surface energy is the energy that must be expended to fracture rocks and minerals. It is the measure of the work that must be done to force the opposing sides of


Figure 15. Stress vs. strain diagrams for brittle rocks in uniaxial compression. Left and centre diagram after [3]. Right diagram after [34]. Class II rocks display axial dilation against loading at the yield strength, indicating the magnitude of the internal dispersive pressure available in rock avalanches of Class II rocks. But even Class I rocks dilate during the unstable fracture propagation that shatters grains in experimental uniaxial compression. These behaviours are independent of the stiffness of the compression testing machine [18].

cracks apart to create fragments. As noted by Bieniawski [3], “it is doubtful whether the breakdown of atomic bonds plays any significant part” in the energy losses associated with the breakup of rock masses. The energy losses are entirely associated with the movement of faces of the extending cracks, leading to unavoidable dilation of the fractured rock mass [3]. As such, “fracture-surface energy” is a part of the work done in spreading rock avalanches, and can not be considered to be a loss in the context of rock avalanche mobility. The efficiency of the cycle of energy through elastic strain to fragmentation is about 50% [2]. 7.

Fragmenting Flow and Enhanced Mobility Due to High Pore-fluid Pressure

The ubiquity of dynamic fragmentation through the major portion of rapid runout of rock avalanches has an important implication for pore pressure within such landslides. The increase in pore volume associated with fragmentation and rock-mass dilation ensures that pore pressure is reduced relative to the static condition. The basal grainsize distribution in rock avalanches suggests a very low permeability to water, and the pressure driving water into the mass is 1 bar at the very most, so the rate of water ingress must be low [23]. In the general case, therefore, water-saturation of a fragmenting layer is a physical impossibility. Thus empirical evidence of intense fragmentation is empirical evidence for an absence of high fluid pressure. If undrained loading is to occur, it cannot take place in or near a rapidly shearing zone of fragmenting clasts (other than in a few peculiarly porous lithologies noted earlier, where shear strain leads to loss of solid volume [20]). The 55x106 m3 source volume of the Falling Mountain rock avalanche [28] bulked up to a deposit volume of about 65x106 m3, but may have been ~70x106 m3 at its peak

Pre-failure avalanche mass Granular flow begins as rock mass collapses into joint-bound blocks along the weaker pre-existing defects. Intergranular forces grow until they initiate fragmentation and fragmenting flow, a highly dispersed granular flow in which many intergranular contacts cause grain breakage, rather than movement of grains around one another.

0 ~5 Beneath a carapace of granular flow lies a fragmenting rock mass with the rheology of a heavy vapour under positive pressure.

~30 The isotropic intergranular dispersive stress from fragmentation dilates the mass and increases spreading with normal internal and basal kinetic friction, but with reduced normal stresses beneath the non-fragmenting carapace.

Time (seconds)

~70 Intergranular forces drop below that required for fragmentation. Normal granular flow resumes throughout mass. Dilated mass begins to settle. The mass grinds to a halt.

~100 Post-avalanche topography

~120 Figure 16. Diagram illustrating the hypothesized fragmentation sequence applied to the runout of New Zealand’s earthquake-triggered Falling Mountain rock avalanche of 1929. Time is based on our numerical simulations [9]. Fall height is 1.2 km, runout is 4.54 km. See Figure 5 for line of profile. In this laterally confined avalanche, the strong isotropic dispersive stress of fragmentation acted both to directly extend the avalanche longitudinally, and to reduce the normal stress, so reducing internal friction losses.



dilation before coming to rest. This implies an increase in pore volume of ~15x106 m3 over the ~100sec duration of the avalanche. Since the greatest degree of fragmentation is near the base of the avalanche deposit, we infer that the greatest dilation and greatest pore-pressure gradients also were near the base. In its descent, the Falling Mountain rock avalanche destroyed alpine meadows on snow-avalanche talus/debrisflow fans, before overwhelming a deeply incised rock gorge flanked by trees. The talus and fan deposits fed a perennial river and we infer a large groundwater reservoir in the upper valley. Mixed soil, colluvium, alluvium and woody debris are found at the base of the distal margin of the deposit. We infer that water was sucked into the basal expanding avalanche debris, but the river can have held only a few thousand cubic metres of water. If the eroded groundwater reservoir was an order of magnitude larger, the total amount of available water was 0.1 km ) flank collapses have dramatically modified more than 200 stratovolcanoes worldwide [25, 26] and pose some of the most sudden, destructive, and life-threatening of volcanic events. An exceptional modern example is the enormous failure at Mount St. Helens, USA, in 1980 [33]. Massive collapses generate rock and debris avalanches that may mobilize into debris flows, thereby creating hazards both on the edifice and in areas far downslope or downstream. Moreover, many of the Earth’s ~700 stratovolcanoes endanger residents of developing nations; thus, techniques are especially needed for rapid and cost-effective hazard assessments. About 20,000 people 1

E-mail of corresponding author: [email protected]

445 S.G. Evans et al. (eds.), Landslides from Massive Rock Slope Failure, 445–458. © 2006 Springer. Printed in the Netherlands.


have been killed by historical flank collapses [26], and understanding how, why, and where volcano slopes collapse is vital to evaluating long-term volcano evolution and immediate hazards. Numerous processes can destabilize an edifice [32], including volcano-specific effects, such as magma intrusion, hydrothermal alteration, or thermal pressurization of pore fluids, as well as more commonly recognized destabilizing effects, such as elevated pore-fluid pressures or earthquake shaking. Although this abundance of contributing factors complicates collapse predictions, gravity always affects collapse. The gravitational stability of volcano edifices is strongly influenced by the interplay of topography, potential failure surfaces, and the 3-D distributions of rock strength and pore-fluid pressure. We have developed a 3-D slope stability analysis that can extensively search the rock mass underlying a digital elevation model (DEM) to determine the locations of minimum stability and the expected volumes of potential failures [21]. Two stratovolcano edifices in the USA, Mount St. Helens and Mount Rainier, have had repeated flank collapses and provide excellent opportunities to test the method. In addition, the failure at Volcan Casita, Nicaragua, that was triggered during Hurricane Mitch in 1998, allows us to examine the utility of reconnaissance stability assessments. Here, we briefly describe the 3-D slope stability method and present results from stability analyses of Mount St. Helens, Mount Rainier, and Casita volcanoes.

2. 3-D Slope Stability Analysis Method Volcano edifices exhibit a wide variety of slope failures, ranging from small rockfalls to massive collapses. Most large stratovolcano flank collapses extend deep into their 3 edifices, producing landslide volumes between 0.1 and 20 km [26]. Initial sliding for historical large collapses at Mount St. Helens, Bandai (Japan), and Bezymianny (Russia) volcanoes occurred along deep, arcuate failure surfaces [32]. Although many smaller failures move along rock discontinuities such as bedding or jointing surfaces, failure may be along arcuate surfaces if discontinuities are closely spaced [11]. Relatively small discontinuities commonly do not influence the shape of large failures. Here, we are interested in assessing massive flank collapse where internal structure may be poorly known. A spherical potential failure mass represents the simplest 3-D geometry unconstrained by internal structures or discontinuities. Thus, in our analyses we assume arcuate failure surfaces consisting of the sector of a spherical surface underlying the topography. We use a 3-D "method of columns" limit-equilibrium slope stability analysis to determine the stability of all parts of a landscape by computing the stability of many potential failures. Potential failures encompass a wide variety of depths and volumes throughout the materials underlying a DEM. The method uses a 3-D extension of the 2D Bishop's simplified limit-equilibrium analysis for rotational failure [1], and it can incorporate variable 3-D material properties, 3-D pore-fluid pressure distributions, and simplistic earthquake shaking effects. 3-D stability methods have been presented elsewhere [12, 13, 16] and 2-D stability methods have been used to search DEMs for unstable slopes [19]; our method combines these ideas. Our analysis is implemented in a computer code named SCOOPS [21].


Bishop's [1] limit-equilibrium analysis assumes that the average shear resistance, acting on a potential failure surface is given by the Coulomb-Terzaghi failure rule:

τ = c + (σn − u) tan φ



where c is cohesion, σn is the total normal stress acting on the failure surface, u is pore-fluid pressure on the failure surface, and φ is the angle of internal friction. To estimate the forces acting on each part of a potential failure surface, we assume a spherical failure surface with the solid failure domain divided into 3-D vertical columns. This division allows easy integration with a DEM where column size is controlled by the DEM spacing. Each spherical trial failure surface is defined by a center located above the DEM and a given radius. The potential failure mass is defined by a spherical trial surface intersecting the materials beneath the DEM and is restricted to the columns within the spherical surface. To obtain a factor of safety, F , for this trial surface, we first compute a vertical force equilibrium acting on the failure surface intersecting each column Horizontal force equilibrium between columns is not explicitly determined. Then, we compute the global moment equilibrium for all columns rotating about an axis through the center of the trial surface. Finally, we iteratively compute F by combining the moment equilibrium with the vertical force equilibrium. Details of this quasi-3-D method are presented in Reid et al. [21]. Instability is reflected in values of F < 1.0; low values of F indicate a propensity for collapse.

Figure 1. Digital elevation model of a volcano edifice showing one layer of search grid points. Complete search grid typically extends throughout many vertically stacked layers. Material from a trial failure has been removed from the topography to show potential failure surface.


We systematically search the DEM using a 3-D orthogonal grid of points located above the DEM (Fig. 1). Each point in this search grid represents the center of rotation of a set of spherical trial failure surfaces with different radii. In addition, because a 3-D potential failure mass can rotate about different axes, we can also compute F for different assigned slip directions of the same trial surface. To perform a thorough search we can specify any volume range to be searched, and we can manipulate the limits of the search grid, the search grid spacing, the spherical radius increment, and the directions of slip movement. After this search process is complete, every DEM grid point of interest will have been included in some potential failure masses. During the search, we retain the minimum F value and the volume associated with the trial surface having that minimum F that affect each DEM point. By aggregating the results, we can create maps portraying relative instability for all regions of an edifice ( F min at each DEM point), as well as provide estimates of potential failure volumes for these minimum factors of safety.

3. Assessing Flank Instability at Stratovolcanoes The 3-D slope stability analysis methods discussed above can be used to assess potential volcano flank instability. Here we present and discuss some preliminary slope stability analyses using our 3-D methods for three different stratovolcanoes (Fig. 2). We begin by examining the well-documented collapse of Mount St. Helens, USA, in 1980 to determine if our methods would predict where failure actually occurred. Then we assess potential instability at Mount Rainier, USA, an edifice where future flank collapse threatens heavily populated areas. Finally, we investigate the instability of Volcan Casita, Nicaragua, where a relatively modest flank failure in 1998 transformed into a debris flow and caused thousands of fatalities.

Figure 2. Map showing locations of the three stratovolcanoes for which we performed 3-D slope stability assessments. Base map from [5].


3.1. MOUNT ST. HELENS, WASHINGTON, USA The well documented, 1980 catastrophic debris avalanche at Mount St. Helens provides an opportunity to test our 3-D stability analysis, and in particular to investigate how well this method might have predicted pre-collapse edifice stability. In March 1980, the north flank of Mount St. Helens began moving northward at a rate of several meters per day, creating a conspicuous bulge about 1.5 by 2 km in area [17], apparently in response to shallow magma intrusion. Finally, on May 18 in association with a M 5.2 earthquake, the north flank failed rapidly and retrogressively in a series of three large blocks shown in Figure 4 (blocks I, II, and III) [10]. The entire rock slide, combining all three slide 3 blocks, had an estimated volume of 2.3 km , which was transformed into a 3 heterogeneous debris avalanche deposit of about 2.8 km [33]. We estimate that the 3 initial slide block I had a volume of about 0.8 km . Most of the slide surfaces for all three blocks were located in older dacitic dome rocks; hydrothermal alteration of edifice rock was minor and localized. Earlier 2-D slope stability analyses conducted by Voight et al. [33] indicated that both high pore-fluid pressures and earthquake shaking were required to trigger collapse of the deformed north flank. Our analyses are not intended to simulate the actual Mount St. Helens collapse on May 18. Instead we investigate, in retrospect, whether potential instability caused by gravity acting on topography (represented by pre-collapse DEMs) could have provided useful assessments of the location and volume of the initial May 18 failure. 3.1.1. Scenarios Analyzed We assessed potential instability by considering three cases: (1) an undeformed edifice based on 1979 topography, (2) a deformed edifice as existed 2 days before catastrophic collapse, and (3) a combination of deformed edifice, elevated pore pressures and earthquake shaking, as might have existed at collapse. To obtain edifice topography, we used undeformed topography from the 1979 U.S. Geological Survey Mount St. Helens 30-m DEM and constructed the deformed edifice topography by combining a 5-m DEM of the north flank bulge based on May 16, 1980, aerial photography with the 1979 30-m DEM for the remainder of the edifice. A shaded relief image of the undeformed topography is shown in Figure 3A. These DEMs were then resampled to a resolution of 100-m [21]; such a resolution is sufficient to analyze large failures. Other researchers have estimated rock strengths at Mount St. Helens [20, 33], and rock properties likely varied between and within geologic units in the edifice. However, because spatial variations in material properties at Mount St. Helens were poorly known prior to collapse, we assumed a strong, homogeneous, pre-collapse edifice with the following properties based on values presented by Voight et al. [33]: φ = 40q; c = 1000 kPa; γ = 3 24 kN/m , where γ is total (rock plus fluid) unit weight. We also examined the potentially destabilizing effects of pore-fluid pressure using a pore-pressure ratio, ru , of 0.3 and a horizontal pseudo-acceleration from earthquake shaking, k , of 0.2, values 6 suggested by Voight et al. [33]. For each scenario, we examined ~29 x 10 potential 3 failure masses, varying in volume from 0.1 to 3.5 km .


Figure 3. 3-D stability analyses of Mount St. Helens, USA. (A) Shaded relief image of the pre-collapse, undeformed edifice using a 30-m DEM. Contour interval is 400 m. (B) Relative stability for each DEM node using a dry, undeformed edifice (scenario 1). White lines are boundaries of least-stable areas. (C) Potential failure volumes associated with the least-stable potential failure surface for each DEM node in scenario 1. (D) Relative stability for each DEM node using a dry, deformed edifice (scenario 2). White line is boundary of least-stable area; dashed white line is outline of 1980 north-flank bulge.

3.1.2. Results Reid et al. [21] described a variety of stability analyses of the pre-collapse Mount St. Helens edifice. Here, we summarize three scenarios. For the first scenario, we assumed an undeformed edifice with dry, homogeneous rock. Figure 3B shows the resulting distribution of F min for each DEM grid point. The least-stable areas are predicted to occur on the NW flank of the volcano, a region with a large area of relatively steep slopes. However, the entire northern flank has F values within 5 percent of the global minimum ( F min = 2.23). Figure 3C shows the predicted failure volumes at each DEM grid point associated with the F min values shown in Figure 3B. Because of topographic 3 detail, two minimum volumes are predicted, one with a volume of 0.4 km , the other 3 with 0.8 km . In this scenario, topography alone was a reasonable, although inexact, predictor of instability. In the second scenario, we used the deformed edifice topography with a bulge on the north flank. Given this deformed edifice and dry homogeneous rocks, the least-stable area (with F min = 2.22) occurs on the north flank,


near the location of slide block I (Fig. 3D). For the third scenario, we used estimates of the conditions that existed at collapse, including deformed topography with a north flank bulge and combined pore-pressure and earthquake shaking effects as estimated by Voight et al. [33]. Here F min decreases to 1.05 on the N flank, with a predicted failure volume of about 1.1 km3. In this more complete case, we obtained good estimates of the actual initial failure block location and volume [21]. We also used the 3-D stability methods to examine potential retrogression of failure into the Mount St. Helens edifice. For this analysis, we assumed an initial failure mass approximately equal to slide block I in a deformed, fully saturated edifice (dashed line a in Fig. 4). This mass was then instantaneously removed from the edifice, creating steeper slopes in the scar and less confining stress on the internal pore-fluid pressures, resulting in undrained unloading. The rapid unloading is undrained because we assume that insufficient time elapsed to allow pore-fluid pressures to come to equilibrium with the new topography. Using a 3-D stability analysis and these modified edifice conditions, we searched for the least-stable surface with F min < 1, and then removed this mass. This process was repeated several times to simulate retrogression (lines b, c, and d in Fig. 4). Although this process promoted retrogression of failure into the edifice approximately equaling slide block II, the amount was less than the entire amount observed during the 1980 event (Fig. 4). This result suggests that other factors beyond modified topography and undrained unloading contributed to the actual retrogression.

Figure 4. 3-D retrogression of slope failure into the Mount St. Helens deformed edifice. Light gray lines show the boundaries of actual slide blocks I, II, and III from Glicken [10]. Dashed line (a) is assumed initial 3-D failure surface approximately equal to slide block I. Lines (b), (c), and (d) are resulting sequential retrogressions assuming undrained unloading.

3.2. MOUNT RAINIER, WASHINGTON, USA More than 55 Holocene volcanic debris flows have originated from Mount Rainier stratovolcano (Fig. 6A) in Washington State [2]. A few of these flows traveled > 70 km into the Puget Sound lowland, and similar future events would devastate now densely populated areas. Some of these debris flows initiated as massive flank collapses (large landslides), whereas others were likely triggered by pyroclastic flows entraining snow and ice, glacial outburst floods, or torrential rains. In contrast to Mount St. Helens, Mount Rainier has large areas of hydrothermal alteration that contain weak rocks. Clay minerals with a hydrothermal origin are abundant in some of the most widespread

452 3

debris-flow deposits, including the massive ~3.8 km Osceola Mudflow of 5600 yr ago 3 3 [3, 30], the ~0.2 km Round Pass mudflow of ~2600 yr ago, and the ~0.26 km Electron mudflow of ~500 yr ago [24]. The presence of these minerals in the deposits indicates that weakening of edifice rocks by acid sulfate-argillic hydrothermal alteration likely helped promote flank collapse [2, 24]. Alteration occurs predominantly in an east-west belt that bisects the edifice; it is most intense adjacent to radial dikes and associated open fractures and increases in extent and degree approaching the volcano's summit (Fig. 5A). The gigantic Osceola Mudflow [30] removed much of the summit and part of the east flank, but subsequent eruptions have rebuilt the cone. The smaller Round Pass and Electron mudflows originated from failure of hydrothermally altered rocks in the Sunset Amphitheater area on the upper west flank.

Figure 5. (A) Generalized surface distribution of basement, fresh, and hydrothermally volcanic rocks at Mount Rainier, USA. A-A’ is line of section in (B). (B) 3-D perspective with cut away along section A-A’ showing best-estimate interpretation of internal-edifice geology from detailed geologic mapping and geophysics.

3.2.1. Scenarios Analyzed For the Mount Rainier edifice, we present stability analyses for two scenarios: (1) homogeneous rock strength, to focus on the effects of topography, and (2) spatially variable rock strength where strength decreases as the degree of acid sulfate-argillic hydrothermal alteration increases. These and other analyses of the Mount Rainier edifice have been further discussed in Reid et al. [22]. Our analysis for the homogeneous Mount Rainier scenario uses properties typical of strong fractured rock ( φ 3 = 40q; c = 500 kPa; and γ = 24 kN/m ). However, acid sulfate-argillic alteration can lower rock shear strength [34] over wide areas [8, 9, 18], so alteration can be important for promoting and localizing large collapses. For the second scenario, we subdivided rock units into pre-Mount Rainier Tertiary basement rocks and three broad categories of edifice rocks distinguished by degree of acid sulfate-argillic alteration: fresh, lightly altered, and highly altered (Fig. 5A). Based on field and laboratory shear tests performed on small specimens by other


investigators [34, 35], we postulated that fresh volcanic rocks are relatively strong (as in the homogeneous case), more indurated basement rocks are slightly stronger than fresh 3 ( φ = 42q, c = 600 kPa, and γ = 24 kN/m ), lightly altered rocks are somewhat weaker 3 ( φ = 35q, c = 400 kPa, and γ = 23 kN/m ), and highly altered rocks are even weaker 3 ( φ = 28q, c = 300 kPa, and γ = 21 kN/m ). Although rock shear strength undoubtedly varies within these broad categories, these values provided a reasonable relative ranking 6 for use in hazard assessments. For both scenarios, we examined ~29 x 10 potential 3 failures encompassing a wide variety of depths and volumes between 0.1 and 3.5 km . Topography was obtained from a 10-m DEM resampled to 100-m.

Figure 6. 3-D stability analyses of Mount Rainier, USA. (A) Shaded relief image of the edifice. Contour interval is 400 m. (B) Relative stability for each DEM node using an edifice with homogeneous rock properties. W is the Willis Wall area (C) Relative stability for each DEM node using an edifice with areas of weakened, hydrothermally altered rock. S is the Sunset Amphitheater area.


We derived a well-constrained estimate of the extent of 3-D alteration, and thereby strength, by combining detailed surface geologic mapping [27] with subsurface geophysical imaging [6]. High-resolution airborne magnetic measurements described by Finn et al. [6] can reveal the subsurface distribution of alteration because alteration substantially reduces the strong magnetization of fresh volcanic rocks. Figure 5B shows the subsurface distribution of alteration in the 3-D edifice using this wellconstrained estimate. 3.2.2. Results For Mount Rainier, our analysis using homogeneous rocks predicted a least-stable region in the large, steep Willis Wall (Fig. 6B), a region where few large landslides have originated in the past. Here, the use of topography and homogeneous rock properties does not identify regions subject to past large collapses, and is likely a poor predictor of future events. Using the 3-D distribution of spatially variable rock properties, our analysis predicted that the least-stable part of the volcano is its upper west flank (Fig. 6C), in the basin of Sunset Amphitheater where intensely altered rocks are widely exposed [4, 7, 8]. This result is consistent with Holocene debris-flow history. Overall, 3 our analyses reveal that large flank collapse (> 0.1 km ) is promoted by voluminous, weak, hydrothermally altered rock situated high on steep slopes [22]. At this volcano, incorporating variable 3-D shear strength was essential to provide a reasonable indicator of instability. 3.3. VOLCAN CASITA, NICARAGUA At many other stratovolcanoes worldwide there is a pressing need for slope stability evaluations, but time and costs preclude developing the comprehensive geologic information available for Mount Rainier. For example, at Volcan Casita in Nicaragua, a landslide was triggered by torrential rainfall from Hurricane Mitch in October 1998. 6 3 This 1.6 x 10 m failure, much smaller than the failures discussed above, transformed into a mobile debris flow that completely destroyed two villages and killed more than 2500 people [14, 15, 23]. The initial failure appears to have occurred as two or more retrogressive failures high on the edifice in fractured lavas and volcaniclastic materials, some of which were hydrothermally altered [29]. The Casita edifice lacks detailed geologic mapping; however, it has numerous modern fumarolic areas on its surface (Fig. 7A). These areas, identified by several investigators [28, 31], are likely to be hydrothermally altered at depth and may include rocks with significantly reduced strengths. In addition to the large failure, numerous smaller, shallower failures were triggered during Hurricane Mitch, primarily in hydrothermally altered rocks. 3.3.1. Scenarios Analyzed Although the initial large failure at Casita was smaller than those we examined at Mounts St. Helens and Rainier, we investigated the utility of our method for reconnaissance stability assessments of smaller potential failures. As with Mount Rainier, we present stability analyses for two scenarios: (1) an edifice with uniform, moderately altered rocks, to focus on the effects of topography, and (2) an edifice with an extensive region of highly altered rocks delineated by mapped fumarolic areas. In both cases we assume that alteration extends from the surface to depths below potential


failure surfaces. Here we assume rock properties based on values derived from field and laboratory shear tests on other volcanic rocks [34, 35]. Moderately altered rocks, ( φ = 3 35q, c = 120 kPa, and γ = 23 kN/m ), represent the entire Casita edifice in scenario 1 and the bulk of the edifice in scenario 2. In scenario 2, the fumarolic areas are assumed 3 to be more highly altered and weaker ( φ = 30q, c = 85 kPa, and γ = 20 kN/m ). Our 6 analyses focused on smaller potential landslides ranging in volume from 0.5 to 5.0 x 10 3 m , and we used a pre-failure 10-m DEM.

Figure 7. 3-D stability analyses of Volcan Casita, Nicaragua. (A) Shaded relief image of the pre-failure edifice showing previously mapped areas of fumarolic activity, inferred alteration zone, and location of the 1998 source scar (denoted by arrow). Contour interval is 100 m. (B) Relative stability for each DEM node using an edifice with homogeneous rock properties. (C) Relative stability for each DEM node using an edifice with areas of weakened, hydrothermally altered rock.



With uniform strength (scenario 1), two areas on the west inside crater wall and one area on the upper southwest flank have reduced stability (Fig. 7B). These areas are topographically steep. In this scenario, the 1998 source scar is not among the predicted


least-stable areas. Using a distribution of weaker rocks based on mapped fumarolic areas (scenario 2), stability is reduced near the steeper, more-altered zones. In addition to the areas defined in scenario 1, areas of lower stability occur on the north, northeast, and southeast flanks (Fig. 7C). The source scar for the large failure triggered by Hurricane Mitch is located on the southeast flank, an area our analysis predicted as having lower stability. In addition, many of the smaller failures that occurred during Hurricane Mitch are located in other areas predicted to have lower stability. At Volcan Casita, reconnaissance stability analysis including weaker, altered rocks was able to provide a reasonable indication of the unstable areas that were triggered by Hurricane Mitch.

4. Conclusions Volcano flank instability is influenced by a wide variety of steady and transient factors. Here, we have focused on 3-D slope stability analyses using the relatively static factors of topography and strength variations. Our results demonstrate that 3-D analysis can predict reasonable locations and volumes of massive gravitational failures at a variety of stratovolcano edifices. However, the amount of information needed for adequate analyses varies among the stratovolcanoes we examined. For the relatively uniform Mount St. Helens edifice, deformed topography alone is a good predictor of the location of the initial 1980 collapse. For volcanoes like Mount Rainier, with large regions of both relatively strong and weak rocks, collapse hazards vary substantially from sector to sector depending on the distribution and intensity of alteration and the local relief. Our results from Volcan Casita indicate that even reconnaissance 3-D slope stability analysis, in this case incorporating variations in strength, could be useful. The successes at St. Helens, Rainier, and Casita volcanoes described above suggest that 3-D analyses of potential rotational failures may be useful for preliminary assessments of flank instability at other stratovolcanoes. Moreover, the method can provide estimates of both potential failure locations and volumes. These quantities are crucial for hazard assessments and they serve as initial conditions for models of debris avalanche or flow runout. However, given the geologic complexities present at many volcanoes, simplistic results cannot be used blindly. Such preliminary analyses can aid, but not necessarily replace, site-specific investigations of potential instability. Although many rock failures on non-volcanic mountains are relatively planar or wedge shaped, some massive failures have arcuate failure surfaces. Our 3-D slope stability methods can be used to assess the potential for massive, rotational rock failures at other mountain massifs.

Acknowledgements Sarah Christian contributed to the stability analysis of Mount St. Helens; Tom Sisson helped analyze Mount Rainier. We thank Richard Iverson and Robert Schuster for helpful reviews.


References 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. 16. 17. 18. 19. 20. 21. 22. 23.

24. 25. 26.

Bishop, A.W. (1955) The use of slip circles in the stability analysis of slopes, Geotechnique 5, 7-17. Crandell, D.R. (1971) Postglacial lahars from Mount Rainier Volcano, Washington, U.S. Geol. Surv. Prof. Paper 677. Crandell, D.R. and Waldron, H.H. (1956) A recent volcanic mudflow of exceptional dimensions from Mt. Rainier, Washington, Am. J. Sci. 254, 349-362. Crowley, J.K. and Zimbelman, D.R. (1997) Mapping hydrothermally altered rocks on Mount Rainier, Washington, with airborne visible/infrared imaging spectrometer (AVIRIS) data, Geology 25, 559-562. ESRI (2000) Data and Maps CD, Environmental Systems Research Institute, Redlands, California. Finn, C.A., Sisson, T.W., and Deszcz-Pan, M. (2001) Aerogeophysical measurements of collapseprone hydrothermally altered zones at Mount Rainier volcano, Nature 409, 600-603. Fiske, R.S., Hopson, C.A., and Waters, A.C. (1963) Geology of Mount Rainier National Park, Washington, U.S. Geol. Surv. Prof. Paper 444. Frank, D. (1995) Surficial extent and conceptual model of hydrothermal system at Mount Rainier, J. Volc. Geotherm. Res. 65, 51-80. Frank, D.G. (1985) Hydrothermal processes at Mount Rainier, Washington, University of Washington, Seattle. p. 195. Glicken, H. (1996) Rockslide-debris avalanche of May 18, 1980, Mount St. Helens Volcano, Washington, U.S. Geol. Surv. Open-File Rep. 96-0677. Hoek, E. and Bray, J.W. (1981) Rock Slope Engineering, 3rd. ed, Institute of Mining and Metallurgy, London. Hungr, O. (1987) An extension of Bishop's simplified method of slope stability analysis to three dimensions, Geotechnique 37, 113-117. Hungr, O., Salgado, F.M., and Byrne, P.M. (1989) Evaluation of a three-dimensional method of slope stability analysis, Can. Geotech. J. 26, 679-686. Kerle, N. (2002) Volume estimation of the 1998 flank collapse at Casita Volcano, Nicaragua: A comparison of photogrammetric and conventional techniques, Earth Surf. Processes and Landforms 27, 759-772. Kerle, N. and van Wyk de Vries, B. (2001) The 1998 debris avalanche at Casita Volcano, Nicaragua investigation of structural deformation as the cause of slope instability using remote sensing, J. Volc. Geotherm. Res. 105, 49-63. Lam, L. and Fredlund, D.G. (1993) A general limit-equilibrium model for three-dimensional slope stability analysis, Can. Geotech. J. 30, 905-919. Lipman, P.W., Moore, J.C., and Swanson, D.A. (1981) Bulging of the north flank before the May 18 eruption: Geodetic data, U.S. Geol. Surv. Prof. Paper 1250, 143-156. Lopez, D.L. and Williams, S.N. (1993) Catastrophic volcanic collapse; relation to hydrothermal processes, Science 260, 1794-1796. Miller, D.J. (1995) Coupling GIS with physical models to assess deep-seated landslide hazards, Env. Engrg. Geoscience 1, 263-276. Paul, A., Gratier, J.P., and Boudon, J. (1987) A numerical model for simulating deformation of Mount St. Helens volcano, J. Geophys. Res. 92, 10,299-10,312. Reid, M.E., Christian, S.B., and Brien, D.L. (2000) Gravitational stability of three-dimensional stratovolcano edifices, J. Geophys. Res. 105, 6043-6056. Reid, M.E., Sisson, T.W., and Brien, D.L. (2001) Volcano collapse promoted by hydrothermal alteration and edifice shape, Mount Rainier, Washington, Geology 29, 779-782. Scott, K.M. (2000) Precipitation-triggered debris-flow at Casita Volcano, Nicaragua; implications for mitigation strategies in volcanic and tectonically active steeplands, in G.F. Wieczorek and N.D. Nasser (eds.), Debris-flow hazards mitigation; mechanics, prediction, and assessment, Proceedings of the Second International Conference on Debris-Flow Hazards Mitigation, A. A. Balkema, Rotterdam, Netherlands, v. 2, pp. 3-13. Scott, K.M., Vallance, J.W., and Pringle, P.T. (1995) Sedimentology, behavior, and hazards of debris flows at Mount Rainier, Washington, U.S. Geol. Surv. Prof. Paper 1547. Siebert, L. (1984) Large volcanic debris avalanches: characteristics of source areas, deposits, and associated eruptions, J. Volc. Geotherm. Res. 22, 163-197. Siebert, L., Glicken, H., and Ui, T. (1987) Volcanic hazards from Bezymianny- and Bandai-type eruptions, Bull. Volc. 49, 435-459.

458 27. 28. 29. 30. 31. 32. 33. 34. 35.

Sisson, T.W., Vallance, J.W., and Pringle, P.T. (2001) Progress made in understanding Mount Rainier's hazards, Eos, Trans. Am. Geophysical Union 82, 113-120. Texas Instruments Inc. (1970) Reporte final, Proyecto de Recursos Geotermicos - Etapa Una. Parte 2, Geologia Regional, Manifestaciones Termales de Nicaragua Occidental. Informe elaborado para el Gobierno de Nicaragua, Ministerio de Economia, Industria y Comercio. Vallance, J.W., Schilling, S.P., Devoli, G., Reid, M.E., Howell, M.M., and Brien, D.L. (2001) Lahar Hazards at Casita and San Cristóbal Volcanoes, Nicaragua, U.S. Geol. Surv. Open-File Rep. 01-468. Vallance, J.W. and Scott, K.M. (1997) The Osceola Mudflow from Mount Rainier; sedimentology and hazard implications of a huge clay-rich debris flow, Geol. Soc. Am. Bull. 109, 143-163. van Wyk de Vries, B., Kerle, N., and Petley, D. (2000) Sector collapse forming at Casita volcano, Nicaragua, Geology 28, 167-170. Voight, B. and Elsworth, D. (1997) Failure of volcano slopes, Geotechnique 47, 1-31. Voight, B., Janda, R.J., Glicken, H., and Douglass, P.M. (1983) Nature and mechanics of the Mount St. Helens rockslide-avalanche of 18 May 1980, Geotechnique 33, 243-273. Watters, R.J. and Delahaut, W.D. (1995) Effect of argillic alteration on rock mass stability, in W.C. Haneburg and S.A. Anderson (eds.), Clay and Shale Slope Instability, Geol. Soc. Am. Reviews in Engineering Geology, v. 10, pp. 139-150. Watters, R.J., Zimbelman, D.R., Bowman, S.D., and Crowley, J.K. (2000) Rock mass strength assessment and significance to edifice stability, Mount Rainier and Mount Hood, Cascade Range Volcanoes, Pure Appl. Geophys. 157, 957-976.

CATASTROPHIC VOLCANIC LANDSLIDES: THE LA OROTAVA EVENTS ON TENERIFE, CANARY ISLANDS M. HÜRLIMANN1 and A. LEDESMA Department of Geotechnical Engineering and Geosciences, Technical University of Catalonia (UPC), Jordi Girona 1-3, 08034 Barcelona. Spain

Abstract Giant volcanic landslides are one of the most hazardous geological processes. On Tenerife, seven large landslides affected the subaerial and submarine morphology during the last ~6 Ma. A comprehensive analysis of the La Orotava events was carried out including site investigation, laboratory tests and stability analyses. The results revealed that the stability of the volcano can be strongly reduced by geologic, morphologic, climatic and volcanological factors. Widespread residual soils might act as potential slip surfaces, while deep erosive canyons probably evolve into the lateral limits of the failures. A high coastal cliff, humid climate and especially persistent dike intrusion contributed to critical stability conditions. Finally, seismic ground acceleration generated by a strong and adjacent tremor triggered the landslides. On Tenerife, a temporal coincidence of large landslides with caldera collapse events suggests that strong earthquakes associated with caldera collapses may have triggered the failures.

1. Introduction Large volcanic landslides can exceed several cubic kilometres in volume and generally surpass the volumes of non-volcanic events by two or three orders of magnitude (Figure 1). Recent studies indicate that catastrophic failures have occurred world-wide at several hundred volcanoes [27, 28]. Siebert [28] calculated that large volcanic landslides have taken place globally by an average of four times a century during the past 500 years, which was thought to be an underestimated frequency considering the large number of failures during the 20th century [22]. Moreover, it is assumed that 75 % of the Andean volcanoes with volcano heights in excess of 2,500 m have experienced 1

E-mail of corresponding author: [email protected]

459 S.G. Evans et al. (eds.), Landslides from Massive Rock Slope Failure, 459–472. © 2006 Springer. Printed in the Netherlands.

460 one or more failures [10], and more than 100 events have been observed at Japanese Quaternary volcanoes [16]. This paper is divided into two parts. After a short introduction, we will summarise the different causes of catastrophic volcanic landslides distinguishing between nonvolcanic and volcanic factors. The objective of this part is to point out the complexity of processes that finally provoke catastrophic volcanic landslides. In the second part of the paper, we will present the results obtained from the geotechnical analysis of the La Orotava events on Tenerife which includes a site investigation, laboratory tests and stability analyses. The objective of this second part is to improve our insight into the processes that generated the catastrophic landslides in La Orotava. When studying volcanic landslides, a multidisciplinary approach is required. On one hand, a conventional geomechanical analysis can be performed, considering typical aspects of mass movements like slip surface geometry, geomorphologic features, groundwater and climate conditions, material properties, etc. On the other hand, an analysis of the dynamics of the volcano can be carried out. In many cases, a combination of both aspects is necessary to understand the mechanisms that produced the landslide.






Canary Islands

Soc Sha

Non-volcanic landslide Continental volcanic landslides Landslides on volcanic islands 0.01 0.01








Volume [km ] Figure 1. Mobility versus volume of non-volcanic and volcanic landslides. Mobility is the ratio between vertical drop H and runout distance L. Some examples form the Hawaiian archipelago and the Canary Islands are indicated. Hel: Mt. St. Helens, Sha: Shasta, Soc: Socompa, Pop: Popocatepetl. Data after [12, 23, 27].

461 2.

Causes of Large Volcanic Landslides

2.1. GENERAL ASPECTS Generally, large landslides originate either by increasing the destabilising forces (shear stress) or by reducing the strength of the materials involved, or both. In contrast to non-volcanic landslides, the causes of volcanic failures are more complex due to the influence of volcanism. The volcanic activity produces many different processes destabilising directly or indirectly the natural slopes. The complexity of factors influencing on the mechanical stability of volcanic slopes can be seen in Table 1, where some of the non-volcanic and volcanic agents are listed. Table 1. Causes of catastrophic volcanic landslides distinguishing between volcanic (v) and non-volcanic (nv) factors. Adapted from [32]. Increase of shear stress

Reduction of shear strength

Removal of lateral or underlying support: - coastal or fluvial erosion - phreatic explosion near base of slope Static loading: - sedimentation of volcanic material - seepage pressure - magma pressure Dynamic loading: - tectonic and volcanic earthquakes Increase of surface slope: - magma intrusion related deformation

Physicochemical factors: - Hydrothermal alteration (v) - Weathering (nv) Pore fluid pressure enhancement: - Pore-pressure changes in aquifers adjacent to magma intrusion or due to hydrothermal processes (v) - Sea level changes (nv) - Enhanced glacier/snow melting due to increased geothermal flux (v) Changes in structure: - Disturbance, remoulding (nv) - Collapse of structure in tephra deposits or residual soils (nv) - Deep seated fracturing associated with magma intrusion, geothermalprocesses (v)

(nv) (v) (v) (nv) (v) (nv/v) (v)

2.2. NON-VOLCANIC FACTORS There are various non-volcanic factors that can affect the stability of volcano flanks including geologic factors, morphologic factors, climatic factors, sea level changes or tectonic seismicity. The geologic materials composing volcanoes influence the stability of the slopes. However, the mechanical properties of volcanic materials are not well known and few geotechnical studies on volcanic deposits exist [14, 34]. Nevertheless, the role of potential weak layers destabilising volcano flanks have been suggested by several authors. Weak pyroclastic rocks were detected at Bandai and Stromboli within the volcano slopes [27] or at the contact with the basement [30]. Other weak materials include hyaloclastite deposits as found at La Réunion, La Palma and Kilauea volcano [6, 7] and a ductile layer of olivine cumulates observed on Hawaiian volcanoes [4]. In

462 addition, water saturated pyroclastic deposits and fractured lava flows are soil and rock types associated with a high risk of landslide initiation [8]. Slope angle has traditionally been assumed to be one of the most important morphologic features which influences the stability of volcano flanks. The slope angles of 55 Quaternary volcanoes, which have undergone major slope failure, indicated that large landslides appear most likely when the slope angle are between 28° and 34º [27]. But mass movements on volcano flanks with an inclination of about 10° have also been observed [17]. The relationship between the climate and the groundwater system is a fundamental consideration in the stability analysis of natural slopes. There is increasing evidence that water plays an important role in the stability conditions of volcanic edifices. Pore water pressure controls the strength of the material and thus slope stability mechanics. The significant effect of pore fluid pressure in the initiation of landslides has mainly been constrained for non-volcanic events, while only theoretical models have been applied to those associated with volcanic events [5, 31, 32]. Many of the largest landslide events are located on volcanic islands (Figure 1) and the destabilising effect of sea level changes on the stability of the volcanic edifice has been proposed by different authors [2, 21]. The principal origins of sea level changes are twofold: 1) the global climatic variations with the associated increase of the ice sheets; and 2) the vertical movement of the volcanic island due to magma loading and unloading. The effects of sea level changes on the stability of volcano flanks include undermining due to coastal erosion, increase of pore water pressure or even the influence on volcanic activity [21]. Seismicity in volcanic terrains can have tectonic, volcanic or volcano-tectonic origins. In contrast to purely tectonic dynamics, where all seismic failures can be ascribed to double-couple systems of shear forces acting on two orthogonal planes, the seismic source at active volcanoes may be highly complex, since it often involves interactions between gas and liquid or liquid and solid [9]. However, while the tectonic shocks can be characterised by shear failures, volcanic earthquakes are generally tensional failures. 2.3. VOLCANIC FACTORS Since the Mt. St. Helens event in 1980 the influence of volcanic activity on the stability of volcano flanks has been intensively studied. Many potential theories have been proposed on the mechanisms provoking a volcano failure, but the relationship between landslide triggers and volcanic activity has not yet been resolved. The following volcanic processes will be summarised below: magma intrusion, hydrothermal alteration, volcanic seismicity and caldera collapse episodes. Most volcanic eruptions are supplied with magma through fractures. Magma filled fractures are referred to as dikes when they are sub-vertical and where repeated dike intrusion occurs, the process is commonly called ‘rifting’. From a mechanical point of

463 view, the intrusion of magma influences in many forms on the stability of volcanoes: 1) horizontal stress perpendicular to the magma intrusion, 2) mechanically and thermally generated pore fluid pressure, 3) hydrothermal alteration, and 4) volcanic tremors. Point 2 includes the pressurisation of pore fluids due to magmatic intrusion [31, 32]. Worldwide data indicate that the axes of the amphitheatres often show a preferred orientation normal to the dominant direction of the dike intrusion or rifting [27]. Hydrothermal alteration of the volcanic material can reduce the shear strength and thus could play an important role in the volcano stability [25]. Hot water and other fluids heated by magma change the material properties of the surrounding rocks and are able to weaken even hard, compact lava into unsound material. On the other hand, pore fluid pressures generated by hydrothermal systems may also influence the mechanical stability of volcanoes [32]. Two main sources of seismic tremors in the vicinity of volcanoes can be distinguished: 1) those generated by the movement of magma or by formation of cracks through which magma can move, and those resulting from gas explosions within a conduit; 2) seismic tremors that result from readjustments of a volcanic edifice following eruption or movement of magma [11]. Earthquakes directly associated with movement or eruption of magma seldom exceed a magnitude of about 5.0 [24], while seismic shocks which have volcano-tectonic origins can reach high magnitudes up to about 7.0 [1]. Caldera collapse episodes are cataclysmic infrequent events, which form large subcircular depressions. The destabilising effects during a caldera collapse episode are complex and include many different processes, but the most important are the seismic tremors caused by the shear stresses on the ring faults that generate the vertical collapse of the roof into the decompressed magma chamber [15]. 3.

The La Orotava Events

3.1. GEOLOGICAL SETTINGS In the Canary Islands, more than 20 catastrophic volcanic landslides have been detected during the last decades [18, 35]. Seven of these events are situated on Tenerife Island, and we selected La Orotava valley at the north flank of Tenerife as the main test site for this study due to the large amount of data available (Figure 2). Tenerife is the largest of the Canary Islands and has a total height of about 8 km starting at the sea floor at 4 km b.s.l. and rising up to 3718 m a.s.l.


Figure 2. Large landslide events around the Canary Islands. Solid lines show main structural axis on the islands. LO for La Orotava, Gü for Güimar, OPS for old post-shield, A for Anaga, ED for East Dorsal, I for Icod and T for Teno Contour lines in km below sea level. Modified from [2, 3].

The evolution of Tenerife includes three main stages: 1) the formation of a basaltic shield with ages between 8 and 4.5 Ma, 2) the construction of the central edifice between ~3 and 0.18 Ma; and, 3) the formation of the current active volcanic complex. The assumed onshore limits of the landslide amphitheatres are illustrated in Figure 2. Moreover, the two most important volcanic features are indicated: the Las Cañadas caldera that has been formed by diverse caldera collapses [20], and the two structural axes that can be seen by the orientation of the eruptive vents. Two of the seven landslides (Anaga and Teno) occurred during the evolution of the basaltic shield, while the other five ones (Icod, La Orotava, Güimar, East Dorsal and “old post-shield”) took place during the second phase of the evolution of Tenerife [2]. An analysis of the landslide ages revealed that a relationship between the slope failures and the caldera collapse episodes may exist [19]. Moreover, Hürlimann et al. [15]

465 recently presented a mechanical explanation that links the initiation of the volcanic landslides with the seismic acceleration related to the caldera collapse. 3.2. SITE INVESTIGATION As stated above La Orotava valley was selected for study (Figure 3). It has a width of almost 10 km, a length between 9 km and 14 km and lateral scarps with heights of up to 500 m. Several techniques were applied during the site investigation. Firstly, all data available were incorporated into a geographic information system. A digital elevation model was studied in order to obtain information on the geomorphology of the valley and of the island. A comprehensive study of aerial photographs and various field surveys were carried out. Several samples of volcanic material were collected during the field trips to be analysed afterwards in the laboratory. The site investigation led to an improved understanding of the failure mechanisms of the volcanic landslides. The results obtained refer to geologic, morphologic, volcanic and climatic features. The geologic and volcanic features include the two structural axes and the Las Cañadas caldera. The location of these two features, adjacent to the main scarps of the landslides, suggests an influence on the failures (Figure 3). In addition, various morphological features were observed that may have influenced the stability of the volcano flanks. Firstly, deep erosive canyons were detected and may have been the lateral limits of the landslides. Furthermore, high coastal cliffs were observed, which may have reduced locally the slope stability and may have developed finally into the seaside limit of the landslides (see “pre-slide topography” in Figure 4). The field surveys revealed that widespread residual soils (also called paleosols) could have been potential slip surfaces for large landslides. Such residual soils are common deposits on the island and represent the only material in the volcano slopes with planar surfaces, weak properties and wide extent. 3.3. LABORATORY TESTS A volcanic residual soil from Tenerife was analysed. Such red coloured soils are generally located at the top of pyroclastic deposits produced by explosive eruptions of phonolitic magmas.


UTM projection


La Orotava valley 3140 Latitude [km]



rs a


id ge





Las Cañadas caldera 3130 00 25



Longitude [km]



Figure 3. La Orotava amphitheatre on Tenerife. Thick solid line indicates the onshore limits of the catastrophic volcanic landslides. Thick dashed lines show the orientation of the structural axis “Dorsal Ridge” and the extension of the Las Cañadas caldera.


Dorsal Ridge

actual topography pre-slide topography model topography model failure surface


2000 1000

sea level

0 -1000 -2000 0







Figure 4. Onshore-offshore profiles of La Orotava landslide and geometry of the model applied during the limit equilibrium analyses for the case of 25 km of model length. Geometric characteristics of model boundaries are described in the text.

467 Here, the most important results of the laboratory tests will be presented, whereas a general description of the residual soil can be found in Hürlimann et al. [14]. The geotechnical tests revealed two types of findings: one refers to the relation between the strength of the soil and the magnitude of the stress applied; and the other concerns the effect of undrained loading. The relation between the strength of the soil and the magnitude of the stress applied was observed during direct shear tests and also during triaxial tests. These tests exposed two very distinct types of mechanical behaviour of the soil. During direct shearing the behaviour depends on the normal stress applied to the specimen and during the triaxial tests, the behaviour depends on the effective confining stress. Figure 5 illustrates the peak shear strengths obtained from direct shear tests applying different normal stresses. Therefore, the Mohr-Coulomb failure envelopes and the strength parameters of the soil could be calculated. The differences of the two behaviour types can be easily seen by the friction angles. A peak friction angle of about 45° has been calculated for low normal stresses and one of less than 30° for high normal stresses (Figure 5). A similar effect could be observed for the results of the triaxial tests. The cause of this particular phenomenon can be found in the structure of the soil. The soil is characterised by a cementation or bonding that unifies the soil particles. This bonding breaks when a large stress is applied and the behaviour of the soil completely changes, decreasing its strength.


shear stress (kPa)







0 0





normal stress (kPa)

Figure 5. Mohr-Coulomb failure planes of the residual soil obtained from peak shear strength values. The calculated friction angles distinguish between low normal stress (dots) and high normal stress (triangles).

468 The second important finding of the laboratory tests is the decrease of soil strength during undrained loading. This behaviour was observed during undrained triaxial tests and indicates the importance of the type of loading [13]. Generally, undrained loading conditions occur if two facts coincide: the soil must be saturated and the process that applies the loading must be rapid. Earthquakes that are common processes in volcanic areas can generate a fast loading and precipitation can saturate the residual soils. In conclusion, the laboratory tests show that two factors can strongly reduce the soil strength and thus influence the stability of the volcano slope. The first factor refers to the magnitude of the initial normal stresses applied to the soil. If these stresses are high, the soil strength strongly decreases. This may have occurred during the large-scale failures on Tenerife, since the slip surface is located at a depth of some hundreds of meters generating initial normal stresses higher than 104 kPa (Figure 5). The second factor refers to the type of loading conditions. If the loading is undrained, the pore fluid pressure increases reducing the soil strength. Earthquakes together with saturated soils can thus generate undrained loading. 3.4. STABILITY ANALYSIS Two types of approaches were applied: firstly, we carried out a comprehensive stability analysis using limit equilibrium method (LEM) and secondly, the stress field of the volcano flank was simulated by finite element method (FEM). At the beginning, we performed a sensitivity analysis by LEM including various parameters such as the material strength, the water table and the slope inclination, amongst others. In the second part, we focussed on two processes that may have influenced the slope stability of the volcano flank. On one side, seismic acceleration, a, due to an earthquake was incorporated into the models. On the other side, the effects of the horizontal stress, Vh, caused by persistent dike intrusion along a structural axis were simulated. The magnitude of Vh is defined as the sum of the magmastatic pressure and the magma overpressure [31]. The 2-dimensional schematic LEM-model of the simplified pre-slide volcano flank was established using the available data on morphology, geology, volcanology and hydrogeology [20, 26, 29] as well as the results presented in the previous Sections. Finally, we approximated the geometry of the La Orotava valley by a wedge, 25 km in length at its base and a constant slope angle of 11º (Figure 4). Smaller lengths were also analysed keeping constant the slope angle. The inclination of the water table is given as 10º and the failure surface was approximated by a straight line with a slope angle of 9º. Homogeneous material properties for fractured lava were assumed as proposed in Voight and Elsworth [30]. The density of the fractured lava and of the intruding magma is given as Ul = Um = 2700 kg m-3, the friction angle as I = 35º and the cohesion, c = 0 kPa. The results obtained from the sensitivity analysis using LEM indicate that conventional factors such as a fully saturated slope or weak material properties cannot initiate the failure. Therefore, additional mechanisms are necessary in order to trigger

469 such large-scale landslides. So, the Factor of Safety (FS) was calculated incorporating a seismic acceleration, a, into the model. The results of these simulations are illustrated in Figure 6a specifying different water table inclinations. The graphs indicate that the FS values decrease sharply with increasing average ground acceleration. Conversely, the angle of the water table plays a minor role. Additionally, an external horizontal stress simulating the effect of dike intrusion was applied at the left-hand limit of the model (Figure 4), considering different model lengths. The results are illustrated in Figure 6b and show that the slope stability mainly depends on the model length. On the other hand, the effect of the magma overpressure, Pmo , is less significant. The FS-values decrease slightly for a constant model length of 25 km and a water table inclination of 10° applying magma overpressures up to 10 MPa. However, a Pmo-value of 10 MPa produces unstable conditions for a model length of about 3 km. a




Pmo 2.5 MPa 5.0 MPa 7.5 MPa 10 MPa

Safety Factor, FS

E 9° 10° 11°








0.0 0.0


0.2 a [g]








model length [km]

Figure 6. Influence of volcanic and non-volcanic factors on slope stability for cohesionless volcano flanks with an angle of friction of 35º. (a) Seismic acceleration, a, and water table inclinations, Eѽ and (b) model length and magma overpressure, Pmo, due to dike intrusion.

In the second part of the stability analysis, we carried out 2D and 3D FEMcalculations. In the 3D model, we analysed the influence of the morphologic features described in the site investigation (deep canyons and high coastal cliff). A horizontal stress due to persistent dike intrusion along the structural axis (“Dorsal Ridge” in Figure 3) was applied at the rear of the 3D model. The material has been assumed as being linear elastic and isotropic. The elastic parameters (Young’s modulus, E = 7.5x103 MPa, and Poisson’s ratio, Q = 0.25) were considered homogeneous and based on [32-34]. The value adopted for the magma/rock density was the same as in the LEM model. The results of the FEM modelling indicate the importance of the two morphologic features. On one side, the coastal cliff produces an accumulation of unusually high shear stresses along the shoreline where the marine erosion formed a steep cliff. On the other side, the deep canyons generate a strong increase of the shear stresses near these topographic features whereas highest values are situated at the base of the canyons

470 indicating, therefore, the importance of such morphologic units as lateral boundaries of landslides. 4.


The site investigation has revealed the presence of residual soils that may have formed the failure surface of the landslides. Additionally, morphologic features such as erosive canyons or coastal cliffs and climatic factors were considered. A temporal relationship between large landslides and caldera collapse episodes was indicated based on geochronological data. In the laboratory, the mechanical behaviour of the residual soil has been analysed. The results of the tests indicate that the soil differs from other volcanic materials by its particular characteristics. The shear strength strongly decreases applying high normal stresses. Furthermore, pore pressure increases significantly during undrained loading. This fact is fundamental in relation to the stability of volcanic slopes since it reduces drastically the strength of the soil. Earthquakes, a common process in volcanic areas, and saturated conditions can generate high excessive pore pressures, which suggest the importance of the regional climate and seismicity characteristics on the initiation of failure. The stability analysis considered two different mechanisms: seismic acceleration and horizontal stress due to dike intrusion. The results indicate that the seismic acceleration reduces significantly the slope stability, whereas the effect of dike intrusion is minor. The horizontal stresses due to dike intrusion may act as a preparing factor destabilising the slope, but are not able to produce the final trigger. The 3D simulations show the destabilising influence of the erosive canyons at lateral margins. Consequently, we propose the following scenario for the evolution of the La Orotava amphitheatre: different factors such as deep canyons, high coastal cliffs, the humid climate and the existence of weak layers changed the mechanical equilibrium of the volcano flank. Persistent dike intrusion decreased the stability to critical conditions and finally, seismic ground motion generated by a caldera collapse event triggered the large volcanic landslides. The temporal relationship between caldera collapse episodes and giant volcano failures on Tenerife support this hypothesis. The results obtained from this study may be applied to other large-scale volcanic failures since many of the analysed factors also exist at other volcanoes. Finally, the laboratory results indicate that the residual soil analysed is a good candidate for generating the failure surface of the large landslides as well as for reducing the stability of the volcano slopes on Tenerife.

471 References 1. 2. 3. 4. 5.


7. 8. 9. 10. 11. 12. 13. 14.

15. 16. 17. 18. 19. 20. 21.



Abe, K. (1992) Seismicity of the caldera-making eruption of Mount Katmai, Alaska in 1912, Bull. Seism. Soc. Am. 82, 175-191. Ablay, G. J., and Hürlimann, M. (2000) Evolution of the north flank of Tenerife by recurrent giant landslides, J. Volc. Geotherm. Res. 103, 135-159. Carracedo, J. C. (1999) Growth, structure, instability and collapse of Canarian volcanoes and comparisons with Hawaiian volcanoes, J. Volc. Geotherm. Res. 94, 1-19. Clague, D. A., and Denlinger, R. P. (1994) Role of olivine cumulates in destabilizing the flanks of Hawaiian volcanoes, Bull. Volcanol. 56, 425-434. Day, S. J. (1996) Hydrothermal pore fluid pressure and the stability of porous, permeable volcanoes, in W. J. McGuire, A. P. Jones, and J. Neuberg (Eds.) Volcano Instability on the Earth and Other Planets, Geological Society Special Publication, London, pp. 77-93. Day, S. J., Carracedo, J. C., Guillou, H., and Gravestock, P. (1999) Recent structural evolution of the Cumbre Vieja volcano, La Palma, Canary Islands: volcanic rift zone reconfiguration as a precursor to volcano flank instability?, J. Volc. Geotherm. Res. 94, 135-167. Duffield, W. A., Stieltjes, L., and Varet, J. (1982) Huge landslide blocks in the growth of Piton de la Fournaise, La Réunion, and Kilauea, Hawaii, J. Volc. Geotherm. Res. 12, 147-160. Eisbacher, G. H., and Clague, J. J. (1984) Destructive mass movements in high mountains: Hazard and management, Geological Survey of Canada84-16. Ferrucci, F. (1996) Seismicity of volcanoes, in F. Barberi, R. Casale, and R. Fantechi (Eds.) The mitigation of volcanic hazards, Office for Official Publication of the EC, Brussels, pp. 225-235. Francis, P. W. (1994) Large volcanic debris avalanches in the central Andes., in International conference on vocanic instability on the earth and other planets, Geological Society of London. Hoblitt, R. P., Miller, C. D., and Scott, W. E. (1997) Volcanic Hazards with Regard to Siting NuclearPower Plants in the Pacific Northwest, USGS Open-File Report 87-297. Hsü, K. (1975) Catastrophic debris streams (Sturzstroms) generated by rock falls, Geol. Soc. Amer. Bull. 86, 129-140. Hürlimann, M., Ledesma, A., and Martí, J. (1999) Conditions favouring catastrophic landslides on Tenerife (Canary Islands), Terra Nova 11, 106-111. Hürlimann, M., Ledesma, A., and Martí, J. (2001) Characterisation of a volcanic residual soil and its implications for large landslide phenomena: Application to Tenerife, Canary Islands, Eng. Geol. 59, 115-132. Hürlimann, M., Martí, J., and Ledesma, A. (2000) Mechanical relationship between catastrophic volcanic landslides and caldera collapses, Geoph. Res. Lett. 27, 2393-2396. Inokuchi, T. (1988) Gigantic landslides and debris avalanches on volcanoes in Japan, in Kagoshima International conf. on Volcanoes, National Institute for Research Administartion, Japan, pp. 456-459. Iverson, R. M. (1995) Can magma-injection and groundwater forces cause massive landslides on Hawaiian volcanoes?, J. Volc. Geotherm. Res. 66, 295-308. Krastel, S., Le Bas, T. P., Alibés, B., Schmincke, H. U., Jacobs, C. L., and Rihm, R. (2001) Submarine landslides around the Canary Islands, J.Geoph. Res. 106, 3977-3997. Martí, J., Hürlimann, M., Ablay, G. J., and Gudmundsson, A. (1997) Vertical and lateral collapses on Tenerife (Canary Islands) and other volcanic ocean islands, Geology 25, 879-882. Martí, J., Mitjavila, J., and Araña, V. (1994) Stratigraphy, structure and geochronology of the Las Cañadas caldera (Tenerife, Canary Island), Geolog. Magaz. 131, 715-727. McGuire, W. J. (1996) Sea-level change and the stability and activity of coastal and island volcanoes, in F. Barberi, R. Casale, and R. Fantechi (Eds.) The mitigation of volcanic hazards, Office for Official Publication of the European Communities, Brussels, pp. 341-363. McGuire, W. J. (1996) Volcano instability: a review of contemporary themes, in W. J. McGuire, A. P. Jones, and J. Neuberg (Eds.) Volcano Instability on the Earth and Other Planets, Geological Society Special Publication, London, pp. 1-23. Moore, J. G., Clague, D. A., Holcomb, R. T., Lipman, P. W., Normark, W. R., and Torresan, M. E. (1989) Prodigious submarine landslides on the Hawaiian Ridge, J. Geophys. Res. 94, 17465-17484.

472 24.

25. 26. 27. 28. 29. 30. 31. 32. 33. 34.


Okada, H. (1983) Comparative study of earthquake swarms associated with major volcanic activities, in D. Shimozurur, and I. Yokoyama (Eds.) Arc Volcanism, Terra Scientific Publishing Company, Tokyo, pp. 43-61. Reid, M. E., Sisson, T. W., and Brien, D. L. (2001) Volcano collapse promoted by hydrothermal alteration and edifice shape, Mount Rainier, Washington, Geology 29, 779-782. Servicio de Planificación Hidráulica (1991) Plan hidrológico insular de Tenerife, Gobierno de Canarias, Tenerife. Siebert, L. (1984) Large volcanic debris avalanches: characteristics of source areas, deposits, and associated eruptions, J. Volc. Geotherm. Res. 22, 163-197. Siebert, L. (1992) Threats from debris avalanches, Nature 356, 658-659. Teide-Group (1997) Morphometric interpretation of the northwest and southeast slopes of Tenerife, Canary Islands, J. Geophys. Res. 102, 20325-20342. Van Wyk de Vries, B., and Francis, P. W. (1997) Catastrophic collapse at stratovolcanoes induced by gradual volcano spreading, Nature 387, 387-390. Voight, B. (2000) Structural stability of andesite volcanoes and lava domes, Phil. Trans. Roy. Soc. London A, 1663-1703. Voight, B., and Elsworth, D. (1997) Failure of volcano slopes, Géotechnique 47, 1-31. Voight, B., Janda, R. J., Glicken, H., and Douglass, P. M. (1983) Nature and mechanics of the Mount St Helens rockslide-avalanche of 18 May 1980, Géotechnique 33, 243-273. Watters, R. J., Zimbelman, D. R., Bowman, S. D., and Crowley, J. K. (2000) Rock mass strength assessment and significance to edifice stability, Mount Rainier and Mount Hood, Cascade Range Volcanoes, Pure and Applied Geophysics 157, 957-976. Watts, A. B., and Masson, D. G. (1995) A giant landslide on the north flank of Tenerife, Canary Islands, J. Geophys. Res. 100, 24,487-24,498.


ROCK SLOPE FAILURES IN NORWEGIAN FJORD AREAS: EXAMPLES, SPATIAL DISTRIBUTION AND TEMPORAL PATTERN L.H. BLIKRA1, O. LONGVA, A. BRAATHEN Geological Survey of Norway, N-7491 Trondheim, Norway E. ANDA The County Council of Møre & Romsdal, N-6404 Molde, Norway J.F. DEHLS, K. STALSBERG Geological Survey of Norway, N-7491 Trondheim, Norway

Abstract Rock avalanches and related tsunamis represent one of the most serious natural hazards in Norway, and during the last 100 years more than 170 people have lost their lives in western Norway. Large-scale rock-slope failures range from sliding of relatively intact masses of rock, to fully disintegrated rock avalanches. A wide variety of features mirror rock avalanches plunging into valleys or fjords. Bouldery fans, lobes and ridges characterize the proximal parts, while thin debris-flow deposits often occur far beyond this zone. Major deformations of valley-fill and fjord sediments are commonly related to the impact of large volumes of rock. The spatial and temporal pattern of rockavalanche events in Norway demonstrates that such events are common and occur within certain regions, and are important data for evaluating background hazard levels. The mechanisms for occurrence and triggering of rock-slope failures are still uncertain, but seismic ground shaking and creep processes are probably important although, in some areas, effects of glacial unloading during the deglaciation phase cannot be excluded. The geographic concentrations of events indicate that relatively large earthquakes may have played a role as triggering mechanisms. This hypothesis is strengthened by the identification of postglacial faults in two of the rock-failure zones.

1. Introduction Large rock avalanches represent one of the most serious natural hazards in Norway, as exemplified by the Tafjord disaster of 1934 when 3·106 m3 rock mass dropped into the fjord. The tsunami generated by the avalanche reached a maximum of 62 m above sea level, and several inhabited villages along the fjord were destroyed (Figure 1). During the last 100 years, 174 people have lost their lives in three such events in a limited 1

E-mail of corresponding author: [email protected]

475 S.G. Evans et al. (eds.), Landslides from Massive Rock Slope Failure, 475–496. © 2006 Springer. Printed in the Netherlands.


region in northern West Norway. A database containing historical avalanche events, clearly shows the high number of rock-avalanche events in the inner fjord areas (Geological Survey of Norway (NGU) - Geohazard database; Figure 2). Although high risk is related to such events, very little attention has been paid to hazard assessment. During the last five years, NGU has investigated the geographic distribution and temporal occurrence of rock-slope failures in order to define background hazard levels [1, 4, 5, 6, 9]. More specific studies have been done on unstable rock-slope areas [7]. The present paper presents some examples of different types of rock-slope failures in Norway. Many studies have focused on the morphology and sedimentology of rockavalanche deposits [10, 14, 15, 20, 27, 30], but lack research on the characteristics of deposits related to rock-slope failures descending into deep valley fills and marine sediments of fjords. The examples shown here describe rock-avalanche deposits in such settings. Attention is also paid to the spatial and temporal patterns of rock-slope failures, and the importance of such data in the context of background hazard levels and the discussion of triggering mechanisms.

Figure 1. The damage caused by the 1934 tsunami in the village Fjøra in Tafjord is shown by photographs before (A) and after (B) the disaster. Photos from Furseth [18].

2. Methods Rock-avalanche deposits have been mapped in Møre & Romsdal, Sogn & Fjordane and Troms counties. Many of these have been visited in the field in order to document special features, while some have undergone detailed investigations in the form of geological mapping, georadar profiling, refraction-seismic profiling and excavations [5, 6, 13, 24, 25]. In addition, some of the areas characterised by extensive gravitational faulting have also undergone detailed structural mapping. As part of NGU projects focusing on rock-avalanche hazard, several fjords have been covered by swath bathymetry, and most of the fjords affected by large rock avalanches are covered by reflection-seismic profiles. Excavations in rock-avalanche deposits on land have been carried out to try to date individual mass-movements. Furthermore, two constrained ages are from Tafjorden, and one from Aurlandsfjorden [22]. The radiocarbon dates are given with one standard deviation, and calibrated to calendar years according to Stuiver et al. [33].


Results extracted from a database comprising historical avalanches, a part of NGUs Geohazard database, are also presented here. These data were collected in cooperation with Astor Furseth, from Norddal in Møre & Romsdal, who has worked on historical avalanches and written a book on the Tafjord accident of 1934 [18]. The database on historical events covers more than 2100 avalanches, and includes numerous large rock-avalanche events.

Figure 2. Historical documented rock-avalanche events in two counties in western Norway (Sogn & Fjordane and Møre & Romsdal) and Troms county in northern Norway, NGU – Geohazard database. Altogether 31 of these recorded events have caused tsunamis. See location of examples described in the text.

3. Examples of Massive Rock Slope Failures in Western and Northern Norway Large-scale rock-slope failures in Norway range from the sliding of relatively intact masses of rock to fully disintegrated and fast-moving rock avalanches. Rock-slope failures that develop into true rock avalanches are the dominating type of event in Norway, mainly due to the high, steep topographic relief of Norwegian fjords and valleys. The examples discussed are focused on these fully disintegrated rock slope failures.


3.1. ROMSDALEN VALLEY One of the largest concentrations of rock avalanches on land is found in Romsdalen (Figures 2 and 18), where more than 15 large rock avalanches cover almost the entire valley floor over a distance of 25 km. One of the source areas for these rock avalanches shows evidence of unstable conditions. Typical features include major gravitational faulting with deep clefts, and with distinct horizontal displacements (Figure 3). An area of the plateau covering an area of more than 2 km by 200 m is marked by fractures, with crevasses more than 20 m deep and horizontal displacements of more than 20 m. Several rock-avalanches have been triggered from this area in the past and a series of rock-avalanche deposits are found in the valley.

Figure 3. Aerial view of the gravitational faults and fractures on Børa and Mannen in Romsdalen. See location in Figure 2. (A) View towards south with Romsdalen valley to the left; (B) Detail of the northernmost bedrock failure at Mannen, showing major vertical displacement of more than 15 m; (C) View towards north showing the deep fractures dissecting the glacial till and underlying bedrock on Børa. The clefts can be followed for more than 2 km and is situated more than 200 m from the slope edge.


Figure 4. Schematic sketch from the Venje rock avalanche in Romsdalen showing sediment distribution and effects of a large rock avalanche impacting valley-fill sediments.

Figure 5. The Venje rock avalanche in Romsdalen: (A) Overview of the outer part of the rock-avalanche deposits; (B) Excavation in one of the mounds in the outer area.

Many of the rock-avalanche deposits are characterised by bouldery cones or lobes, commonly revealing a chaotic morphology with ridges, mounds and intervening basins/ponds. The Venje rock avalanche in Romsdalen exemplifies typical morphological and sedimentological features formed when large masses of rock impact a sediment-filled valley floor (Figures 4, 5 & 6): 1) a bouldery deposit close to the mountain slope; 2) extensive deformation in form of folds and faults up to 20 m below the surface (evidenced by penetrating radar profiling and excavations); and 3) a thin, diamict unit capping the deformed sediments. The complex nature of such rockavalanche deposits has been presented in several papers [14, 20]. The diamict unit is interpreted to be formed by debris-flow processes resulting from remobilization and liquefaction of fine-grained valley-fill sediments during the rock avalanche impact (Figure 6). Such deposits are found in extensive areas more than 1 km outside the bouldery cone, and their run-outs are often limited by the opposite valley slope. These processes have probably been highly influenced by lubrication of water extruded from saturated valley fill (see also discussion in Erismann & Abele [14]). Extensive


deformation is found below and outside the areas covered by the secondary debris-flow deposits (Figures 5, 6), a situation that is quite similar to those documented by Hewitt [20] from the Karakoram Himalaya. Sometimes it may be difficult to distinguish the deformation of rock-avalanche impact from that of glaciotectonic origin. In the present case, the tectonized sediments are fine-grained marine fine-sand, silt and clay of Holocene age. They are the result of the postglacial Holocene deltaic and marine infill, and a glacial origin can thus be excluded.

Figure 6. Details deformations and diamict units in rock-avalanche deposits: (A) Deformed sand and silt of the valley-fill below a thin diamict unit (debris-flow deposit). Detail from the excavation shown in Figure 5B; (B) Deformations in shallow-marine sediments marginal to a rock-avalanche deposit in Romsdalen. The section is c. 3,5 m high.

Figure 7. Reconstructed sea-level curve from Romsdalen based on Svendsen & Mangerud [34]. The maximum ages of individual rock avalanches are plotted on the basis of altitude of undisturbed surface of the deposits. The altitude has been compensated for a river gradient from the deposits down to the supposed former delta area (c. 5 m).


Figure 8. Rock-avalanche deposits and scars in Tafjorden. The position of the disastrous 1934 Tafjorden avalanche, which caused a major tsunami, is shown. The data is based on satellite images onshore and a shaded relief map of the fjord based on swath bathymetry. Seismic profile is shown in Figure 10 and the core is presented in Figure 11. See location in Figure 2.


Many of the complex and chaotic rock-avalanche deposits have earlier been misinterpreted as the product of glacier-related transport and deposition, a problem also encountered in other glaciated regions of the world [20]. In Romsdalen, the rockavalanche deposits are situated well below the marine limit of the area (70-80 m a.s.l., Figure 7). The fact that they are of subaerial origin, not capped by marine or fluvial sediments, indicates that they are quite young and not of glacial origin. When we plot the altitude of the deposits into a reconstructed sea-level curve of the area, we achieve maximum ages of the different deposits (Figure 7). The Venje event is then younger than 2,000 calendar years BP, and many of the others were deposited several thousand years after the deglaciation. The young age of the Venje events is documented by charcoal found below the rock-avalanche deposits, yielding ages of 1,992-1,882 and 1,413-1,352 calendar years BP.

Figure 9. A 3D image of bouldery rock-avalanche deposits in the outer part of Tafjord (see location in Figure 8). The image is based on detailed swath bathymetry.

3.2. TAFJORDEN The highest hazard to humans in Norway is related to rock-slope failures in the fjords. With this in mind, a selection of fjords has been mapped by swath bathymetry and by a reflection-seismic net in order to identify rock-avalanche deposits. The Tafjord area has been investigated in great detail because this fjord experienced one of the most serious natural disasters in Norway (see Figure 1 and location in Figure 2). Geological mapping reveals many slide scars in the slopes above the fjord, and about 10 large rockavalanche deposits have been detected at the bottom of the fjord (Figure 8). The swath bathymetry reveals the morphology of the individual deposits (Figures 8 & 9). Some of


the largest avalanches have a cone-shaped morphology with a bouldery surface, while others are smaller and can be seen as distinct concentric wavelike features around a core of rock debris. These rings are interpreted to be caused by severe deformation (extrusion) of underlying sediments. The extensive folding and faulting of soft sediments is similar to the observations in outcrops and in georadar profiles associated with rock avalanches on land (e.g. at Venje, Figures 4-6), and those described by Hewitt [20] from the Karakoram Himalaya, northern Pakistan. He described large mounds, 1030 m high, which are arranged concentrically down from bouldery rock-avalanche deposits. From the 1934 Tafjord event we can see the boulder cone, but also an outer splash zone interpreted to be formed by secondary mass flows (Figure 8). These may be comparable to the thin debris-flow deposits seen in excavations in distal parts of some rock-avalanches onshore (Figures 4 & 6).

Figure 10. A reflection seismic profile from Tafjorden, showing the relationship between rock-avalanche deposits and finer-grained marine sediments [5, 6]. See the location of seismic line in Figure 8.

Figure 11. A core from Tafjorden. See locality in Figures 8 and 10. See Lepland et al. >22@ for details.


An enormous rock avalanche, estimated to include more than 100·106 m3, occurred in the valley near the Tafjord community (Figure 8). The slide scar is on Kallskaret, with its topmost part at about 1300 m.a.s.l. The boulder fan lies on land, but the avalanche debris can be tracked into the fjord with an outrun distance of more than 2 km along the fjord bottom. This rock-slope failure also dammed a lake. The outburst of this dam formed a large boulder fan, which can be followed down to the fjord. The Quaternary stratigraphy of Tafjord, established from a dense net of reflectionseismic profiles (Figure 10), and calibrated with a few sediment cores (Figure 11), demonstrates that individual rock-avalanche events are common throughout the postglacial period, but with a higher frequency in the second half of the Holocene. Below the 1934 slide scar, at least three individual rock-avalanche deposits have been identified, with the 1934 event being the smallest (Figures 8 & 10). Two samples from a 1.2 m long core of Tafjorden have been dated (Figure 11). The core shows several sand layers with erosive boundaries. The upper unit with an age of 658-564 calendar years BP, lays below a relatively thick, normally graded turbidite deposit, which may be related to the 1934 event. The lowermost, normal graded unit has isolated wood fragments, probably from a tsunami transporting terrestrial material into the fjord. Altogether four events are recorded in the core, indicating that a series of rock avalanches has occurred since 3,200-3,300 calendar years BP.

Figure 12. Gravitational fractures and slide scars in Flåmsdalen and Aurlandsfjorden, located in Figure 2. The image is a 3D DEM based on satellite and topographic data, viewed towards southeast.

3.3. FLÅMSDALEN AND AURLANDSFJORDEN A range of slope-instability features has been mapped in areas of schist bedrock in the Aurland-Flåm region in western Norway [6] (see location in Figure 1). Geological mapping shows that many rock avalanches and gravitational bedrock fractures occur in a zone that is nearly 11 km long (Figures 12 & 13). The zone is bounded to the east by a normal fault, traceable for 4 km and seen as a 1-2 m high scarp, at places showing local sink hole development. In the northern part of the area, slide structures in bedrock and glacial till are associated with the fault zone. It has been concluded that the entire


mountain area west of the eastern boundary faults has moved several metres down-tothe-west [6]. Indicated thickness of the slide block is 300-500 metres, which suggests a total volume of between 900-1500·106 m3. Rock-avalanche deposits found along Flåmsdalen and in Aurlandsdalen are not included in this estimate. It cannot be excluded that the faults are of a tectonic origin, but gravity most likely was the controlling factor. A series of prominent slide scars are found along the entire eastern mountainside of Flåmsdalen and Aurlandsfjorden (Figure 13). Large portions of the eastern slopes are characterised by 100-700 m wide, imbricated bouldery lobes, with characteristic steep fronts and internal flatter areas. They are interpreted to be formed by slow creep in deposits originating from older rock avalanches. At least some of these lobes are still active today, and a sudden increase in flow velocity during periods of high precipitation and snowmelt has been observed [12]. Slide features in bedrock and glacial tills on relatively moderate slopes (16-24º) have been observed in these mountains. Such

Figure 13. Gravitational fractures and slide scars in Flåmsdalen and Aurlandsfjorden. See location in Figure 2. The topographic map shows parts of the area locating major faults, slide scars and the distribution of rockavalanche deposits. Modified from Blikra et al. [6].


Figure 14. Bathymetric data from Aurlandsfjorden showing the main geomorphic features and sediment thickness above slide deposits. Locaton in Figure 2, and the seismic line presented in Figure 15 is shown. Modified from Blikra et al. [6].

gravitational failures are not common in moderate slope gradients. Weak schist bedrock is probably the main reason for the occurrence of both the extensive faults and fractures on the mountain plateau, and for the active creep processes on the mountain slopes. The fjord basin below the phyllitic slope shows large ridges, basins, and hummocky deposits (Figure 14). The large ridges in the western part of the fjord are bedrock features, while the hummocky surface in the eastern sector of the fjord represent one or more avalanche deposits. Their source area is on the eastern mountain slope. A dense grid of seismic profiles demonstrate that large parts of the inner fjord basin are filled by slide deposits [6]. Fine-grained sediments are deposited in depressions on top of and in between slide units (Figure 14), and only 2-4 m of younger soft sediments cover the avalanche deposits. An age of 2,840-2,720 calendar years


before present at 1.5 m depth in a core from Aurlandsfjorden indicates that the slide below is approximate 3000 years old. In the basin between the two bedrock features, the soft sediments are more than 12 m thick, which infers that the avalanche did not enter this basin (Figure 14). 4 km further out in the fjord a bed of slide deposits, interpreted from the seismic survey, is intercalated in fine-grained marine sediments. Depth of burial is 5-8 m (Figure 14 & 15). In a normal development, the sediment thickness should decrease away from the delta area of Flåm, which is the main sediment source to the system. The sediment thickness above slide deposits is only 2-4 m at a distance of less than 2 km from the delta, but more than 5 m at a distance of more than 4 km from the delta (Figure 14). This indicates that there are several avalanche events in the fjord that have disturbed the normal sedimentary evolution of the fjord basin, with the outer slide as the oldest. If the sedimentation rate shown in the dated core is representative, the Holocene sediments are between 4 and 5 metres thick. The sediment thickness capping the slide in the outer part of the fjord (Figures 14 & 15) is consistent with this, and the largest and oldest slide seem to have occurred shortly after the deglaciation at c. 11,000 calendar years BP. We speculate that the large bedrock failures and the formation of widespread gravitational faults, as well as rock avalanches in Flåmsdalen and along Aurlandsfjorden, have been formed during this oldest event.

Figure 15. Reflection seismic profile from the northernmost part of the Aurlandsfjorden site. The profile is located in Figure 15. Modified from Blikra et al. [6].

3.4. GROVFJORDEN AND BALSFJORDEN, NORTHERN NORWAY Numerous rock slope failures are mapped in Troms County, northern Norway. The largest occurs in Grovfjorden (Figure 16; for location see Figure 2 & 19). The rock


avalanche was triggered from about 1000 m.a.s.l., crossed the fjord and dammed the lake Saltvatnet. The total run-out is more than 4 km and the total volume is estimated to be between 60 and 100·106 m3. The morphological features are characteristic of rock avalanches, including bouldery ridges and mounds, and individual boulders that are up to 20 m in diameter (Figure 16).

Figure 16. (A) Rock-avalanche deposits in Grovfjorden, Troms County. See location in Figures 2 and 19; (B) View towards north showing the hummocky surface and the bouldery deposits. Individual blocks are up to 20 m in diameter (circle).


Some rock-avalanche events in this region have been radiocarbon dated. A rock avalanche in Balsfjord, in the middle part of Troms County, is one example. Here, excavations during road construction exposed some peculiar sediments at 50 meters above the present sea level. The sediments were composed of large boulders, up to 10 m in diameter, with interstitial areas filled by stratified and laminated sediments consisting of silt and fine sand to gravel (Figure 17). The sediments showed no sign of deformation, thus demonstrating sediment infill after the deposition of the blocks. The surface of the boulders were covered by Balanus shells (see detail in Figure 17), another demonstration of an originally openwork texture. The blocky deposits are interpreted to be formed by a rock avalanche. The shells indicate that the rock slope failure occurred when sea level was at least 50 m above present sea level, suggesting that the rock-slope failure occurred shortly after deglaciation (ca. 11,000 calendar years BP). The upper part of the deposit must have had an openwork texture, a good habitat for molluscs and gastropods. The voids between boulders were later infilled by silt, sand and gravel supplied by nearby rivers, which finally capped the entire bouldery deposit. Several radiocarbon dates yielded ages of ca. 10,600 calendar years BP, confirming that the event occurred shortly after the deglaciation, between 11,000 and 10,600 calendar years BP. It is still not established whether or not the avalanche occurred while glacier ice still occupied parts of the run-out zone.

Figure 17. Rock-avalanche deposits in Balsfjord, Troms. See location in Figures 2 and 19. The detail shows Balanus shells covering large blocks. See text for discussion.


4. Spatial Distribution Geological mapping on land and in fjords of western and northern Norway has identified a high frequency of rock-slope failures throughout the last 10,000 years. In addition, the geographic distribution of events shows a clustering in specific zones (Figures 18 & 19). In Troms County, northern Norway, the most frequent features are large-scale rock avalanches (Figure 16) and rock glaciers of rock-avalanche origin. More than 150 such features have been mapped in the northeastern part of the County, covering an area of ca. 7000 km2. A less pronounced grouping of rock avalanches also occurs in the southwestern part of the County, east of Harstad (Figure 19). In the Møre & Romsdal County in northern West Norway, almost 200 individual events have been mapped, with distinct concentrations in the inner fjord areas (Figure 18). Furthermore, some minor clusters also occur in the outer coastal parts. The largest concentration on land is found in Romsdalen, were more than 15 large rock avalanches cover almost the entire valley floor over a distance of 25 km (see above). In Tafjorden, more than 10 rock-avalanche deposits have been mapped in the fjord covering a distance of less than 7 km. Also, within a short timeframe, the historical sources evidence a high number of events in the inner fjord areas (Figure 2).

Figure 18. Distribution of rock-avalanche events and gravitational fractures in Møre & Romsdal county, western Norway (modified after Blikra et al. [6]. The postglacial Berill Fault is shown.


5. Temporal Pattern The dataset consists of events that have been radiocarbon dated, or indirectly dated by use of seismic stratigraphy or sea-level curves (Table 1 and 2). Dated events in northern Norway show that they are old, and occurred in a period shortly after deglaciation. The clusters of rock glaciers, originated from massive rock-slope failures, are restricted to areas outside the Younger Dryas ice margin or to nunataks within the ice sheet. This indicates that they were formed during or before the Younger Dryas period (13,000– 11,500 calendar years BP), as is also suggested by Sollid and Sørbel [30]. Rock glaciers require permafrost conditions to evolve and the only actual cold period is the Younger Dryas period, or possibly, some older phases. Radiocarbon dated events show that they occurred shortly after deglaciation, 10,500-11,000 calendar years BP (Table 2). Only a few rock-avalanche events are recorded in historical times (Figure 2).

Figure 19. Distribution of rock avalanche events, gravitational fractures and rock glaciers in Troms County, northern Norway. The data are partly extracted from a Quaternary geological map [32].

It has thus been postulated that most rock avalanches occurred shortly after the last deglaciation. This might be correct for some districts in Norway, but the present studies show that this idea has to be modified. The data from western Norway indicate that many of the avalanches there occurred during the last 5,000 years, possibly with a peak


of activity around 3,000 calendar years BP (Table 1). The general stratigraphy based on seismic data in Tafjord, where a series of rock-avalanche events can be documented (see above), demonstrates that individual events are spread throughout the postglacial period, but with a higher frequency in the latter half of the Holocene. In the outer coastal area of northern West Norway, the time constraint of the gravitational fractures and rock avalanches is poor, but there are indications that most of them occurred shortly after the deglaciation, 15,000-14,000 calendar years BP [6]. Table 1. Dated rock-avalanche deposits in Møre & Romsdal and Sogn & Fjordane counties. Historical events are not listed. For more details see Blikra et al. [6]. Locality Age (Calendar years BP) Dating method Dated material 14 Venje, Romsdalen 2,300 14 C Palaeosols Innfjorden 11,500 Sea level (YD frost-shattering) Skorgeura, Ørsta >11,500 Sea level (YD frost-shattering) Øtrefjellet >11,500 Lake core (Vedde ash layer) Sørdalen, Syvdsfjorden 11,500 Rock glaciers Grytøya >11,500 Rock glaciers

6. Discussion 6.1. BACKGROUND HAZARD LEVEL The spatial distribution and the temporal pattern of rock-slope failures are of major importance in evaluating background hazard levels. In Troms County of northern Norway, there is only one documented historical event. This was a rock avalanche in Lyngen in the year 1819, which reached the fjord. The resulting large tsunami destroyed three farms, and 14 people were killed. Although a large number of rock avalanches have occurred in this area (Figure 19), the available age data suggest that most failures were restricted to a short period before, during, or shortly after the Younger Dryas period (13,000-11,500 calendar years BP). Likely key triggering mechanisms could have been high-magnitude earthquakes or permafrost melting, conditions which were


probably quite different from those of today. The spatial distribution of events therefore cannot be directly used in estimating background hazard levels for this region. Altogether, the limited documentation of historical events and the fact that most older events seem to be related to a period shortly after deglaciation, suggest that the probability for large rock failures in Troms county is relatively low. However, some of the gravitational faults and fractures observed on mountain plateaux may still be unstable and under development. If so, the evolution relates to ongoing rock slope movement and possibly melting of existing permafrost. A somewhat different picture can be established from the spatial and temporal pattern of rock-slope failures in the inner fjord areas of Møre & Romsdal County, western Norway (Figure 18). The high concentration of rock-avalanche deposits, together with the age data, demonstrate that they have occurred throughout the Holocene. In fact, the frequency seems to be higher in the second half of the Holocene. The regional risk is also indicated by a high number of documented historical events (Figure 2). The highest risk areas of Norway related to rock-slope failures are thus the fjords of western Norway, as illustrated by the dataset, which indicate a return interval of more than 1 event in 1,000 years in some of the fjords. The risk is especially large because of the destructive tsunamis related to such events. The age control is weaker for the less pronounced clusters of rock-slope failures in the coastal zones of the Møre & Romsdal County. There are indications that these events could be much older, probably related to a phase shortly after deglaciation. These features may thus be an effect of high-magnitude earthquakes or result from permafrost melting, a situation that is quite different from today. The probability of large bedrock failures in these areas thus seems to be much smaller, in agreement with few documented historical events (Figure 2). 6.2. GEOLOGICAL CONTROL AND TRIGGERING MECHANISMS Pre-existing planar bedrock structures are important for the occurrence of rock slope failures, especially the foliation of the rocks as it constitutes potential planes of reactivation, and thereby affects the geometry of the failure [9]. Rock avalanches initiate when the driving forces overcome the shear strength of the rock mass. Important factors for rock mass stability include glacial debutressing, seismic activity, water pressure, and the role of frost wedging and permafrost melting. The effects of removal of the support of adjacent glacier ice during periods of down-wastage (glacial debutressing) and the consequent stress-release are thought to be of major importance for rock-slope stability [3, 15, 26]. These authors argue that the loading by overlying ice induces internal high stress levels both on the valley floor and within the valley slopes. This hypothesis postulates that the release of elastic strain energy during periods of ice down-wasting results in propagation of the internal joint networks which may cause rock-slope failures. Very few of the rock-avalanche deposits mapped in Norway show evidence of being deposited on top of glaciers, indicating that there were very few immediate rock-slope failures following the gradual deglaciation. This indicates that the debutressing theory needs to be validated by age control. Also of importance is that this process can be considerably delayed due to time-dependent dissipation of residual stresses within the rock mass [3, 37]. The relatively young ages


of the rock-slope failures in the inner fjord areas of Møre & Romsdal County is consistent with effects other than glacial deloading. However, it may be important for the older events in the coastal area of this region, and for the occurrence of rock failures in northern Norway. The possibility that one or more large earthquake(s) have caused the observed clusters of rock-avalanche events cannot be excluded (Figures 18 and 19). The concentrations of rock avalanches in certain zones [6, 7], seems to suggest a regional driving force, such as earthquake. Earthquakes have triggered numerous rock avalanches in historical times [3, 19, 21, 23, 36]. However, there are no certain examples of avalanches triggered by earthquakes in Norway, although this mechanism has been suggested for the offshore Storegga slide. The recently documented neotectonic Berill fault [2] in the Møre & Romsdal area (Figure 18) shows development of a complex rock-slope failure area in the hanging wall, suggesting a link between major rock collapse and a pre-historic earthquake. This is also indicated by a high number of rock slope failures, which occur close to this young fault. In Troms County, a postglacial normal fault occurs within the northern clusters of rock-slope failures [11], also pointing to a possible link between rock avalanches and seismic activity. However, with the present knowledge and dataset, we are not able to distinguish between single, major palaeoseismic events, or other factors resulting in temporal repetition, as the cause for pre-historical rock avalanches. Water is regarded as an essential component in gravitational mass movements; either as an important agent to mineral breakdown or growth, as a contributor to lubrication, or as an active force [35]. In Norway, water pressures increase during periods of snow melting and heavy seasonal rain, especially within well-developed fracture networks. Frost wedging, i.e., expansion due to freezing of water, creates a significant force. For fractures with a small aperture, frost wedging can cause further opening. Hence, ice wedging may act as an important factor in the formation of rockfalls and rockslides. Permafrost conditions existed in large parts of mountainous areas of Norway in past periods. During the cold Younger Dryas, permafrost probably occurred down to sea level in western Norway [8]. The stability of the rock slopes can thus have been influenced by permafrost melting shortly after the deglaciation. Highmountain areas of Norway are still characterized by permafrost conditions. Areas above 1500 m.a.s.l. in western Norway and 1000 m.a.s.l. in northern Norway may have permafrost to considerable depth [17]. Pre-failure movement is a relatively slow process that in many cases seems to occur prior to rock-avalanches [28]. This is well illustrated by the disastrous avalanche in Vaiont, Italy [14], but was also observed prior to the Tafjord and Loen events of Norway. Pre-failure movements are mainly controlled by two common mechanisms, increased pore fluid/water pressure along the shear surface, and weathering and/or abrasion along the shear surface. Slow pre-failure movements or creep is thought to be important for several of the rock-slope failures in Norway [9]. Monitoring of potential movement is the only way to evaluate if creep is active or not, and is therefore fundamental for hazard assessments.


7. Conclusions Large-scale rock-slope failures in Norway range from sliding of relatively intact masses of rock to fully disintegrated rock avalanches. A wide variety of diagnostic features characterize rock avalanches that impact on valleys or fjords, including bouldery fans, lobes and ridges, deformation of valley-fills and an outer thin zone of debris-flow deposits. The spatial and temporal pattern of rock-avalanche events in Norway demonstrates that such events are common and occur in clusters; they form important data for evaluating background hazard levels. The data indicate a return interval of more than 1 event each 1,000 years in some of the fjords, where the risk is especially high due to major consequences related to tsunamis. The reason for the occurrence and the nature of triggering mechanisms of rock-slope failures are still poorly understood. Key factors likely include seismic triggering and creep processes, although, in some areas, the effect of glacial unloading during deglaciation may have played a major role. Identification of postglacial faults in two avalanche zones further indicate that relatively large earthquakes may have contributed to the triggering of massive rock-slope failures.

Acknowledgements The paper builds on data collected during projects supported by the Geological Survey of Norway (NGU), County Councils, Norwegian Geotechnical Institute, the National Fund for Natural Damage Assistance, and Norsk Hydro ASA. We wish to thank the editors for reviewing the paper and suggesting improvements.

References 1.

Anda, E. and Blikra, L.H. (1998) Rock-avalanche hazard in Møre & Romsdal, western Norway. Norwegian Geotechnical Institute Publication 203, 53-57. 2. Anda, E., Blikra, L.H. and Braathen, A. (2002) The Berill fault – first evidence of neotectonic faulting in southern Norway. Norsk Geologisk Tidsskrift 82, 175-182. 3. Ballantyne, C.K. (2002) Paraglacial geomorphology. Quaternary Science Reviews 21, 1935-2017. 4. Blikra, L.H. and Anda, E. (1997) Large rock avalanches in Møre og Romsdal, western Norway. Geological Survey of Norway Bulletin 433, 44-45. 5. Blikra, L.H., Anda, E. and Longva, O. (1999) Fjellskredprosjektet i Møre og Romsdal: Status og planer. Geological Survey of Norway Report 99.120. 6. Blikra, L.H., Braathen, A., Anda, E., Stalsberg, K. and Longva, O. (2002) Rock avalanches, gravitational bedrock fractures and neotectonic faults onshore northern West Norway: Examples, regional distribution and triggering mechanisms. Geological Survey of Norway Report 2002.016. 7. Blikra, L.H., Braathen, A. and Skurtveit, E. (2001) Hazard evaluation of rock avalanches; the Baraldsneset-Oterøya area. Geological Survey of Norway Report 2001.108. 8. Blikra, L.H. and Longva, O. (1995) Frost-shattered debris facies of Younger Dryas age in the coastal sedimentary successions in western Norway: palaeoenvironmental implications. Palaeogeography, Palaeoclimatology, Palaeoecology 118, 89-110. 9. Braathen, A., Blikra, L.H., Berg, S.S. and Karlsen, F. (in press) Rock-slope failures of Norway; type, geometry, deformation mechanisms and stability. Norwegian Journal of Geology (NGT). 10. Cruden, D.M. and Hungr, O. (1986) The debris of the Frank slide and theories of rockslide-avalanche mobility. Canadian Journal of Earth Sciences 23, 425-432. 11. Dehls, J.F., Olesen, O., Olsen, L. and Blikra, L.H. (2000): Neotectonic faulting in northern Norway; the Stuoragurra and Nordmannvikdalen faults. Quaternary Science Reviews 19, 1447-1460.

496 12. Domaas, U., Rosenvold, B.S., Blikra, L.H., Johansen, H., Grimstad, E., Sørlie, J.E., Gunleiksrud, O., Engen, A. and Lægreid, O. (2002) Studie av fjellskred og dalsidestabilitet i fyllittområder (Report to the Norwegian Research Council). Norwegian Geotechnical Institute Report 20001132-32. 13. Elvebakk, H. and Blikra, L.H. (1999) Georadarundersøkelser I forbindelse med undersøkelser av fjellskred I Romsdalen, Møre og Romsdal. Geological Survey of Norway Report 99.025. 14. Erismann, T.H. and Abele, G. (2001) Dynamics of Rockslides and Rockfalls. Springer-Verlag Berlin Heidelberg New York. 15. Evans, S.G. and DeGraff, J.V. (eds.) (2002) Catastrophic Landslides: effects, occurrences and mechanisms. Geological Society of America Reviews in Engineering Geology 15, 412p. 16. Evans, S.G., Hungr, O. and Enegren, E.G. (1994) The Avalanche Lake rock avalanche, Mackenzie Mountains, Northwest Territories, Canada: Description, dating, and dynamics. Canadian Geotechnical Journal 31, 749-768. 17. Etzelmüller, B., Berthling, I. and Sollid, J.L. (1998) The distribution of permafrost in southern Norway – a GIS approach, in Proceedings of the 7th International Conference on Permafrost, Yellowknife, Canada, 23-27 June. Nordicana 57, 251-258. 18. Furseth, A. (1985) Dommedagsfjellet – Tafjord 1934. Gyldendal Norsk Forlag A/S. 19. Hermanns, R. L. and Strecker, M. R. (1999) Structural and lithological controls on large Quaternary rock avalanches (sturzstroms) in arid northwestern Argentina. Geological Society of America Bulletin 111, 934-948. 20. Hewitt, K. (1999) Quaternary moraines vs catastrophic rock avalanches in the Karakoram Himalaya, northern Pakistan. Quaternary Research 51, 220-237. 21. Keefer, D. K. (1984) Landslides caused by earthquakes. Geological Society of America Bulletin 95, 406421. 22. Lepland, A., Bøe, R., Sønstegaard, E., Haflidason, H., Hovland, C., Olsen, H. and Sandnes, R. (2002) Sedimentological descriptions and results of sediment cores from fjord and lakes in northwest Western Norway – final report. Geological Survey of Norway Report 2002.14. 23. Maharaj, R. J. (1994) The morphology, geometry and kinematics of Judgemment Cliff rock avalanche, Blue Mountains, Jamaica, West Indies. Quarterly Journal of Engineering Geology 27, 243-256. 24. Mauring, E., Blikra, L.H. and Tønnesen, J.F. (1997) Refraksjonsseismiske malinger i Tafjord, Møre og Romsdal. Geological Survey of Norway Report 97.186. 25. Mauring, E., Lauritsen, T. and Tønnesen, J.F. (1998) Georadarmålinger i forbindelse med undersøkelser av fjellskred i Tafjord, Romsdalen, Hellsesylt og Innfjorden, Møre og Romsdal. Geological Survey of Norway Report 98.047. 26. McSaveney, M.J. (1993) Rock avalanches of 2 May and 6 September 1992, Mount Fletcher, New Zealand. Landslide News 7, 2-4. 27. Nicoletti, P.G. and Sorriso-Valvo, M. (1991) Geomorphic controls of the shape and mobility of rock avalanches. Geological Society of America Bulletin 103, 1365-1375. 28. Petley, D.N. (2002) Patterns of acceleration for large slope failures, in S.G. Evans and S. Martino (eds.), Massive rock slope failure: new models for hazard assessment. NATO advanced research workshop, Italy, June 2002. 29. Schuster, R. L., Nieto, A. S., O´Rourke, T. D., Crespo, E. and Plaza-Nieto, G. (1996) Mass wasting triggered by the 5 March 1987 Ecuador earthquakes. Engineering Geology 42, 1-23. 30. Shreve, R.L. (1968) The Blackhawk landslide. Geological Society of America Special Paper 108. 31. Sollid, J.L. and Sørbel, L. (1992) Rock glaciers in Svalbard and Norway. Permafrost and Periglacial Processes 3, 215-220. 32. Sveian, H. (in press) Quaternary geological map of Troms County. Scale 1 : 250 000. Geological Survey of Norway. 33. Stuiver, M., Reimer, P.J., Bard, E., Beck, J.W., Burr, G.S., Hughen, K.A., Kromer, B., McCormac, G., van der Plicht, J. and Spurk, M. (1998) INTCAL98 Radiocarbon age calibration 24,000-0 cal. BP. Radiocarbon 40, 1041-1083. 34. Svendsen, J.I. and Mangerud, J. (1987) Late Weichselian and Holocene sea-level history for a cross section of western Norway. Journal of Quaternary Science 2, 113-132. 35. Terzaghi, K. (1962) Stability of steep slopes on hard unweathered rock. Geotechnique 12, 251-270. 36. Wieczorek, G. F. and Jäger, S. (1996) Triggering mechanisms and depositional rates of postglacial slope-movement processes in the Yosemite Valley, California. Geomorphology 15, 17-31. 37. Wyrwoll, K-H. (1977) Causes of rock-slope failure in a cold area: Labrador-Ungava. Geological Society of America Reviews in Engineering Geology 3, 59-67.

ROCK AVALANCHING IN THE NW ARGENTINE ANDES AS A RESULT OF COMPLEX INTERACTIONS OF LITHOLOGIC, STRUCTURAL AND TOPOGRAPHIC BOUNDARY CONDITIONS, CLIMATE CHANGE AND ACTIVE TECTONICS R.L. HERMANNS1, S. NIEDERMANN Geological Survey of Canada 101-605 Robson Street Vancouver, British Columbia, Canada V6B 5J3 A. VILLANUEVA GARCIA Universidad de Tucumán 4000 Tucumán, Argentina A. SCHELLENBERGER Institute of Geography, University of Berne CH-3012 Berne, Switzerland

Abstract In NW Argentina rock avalanching occurs in two geomorphic settings: A) narrow valleys draining large basins and B) mountain fronts bordered by wide piedmont areas. In the narrow valley environment, the deposits are relatively young. Landslide events concentrate during humid climate periods, and they occur with recurrence intervals of a few ka. In contrast, the deposits in piedmont settings are significantly older, while rock avalanches occur with recurrence intervals of several tens ka, and do not show any direct relation with climate change. Common to the regional distribution in both settings is the influence of lithology, structural control, as well as tectonic activity. Rock avalanches have occurred only in granites, low-grade metamorphic rocks and coarse clastic sedimentary rocks. These lithologies are competent enough to form steep slopes and provide planar structures that dip towards the valley. Because all rock-avalanches originated in the hanging wall of reverse faults with important Neogene displacement causing mountain-front oversteepening, it is inferred that most collapses were tectonically conditioned and/or triggered. However, only at a few sites detailed sedimentologic studies of related sediments show unequivocally that strong seismic activity triggered landsliding. Geological evidence and comparison with empirical data suggest that these earthquakes have been either crustal and of magnitude > M 7 or very shallow and of a magnitude > M 5.5.


E-mail of corresponding author; [email protected].

497 S.G. Evans et al. (eds.), Landslides from Massive Rock Slope Failure, 497–520. © 2006 Springer. Printed in the Netherlands.

498 1.


In NW Argentina catastrophic bedrock collapse of the rock avalanche type is frequent along the margins of intramontane basins and valleys [30]. Failure is limited to tectonically active, lithologically and topographically predisposed mountain fronts [13, 30] in two morphologic settings: 1) fault-bounded mountain fronts bordered by wide piedmonts and 2) narrow valleys, where mountain-bounding faults coincide with allochthonous rivers draining the surrounding high ranges [29]. All dated rock avalanches in the first setting are relatively old with ages larger than 100 ka up to several hundreds ka [29]. In the valley setting they are generally younger with Holocene and late Pleistocene ages of a few ka up to a few tens ka [72, 29, 71]. While in the piedmont environment the conditioning of mountain-front oversteepening could be related only to tectonic activity [28], in the narrow valleys it has been related to a combination of mountain-front uplift, structural complexity, and subsequent erosional undercutting [65, 30, 29]. Although rock-avalanching in the narrow valleys concentrates during climatic periods characterized by enhanced humidity in the Central Andes and in subtropical South America [72, 29, 71], no rock avalanches occurred along tectonically inactive segments of mountain fronts showing the dominant influence of tectonics for slope oversteepening and/or triggering of mountain-front failures. Here we show examples from both morphologic settings which highlight the importance of static and dynamic conditioning factors controlling the spatial-temporal distributions of mountain-front failures. We include new sedimentologic and structural data which further underline the role of tectonic activity of affected mountain fronts, and indicate that strong seismic activity and landsliding occurred nearly coevally. In addition, new tephrochronologic data from the Quebrada del Tonco [27] demonstrate unequivocally that multiple coeval mountain-front failures have been triggered 14 seismically. Two new AMS C-ages help to better determine the age of rock avalanches and indicate that rock avalanches not only occurred in the Minchin and Titicaca wet periods as earlier suspected [71] but in most periods characterized by more humid climate conditions. 2.

Geological Setting

Mountain fronts bordered by wide piedmonts are located in the Puna Plateau and the northwestern Sierras Pampeanas (Figure 1a and b). These are characterized by reversefault bounded ranges composed of late Precambrian-Palaeozoic low- to high-grade metamorphic rocks and Palaeozoic granites e.g. [15, 53] surrounding intervening basins which contain Tertiary to Quaternary basin-fill deposits e.g. [66]. The narrow valleys lie to the southeast within the Cordillera Oriental geologic province (Figure 1), which is a fold-and-thrust belt of Precambrian basement and overlying unmetamorphosed Cambrian to Tertiary sediments e.g. [45, 55]. Because of the position of the study area (Figure 1a) west of the easternmost ranges with elevations higher than 3000 - 4000 m it is protected from the moisture-bearing winds from the Atlantic and has therefore been characterized by arid climatic conditions since the Pliocene [38] when exhumation of these ranges accelerated [38, 63]. However, at present this area is characterized by seismicity of low frequency as indicated by recorded seismicity of the past 38 years [8].

499 There is only one structure in the study area, and two structures in the Puna Plateau immediately west of the study area, where stronger earthquakes have occurred since instrumental recording (Figure 1a). Most earthquakes occur along the eastern deformation front of the Andes in the Santa Barbara Province and the Subandean Belt to the north and east or in the Sierras Pampeanas to the south. The historic seismic record of the past 200 years shows that three stronger earthquakes have occurred in the study area [6]. 3.


Rock-avalanche deposits were first identified using Landsat TM imagery, using spectral bands 5, 4, and 2 [30]. Typical recognition criteria include lobate forms in piedmont environments and spectral contrasts with valley deposits in narrow valley environments where lobate morphologies could not be attained due to a restricting topography preventing long run-out distances. In a second step, the inferred landslide origin of these deposits was verified on stereopairs of aerial photographs, which further helped to constrain landslides of different ages based on surface morphology. Finally, about 90% of the landslide sites were visited in the field to verify the (landslide) type of the deposit by facies analyses, to study the breakaway scars in detail, to investigate the stratigraphic relations with adjacent deposits, and to sample volcanic ashes for tephrostratigraphic correlations, take surface samples for cosmogenic nuclide dating, and organic material for radiocarbon dating. 3.1. TEPHROCHRONOLOGY Initially, tephra samples were taken from nine different Quaternary sedimentary sequences associated with landslide deposits and grouped using the geochemical composition of glass shards (electron-microprobe analysis, EMA), their morphology (scanning electron microscope analysis) and the mineralogy of the tephra [29]. These groups were defined as deposits of single eruptions. More recently, new tephra samples were taken from five additional stratigraphic sequences in the Tonco valley (cf. Figure 5) and from a well dated profile south of the Cumbres Calchaquíes (asterisk in Figure 1b) and analysed by EMA using a new analysis procedure that gives results closer to the true composition of the glass shards [46]. Changes in procedure have been a reduced counting time for Na2O (10 s instead of 20 s) and the use of an additional spectrum allowing to analyse SiO2 simultaneously with Na2O. This allows to reduce the effects of “Na (and K) loss” and secondary “Si (and Al) grow-in”. For comparison the samples with numbers g07, g31 (Cerro Paranilla Ash), g29, g39 (Quebrada del Tonco Ash), and g04 (El Paso Ash) have been reanalysed (Table 1). Changes of analytical results caused by the change of analytical procedure show that the “Na loss” is the strongest effect and that the new Na2O results are by 5.4 to 43% higher compared to composition reported earlier [29] (Table 1). Changes in K2O, SiO2 and Al2O3 vary depending on the sample and lie for K2O between 5.2 and –1.2%, and for SiO2 and Al2O3 between 0.7 and –2.8%. MgO, CaO, TiO2, MnO, and FeO do not


Figure 1a. Topographic map showing distribution of mountain ranges, intramontane basins, narrow valleys, and the epicentres (circles) of instrumentally recorded earthquakes > Mb 4 since 1964 [8] as well as epicentres of historical earthquakes (stars) with an estimated intensity MM > 6 [6]; box corresponds to location of Figure 1b. Inset shows tectonic provinces of the Andes after [34]; box corresponds to location of map.


Figure 1b. Generalized geologic map of the study area and distribution of landslide deposits after [30] and references therein. Grey circles denote rock-avalanche deposits in narrow valleys, filled circles correspond to rock-avalanche deposits in piedmont environments. Asterisk denotes locality of dated profile in the Valle de Tafí south of Cumbres Calchaquí containing the El Paso ash (sample S 46 in Table 1, see text).

502 seem to be affected as differences between results of both procedures lie within the variance of the analysis. The identical geochemical composition of all the new tephra samples with previously defined tephra indicates that these samples belong to either the Cerro Paranilla Ash, the Quebrada del Tonco Ash, or the El Paso Ash (Table 1). Table 1. Median glass shard major element composition* of tephra layers in Tonco valley (NW Argentina). Sample Na2O SiO2 Cerro Paranilla Ash g07 3.39 76.59 g07 2.30 77.96 g31 3.50 76.42 g31 3.12 77.12 g75 3.48 76.53 g77 3.52 76.41 g79 3.17 76.72 g81 3.34 76.93 Quebrada del Tonco Ash g29 3.80 78.09 g29 3.61 77.94 g39 3.84 78.0 g39 3.63 77.80 g83 3.96 77.3 El Paso Ash g04 3.33 78.40 g04 2.39 79.01 S46 3.08 78.44








0.09 0.09 0.09 0,09 0.10 0.09 0.09 0.09

12.88 13.06 13.06 12.97 13.08 13.05 13.11 12.89

5.17 4.94 5.21 4.95 5.22 5.30 5.24 5.09

0.72 0.73 0.71 0.73 0.72 0.71 0.70 0.70

0.13 0.13 0.14 0.14 0.13 0.13 0.14 0.14

0.06 0.06 0.06 0.06 0.05 0.05 0.05 0.04

0.81 0.76 0.75 0.78 0.75 0.78 0.78 0.77

0.04 0.04 0.05 0.05 0.04

12.63 13.06 12.77 13.12 13.15

4.35 4.36 4.26 4.31 4.45

0.49 0.50 0.49 0.50 0.49

0.08 0.08 0.06 0.08 0.07

0.09 0.10 0.11 0.09 0.10

0.45 0.42 0.45 0.49 0.48

0.06 0.05 0.06

12.55 12.90 12.39

4.23 4.16 4.56

0.80 0.83 0.82

0.08 0.08 0.09

0.05 0.05 0.04

0.56 0.54 0.57

* calculated from 13 to 37 analyses per sample. Data from Hermanns et al. [29] in italics, new data in standard letters.

3.2. COSMOGENIC NUCLIDE DATING Eight laterally superposed rock-avalanche deposits sourced in the same sector of southwestern Sierra Laguna Blanca (Figure 1) were dated using cosmogenic 21Ne in quartz separates obtained from clasts derived from the metamorphic and granitic basement rocks (see [28] for sampling procedure and [49, 48] for technical details and production rate of 21Ne analyses). Although attention was paid to the fact that sampled clasts had not been redeposited, ventifacts indicate that some erosional loss took place after landsliding. Therefore the obtained surface exposure ages represent minimum estimates of the true ages. 3.3.



Here we report the AMS radiocarbon ages as radiocarbon years before present (“present” = 1950 AD), using the Libby 14C half life of 5568 a. The ages < 24 ka were

503 calibrated to calender years (‘CAL BP’) using [68] and are documented with 2 sigma uncertainties. Two new ages of organic material are reported: one from a rockavalanche-dammed lake in Las Conchas valley (asterisk in Figure 3), and one from a 10-m-profile of Holocene deposits containing the El Paso Ash directly overlying the dated material in the Valle de Tafí (Figure 1b). These ages better bracket the timing of related landslides and/or tephra deposits but also the age of a seismite which is stratigraphically associated with the organic material from the lake in Las Conchas valley. One age was determined by Beta Analytic Inc., Miami, Florida, and the other one by Ångström Laboratory, 14C-Lab, Uppsala University, Sweden. 4.

Regional Examples of Massive Rock Slope Failures

In the following, representative examples of rock avalanches from narrow valley environments and from piedmont environments are given. These examples represent either the best documented sites and/or sites where significant new unpublished information was obtained allowing further understanding of the nature of mountainfront collapse in NW Argentina. A detailed list containing information on all rockavalanche deposits shown in Figure 1b and detailed descriptions of the sites are given in [30, 25]; stratigraphic relations between landslides, lake deposits and tephra layers are given in [29]. 4.1. CERRO ZORRITO ROCK AVALANCHES Cerro Zorrito consists of Cretaceous conglomerates, which were uplifted during Andean deformation in two kinematic phases (Figure 2) [19, 65]. In a first phase uplift occurred along the Picacho fault in Neogene time. To the southeast this thrust was linked to the Chacras thrust fault via a left-lateral transfer fault (El Zorrito fault). During the younger kinematic regime the El Zorrito fault was reactivated as a reverse fault and brought the Mesozoic conglomerates into thrust contact with Quaternary piedmont sediments [19, 65]. The present relief between the Río de las Conchas and the highest point on Cerro Zorrito is 1700 m (Figure 2). The conglomerates dip 34° east on the east side of the mountain, and 21° west on the west side. Additional planar structures are 40°-75° southeast-dipping exfoliation joints and 70°-89° southwest-dipping extension fractures (A and B in inset of Figure 2, respectively) that are well developed on the southern Cerro Zorrito slopes. Two rock avalanches each formed on the west, south, and east sides of the mountain. The deposits of the rock avalanches at the east side (Casa de Los Loros = C.L.) and at the south side (El Paso = E.P.) [30] formed natural dams in Las Conchas valley which impounded lakes upriver and were later eroded. A first rock avalanche deposit at C.L. is largely covered by a more extensive younger rock-avalanche deposit and can only be observed along a small road cut. The younger rock-avalanche deposit at that site dammed the valley up to an altitude of 1700 m as indicated by lake sediments upriver which are preserved up to this contour. An older rock-avalanche at E.P. overlies in direct contact terrace deposits of unknown age comprising mainly clasts derived from basement outcrops upriver. This rock-avalanche deposit is overlain partially in up- and


Figure 2. Geologic map of the southwestern part of the Río de las Conchas valley after [30] and references therein with fault-plane analysis represented as pseudofault-plane solutions (after [18]); grey pattern corresponds to extension of old kinematic regime, black represents extension of young kinematic regime. Inset shows A) exfoliation joints and B) extension fractures in lower hemisphere projection representing the planes of movement and breakaway surfaces, respectively, at the south side of Cerro Zorrito; box illustrates the location of Figure 3. E.P. = El Paso, C.L. = Casa de los Loros.

downstream direction by lake deposits up to 45-m-thick [72] containing the Quebrada del Tonco Ash (g29) and in a small depression by swamp deposits containing the El Paso Ash (g04) [29] (Figure 2, Table 1). The lake deposits in turn are overlain partially by younger rock-avalanche deposits also overlying at some locations the older rockavalanche deposit. During this second landslide event at E.P., the lake sediments must

505 have been water saturated because they were folded and injected into the landslide mass [30]. No lake deposits occur on top of the younger rock avalanche, which is overlain by up to 10-m-thick carbonate-rich swamp deposits in small depressions [72]. The rockavalanche deposits at E.P. divided the lake dammed by the landslide at C.L. in two basins (Figure 2). The upper one had a surface area of > 600 km2 [30], the lower one of 8 km2. In both basins lake deposits overlie in direct contact mainly Tertiary sediments composed of silt- to sand-sized clastic units and extensive marl and minor salt deposits [18]. The basins were at least temporarily connected through a spill way at the southern end of the old rock-avalanche deposit at E.P. which is below the contour of 1700 m as indicated by lake deposits at that site. The entire sequence of rock-avalanche and lake/swamp deposits was previously defined to 28,990 ± 150 a B.P. by an AMS-14C age of carbonate bivalves from the upper part of a sediment profile ~5 km upriver from the landslide barrier at E.P. [72]. A second age of 35,650 ± 380 a B.P. was obtained from freshwater snails of the 10-mthick swamp deposit on top of the youngest rock avalanche [72]. However, this older age was contradictory to the younger one because the swamp deposits represent the youngest deposits of the entire stratigraphic column, therefore this age was interpreted to be erroneous and related to a hardwater effect [72]. The temporal landslide-lake sediment distribution is further defined by a sequence of 40-m-thick laminated lake deposits intercalating at El Paso both rock-avalanche deposits. These lamina are cyclic and indicate strong El Niño/Southern Oscillation (ENSO) and Tropical Atlantic Variability (TAV) influence [71]. They are interpreted to be varves [72] and lamina counts indicate a time difference of 6700 ± 700 a [70] between landslide damming at C.L. and the younger rock avalanche at E.P. The end of the lake phase of the upper basin is defined sedimentologically by gravels overlying the lake deposits [70]. A new AMS-14C age of 7,500 ± 70 a ‘CAL BP’ of organic sediment (Beta-152180) sampled from a deformation horizon in the lake sediments 200 m southwest of the older landslide barrier at E.P. (asterisk in Figure 3) indicates that this lake existed in the Holocene. This age is supported by a new age determination of the El Paso Ash (S 46) in a profile in the Valle de Tafí (asterisk in Figure 1b) which indicates that this ash is younger than 10,870 ± 290 a ‘CAL BP’ (Ua-17960). At E.P. this ash occurs within swamp deposits in an internal basin on top of the older rock avalanche deposit, indicating that the porous rock avalanche deposits were water saturated and therefore that the landslide dammed lake most likely existed at the time of swamp formation and tephra deposition. The laminated profile at E.P. contains deformation horizons; such horizons also occur in the upper lake basin. For example, in Las Conchas sub-basin (Figure 3) two strongly folded and disrupted layers exist which are separated by a sequence of 11 m of undisturbed lake deposits (the base of the deposits is not exposed at that locality). At the western rim of the El Paso sub-basin these deformation horizons start along a line characterized by overturned folds within the lake deposits which in turn lie within the strike of two reverse faults offsetting lake deposits. These fault offsets are separated ~ 10 m laterally and provide buckle folds in the footwall and slump folds in the hanging wall indicating that they formed when the lake deposits were water saturated. Kinematic analyses of slicken sides on these faults (performed with the program FAULT KINEMATICS in accordance with [44]) indicate a SW-NE directed shortening


Figure 3. Topographic map of the El Paso and Las Conchas sub-basins showing the distribution of lake deposits and seismic deformation structures. Symbols correspond to those in Figure 4. Inset shows slickensides of striae on faulted lake sediments and calculation of compressional direction (P), and extensional direction (T) of related earthquakes. Asterisk denotes location of seismite associated with organic material with an age of 7,500 ± 70 a ‘CAL BP’.

Figure 4. Grain size distribution of soft sediment deformation structures, in relation to known liquefiable domains [40, 37].


direction which is parallel to the regional compression direction of Quaternary deformation in NW Argentina [43] (Figure 3). Deformation horizons do not occur in this quantity in the centre of the El Paso sub-basin which is characterized by purely silty lake deposits. Only one folded deformation horizon was found there which is the layer containing visible organic material used for AMS-14C dating (see above). The deformation horizons, the folded lake sediments as well as the faulted lake sediments were sampled for grain-size analyses. Results are summarized in Figure 4 and show that localities with deformation horizons are strongly limited to sections with alternating silty and sandy layers, which have grain size distributions similar to sands liquefied during earthquakes in Japan or during laboratory tests simulating earthquakes [40, 37]. 4.2. TONCO ROCK AVALANCHES Five rock avalanches have occurred along Río Tonco in the hanging wall of a westvergent thrust, which places Cretaceous to Upper Tertiary rocks against Andean foreland-basin strata [27, 30]. Bedrock failure leading to avalanche formation occurred in the folded Cretaceous conglomerates that are part of the western limb of an asymmetric syncline (Figure 5). Two of the rock-avalanche deposits are located within the syncline in front of a remarkable scarp. The breakaway zones lie between 100 m above the valley floor and the 3185-m-high crest of the mountain range. A block with a volume of 70 x 106 m3, 50 m thick, 1 km long and 1,4 km wide collapsed along the 35° east-dipping bedding plane (I). A second minor rock avalanche (II) occurred along the northern breakaway wall of the first event and resulted in a lobe being deposited on top of the older avalanche deposit. Remnants of additional rock-avalanche deposits (III – V) are found to the south. These are only partially preserved because of their position in the main course of the river and related erosion. The temporal distribution of the landslides is still poorly defined. Rock-avalanche deposit I and IV overlie in direct contact the Cerro Paranilla Ash (g7, g31, g75). This ash is in addition overlain in direct contact by a rock-fall deposit with a volume of a few cubicmeters south of landslide deposit I (g77) and by a rock-fall deposit of similar volume originated by the secondary collapse of rock-avalanche deposit V (g79) (Figure 5). At all sites this ash is pure and does not show any structures indicating redeposition. The Cerro Paranilla Ash also occurs within terrace deposits 40 m above the actual river level (g81), which are in turn overlain by a 12-m-thick pile of lake sediments (Figure 5). These lake deposits were most likely also associated with a landslide, although no such deposits are preserved. The lake-sediment sequence was found upstream of a deeplyincised canyon, only ~1 m wide in some segments, where landslide deposits could have been easily eroded during the last millennia. From detailed mapping of the geology and the geomorphology of this region, a landslide-dammed lake is the best interpretation for this site. Cerro Paranilla Ash was previously dated with a biotite-multi-crystal 40Ar/39Ar age at 723±89 ka [29], however morphologic observations suggest that this age is too old. Based on a comparison of the morphology and the break away scarps of the slides with those of the Cerro Zorrito rock avalanches a similar age range is estimated. The significantly too old age of the 40Ar/39Ar method is likely due to contaminations of the multi-crystal sample as indicated by a K/Ca ratio < 1 for most of the analytical steps

508 (M. McWilliams, Personal Communication, 2002). A new sanidine-multi-crystal 40 Ar/39Ar age of 483±16 ka supports the interpretation that the previously published age does probably not give the age of tephra eruption and deposition. However, the fact that also the latter age is significantly too old is indicated by large amounts of excess 40Ar trapped in the crystals (A. Deino, Personal Communication, 2002). A late Pleistocene / Holocene age of the landslides is inferred by the Quebrada del Tonco Ash, which occurs in the central part of the former lake at its bottom (g83). Here it is overlain by rock fall deposits of Cretaceous siltstone. Finally this ash (g39) overlies landslide deposit I within swamp deposits. Due to stratigraphic relations of the Quebrada del Tonco Ash in the El Paso section (Figure 2), this ash has an apparent age of latest Pleistocene / Early Holocene similar to those of landslides at Cerro Zorrito and therefore all landslides associated in Quebrada del Tonco with this ash are within the same range of age.

Figure 5. Geologic map of the Tonco slides in the folded Mesozoic rocks in the Cordillera Oriental modified after [30, 27]. Numbers refer to tephra samples as listed in Table 1.

4.3. SIERRA LAGUNA BLANCA ROCK AVALANCHES The western mountain front of the 6000-m-high Sierra Laguna Blanca in the arid southern intra-Andean Puna Plateau is representative for the piedmont type of setting.

509 Ten rock-avalanche deposits occur on the western piedmont of this range. Despite the high elevation no evidence of Pleistocene or Holocene glaciations exists on the west flank where rock avalanches were generated [14, 22], indicating overall sustained arid conditions. In addition, trunk streams which could have potentially influenced oversteepening of slopes are 14 km away.

Figure 6. Geologic map of the western Sierra Laguna Blanca after [17, 30, 28].

Sierra Laguna Blanca comprises late Proterozoic to Palaeozoic low-grade metamorphic rocks in the south and granites in the north (Figure 6) which are faulted against Neogene coarse clastics and Quaternary boulder conglomerates on its western

510 side [17]. In contrast, along the eastern flanks the basement rocks are partially covered by volcanic rocks [73]. The metamorphic basement rocks are characterized by a penetrative schistosity dipping 40°-80° northeast to east, a second minor schistosity dipping 15° southeast to east, exfoliation joints dipping 45°-70° west to west-northwest, and minor thrusts with 55°-70° west-southwest dipping fault planes [30]. Deformation at Sierra Laguna Blanca has propagated westward from the mountain front involving the western piedmont, indicated by a sequence of north-south striking reverse faults affecting successively younger strata (see also profile in Figure 6) [28]. For example, Pleistocene reverse faulting has caused a 120-m-high tectonic scarp in the piedmont gravels and underlying Tertiary units (inset in Figure 6). This scarp passes under the proximal end of rock-avalanche deposits aiding to the protection of the distal parts of these deposits against mountain-stream erosion (Figure 6). The southwestern piedmont of Sierra Laguna Blanca hosts eight superimposed rockavalanche deposits (I – VIII) each involving several hundred million cubic meters of bedrock debris [30]. Farther north, additional rock-avalanche deposits (IX and X) resulted from the collapse of terrace deposits uplifted along the mountain-bounding reverse fault which form a tectonic scarp ~ 400-m-high (Figure 6). In addition, multiple amphitheater-shaped breakaway scarps occur in the basement rocks along the steep west-facing slope suggesting sustained landsliding activity (Figure 6) [30]. The temporal distribution of the eight partially superposed rock-avalanche deposits as well as the age of the uplifted terrace deposit is defined by 21Ne cosmogenic nuclide dating [28] (Figure 7). Within uncertainty the minimum ages correspond to the stratigraphic relations between the individual landslide deposits, suggesting that seven rock avalanches occurred between 152 +17/-24 ka and 317 +38/-45 ka. Due to the overlapping uncertainties of the age determinations, precise recurrence intervals of rock avalanching cannot be determined. Rock avalanche deposit I has a minimum age of 431+18/-26 ka and is thus significantly older. The terrace surface has ages between 85 +9/-27 and 130 +7/-30 ka. However, these latter ages are only estimates of the true age of the terrace surface, as sampled clasts may have suffered cosmogenic irradiation not only during exposure on the terrace surface but also at the mountain front prior to erosion and during transport. Abundant ventifacts on this surface also indicate erosional loss due to abrasion. 5.


A combination of various conditioning factors control the spatial and temporal distribution of rock avalanches in NW-Argentina. Conditioning factors which do not change in time, such as lithological and structural prerequisites for the formation of giant collapses, are static boundary conditions and mainly control the regional distribution of rock avalanches. Those which change the stability of a mountain front in time are summarized as dynamic boundary conditions and control the temporal distribution of landslides. The ultimate trigger mechanism of a mountain-front collapse is often difficult to define for palaeolandslides. However there is strong evidence that at least some of the northwest Argentine rock avalanches were triggered seismically.


Figure 7. Stratigraphic position and ages of rock-avalanche and terrace deposits from the western pediment of Sierra Laguna Blanca after [25] compared to oxygen isotope stages as defined by Imbrie et al. [32].

5.1. STATIC BOUNDARY CONDITIONS One of the most important controls on massive rock slope failure in the Argentine northwest is lithologic. Although the majority of ranges in this region are composed of a variety of low to high grade metamorphic rocks, granites, and fine- to coarse-grained sedimentary units with minor units of biogenic and chemogenic sediments, rockavalanches are restricted to low-grade metamorphic rocks, granites, and coarse clastic sedimentary units. These lithologies are competent enough to form steep slopes and provide discontinuities that represent potential surfaces of failure. These discontinuities are generally inclined directly towards the valley or alternatively two sets of discontinuities exist that are inclined obliquely towards the valley thus allowing wedge failures [30, 25]. Within the granites these planes are exfoliation joints; within the lowgrade metamorphic rocks these planes are at least one set of the well-developed

512 crenulation cleavage (Sierra Laguna Blanca rock avalanches) [30]. In the coarse-clastic sedimentary units discontinuities are bedding planes and exfoliation joints. These planes either intersect with the valley wall due to river incision (Cerro Zorrito rock avalanches) or breakaway/bedding planes are valley-wall parallel and connected by a gently inclined planar detachment surface cutting obliquely to the bedding with the slope (Tonco rock avalanches) [30]. These types of detachments are reminiscent of massive rock slope failures in the Canadian Cordillera [10, 9], the Apennines [56], and those described recently in great detail by Fauqué and Tchilinguirian [12]. Another factor controlling the occurrence of large landslides is the repeated activity of mountain-bounding faults which are characterized by inherent structural complexity resulting in numerous minor faults and fracture planes that significantly weaken the coherence of bedrock in the hanging walls. This type of weakening also seems to have favored catastrophic failure in other tectonically active regions [75, 58]. Common to all rock avalanches in massive rock slopes in NW Argentina is the topographic constraint that vertical relief contrasts between the top of the breakaway zone and the foot of the mountain front must exceed a threshold of 400 m. If this relief contrasts is less than 400 m, cliff collapses result in deposits with characteristics of both rock avalanches and block-glide rockslides [25, 30, 12]. This is a distal part of the deposit having typical facies of rock avalanches such as large chaotic blocks at the surface overlying a mass of completely disintegrated material at the bottom and a proximal part of the deposit composed of a strongly fractured block with undisturbed stratigraphy [30, 12]. 5.2. TIMING OF ROCK AVALANCHING The morphologies of landslide source areas and deposits indicate that in the piedmont setting, rock avalanches are significantly older than in the narrow valley setting. Typically, breakaway scarps and sliding surfaces are eroded or poorly preserved. The inference of an old age for these deposits is also supported by the degree of soil development on the avalanche deposits [64] and by cosmogenic nuclide dating of a series of 8 superposed rock-avalanche deposits on the western Sierra Laguna Blanca piedmont. Surfaces of these deposits have exposure ages between 152 +17/-24 ka and 431 +18/-26 ka, indicating an average recurrence interval of large gravitational mountain-front failures of ~28 ka for the last seven landslides (Figure 7) [28] whereas the first one occurred at least 100 ka earlier. Rock avalanche breakaways and deposits in narrow valleys are much better preserved, coinciding with the late Pleistocene and Holocene ages of the landslides [72, 29, 71]. These ages indicate clustering of events between 35 and 25 ka and after 5 ka. New ages of 7,500 ± 70 a ‘CAL BP’ and 10,870 ± 290 a ‘CAL BP’ for sediments (organic material from lake deposits and the El Paso Ash, respectively) stratigraphically related with the older rock-avalanche deposit at E.P. (Cerro Zorrito rock avalanches) given in this paper are either contradictory to the previously published age of ~ 30 ka of these sediments [72, 29, 71] or indicate that the lake existed significantly longer than believed [70]. The age of ~ 30 ka was based upon dating of carbonate bivalves. Because a hardwater effect on this bivalve age is suggested by a reversal of ages within the stratigraphic profile [72] and can be explained by the extensive Tertiary marl deposits

513 underlying directly the lake sediments, it is most probably too old. This is similar to other lake systems in the region which are underlain by carbonate bearing deposits and also show strong reversals and apparent ages too old by > 14 ka [47]. Therefore, we infer that our organic sediment AMS-14C age of 7,500 ± 70 a ‘CAL BP’, together with the newly defined age of the El Paso ash (10,870 ± 290 a ‘CAL BP’) overlying the older landslide at E.P., better dates the true age of the landslide-lake sediment sequence in Las Conchas valley. This is further supported by a preliminary 10Be exposure age of the breakaway scarp of the young E.P. rock avalanche of ~4.7 ka for that landslide [26] (Figure 8). Taking into account the time interval of 6700 ± 700 a between landslide damming at C.L. and the younger rock avalanche at E.P. given by Trauth et al. [70] and the stratigraphic relations between the old and the young rock avalanche at E.P., the old rock-avalanche deposit at E.P. and the young rock-avalanche at C.L. are about 10,500 to 12,500 a old (Figure 8). On the one hand in this dynamic narrow-valley setting it is likely that landslides older than 35 ka have occurred, however, with exceptions [12] they were most likely obliterated form the stratigraphic record. On the other hand it is difficult to estimate when the most recent rock avalanche occurred. The least altered morphology is observed for the deposit of the youngest rock avalanche in the Tonco valley (deposit II in Figure 5), a rock avalanche which formed the still existing lake at Brealito and a rock avalanche composed of granitic debris at Cerro Paranilla [29, 30] (see Figure 1b for locations). Although most young rock avalanches in the study area are overlain at least at small spots by either the Buey Muerto or the Alemanía ash (3.6 – 4.4 ka and < 3.6 ka old, respectively), this is not the case for these three deposits, indicating that they are most likely younger than 3.6 ka (Figure 8). However, they are obviously older than historic times. The best preserved rock avalanche scar in the study area - as indicated by preserved groove marks on the sliding plane – is that of the Brealito rock-avalanche [25, 29] which is partially covered by petroglyphs painted in this region in prehispanic time. 5.3. DYNAMIC BOUNDARY CONDITIONS Landslide deposits in the piedmont environment have a much higher preservation potential. This is because of the large distance between the mountain fronts, rockavalanche deposits and trunk streams. Higher discharge of rivers during wetter climatic conditions cannot affect the deposits or the stability of mountain fronts. Therefore it is likely that these environments are insensitive to climatic change. This interpretation is corroborated when landslide ages are compared to Andean [31] or global climatic records (Figure 7, [32]). In addition, the landslide ages do not correlate with any other climate-driven morphologic changes in the Argentine Andes [62]. Instead, ages of enhanced slope instability and fault activity at Sierra Laguna Blanca suggest that mountain-front stability is strongly influenced by tectonic activity [28]. As long as tectonic activity concentrated along the mountain-bounding fault rock avalanching took place with an average recurrence interval of ~28 ka. When deformation propagated away from the mountain into the piedmont region at ~130 ka, as indicated by the exposure age of the uplifted terrace deposit (Figure 7), no further mountain-front failures have occurred along that segment of the range. Mountain front oversteepening by active tectonics in piedmont regions of the Sierras Pampeanas is also suggested by

514 10

Be and 26Al exposure ages of rock avalanche deposits in the piedmont of the eastern mountain front of the Sierras de San Luis [16]. At that location the mountain-bounding fault is still active and two rock avalanches occurred in a small segment, one (Potrero de Leyes) at ~35 ka ago and the other one (Las Cañas) at ~60 ka ago. These ages show a similar recurrence interval of landsliding as the deposits at Sierra Laguna Blanca do. Rock-avalanching in narrow valleys is also bound to tectonically active mountain fronts and clusters along those segments where structural complexity exists [65, 30]. At these sites the reactivation of steeply dipping strike-slip faults as reverse faults resulted in an enhanced uplift of mountain-front segments causing extreme local relief contrasts [65]. However, in addition to tectonic destabilisation in the narrow valley environment, river incision along river crossings of reverse faults keeps pace with uplift contributing to valley-wall oversteepening. The effectiveness of this process is illustrated by the ages of rock avalanches in the narrow valleys of NW Argentina [29, 69]. They cluster in three time periods between 30 and 26 ka, ~ 11 ka and after 5 ka (Figure 8). These clusters of enhanced landsliding are coeval with increased humidity and runoff in this part of the Andes e.g. [76, 74, 20, 69, 4, 5, 57]. Although the late Pleistocene and Holocene humid periods (Figure 8) have been dated controversially by different methods [69, 4, 5], landslide ages correspond to wet phases irrespective of the method applied. With respect to the total number, the most important landslide clusters in

Figure 8. Correlation of landslide and lacustrine events in NW Argentina with tropical and subtropical South American wet episodes changed after [71]; data from [29, 26, 71, 70, 12] and this work (see text). Black horizontal bars denote landslide events. Lacustrine phases are indicated by grey rectangles. Estimated age of older rock avalanche at Villa Vil from [12]. Chronology of the Late Pleistocene wet periods has been controversial [4, 5, 69]. Here the landslide ages are compared to the timing given in [74, 76, 39, 20, 69, 57], however landslide ages also correspond to humid periods given by [4, 5]. See Figure 1b and 2 for localities.


valleys lie in the Las Conchas valley and in the Quebrada del Toro. Both valleys are characterized by large catchment areas of up to 19,800 km2 and important runoff [11, 41] originating in regions with elevations > 4000 m that were also affected by multiple late Pleistocene glaciations [22]. Glacial advances have be dated in the Cordillera Oriental at ca. 27,970 ± 190, 9,265 ± 370, and after 5,280 ± 200 a [78]. Higher runoff in the course of climate change would have resulted in enhanced scouring, undercutting, and landsliding along tectonically-preconditioned valley walls. Similar relations have been demonstrated for enhanced landsliding activity along the Rio Grande, New Mexicco, by Reneau and Dethier e.g. [54]. 5.4. TRIGGERING FACTORS Seismic shaking was interpreted to be the most likely trigger mechanism of rock avalanching in NW Argentina because all such collapses occurred along tectonically active mountain fronts [30]. Although steep tectonically inactive mountain fronts, with high relief contrasts comprising lithologies prone to large collapses, do exist in the study area, no such mountain front is related with rock-avalanche deposits. In addition, the largest rock-avalanche clusters exist either a) along mountain fronts with structural complexity where the kinematic change of Andean deformation is likely to have caused a build-up of stress along older reactivated faults and may ultimately have ruptured in stronger earthquakes (e.g. Cerro Zorrito slides) [30, 65], or b) along range segments where faults document important late Pleistocene offsets (e.g. Sierra Laguna Blanca slides) [29, 28]. The new sedimentologic data from Las Conchas valley and the tephrochronologic data from Tonco valley reported here support that interpretation. These data show that deformation of sandy lake layers and fault offsets occurred in the same period of time as rock-avalanching or that multiple landsliding took place coevally. Deformed sandy layers in lake sediments in Las Conchas valley have grain size distributions similar to sands liquefied during earthquakes in Japan [37] or to results of laboratory tests simulating liquefaction of sands during earthquakes in California [40] (Figure 4). In addition, some of these deformed layers lie in the continuation of overturned folds which by themselves lie in strike with reverse faults offsetting the lake deposits (Figure 3). Buckle folds in the lake sediments of the foot wall of these reverse faults and slump folds in their hanging wall, indicate that lake sediments were water saturated during fault offset. The deformation horizons can thus be interpreted as seismites as described previously in other tectonically active regions e.g. [24, 61, 60, 42]. This indicates that at least two earthquakes with surface rupture occurred during the 6700 ± 700 a long phase of this lake along the mountain-bounding fault of Cerro Zorrito. The young E.P. rock avalanche of Cerro Zorrito occurred within the same time frame as these tectonic events. This situation is similar to landslide- and liquefactionfeature generation in the New Madrid seismic zone [52, 33] or in the Ottawa valley, Canada [3], where liquefaction features and voluminous landslides could be dated to have occurred in the same range of time. Although landslides and liquefaction features are significantly better constrained temporarily in the New Madrid seismic zone, rock avalanching and liquefaction at Cerro Zorrito is inferred to have originated in the same way, i.e. by strong seismic activity.

516 In the Tonco valley, the outcrops identified recently and the new tephrochronologic data given here, allow to relate contemporaneous rock-avalanche and rock-fall deposits to a single event. The unredeposited character of the Cerro Paranilla Ash at all sites, where it lies in direct stratigraphic contact below landslide deposits, indicates the contemporaneous age of landsliding because air-fallen tephra deposits are easily eroded by wind- or water-driven processes which result in detritic contamination of tephra. In addition, the purity of the tephra also indicates that strong rainfall could not have produced the multiple contemporaneous landslides because rainfall-driven redeposition of a few dm thick tephra is likely to occur before collapse of a 50-m-thick bedrock block. Hence, in correspondence to other places in tectonically active mountains where multiple coeval landsliding was interpreted to indicate a strong earthquake as common trigger mechanism [e.g., 50, 1, 59], these stratigraphic relations indicate that seismicity is the only likely trigger mechanism for the coeval landslides although they are still poorly defined temporally. However, unredeposited tephra layers better confine the coeval age of landslides than most Quaternary dating techniques because redeposition is likely to occur during the first strong rain-fall event and therefore in most climate regions (as for example the present NW Argentina) within a time span significantly smaller than uncertainties of most radiometric or other methods basing upon geochemical or physical analyses. In addition, the conclusion of seismic triggering is further corroborated by topographic fingerprints such as detachments high above valley bottoms and close to the ridge crests [7] which also in the Tonco valley suggest that seismic shaking is the most likely trigger mechanism. 5.5. MAGNITUDE OF TRIGGERING PALAEOEARTHQUAKES It is difficult to estimate the magnitude of earthquakes triggering landslides which are several thousand years old using empirical relations of regional landslide distribution versus magnitude of triggering event as given by Keefer [36]. Small landslides are not preserved in the geologic record, and the contemporaneousness of landslides is difficult to be tested if stratigraphic marker horizons are absent or datable material cannot be found. A better approach allowing an estimate of minimum magnitude is the use of empirical data of total landslide volume versus magnitude of triggering event also given by Keefer [35]. This relation indicates that the coeval landslides in the Tonco valley and the young E.P. rock avalanche of Cerro Zorrito were triggered by earthquakes of magnitudes M 7.1 and M 7.5, respectively. Presuming paleoearthquakes as triggers for the Sierra Laguna Blanca rock avalanches, these had magnitudes of M ~7.5. However, except at the southern end of the Cumbes Calchaquí mountain front (Figure 1) instrumental records of earthquakes do not show events > M 4 along the collapsed mountain fronts for the last 38 years [8] nor have there been reports of earthquakes with intensities > MM 6 in historic records of the last ~ 200 a [6] along most of these ranges. Indeed, faults with large offsets in the late Pleistocene indicate strong tectonic activity [29, 28]. However, at these sites it was not possible to determine fault offsets which occurred during single palaeoearthquake events. In addition, mountain-bounding reverse faults are strongly segmented [66, 19] and lie between basement or Mesozoic strata and Neogene basin-fill deposits. The only place where displacements by a single event can be estimated lies 30 km SSW of Cerro Zorrito. At that site, the Las Bañadas

517 fault offsets Holocene alluvial fan deposits on a segment > 4 km long with a maximum displacement of ~5 m [66]. Compared to empirical data [77, 67] these offsets also indicate related earthquakes with magnitudes > M 7 or > MS 6.5, respectively. The absence of rock-avalanches along mountain fronts which are also lithologically and topographically preconditioned for large failures but, which in contrast to the collapsed mountain fronts are not characterized by Quaternary tectonic activity and also are separated from the active faults by a few tens of km [30], suggests that epicentres of triggering earthquakes were shallow and located close to the location of mountain collapses. Under these conditions topographic effects such as concavity are likely to result in an amplification of earthquake waves and concentration of landsliding along ridge crests as seen in relation with the 1974 Guatemala earthquake [21] or modelled by computer for Central Asian earthquake-triggered landslides [23]. In addition, shallow underground nuclear explosions, with local surface effects similar to earthquakes of magnitudes 5–6 (V.V. Adushkin, Personal Communication, 2002) have also triggered landslides of volumes up to 0.1 km3 [2] making shallow epicentre earthquakes of magnitudes M 5-6 likely as triggers of the NW Argentine rock avalanches. Moreover, such magnitudes have been documented as necessary to produce deformation features such as the seismites in Las Conchas valley [51]. In conclusion, earthquakes which plausibly triggered rock avalanches in NW Argentina were either crustal and of a magnitude > M 7 or they were very shallow and of a magnitude > M 5.5. Both types of earthquakes have not been observed instrumentally along these ranges for the past 38 years. Only at two localities landslide deposits coincide with historic earthquakes (Figure 1 a and b) which might have had a magnitude of M 5.5, however during these historic earthquakes no rock avalanches were triggered. Hence, our data on palaeoearthquakes suggest that earthquake hazard in these valleys and basins is higher than expected from historical and instrumental records.

Acknowledgements The authors would like to thank A. Alonso for logistical support in Argentina, U. Bastian for help with grain size analyses, M. Trauth and M. Strecker for fruitful discussions and G. Borm and B. Merz for their encouragement and help. Supported by the GeoForschungsZentrum Potsdam and the Graduate College grant 450 to Hermanns. We gratefully acknowledge thorough reviews by E.G. Stevens and A. Strom. References 1. 2. 3. 4.

Adams, J. (1981) Earthquake-dammed lakes in New Zealand, Geology 9, 215-219. Adushkin, V.V. (2000) Explosive initiation of creative processes in nature, Comb. Expl. and Shock Waves 36, 695-703. Aylsworth, M., Lawrence, D.E. and Guertin, J. (2000) Did two massive earthquakes in the Holocene induce widespread landsliding and near-surface deformation in part of Ottawa Valley, Canada?, Geology 28, 903-906. Baker, P.A., Rigsby, C.A., Seltzer, G.O., Fritz, S.C., Lowenstein, T.K., Bacher, N.P. and Veliz, C. (2001a) Tropical climate changes at millennial and orbital timescales on the Bolivian Altiplano, Nature 409, 698-701.

518 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. 16. 17. 18. 19. 20. 21. 22. 23. 24. 25. 26. 27. 28.

Baker, P.A., Seltzer, G.O., Fritz, S.C., Dunbar, R.B., Grove, M.J., Tapia, P.M., Cross, S.L., Rowe, H.D. and Broda, J.P. (2001b) The history of South American tropical precipitation for the past 25,000 years, Science 291, 640-643. Castano, D.E. (1997) Teremotos históricos, sismicidad y tectónica en el noroeste Argentino, VII Congreso Geológico chileno. Antofagasta, 665-669. Densmore, A.L. and Hovius, N. (2000) Topographic fingerprints of bedrock landslides, Geology 28, 371-374. Engdahl, E.R., van der Hilst, R. and Buland, R. (1998) Global teleseismic earthquake relocation with improved travel times and procedures for depth determination, Bull. Seis. Soc. Am. 88, 722-743. Evans, S.G. (1989) Rockavalanche run-up record, Nature 340, 271. Evans, S.G., Aitken, J.D., Wetmiller, R.J. and Horner, R.B. (1987) A rock avalanche triggered by the October 1985 Nahami earthquake, District of Mackenzie, N.W.T., Can. J. Earth Sci. 24, 176-184. EVARSA (Evaluación de Recursos Sociedad Anonima) (1991) Estadística Hidrológica (vols. 1, 2), Buenos Aires, Secretaría de Energía, pp. 1-651. Fauqué, L. and Tchilinguirian, P. (2002) Villavil rockslides, Catamarca Province, Argentina in Evans, S.G. and DeGraff, J.V. (eds.), Catastrophic rockslides: Effects, occurrence, and mechanisms, Geological Society of America Reviews in Engineering Geology 15, 303-324. Fauqué, L. and Strecker, M.R. (1988) Large rock avalanche deposits (Sturzstrome, sturzstroms) at Sierra Aconquija, northern Sierras Pampeanas, Argentina, Ec. Geol. Helv. 81, 579-592. Fox, A.N. and Strecker, M.R. (1991) Pleistocene and modern snowlines in the Central Andes (2428°S), Bamb. Geogr. Schriften 11, 169-182. González Bonorino, F. (1951) Descripción Geológica de la Hoja 12e, Aconquija (CatamarcaTucumán), Dirección Nacional de Minería, pp. 50. González Diaz, E.F., Costa, C., Giaccardi, A., Bierman, P.E. and Caffee, M. (1999) Edades de cosmonuclídeos 10Be y 26Al en avalanchas de rocas de la Sierra de San Juan: Implicancias paleosismológicas, XIV Congreso Geológico Argentino. Salta, 81-82. González, O.E., Hongn, F.D. and Mon, R. (1991) Estructura de la Sierra Laguna Blanca y zonas aledañas, Provincia de Catamarca, Rev. de la Asoc. Geol. Arg. 46, 299-308. Grier, M.E. (1990) The influence of the Cretaceous Salta rift basin on the development of Andean structural geometries, NW Argentine Andes [PhD thesis], Faculty of the Graduate School, Cornell University, pp. 178. Grier, M.E., Salfity, J.A. and Allmendinger, R.W. (1991) Andean reactivation of the Cretaceous Salta rift, northwestern Argentina, J. of S. Am. Earth Sci. 4, 351 - 372. Grosjean, M., Valero-Garcés, B.L., Geyh, M.A., Messerli, B., Schotterer, U., Schreier, H. and Kelts, K. (1997) Mid and late Holocene limnogeology of the Laguna del Negro Francisco, northern Chile, and its paleoclimatic implications, The Holocene 7, 151-159. Harp, E.L., Raymond, C.W. and Wieczorek, G.F. (1981) The Guatemala earthquake of February 4, 1976 - Landslides from the February 4, 1976, Guatemala earthquake, Geol. Surv. Prof. Pap. 1204-A, 135. Haselton, K., Hilley, G. and Strecker, M.R. (2002) Average Pleistocene climatic patterns in the southern Central Andes: Controls on mountain glaciation and paleoclimate implications, J. of Geol. 110, 211-226. Havenith, H.-B., Jongmans, D., Abdrakmatov, K., Trefois, P., Delvaux, D. and Torgoev, I.A. (2000) Geophysical investigations of seismically induced surface effects: case study of landslides in the Suusamyr valley, Kyrgyzstan, Surv. in Geophys. 21, 349-369. Hempton, M.R. and Dewey, J.F. (1983) Earthquake-induced deformational structures in young lacustrine sediments, East Anatolian Fault, southeast Turkey, Tectonophys. 98, T7-T14. Hermanns, R.L. (1999) Spatial-temporal distribution of mountain-front collapse and formation of giant landslides in the arid Andes of northwestern Argeintina (24-28° S, 65-68° W) [PhD thesis], Mathematisch-Naturwissenschaftliche Fakultät, Universität Potsdam, pp. 123. Hermanns, R.L., Niedermann, S., Ivy-Ochs, S., Kubik, P.W. and Strecker, M.R. (2002a) Paleoseismic triggering of multiple paleolandslides in the NW-Argentine Andes, EGS. Nice. Hermanns, R.L., Niedermann, S., Strecker, M.R., Trauth, M.H., Alonso, R.N. and Fauque, L. (2002b) Prehistoric rock avalanches in the NW-Argentine Andes: boundary conditions and hazard assessment, 1st European Conference on Landslides. Prague, 199-205. Hermanns, R.L., Niedermann, S., Villanueva Garcia, A., Sosa Gomez, J. and Strecker, M.R. (2001) Neotectonics and catastrophic failure of mountain fronts in the southern intra-Andean Puna Plateau, Argentina, Geology 29, 619-623.

519 29. 30. 31. 32.

33. 34. 35. 36. 37. 38. 39. 40. 41. 42. 43.

44. 45. 46. 47. 48. 49. 50. 51. 52.

Hermanns, R.L., Trauth, M.H., Niedermann, S., McWilliams, M. and Strecker, M.R. (2000) Tephrochronologic constraints on temporal distribution of large landslides in northwest Argentina, J. of Geol. 108, 35-52. Hermanns, R.L. and Strecker, M.R. (1999) Structural and lithological controls on large Quaternary rock avalanches (sturzstroms) in arid northwestern Argentina, Geol. Soc. of Am. Bull. 111, 934-948. Hooghiemstra, H. (1995) Environmental and Paleoclimatic Evolution in Late Pliocene-Quaternary Colombia in Vrba, E.S., Denton, G.H., Partridge, T.C. and Burckle, L.H. (eds.), Pateoclimate and Evolution, with Emphasis on Human Origins, Yale University Press., 249-261. Imbrie, J., Boyle, E.A., Clemens, S.C., Duffy, A., Howard, W.R., Kukla, G., Kutzbach, J., Martinson, D.G., McIntyre, A., Mix, A.C., Molfino, B., Morley, J.J., Peterson, L.C., Pisias, N. G., Prell, W.L., Raymo, M.E., Shackelton, N.J. and Toggweiler, J.R. (1992) On the structure and origin of major glaciation cycles: 1. linear responses to Milankovitch forcing, Paleoceanogr. 7, 701-738. Jibson, R.W. and Keefer, D.K. (1993) Analysis of the seismic origin of landslides: Examples from the New Madrid seismic zone, Geol. Soc. of Am. Bull. 105, 521-536. Jordan, T.E., Isacks, B.L., Allmendinger, R.W., Brewer, J.A., Ramos, V.A. and Ando, C.J. (1983) Andean tectonics related to geometry of subducted Nazca plate, Geol. Study of Am. Bull. 94, 341-361. Keefer, D.K. (1994) The importance of earthquake-induced landslides to long-term slope erosion and slope-failure hazards in seismically active regions, Geomorph. 10, 265-284. Keefer, D.K. (1984) Landslides caused by earthquakes, Geol. Soc. of Am. Bull. 95, 406-421. Kishida, H. (1970) Characteristics of liquefaction of level sandy ground during the Tokachioki earthquake, Soils and Found. 10, 103-111. Kleinert, K. and Strecker, M.R. (2001) Climate change in response to orographic barrier uplift: paleosol and stable isotope evidence from the late Neogene Santa Maria basin, northwestern Argentina, Geol. Soc. of Am. Bull. 113, 728-742. Ledru, M.P., Braga, P.I.S., Subiès, F., Fournier, M., Martin, L., Suguio, K., and Turcq, B. (1996) The last 50,000 years in the Neotropics (Southern Brazil) evolution of vegetation and climate, Palaeogeogr. Palaeoclimatol. Palaeoecol. 123, 239-257. Lee, K.L. and Fitton, J.A. (1969) Factors affecting the cyclic loading strength of soil, Am. Soc. for Test. Mat. Spec. Tech. Pub. 450, 71-95. Malamud, B.D., Jordan, T.E., Alonso, R.A., Gallardo, E.F., Gonzalez, R.E. and Kelley, S.A. (1996) Pleistocene lake Lerma, Salta Province, NW Argentina. XIII Congreso Geológico Argentino y III Congreso de Exploración de Hidrocarburos. Buenos Aires, 103-114. Marco, S., Stein, M. and Abnon, A. (1996) Long-term earthquake clustering: A 50,000-year paleoseismic record in the Dead Sea Graben, J. Geophys. Res. 101, 6179-6191. Marrett, R.A., Allmendinger, R.W., Alonso, R.N. and Drake, R.E. (1994) Late Cenozoic tectonic evolution of the Puna Plateau and adjacent foreland, northwestern Argentine Andes, J. of S. Am. Earth Sci. 7, 179-207. Marrett, R. and Allmendinger, R.W. (1990) Kinematic analysis of fault-slip data, J. of Struc. Geol. 12, 973-986. Mon, R. (1976) The structure of the eastern border of the Andes in north-western Argentina, Geol. Rundschau 65, 211-222. Morgan, G.B. and London, D. (1996), Optimizing the electron microprobe analysis of hydrous alkali aluminosilicate glass, American Mineralogist 81, 1176-1185. Müller, A.B. (2001) Die Änderungen von Niederschlagsverteilungen und deren Einfluß auf die Provenienz von Seesedimenten während der letzten 30.000 Jahre in den nordwestargentinischen Anden [Dipl thesis], Institut für Geowissenschaften, Potsdam, pp. 80. Niedermann, S. (2000) The 21Ne production rate in quartz revisited, Earth Planet. Sci. Lett. 183, 361364. Niedermann, S., Graf, T., Kim, J.S., Kohl, C.P., Marti, K. and Nishiizumi, K. (1994) Cosmic-rayproduced 21Ne in terrestial quartz: the neon inventory of Sierra Nevada quartz separates, Earth Planet. Sci. Lett. 125, 341-355. Nikonov, A.A. and Shebalina, T.Y. (1979) Lichenometry and earthquake age determination in central Asia, Nature 280, 675-677. Obermeier, S.F. (1996) Using liquefaction-induced features for paleoseismic analysis in McCalpin, J. P. (ed.), Paleoseismology, Academic Press, 331-396. Obermeier, S.F., Jacobson, R.B., Smoot, J.P., Weems, R.E., Gohn, G.S., Monroe, J.E. and Powars, D.S. (1990) Earthquake-induced liquefaction features in the coastal setting of South Carolina and fluvial setting of the New Madrid seismic zone, U.S. Geol. Surv. Prof. Pap. 1504, 1-44.

520 53. 54. 55. 56. 57. 58. 59. 60. 61. 62. 63. 64. 65. 66. 67. 68. 69. 70. 71. 72. 73. 74. 75. 76. 77. 78.

Rapela, C.W. (1976) El basamento metamórfico de la region de Cafayate, Provincia de Salta. Aspectos petrológicos y geoquímicos, Rev. de la Asoc. Geol. Arg. 21, 203-222. Reneau, S.L. and Dethier, D.P. (1996) Late Pleistocene landslide-dammed lakes along the Rio Grande, White Rock Canyon, New Mexico, Geol. Soc. of Am. Bull. 108, 1492-1507. Salfity, J.A. and Marquillas, R.A. (1994) Tectonic and Sedimentary Evolution of the CretaceousEocene Salta Group Basin, Argentina in Salfity, J.A. (ed.), Cretaceous tectonics of the Andes, Friedrich Vieweg und Sohn, 266-315. Scarascia-Mugnozza, G., Bianchi-Fasani, G., Esposito, C., Martino, S., Saroli, M., Di Luzio, E., and Evans, S.G. (2003). Rock avalanche and mountain slope deformation in a convex dip slope - the case of the Maiella Massif, Central Italy. This volume. Schäbitz, F., (1999) Vegetation and climate history on the eastern flank of the Sierra de Santa Victoria, Jujuy Province, NW-Argentina (first results), Zbl. Geol. Paläont. Teil I, 7/8, 969-984. Scheidegger, A.E. (1998) Tectonic predesign of mass movements, with examples from the Chinese Himalaya, Geomorph. 26, 37-46. Schuster, R.L., Logan, R.L. and Pringle, P.T. (1992) Prehistoric rock avalanches in the Olympic Mountains, Washington, Science 258, 1620-1621. Scott, B. and Price, S. (1988) Earthquake-induced structures in young sediments. Tectonophys. 147, 165-170. Seilacher, A. (1984) Sedimentary structures attributed to seismic events, Marine Geol. 55, 1-12. Siame, L.L., Bourlès, D.L., Sébier, M., Bellier, O., Castano, J.C., Araujo, M., Perez, M., Raisbeck, G.M. and Françoise, Y. (1997) Cosmogenic dating ranging from 20 to 700 ka of a series of alluvial fan surfaces affected by the El Tigre fault, Argentina, Geology 25, 975-978. Sobel, E.R. and Strecker, M.R. (2002) Contrasting histories from vertical profiles in the Calchaquies and Aconquija ranges, northwest Argentina. Geotemas 4, 139-142. Strecker, M.R. (1987) Late Cenozoic landscape development, the Santa Maria valley, northwestern Argentina [PhD thesis], Faculty of the Graduate School, Cornell University, pp. 262. Strecker, M.R. and Marrett, R.A. (1999) Kinematic evolution of fault ramps and role in development of landslides and lakes in northwestern Argentine Andes, Geology 27, 307-310. Strecker, M.R., Cerveny, P., Bloom, A.L. and Malizzia, D. (1989) Late cenozoic tectonism and landscape development in the foreland of the Andes: Northern Sierras Pampeanas (26°-28°S), Argentina, Tectonics 8, 517-534. Strom, A.L. and Nikonov, A.A. (1997) Relations between the seismogenic fault parameters and earthquake magnitude. Phys. of the Solid Earth 33, 1011-1022. Stuiver, M., Reimer, P.J. and Braziunas, T.F. (1998) High-precision radiocarbon age calibration of terrestrial and marine samples, Radiocarbon 40, 1127-1151. Sylvestre, F., Servant, M., Servant-Vildary, S., Causse, C., Fournier, M., and Ybert, J.P. (1999) Lakelevel chronology on the southern Bolivian Altiplano (18° - 23° S) during Late-Glacial time and Early Holocene, Quat. Res. 51, 54 -66. Trauth, M.H., Bookhagen, B., Müller, A.B. and Strecker, M.R. (2003) Late Pleistocene climate change and erosion in the Santa Maria basin, NW Argentina, J. Sed. Res. 73, 82-90. Trauth, M.H., Alonso, R.A., Haselton, K.R., Hermanns, R.L. and Strecker, M.R. (2000) Climate change and mass movements in the Argentine Andes, Earth Planet. Sci. Lett. 179, 243-256. Trauth, M.H. and Strecker, M.R. (1999) Formation of landslide-dammed lakes during a wet period between 40,000 and 25,000 yr BP in NW Argentina, Palaeogeo. Palaeocl. Palaeoecol. 153, 277-287. Turner, J.C.M. (1973) Descripción geológica de la hoja 11d, Laguna Blanca, Provincia de Catamarca, Dirección Nacional de Geología y Minería Boletin, pp. 71. van der Hammen, T. and Absy, M.L. (1994) Amazonia during the last glacial, Palaeogeo. Palaeocl. Palaeoecol. 109, 247-261. Weidinger, J.T., Schramm, J.-M. and Surenian, R. (1996) On prepatory causal factors, initiating the prehistoric Tsergo Ri landslide (Langthang Himal, Nepal), Tectonophys. 260, 95-107. Wirrmann, D., and Mourguiart, P. (1995) Late Quaternary spatiotemporal limnological variations in the Altiplano of Bolivia and Peru, Quat. Res. 43, 344-354. Wells, D.L. and Coppersmith, K.L. (1994) New empirical relationships among magnitude, rupture length, rupture width, rupture area, and surface displacement, Bull. Seis. Soc. of Am. 84, 974-1002. Zipprich, M., Reizner, B., Zech, W. Stingl, H., and Veit, H., (1999) Upper Quaternary landscape and climate evolution in the Sierra Santa Victoria (nort-western Argentina) deduced from geomorphologic and pedologic evidence, Zbl. Geol. Paläont Teil I, 7/8, 180-181.

ROCK AVALANCHES WITH COMPLEX RUN OUT AND EMPLACEMENT, KARAKORAM HIMALAYA, INNER ASIA K. HEWITT1 Cold Regions Research Center and Department of Geography and Environmental Studies, Wilfrid Laurier University, Waterloo, Ontario, Canada, N2L 3C5

Abstract The paper concerns rock avalanches and their deposits reflecting strong interactions with rugged terrain, deformable and wet substrates. Of 186 rockslide-rock avalanche events identified in the Karakoram Himalaya, over 160 show singularities related to run out in rugged terrain. Twelve record complex interactions with substrate materials over large areas of intermontane basins. Other complications arise from travel on to glaciers. 'Complexity' appears in diverse and irregular plan forms recording divergence and splitting of debris streams into multiple lobes. Deposits often have highly asymmetrical long and cross profiles, including Heim's brandung or 'surge' forms against opposing slopes. Morphological complexity is reflected in deposit facies and fabric. Rock avalanches encountering erodible and deformable substrate developed extensive longitudinal ('digitate') and transverse ridges. Substrates display complicated folding, faulting and dislocations ('landslide-tectonized' forms). Substrate material penetrate the rock avalanches in multiple dykes, flame, diapir and injection structures. Some debris streams disintegrated to a 'chaos' of substrate and rock avalanche materials. Field diagnostics are emphasised due to a history of misidentification involving, especially, features associated with complex run out. Diagnosis is further complicated by post-emplacement erosion and burial, associated with massive disturbance of Holocene landform development, notably by large rock avalanche dams of which over 90 are identified. 1. Introduction The catastrophic mass movements of interest here belong to the class of rock slide-rock avalanches. They originate in massive rock wall failures. The main bodies of their deposits record the high-speed runout of fractured, crushed and pulverised bedrock [20, 34]. A large-scale unity of emplacement reflects the single catastrophic event. However, the focus is on distinctive and diverse 'styles' of event that result from their interactions with rugged terrain or deformable substrates [26, 30]. In the world’s highest and most rugged mountains, especially, terrain has a major influence on the run out and 1

E-mail: [email protected]

521 S.G. Evans et al. (eds.), Landslides from Massive Rock Slope Failure, 521–550. © 2006 Springer. Printed in the Netherlands.

522 emplacement of rock avalanches. Where there are extensive intermontane sediments, important modifications arise from interactions of rock avalanches from adjacent mountain walls, with deformable or wet substrates. Travel and deposition over or around ice masses can further complicate rock avalanche behaviour, and our ability to recognition of past events in the landscape. Terrain interference, and rock avalanche responses to it, are evident in a wide range of emplacement morphologies, some on scales comparable to the classic 'elements'. While these overall complexities are the largest in scale, other features that may be relatively small occur in large numbers, including some that are easily mistaken for products of other processes. In addition, distinctive sedimentary features arise in the rock avalanche debris itself, in substrate and rim materials, and through inter-penetration or mixing of rock avalanche and entrained material. Some or all of these phenomena are identified in rock avalanches with 'complex run out and emplacement' (Figure 1).


Figure 1. Rock avalanches with complex emplacement: plan forms and longitudinal cross-sections of selected examples form the Karakoram Himalaya. In the plan views, transverse and longitudinal ridges or thickening in excess of 10m vertical relief are suggested schematically by shading. Haldi event, Hushe-Saltoro valley junction, Eastern Karakoram, Baltistan [27, 30], Dulung-Bar-Darkot event, Gilgit River, Western (see Figures 3,4,5), Naltar Lakes event, Hunza River basin, Central Karakoram (see Figure 10) [28, 29], Gol-Ghone 'A' and ''B' events, Indus valley, Baltistan, Central Karakoram (see Figure 9) [26].

In what may be described as the basic or 'classic' case just two universally present elements are considered – massive rock wall failure and the rock avalanche. The first involves a detachment zone, where a large mass of in situ and more or less intact bedrock is released, slides or collapses down the source slope. The second element involves high speed run out and emplacement of thoroughly disaggregated bedrock, sometimes called the sturzstrom or flow-slide [32, 46]. Hungr [33] refers to these as the main 'geometric elements'. The same elements are identified in the 'intermediate-mobility rock avalanche' or Type B of Nicoletti and Sorriso-Valvo [41], and Strom’s [50] 'primary'

524 case 1a – "expanding over an unbounded surface". In these models, and examples of the classic case, rock avalanche debris is emplaced as a single, thin and tongue-like deposit. It is much greater in length than width, and has little variation in thickness or surface relief. Little or no erosion of the underlying surface is involved. Some studies recognise a third element, a transport or 'transition' zone linking, but distinct from, the detachment and deposition zones [6]. All events must go through this transition phase, which involves major changes in the geometry, composition and dispersal of the original failed bedrock mass. However, some or all of the changes are, indeed, transitions that may be inferred from properties of the other two elements but are not themselves apparent in the post-deposition landscape. It will be suggested later that when events are strongly modified by terrain, some of the complexities record transitional developments that 'stall' or are incomplete. Nicoletti and Sorriso-Valvo [41] and Strom [50] recognise additional morphological types related to topographic constraints, the former emphasising plan form, the latter longitudinal distribution of deposit mass. The examples and evidence discussed here, from the Upper Indus Basin, include each of the situations identified by these researchers. In many cases most or all occur in different parts or debris streams of the same events. Moreover, further, large scale or singular developments are found in the 'complex run out and emplacement' events discussed. They suggest the need for further 'elements' or styles of event associated with distinctive plan forms, distributions of deposit mass, internal features, and interactions with substrates and surrounding terrain. They are distinct from, but may occur with, rock avalanches that are transformed into debris avalanches or debris flows. The latter occur when sediments and/or moisture from the run out path, are entrained in quantities sufficient to alter the basic mechanics of movement and the nature of the deposits [1, 36, 44]. They are 'compound' as well as complex events. The examples described here are from the Karakoram Himalaya, Inner Asia, where complex emplacement has occurred in the great majority of known rockslides – rock avalanches. The ability to recognise complex features has been essential in developing an adequate inventory of massive rock wall failures in the region and the scale of individual events. This, in turn is necessary to assess their role in development of the high mountain landscape and the risks such events pose to human communities along the valleys of the transHimalayan Upper Indus. 2. Catastrophic Rock Slides in the Karakoram Himalaya To date, a total of 186 rockslide-rock avalanche events have been identified in the region (Figure 2). More than 160 exhibit singularities related to run out in rugged, high relief terrain. The deposits of twelve examples that travelled over intermontane basins are extensively modified by interactions with substrate materials. Many others are locally affected by this. Seven events, including the only twentieth century examples, descended onto glaciers. They display a combination of complex interactions with glacierised terrain and reworking or removal of the deposits by ice. On the basis of recent events, it seems likely that the largest number of rock avalanches has and will occur in the extensive glacier basins. These cover some 60 per cent of the Karakoram Range itself, and 20 per cent of the transHimalayan Upper Indus Basin. However, materials

525 deposited wholly in the glacio-nival environment are reworked and removed by the ice in a few years [18]. The recognition of ice margin remnants, or deposits of glacially modified rock avalanche materials, is in its infancy and no more will be said about it here. Most events discovered to date descended into and along stream valleys in the zone where permanent settlements occur. The valleys were formerly glaciated, and the rock avalanches seem to derive mainly from glacially over-steepened rock walls. Reconstruction of ice heights and movements during the last glacial maximum suggests that debutressing, associated with deglaciation, was an important factor preparing rock slopes for these massive failures [3]. In accessible cases, features related to gravity-induced, post-glacial rock creep and fracture, or sackungen, were seen around the head walls of the detachment zones [28, 29]. All the events exceeded 10 million cubic meters in volume, those discussed here more than 200 and some over one billion. Vertical displacements, from the head of the detachment zone to the farthest run out of debris, were at least 1000 m, in some more than 2000 m. Source slopes are in excess of 40º, often over 60º. Maximum horizontal displacements exceeded 6 km, in some cases more than 10 km [26-28]. Rockslide - rock avalanches were found in every elevation zone between 1,800 and 7,200 m asl. and in all geological terraines. They are divided almost equally among igneous, metamorphic and metasedimentary rock types [29]. Emphasis is given to field identification partly because of a history of misidentification. More than fifty Karakoram rock avalanches were formerly classified as glacial deposits, many of the features discussed here being assigned to glacial action [27]. As Heim [21] showed in the European Alps, it is usually the complex run out and emplacement features that are attributed to other processes. Meanwhile, many of the rock avalanche deposits have been subject to considerable erosion or burial. The remnants that survive, or are more readily identified in the landscape, most often consist of materials thickened, strengthened or protected by interaction with rugged terrain or deformable substrates. Erosion and burial can also destroy continuous or readily traced connections between parts or all of a rock avalanche deposit and its detachment zone. With complex run out the two are often not inter-visible. These conditions add to the general difficulties of deposit identification in rugged terrain, especially since there are many other and varied forms of coarse, poorly or unsorted sediments, including reworked rock avalanche debris. 2.1. ILLUSTRATIVE CASE STUDY I: DULUNG BAR - DARKOT This rock avalanche (Table 1; Figure 3) originated in collapse of a glacially oversteepened spur at around 4.400 m asl. in the upper, hanging valley of the Dulung Bar. The main slide surfaces are bedding planes in metasedimentary limestones of the Asian Plate, Permo-Carboniferous Darkot Group [49]. These dip steeply towards the valley but are undercut below, an example of the 'overdip slope' of Cruden [8]. On impact with an ice-free valley floor, the debris split into several rock avalanche streams. Those, which moved up and across the valley are recorded in 'digitate' debris ridges, and impound the high part of Dulung Bar (Figure 4).


Figure 2. Location map and distribution of rock slide-rock avalanches identified in the Karakoram Himalaya to date.

527 Table 1. Parameters of the Dulung Bar-Darkot rockslide - rock avalanche. Name Location Valley Mt. range Source lithology Orientation of the initial slope failure

Dulung Bar 36º 40' N; 73º 25' E Darkot (Gilgit) Hindu Raj carbonate (metasediments) of the Permo-Carb. Darkot Group North

Runout dimensions: Area of deposit (exposed now / estimated original) Volume of deposit (estimated now / estimated original) Elevation (Highest / Lowest) Maximum vertical drop, H Maximum horizontal travel, L Fahrboschung (angle) Coefficient of friction H/L Highest run up, h

4.2 / 5.5 km2 250 million / 400 million m3 4,600 / 2,700 m asl. 1.900 m 8.2 km 13º 0.23 230 m

Figure 3. The Dulung Bar-Darkot complex rock avalanche event. Landslide survey superimposed on magnified portion of the 1:250,000 topographic map [52].

528 The main volume travelled into and down the steep, narrow outlet from the hanging valley, a gorge that descends 600m from the upper hanging valley to the main Darkot River. 'Caroming' flow, as the debris moved down the gorge, is recorded in ridges of material ramped up alternately on one flank and the other, and in trim-lines as much as 100m above the floor. There is a gap of more than 1 km between the upper and lower segments of the deposit. Some it is due to removal by erosion, but the outer mass may have accelerated through the narrowest, steepest part of the gorge to become detached from the upper part. This kind of topography-induced separation in narrow canyons has been observed in other Inner Asian ranges (Strom, Personal Communication, 2002). Five kilometres from its source, the rock avalanche entered Darkot valley at right angles spreading to form a cone of debris more than 100m thick in its central part. At depth the deposit consists of thoroughly crushed carbonate, like white 'rock meal', exposed where the debris fan is excavated and segmented by the Darkot and Dulung Bar streams. However, undisturbed areas of the surface are covered in megaclasts, including large, intact units of bedded limestone (Figure 5). Moving directly out of the tributary canyon the material created a high, cross-valley ramp, and climbed about 150m up the opposing east slope. Here, the deposit formed a barrier, which dammed the upper Darkot valley to a depth of over 100m. The lake has now drained, but breaching occurred across the rock spur on the east flank, to form a superimposed or 'epigenetic' rock gorge – a typical post-emplacement development where rock avalanches have dammed Upper Indus streams [27]. The original valley remains buried under undisturbed rock avalanche material. The lowest reach of the event, towards the site of Darkot village, is lost to burial and erosion, but remnants were found to three kilometres beyond the Dulung Bar junction, recording a total run out of at least 8 km. About 200 million m3 of debris was deposited in the Darkot valley, while the total volume exceeded 400 million. If not the most complex, the deposits illustrate a range of distinctive terrain-related features in a single event. Using the language of Heim [20] for the Elm Event, this was a drama in five rather than three acts: the initial detachment and fall, the air launch or jump to the valley floor, and three rock avalanche 'acts', each involving complex run out and emplacement [32]. 'Act 3', in the upper valley, displays splitting into up-, down- and cross-valley lobes, and digitate emplacement of debris in longitudinal ridges. This is due mainly to the interaction with deformable and wet substrate sediments. There are also brandung features emplaced against the opposing valley wall. 'Act 5', represented by the lower, outer section, 4km distant, displays debouchement features at a valley junction with radial spreading over a sediment fan, development of the ‘Deformed Tshape' of Nicoletti and Sorriso-Valvo [41], and a further set of impact slope or brandung features. Between these two, 'Act 4' records the effects of transverse confinement, funneling or channelising of debris over 4 km through a (relatively) narrow, precipitous gorge. There are typical remnants of valley-side 'caroming flow' [15, 45], also called 'swash' features [53] or 'throwing' [50]. Both the upper and lower 'main' deposits impounded the valleys above, and several small tributaries were blocked. The subsequent history of the barrier on the Darkot illustrates a complicated pattern of breaching, segmentation and excavation of the cross-valley deposit.


Figure 4. The upper zone of the Dulung Bar-Darkot complex rock avalanche event. View of the valley deposits showing series of 'digitate' ridges of rock avalanche debris (arrows) below the detachment zone. The break out zone, with smooth slabs dipping steeply towards the valley, is clearly visible in the left background. Arrow indicates small 'shepherds' hut in the foreground approximately 7 x 10 x 2 m in size. Distributaries of the landslide dammed Dulung Bar streams meander from right to left through the photo and have partly buried the rock avalanche. The photo station is on debris that climbed the impact slope here.

2.2. CASE STUDY II: THE GOL-GHONE On the main stem of the Indus, in the gorge above Skardu Basin, Baltistan, are overlapping remnants of two prehistoric rockslides – rock avalanches (Table 2; Figure 1). They infixed the valley over a distance of more than 11 km. They formed barriers as much as 550m high that dammed the Indus for many decades at least, to judge from exposed lacustrine beds upstream, possibly for centuries. The rock wall failures occurred in a zone of mountain wall sagging and collapse some 1,200 m above the valley floor on its left/west flank. The older, more northerly, Gol-Ghone 'A' has the thicker and larger [29]. The debris of both events climbed the opposing impact slope to emplace the bulk of the deposits, or 'main accumulations' (Heim’s 'Hauptanlage'), against it. These are 400550 m thick masses of crushed bedrock. A ridge of rock avalanche debris (Heim’s 'Brandung'), as much as 50 m high and up to 720 m above present river level, records the highest run up of Gol-Ghone 'A'. The Indus has not yet cut down to the pre-landslide valley floor, so the maximum run up was greater still [12, 30] (Figure 6). Subsidiary lobes split off and travelled as much as 5-6 km down valley and 3-4 km up valley. In deposits of the latter, where the valley floor is more open, are preserved multiple longitudinal and transverse pressure ridges with surface relief up to 50m. Along the opposing wall down valley of the brandung, raised areas of rock avalanche

530 debris, plastered against the slope, decline progressively in height through the site of Ghone, they record the geometry of deflection of debris by valley walls as defined by Heim [21] and related 'caroming' flow. The result, in each case, is an extremely complex plan form and system of constructional landforms, further complicated by the overlapping of the two events and partial reworking of Gol-Ghone 'A' by 'B'.

Figure 5. The lower zone of the Dulung Bar-Darkot complex rock avalanche event. Debris issuing from the Dulung Bar tributary at the left buried and dammed the Darkot valley. R - undisturbed surface of deposit covered by megaclasts. X - section cut though crushed and pulverised carbonate debris by the Darkot River. E - 'epigenetic' gorge cut through spur buried by rock avalanche, to release former landslide-dammed lake behind the deposit. Table 2. Parameters of the Gol-Ghone rockslides - rock avalanches. Name Location Valley Source lithology Orientation of the initial slope failure

Gol-Ghone 'A' and 'B' 35º 16' N; 75º 50' E Indus (Baltistan) Plutonic/metamorphic of the Ladakh Pluton [49] East-North-East

Runout dimensions: Area of deposit (exposed now / estimated original) Volume of deposit (estimated now / estimated original) Elevation (Highest / Lowest) Maximum vertical drop, H Maximum horizontal travel, L Highest run up, h

12.5 / 25.0 km2 600 million / 1100 million m3 3,900 / 2,300 m asl. 1.600 m •7 km 740 m

River cliffs cut through the barrier to reveal full-depth cross-sections in the rock avalanche material. They show deformation, shearing and brecciation of debris that vary in type and intensity with depth, and with distance from the source slope (Figure 7).


Figure 6. The Gol-Ghone rock avalanche complex, Baltistan. View along the brandung ridge at the highest limit of climb up the valley wall opposite the detachment zone. I - indicates location of the Indus River, 750m below the photo station. R - pressure ridges in rock avalanche debris at the village of Gol left by debris streams traveling up-valley.

The lowest sections, some 100s meters below the rock avalanche surface exhibit extreme pulverising, smearing out, folding and micro shearing of original bedrock materials. Remarkably, the lithological units themselves deform but do not mix. They represent a complex variation of the phenomenon remnant stratigraphy, well known in much thinner and 'simple' rock avalanche deposits [38, 39, 24]. Where the base of the rock avalanche is exposed, many penetrative interactions with the substrate are visible. They include clastic dikes, 'stringers' and folds of finer grained alluvium incorporate into the rock avalanche material – but without mixing. They are typical features of Yarnold and Lombard’s [55] 'mixed basal zone' involving as much as the lower 10-15 m. Where somewhat thinner rock avalanche lobes, 10-15m thick, travelled up valley over pre-existing fill and river channels, some remarkable 'escape structures' pass right through them (Figure 8). They consist of 'pipes' or 'dykes' of coarse river gravels in a sandy matrix, including well-rounded boulders. Presumably these were forced up into, indeed, 'hosed' through, the rock avalanche by the great hydraulic pressures generated in the trapped moisture. Possibly they relate to 'pull-apart' openings Abele [2] hypothesised where a rock avalanche rides over mobilised valley fill. Of course, these debris sheets have no tensile strength so that, if large pressure differences develop locally beneath the debris sheet, rather than stretching they could produce short-lived openings, or Abele’s 'faulting', or perhaps allow water under high pressure to force its way through. Whatever the exact mechanism, these features indicate complex interactions with the substrate.


Figure 7. Crushed bedrock facies at depth in selected rock avalanches: a) Gol-Ghone 'B' debris exposed at 150m below the surface showing jigsaw and crackle breccia. Note the two distinct lithologies, light granitic rock to the light and darker, greenish intrusive to the left. These do not mix and responded somewhat differently to the crushing forces. b) Gol-Ghone 'B' debris exposed at 200m below the surface, showing complicated crushing, deformation, stretching and microshearing of bedrock, while original lithologies do not mix. The direction of movement was into the cliff face here and at a high angle up the opposing slope. c) Chalt rock avalanche, Hunza basin, central Karakoram, 40km north of Gilgit. Crushed and smeared out carbonate lithologies in the medial run out zone at a depth of 20m. The exposed thickness of this facies is up to 12m, and at least 10m below the rock avalanche surface. d) Haldi rock avalanche, Saltoro River mouth, Eastern Karakoram. Crushed bedrock facies in the medial run out zone at a depth of 25m and more, combing rock avalanche, entrained bedrock and 'stringers' of substrate smeared out together but not mixing.

Both landslides dammed the Indus as well as tributaries from the Gol and Ghone valleys. The barrier of the younger Gol-Ghone 'B' was over 500 m high impounding a lake that was more than 90 km long. In addition to up valley lacustrine deposits, the lake’s height is recorded by abandoned spill ways across the deposit opposite Gol, and up to 530m above present river level [27]. More than half of the mass of each rock ava-


Figure 8. Injection structure of coarse river gravels through a lobe of the Gol-Ghone 'B' rock avalanche, upstream of Gol and 3.5 km from the detachment zone. Insert shows detail. Note the surrounding, intact rock avalanche debris and well-define edge of the 'pipe'.

lanche has been removed, but they continue to control flows and the movement of bedload. Catastrophic flood deposits, consisting of megaclasts of reworked rock avalanche debris are found downstream of the barriers, especially on the river terraces opposite and down valley of Ghone, and into Skardu Basin. In Gol they speak of there having been a lake on the Indus here which was destroyed when a great flood came from the Shyok valley. The history and patterns erosion and survival of barrier materials, further complicate the reconstruction of the events. Until now, these deposits were erroneously interpreted as moraines of a glacier entering and damming the Indus from the Gol tributary [9, 27, 30]. 2.3. CASE STUDY III: GHORO CHOH I This is one of three rock avalanches from the west/right wall of the upper Shigar valley, near the settlement of Tisar [27]. They involved massive rock wall failures in a high angle igneous intrusion, the tonalite of Zanettin [56]. Each travelled over the deep fill of the Shigar valley (Table 3). In the run out zone, Ghoro Choh I illustrates a partitioning of movement styles, rather than successive 'acts'. The deposits illustrate, in particular, complexities that arise from the interaction with mobile and erodible substrates, creating distinctive morphological and facies features in the rock avalanche and substrate materials. Their mixing

534 and the entrainment of moisture, also served to transform the rock slide into a compound mass movement. Table 3. Parameters of the Ghoro Choh I complex rockslide - rock avalanche. Name Location Valley Mt. range Source lithology Orientation of the initial slope failure

Ghoro Choh I 35º 40' N; 75º 28' E Shigar (Indus) Karakoram Plutonic of the Indus Suture Zone [49] North-East

Runout dimensions: Area of deposit (exposed now / estimated original) Volume of deposit (estimated now / estimated original) Elevation (Highest / Lowest) Maximum vertical drop, H Maximum horizontal travel, L Fahrboschung (angle) Coefficient of friction H/L Highest run up, h

10 / 14 km2 60 million / 120 million m3 3,800 / 2,500 m asl. 1.300 m 7 km 11º 0.19 150 m

The more salient features record how the rock avalanche split into five major longitudinal streams, radiating fan wise across the valley floor from below the source slope (Figure 9). It seems likely the development followed an air launch from the lip of the detachment scar, 200 m up the valley wall, such that the material plunged into the valley fill at a steep angle. The ridges emerge and fan out from this impact zone as if fed along lines of least resistance, while in between them the debris either stalled or plunged into the valley fill. The developments resemble those in the upper Dulung Bar deposit, but on a larger scale. The longitudinal ridges are 10-35 m high, 2-3 km long, and with very steep flanks. They consist entirely of rock avalanche debris. Sections through the ridges exposed by erosion show them to consist of coarse, angular rubble, clast-supported, in a tightly packed matrix of sand-and silt-sized material. Samples analysed in the laboratory show the material, in all size fractions, to consist wholly of a single lithology, identical to bedrock outcropping in the detachment zone. The ridge surfaces are covered in megaclasts of the same lithology, many exceeding 10m in diameter, the largest more than 30m. The heights of the ridges generally increase outwards from the source, and they end suddenly like tip-heaps. Some of the largest clasts are found at or near the outer termini of these ridges. The volume of debris in the ridges comprises over two thirds of the visible deposit. It is not known how much material descended into and under the softer valley fill, nor how much is buried beneath post-emplacement river flood and lacustrine sediments. The ridges had been misidentified as terminal moraines by past observers [9, 42]. However, the huge, relatively fresh detachment scar of the rock slide is clearly visible immediately above the point from which the ridges fan out. As noted, they consist entirely of crushed and pulverized rock of the lithology outcropping there. Organic material carried by the rock avalanche has yielded a 14C date of 7110±80 yr. B.P., well after the time when glaciers retreated from the main Shigar valley [27]. While there was no mixing of materials in the digitate ridges, vast quantities of valley fill sediments were mobilised between and beyond them and, in other parts, thoroughly mixed with rock avalanche materials. The deformation and mobility of valley

535 fill varied in part with its composition, in part with the presence and amounts of moisture. Some fine-grained alluvium responded with complex folding. Large units were detached and transported across the valley floor, coming to rest in the low areas. This implies that rock avalanche material plunged beneath the valley fill between the rock avalanche ridges, and carried it forward from below. Just beyond the termini of the digitate rock avalanche ridges, coarse river gravels were thrust into a series of arcuate mounds forming a distinctive zone in the mid-region of the deposits. Finally, the outermost regions of the deposits record transformation of the mass movement to a (wet) debris flow. Here, there was a complete disintegration and mixing of fluvial, lacustrine and rock avalanche materials. recorded in a sheet of mixed material, thinning outwards from 2-3 m on the flanks of the ridges in coarse river gravels, to a few centimetres in the outermost areas. However, many large angular megaclasts of the tonalite were carried to the distal rim. They sit at the base of the opposing valley wall, surrounded by debris flow materials, 6-7 km from the source. The material also climbed the opposing slope up to 150 m, evident in a thin brandung of mixed materials containing the unmistakable, green, angular fragments from the rock avalanche, over bedrock of entirely different lithologies.

Figure 9. Ghoro Cho I rock avalanche. A view looking across the deposits across th3e floor of the Shigar valley. The 'digitate' ridges are clearly shown. Single arrows identify the arcuate ridges of thrust up and bulldozed coarse river gravels, beyond which the deposits are debris flow materials of mixed rock avalanche and valley fill sediments. 'D' marks the highest climb of the opposing valley wall, 6 km from the detachment zone.

Identification and reconstruction of the Ghoro Choh I deposits is made more difficult by overlap with the two other rock avalanches, one older and one younger, and a complex history of erosion and sedimentation in the Shigar valley. All three rock avalanches have been reworked and segmented by erosion and large magnitude flood events. Considerable areas are buried by fluvial, lacustrine and aeolian sediments. Each event formed low but extensive impoundment, together covering an area of some 15

536 km3 , and forming a 2-4 km wide, irregular weir which creates a strong hydrophysical break where the Braldu and Basha Rivers join. There has been considerable shifting of the channels from the east to the west side of the valley and back, and there at least six different, abandoned channels large enough to accommodate full summer flows crossing the rock avalanches. Moreover, these streams drain the most heavily glacierized part of the central Karakoram, carry huge sediment loads, and have a history of catastrophic floods, the largest ones from landslide and glacier dam bursts [23]. However, the post-emplacement history of the Ghoro Choh barriers has had less to do with the behaviour of the Braldu and Basha Rivers, than with rock avalanche events downstream. In the Holocene, an enormous episode of intermontane sedimentation has affected the whole Shigar valley and Skardu Basin below it, deposited behind rock avalanches down the Indus, most notably the Katzarah event at the exit from Skardu Basin [26]. More than 25 km3 of material has been deposited behind the latter. It led to burial of the southern flank of the Ghoro Choh I rock avalanche in the sediment wedge of an aggrading Shigar River. From early in the twentieth century renewed incision has been apparent, with several shifts in the channels and trenching of the rock avalanche and associated sediments. Moreover, at least seven other rock avalanches in the Shigar and Skardu Basins exhibit similar and even more complicated interactions over tens of square kilometre where they ran out onto the basin floors [27]. 3. Complex Run out and Emplacement Phenomena The rock slide-rock avalanches defined as complex here, show marked departures from the proportionalities commonly observed where terrain constraints are minor. For any given volume, there is a wide scatter of deposit areas, deposit thicknesses, and maximum run out [30]. It is not entirely clear whether the fahrboschung or 'apparent coefficient of friction' [32] is necessarily greater – that is, recording frictionally reduced run out – though it often is. Complications arise because confinement in narrow, steep gorges can increase mobility, at least locally, as may entrainment of moisture, wet sediments, or movement over a glacier [14]. The events of interest lie in the extreme range of Nicolleti and Sorriso-Valvo's [41] 'high-dispersive-energy' category. Furthermore, the post-failure redistribution of potential energy is usually partitioned among several lobes, and may differ considerably in each. Kinetic and frictional energy, or their relative roles, play out differently in individual lobes and sections of the run out path. Where the debris is trapped in relatively narrow gorges, 'disaggregation' energy seems to continue as a large factor at depth. In cases like the Gol-Ghone events, perhaps it is more accurately described as continued strong rock deformation and cataclasis. Furthermore, where rock avalanches travel over the deep riverine and other intermontane sediments, large amounts of energy may also be transferred to the substrate, deforming and entraining material or to the transport of snow and ice. Large quantities of heat may also be absorbed in melting the latter. Of course, these processes introduce enormous complications for modelling efforts and have generally been neglected in them. However, their scope and prevalence in high mountains demand attention there. With respect to run out and emplacement complexities, it should be said that few, if any, rock avalanche deposits lack features relating to terrain roughness, or some interac-

537 tions with the substrate. They provide useful diagnostic features. However, most observers have found them to be insignificant in the mechanics or energetics of the 'excessive' run out process, the dispersal of mass and defining morphology of these deposits. By contrast, in rugged mountains, terrain constraints generate a variety of distinctive features of sufficient dimensions or frequency of occurrence to require attention. They may be divided into those involving topography or the geometry of the run out zone, and those due to interactions with deformable and varying substrates. The results may be seen in deposit morphology relating to topography or substrate constraints, in the composition or facies of the rock avalanche or disturbed substrate materials, and in transformations related to mixing of the two materials or entrainment of moisture. 3.1. INTERACTIONS WITH RUGGED TERRAIN Topographic interference involves both longitudinal and transverse confinement, the effects of sudden changes in slope, valley form and bifurcation. Several different topographical effects are often combined in different parts of the same event. They can result in blocking or stalling of debris, channeling or divergence and splitting of debris streams, sometimes their converging again along the run out path. In this case, 'complexity' is evident in a diversity of deposit plan forms and related constructional landforms. Multiple major and minor lobes are found, and deposit forms that record channeling along, into and out of, narrow valleys. There is spreading and secondary splitting at valley junctions. Plan form complexity is combined with marked thickening and thinning of deposits in relation to terrain encountered, but unconformable with it. This involves an important difference from wet mass which tend to fill in and thicken at depressions, thin at rises and go around salients. Irregular or asymmetrical long and cross profiles, or ramping up in longitudinal and transverse pressure ridges, occur in most Karakoram examples. There are dozens of cases where relatively large volumes of debris travelled across relatively narrow gorges, to deposit the bulk of the material against the opposing slope as at Gol-Ghone. The lowest part of the cross-valley deposit often occurs below the source slope. At the limit of run up against an impact slope more or less at right angles to movement debris can be emplaced in what Heim [21] termed brandung or 'surge' forms. A number of examples in the Karakoram, record climbs of over 700 m above the intervening valley floor as at Gol-Ghone and, at Rondu-Mendi in the middle Indus gorge, 1100 m [26, 30]. The brandung phenomenon may occur in conjunction with, or be continued in, 'trim-lines' from caroming flow as debris is deflected to right or left of the impact slope. Well-preserved examples occur at the Haldi (Baltistan) and Batkor (Gilgit) events [28, 30]. It should be added that it has been common, and readily understandable, to misread these asymmetrical deposits, especially brandung phenomena, as lateral moraines [27]. Since these morphological features of complex emplacement have been dealt with at length elsewhere [30] they need no further discussion here. 3.2. DEPOSIT FABRIC AND FACIES The complex morphology of rock avalanches in rugged terrain is reflected at depth in their composition, fabric and facies. Longitudinal and, lateral confinement restrict or

538 prevent rapid dispersal to a thin sheet of debris, resulting in emplacement of deposits tens to hundreds of meters thick. Where this occurs in Karakoram examples, distinctive fabric and facies characteristics are seen at depths below 10 -15 m. Presumably they reflect the much greater confining pressures and apportioning of strain in relation to obstructions. At depths exceeding 50 m, and especially near to the source slope, fabrics seem more like continuing bedrock cataclasis than a rock avalanche process, with complex crushing and deformation of lithological units, their folding, attenuation and microshearing (see figure 7). The high confining pressures ensure. As they deform or disintegrate, perhaps 'flow', however, lithologies do not mix but remain in the same relative positions as the original bedrock, except for offsets along shear planes. Nevertheless, these materials are generally found to have been subject to thorough disaggregation. Their removal from highly compacted in situ deposits reveals that weaker units have been reduced to fine powder. Stronger ones display crackle-, and jigsaw brecciation, and seemingly large units of bedrock will disintegrate to cobble-sized and smaller clasts on removal of support [39, 55]. At these deeper levels, presumably the forces tending to comminute the materials are largely crushing or compressional, as opposed to impact shattering or abrasional forces, important in the descent and collapse down the source slope and in thinner, more dilated, debris streams. The extent of pulverising, crushing and fracturing at depth seems to reinforce Heim’s [21] view that the rock avalanche, or its 'debris stream' is also a 'powder mill'. However, it seems at variance with models which attribute the mobility of rock avalanches to negligible frictional forces in the swiftly moving debris itself. In general it seems likely that topographic interference will not only tend to maintained a thicker debris stream, but greater crushing and pulverising forces within the mass. Likewise, sudden stalling and frictional 'freezing' against obstacles introduce or enhance large crushing forces. Extensive shear planes are also found at depth, and may be highlighted by a centimetre or two of 'gouge' [30]. It seems these represent planes of separation in a vertical sense, of upper, more mobile or later-arriving materials, that split around slowing or stalled material below and in front. They may be related to separate lobes in plan [38]. 3.3. INTERACTION WITH DEFORMABLE SUBSTRATE MATERIALS Most of the Karakoram events began as massive rock wall failures below 4,500 m asl, and emplaced deposits in valleys containing large rivers, sometimes lakes, fed by snow and glacier meltwaters from the humid, heavily glacierized high altitude zones [25]. In many cases, therefore, the rock avalanches moved across and into more or less extensive and deep sedimentary fills. This introduces complexities related to interactions with deformable and wet substrates, modifying the behaviour of the rock avalanche and producing distinctive constructional landforms. There can be large scale deformation and remoulding of substrate sediments, their entrainment and deposition in distinctive forms. There are features due to inter-penetration of substrate and rock avalanche materials without mixing, and a spectrum of mixed facies, including 'chaos' deposits in which both have disintegrated and been throughly reworked (see below).

539 In the rock avalanche material itself a range of morphologies are observed from longitudinal ('digitate') and/or transverse ridges, 'ramping up' over mobilised and compressed substrates, and in stalled distal rim materials. Interactions with soft sediment and entrapped moisture may decrease mobility or, as has been argued elsewhere, may increase it. The Dulung Bar and Ghoro Choh I examples introduced some of these features. At the Naltar Lakes event, just north of Gilgit, valley fill and water trapped beneath the rock avalanche were major factors in the extent of run out, the morphology and composition of resulting deposits [28]. Over a distance of some 5 km down the main valley transverse and longitudinal ridges and isolated mounds of rock avalanche debris rise 10-40m above the general surface. Between the depositional ridges are many closed depression, including the basins of the Naltar Lakes, and many others now infilled or captured by the river. The features suggest interaction and break up of the rock avalanche over wet valley fill in the manner described by Abele [2] in the European Alps (Figure 10). 3.4. LANDSLIDE-TECTONIC FORMS IN SUBSTRATE MATERIALS Where a rock avalanche moves over deformable substrate materials strain may be transferred to them creating regions of complex folding, faulting and perturbations in soft sediments (Figure 11). In the Skardu and Shigar intermontane basins, introduced above, most rock avalanche complexities relate to these interactions. The structures observed here include every conceivable style of folding including kink bands, compressive crenulations, boudinage, and faulting or micro shearing, slumps, micronappes, collapsed anticlines, diapiric forms, and wedge-shaped intrusive forms. Whether and how these developments occur obviously depends upon the nature of the substrate materials. They range from sensitive, well-developed structures in fine-grained, stratified sediments, to chaotic masses of coarse gravels. Differences moisture content and interaction with adjacent strata introduce further complexities. In wet sediments, some strata simply disintegrate, undergo diapir-like extrusion or are otherwise remoulded. Drier, coarser or more rigid strata may break up into blocks of intact stratified material in a remolded matrix. A further set of internal fabrics arise from the inter-penetration and mixing of rock avalanche and substrate materials. Slices of deformed substrate may be carried along in the base of the rock avalanche. Clastic dikes, even folds, may penetrate the rock avalanche material – but without mixing. These are typical features of Yarnold and Lombard’s [55] 'mixed basal zone'. Some are features indicative of active penetration by substrate materials into and through the moving rock avalanche. They include dykes, 'stringers', diapirism and high-pressure injections of substrate sediment with entrapped moisture. Similar features have been described elsewhere [37, 54, 51, 2, 22]. The distinctive character of 'complex' events, involving these tectonised substrate forms, lies in the sheer scale and scope of disturbed and transported materials. In all cases observed they are also associated with some or all of the other terrain, and substrate-constrained features in rock avalanche materials, discussed above. This is a question both of the local scale of the tectonised facies and their total extent as part of the architecture of the whole landslide-emplacement event. In the Yarbah Tshoh, Baltistan event, for example, landslide tectonised substrate facies cover an area of some 7 km2,

540 crossing the entire, 3.5km wide Shigar flood plain. Sections exposed by erosion range in thickness from 5 to 15 m, and the full depth of disturbed substrate may be much greater. A roughly similar area of tectonised fan sediments exists from the Satpara-Skardu event. The lateral extent around Skardu town site is some 5km and in places the thickness of disturbed facies exceeds 30m. In both cases the total volume of disturbed and mobilised substrate materials is measured in the millions, possibly tens of millions, of cubic meters. Similar orders of magnitude apply to half a dozen other events in the Skardu and Shigar basins. Similar vertical scales of disturbance and mixing, such as are shown in the photographs from Gol-Ghone and Tsok-Dumordo, have been observed in local exposures at the base of dozens of other events that entered areas of valley fill.

Figure 10. Naltar Lakes rock avalanche, Hunza River Basin, 30 km north of Gilgit. The view looks across the upper Naltar valley towards the detachment zone of the rock avalanche, to the right of, and above, the photo. the complicated system of ridges and depressions in then rock avalanche record interactions with valley fill materials. B - brandung in tributary valley opposite the detachment scar. The bulk of the debris moved to the left down the Naltar valley (cf. Figure 1).

In terms of field identification, the main problems here involve close similarities with glacitectonic forms, and the way the geometry, scale and direction of a rock avalanche lobe may resemble an ice tongue. In fact, almost all the more readily observed Karakoram examples were previously attributed to glacier action, some as far back as the mid-nineteenth century [10, 42, 7, 43, 27]. Associated rock avalanche materials were often classed as terminal moraines. One might expect major differences, given the forces involved in the deformations due to fast-moving rock avalanches. However, in scale, the landslide-generated forms, notably in the Skardu and Shigar Basins, are comparable to glacitectonic features around the margins of large ice sheets, notably those reconstructed around the large Quaternary ice sheets. And it is these conditions and ex-

541 amples past observers in the Karakoram seem to have had in mind, rather than comparatively small and thin valley glacier tongues. Meanwhile, there seems to be no work in micro-tectonics and their modeling to show, definitively, how to distinguish between landslide-, and glaciotectonism.

Figure 11. Landslide-tectonized substrate materials: a) fluvial materials forced up into the base of the GolGhone rock avalanche. Rock avalanche debris at right of photo. Note the complex folding of the finer grained materials. b) Diapir-like injection of silty-sand, valley floor materials into the base of the Tsok-Dumordo rock avalanche, 150m below surface, Panmah Glacier basin, central Karakoram [30]. c) Mixed complex folding and thrust faulting of fine-grained, stratified alluvium inn the proximal zone of the Yarbah Tsho event, Shigar valley, Baltistan. d) Small, localised folds of various styles in fine-grained alluvium in the distal region of the Yarbah Tsho event.

542 3.5. 'CHAOS' DEPOSITS There are rare cases where total disintegration of rock avalanche lobes occurs. In some, this generates 'melange-like' regions of chaotically interspersed and mixed rock avalanche and substrate materials. Exposures in parts of the Yarbah Tshoh, Gombah Thurgon and Ghoro Tsar events are remarkable in this respect [27, 30]. In the extreme, the former alluvium may be totally remoulded, progressively breaking up and mixing with rock avalanche material until both lose their integrity. Yet, the confused and disordered remnants retain certain features of rock avalanche and substrate sediments, and perhaps represent intermediate stages towards the generation of the other types of mass movement. They are, however, distinct from those recording transformation to debris avalanche and debris flow processes, such as in the Ghoro Choh I outermost deposits. In these contexts, large boulders have detached from the debris streams of the rock avalanche and 'surfed' over, or bulldozed into, the substrate (Figure 12). They may 'knife' into soft sediments, something widely observed in the Yarbah Tshoh event. In some cases, including parts of the Satpara-Skardu, Yarba Tshoh and Gomba Thurgon events, substantial sheets of rock avalanche material appear to have 'dived' into or under substrate materials, which they carried it forward from below, eventually disintegrating into mixed facies in distal areas This can be observed in the face of the terrace cut by the Satpara stream under the western sector of Skardu town. These disintegration forms occur especially where large volumes of rock avalanches – 10s-100s of millions of cubic meters – moved into existing shallow lakes, over former lake beds, or braided river reaches. They also occur over the outer, finer grained materials of sediment fans or fan deltas, with features of the kind well-displayed on the Satpara-Skardu fan where it is cut by the Indus terraces at Skardu town site – sites widely described but attributed to glacial action [49]. 4. General Features and Classes of Complex Events The complex event involves the interrelations of four sets of phenomena affecting the rock slide-rock avalanche: 1. Terrain variables, topography and substrate conditions of dimensions sufficient to constrain and modify the rock avalanche in the following aspects, 2. Runout behaviour, the responses of the debris mass and stream(s) to terrain constraints, leading to magnified or different features compared to the 'simple' runout process, 3. Emplacement morphology, geometry and features of the rock avalanche deposit that reflect run out behaviour in response to terrain variables, including distinctive constructional landforms developed within the body of the deposit, 4. Sedimentology relating to the above conditions and including larger, multiplied or distinctive features of the composition, fabric, facies or sedimentary architecture of: a) the rock avalanche debris; b) substrate materials modified by the rock avalanche and associated with its emplacement, specifically what are called 'landslide-tectonic' features here; c) products of the interaction and catastrophic mixing of rock avalanche and substrate materials.


Figure 12. Rock avalanche boulders which 'surfed' and bulldozed fine-grained fluvial and lacustrine sediments near the distal rim of the Yarrbah Tshoh event, Shigar valley Baltistan. Note 'bow-wave' of crumpled alluvium in front of largest boulder. Rock avalanche moved diagonally here from left to right.

The investigations, on which the discussion is based, have looked mainly at field features of the third and fourth sets of phenomena, and attempted to relate them to the first. In addition, attention is directed to problems of identifying forms and materials that survive extensive erosion, reworking or burial of catastrophic landslide deposits. While the focus here is on terrain-related complexities in run out zones, it should be said that they can affect each of the three 'elements' of the classic case defined earlier. Two or more massive rock wall failures may originate from the same break-out zone, whether at different times or as semi-independent components of the same event. They may behave differently, and interact so as to complicate each other's development [40]. In the Karakoram several examples involve two or more large rockslides, one or more of which developed into a rock avalanche, while the other(s) did not. This can be seen at the Ghomboro Complex in the Braldu Gorge [26], and the Upper Henzul Complex on the lower Gilgit River [28]. In these events, in addition to complex rock avalanches, large blocks (>200m in diameter), were lodged at the base of the source slope without breaking up, while others continue to move slowly or intermittently down the transport slope. The detachment zones of these events have also been sources of debris flows. Again, the Nomal Complex [28], one of the largest catastrophic landslide events identified in the region, originated as a contiguous series of mountain wall collapses that moved along varying trajectories, constraining and deflecting each others’ run out and emplacement processes. This illustrates complexities that arise in the 'transition zone' or transport path as well as the detachment zone. Evidently, these are developments that can also affect the run out and emplacement of the rock avalanche beyond the source slope. Further complications in the run out zone can arise from interactions of two or

544 more rock avalanches coming from different detachment zones, or the same detachment zone at different times. The Gol Gone 'A' and 'B' events originated as separate rock wall failures, but their run out zones overlapped, as did the Ghoro Choh I-III events [26, 27]. It is in relation to such developments that several areas of massive rock wall failure in the Karakoram are classed as landslide 'complexes'. These combine two, at least, of the four aspects of complexity defined above, and two or more classes of movement, at least one of which is a rock avalanche, or at least two distinct 'styles' of rock avalanche [30]. They also include debris avalanche or debris flow events and multiple rock falls. However, in focusing on complexity in the run out and emplacement of rock avalanche deposits, the concern is mainly with unusual developments in compared to other rock avalanches. Some features reported in 'simple' events are developed on a much greater scale, or in a much larger number of manifestations. These include longitudinal and transverse pressure ridges, raised distal rims and sedimentary features of the mixed, basal zone. Meanwhile, partitioning, splitting and 'steering' of debris in terrainconstrained run out results in complex plan and deposit forms. Materials deposited and the behaviour of some or all individual debris lobes need not differ greatly from the classic rock avalanche. There are also features that seem likely to belong to the 'transitional stage' [6], and which are lost where emplacement is completed with minimal interference by terrain. Thickened masses in front of the source slope are, in part at least, stalled intermediate stages of the debris dispersal process. Where parts of the debris stream are stalled in gorges or against opposing slopes, incomplete break up of the original rock mass is also observed which appears to represent terrain- impeded disaggregation. Large units of bedrock - several tens of meters in diameter, sometimes more than 100m – survive below, even well out from source slopes [30]. They may be largely intact, or crackle- and jigsaw brecciated. On the other hand, at depth in debris streams that remain very thick through lateral confinement or opposing slopes, there is intensified and extended disaggregation or pulverising. This is indicated in cases such as Gol-Ghone and Haldi. At depths below 100 m some debris horizons are totally crushed to sand-, and silt- sized particles, while in the upper 10-50 m a full range of coarse clasts exist, and in the upper 5 m and at the surface are regions dominated by megaclasts in excess of 5m in diameter [30]. This suggests a crude vertical 'sorting' related to crushing forces. As the descriptions and Figure 7 showed, there appear to be singular developments of fabric and facies at depth. On the one hand these seem more closely related to cataclastic processes or brecciation in fault zones. On the other, they contribute to a distinctive, over all sedimentary architecture of the thickened rock avalanche body, related to the interaction with topography. Likewise, interactions with deformable and erodible substrates, and the 'landslidetectonic' features they generate, seem to be unique aspects of complex emplacement events. While some observations relating to run out behaviour in complex events have been identified, how they may relate to existing or possible models of rock avalanche mechanics remains to be resolved (but see Strom, this volume). Most modelling seems preoccupied with how friction might be radically diminished to explain the high mobility of these materials. The unaided high speed ‘flow’ of such coarse, angular rock debris has seemed improbable to many observers. Some of them have sought to resolve the

545 problem with the aid of an inferred, or a 'surrogate', medium – air, moisture, vapour, 'dust' – or by invoking a ‘sliding’ surface or layer at the base, whether of air, wet sediment, or more finely crushed basal material. While these may point to important considerations, some of the characteristic features of complex emplacement seem to raise difficulties for such notions. First there is the apparent, continued, strong crushing, grinding and micro shearing of materials in the run out zones. Second, there is the behaviour of material that climbs hundreds of meters up slope to emplace large, compact and stable masses of material, notably the brandung phenomenon. If this depended upon a medium, it is difficult to envisage how they climb to the limit of their run up energy and then undergo apparently instantaneous frictional freezing in place. At least it requires that the 'medium' or sliding surface disperses or attenuates in concert with the upward momentum, and ceases to be a factor at its highest reach. In this regard it will be recalled that many observers have remarked upon similarities between rock avalanche deposits and lava flows, glaciers or debris flows [21, 31, 36]. On the one hand, these do share some common deposit features with rock avalanches. Thus, the direction of sedimentation in the rock avalanche is almost never downwards, but more or less horizontal. This is similar for lava flows, and for some glacier and debris flow processes. However, the emplacement of rock avalanche deposits discussed here is often at a more or less steep angle above the horizontal, resulting in constructional landforms unknown in the other processes mentioned. As noted, rock avalanches not only climb hundreds of meters up steep slopes, clearly at high speeds. Having done so they stall or 'freeze' in place. Ice, lava or wet debris flows do not create or sustain any of these forms. They either show a relatively minor run up ability, collapse back from the limit of climb. In the case of most moraine ridges, they develop with support of the adjacent ice mass and most involve extended periods of time or multiple deposit events, and a variety of sedimentation processes, to create them. Likewise, in rock avalanches, interactions with terrain or substrate create longitudinal and transverse constructional ridges that are much higher and steeper than those known in wet debris, ice or lava, or anything these generate as a single emplacement event and sedimentary unit. In sum, these would seem to be aspects of massive rock wall failures that differ from other processes, and must be predicted by models the mechanics of rock avalanche. For features of the complex rock avalanche which are present on a smaller scale or small numbers of manifestations in 'simple' cases, it is necessary to decide just how 'strong' they have to be, or how frequent, to separate complex events from others. Morphological criteria or dimensions, for example, could depend upon the number of independent lobes – say, at least three, of which any one is at least 25% of the plan form, or more than three. Other criteria might include the presence or extent of lateral spreading and/or narrowing along the run out path; the degree of asymmetrical thickening along transverse or longitudinal cross-sections, and the partitioning of the landslide mass among distinct constructional landforms. Special or uniquely well-developed features include the brandung phenomenon, trim-lines or pressure ridges recording caroming flow, run up and overtopping of opposing slopes and interfluves. Another set of criteria refer to the size or frequency of landslide-tectonic forms in substrate, the extent and thickness of structural or facies features within the rock avalanche debris not seen with dispersal to a thin sheet, and mixing features of substrate and rock avalanche materials.

546 It would seem, however, that decisions about what criteria or dimensions might be used to distinguish complex events, or have general significance, need to be considered in relation to more than just one set of events and mountain range. Dialogue with colleagues working in other regions and a comparative assessment of mass rock wall failure in various high mountain regions seems required before deciding upon more formalised criteria. 5. Concluding Remarks Some complex events, or features of complex run out and emplacement, are reported wherever massive rock slope failures have been recognised in the Himalayan and Inner Asian [16, 50, 35]. As indicated in the introduction, there seem to be examples in most of the world’s highest mountains. Some events recognised in the geological record of the American Southwest also suggest 'complex' rather than 'classic' events, usually through interaction with mobile substrates [38, 54, 17, 51]. It is not yet clear whether the much higher incidence in the Karakoram Himalaya is unique to that region, or a result of more detailed surveys. In most regions, to date, however, these phenomena have been treated more as singularities or unique events in rock avalanche investigations. Moreover, they are most widely developed in terrain of the highest relief, often high altitude, or with rock avalanches incorporated into complex intermontane basin sedimentation. Not only are most examples of very large size, they involve logistically and technically difficult conditions, often remote from research centres. The ability to recognise complex features in these situations remains problematic, and in places controversial. Only rarely can their presence be reliably determined on remotely sensed imagery, although this can be an invaluable guide for fieldwork, or once an event is detected on the ground. The need to establish diagnostic criteria, beginning with field recognition, is thus of major importance in this context. However, by way of conclusion some of broader issues raised by the evidence from the Karakoram need to be addressed. The numbers and scale of events, and the downward transport of material by them, imply a very significant role in denudation, and specifically in erosional developments since the last major glaciation. Each event represents enormous, sudden liberation and comminution of bedrock. Many individual cases represent a larger volume of sediment than the total yield from the Upper Indus basin in a year, some being equivalent to several year – and this river has one of the highest yields in the world [47, 29]. The massive rock wall failures play an important role in the morphogensis of the mountains, specifically in unroofing of the many plutons. Batholiths and other igneous masses form the core of the Greater Karakoram, and most lesser ranges [49]. The rock avalanches serve both by strip away metamorphic and metasedimentary covers, and to remove large volumes of plutonic rock from the flanks of exposed plutons. Almost one third of the rock avalanche deposits consist of plutonic lithologies emplaced over valley floors and opposing slopes of other bedrock types. Meanwhile, former rock avalanches are large, local sources of debris moved in debris flows and other mass movements, by streams, ice and wind. Except for the large quantities of dust generated by the rock avalanche, removal mainly occurs in higher

547 energy and high magnitude events which can overcome the resistence of coarse, angular, overcompacted debris Thus, as below the Gol-Ghone event, many of the wellknown catastrophic flood deposits along the Upper Indus streams – conspicuous swathes of large boulders or megaripples on river terraces and in intermontane basins – turn out to be composed mainly or wholly of lithologies from the nearest upstream, cross-valley, rock avalanche deposit. Some may represent dam break floods from the latter, but not necessarily. High magnitude floods due to storms, intense episodes of glacier ablation, or glacier dam burst floods, are more likely than more common high flows to overcome the erosional resistance of rock avalanche deposits. They generate large, local pulses of coarse debris, where they pass over or through already breached rock avalanche barriers [23]. Perhaps more remarkable, in this regard, is the role of these short-lived catastrophic events in contributing to sediment storage in the mountains, and temporarily reducing sediment yields, at least on the Holocene time frame. Once in place, many have formed long-lived barriers and more than 100 events are known to have formed relatively longlived landslide dams, as indicated by extensive lacustrine deposits [4, 29]. In a few cases the lakes were 100s of meters deep at the barrier and more than 100 km long when full. Some individual examples have impounded a third or more of the Upper Indus drainage. Two or more on different tributaries may have coincided in time to impound much more of the basin. The result has been to starve downstream areas of sediment for decades or centuries, and massive intermontane deposition. The sediment volumes impounded can be two or more orders of magnitude greater than the rock avalanche deposit itself [29]. Of course, once the rock avalanche barriers are breached, these temporary stores become significant sources of sediment, generally more readily removed than the resistant rock avalanche deposits. The barriers also act as the main causes of what I have termed a 'chronically fragmented river system' [26]. Although not the focus here, the significance of these events as natural hazards is of obvious. The numbers of rock avalanches now known, most of which are of complex type, have serious implications for landslide risk assessment. While they come from surveys of barely 20% of the Karakoram, the events are known to have affected almost all of the zone of permanent settlement along the Indus streams. Dozens of settlements, including some of the main towns and visitor destinations, lie on or beside rock avalanche deposits. Almost all settlements, most of the roads and other infrastructure, lie on intermontane sedimentary features controlled by rock avalanche barriers [26]. Many of the worst sites of slope instability and recurring debris flows along the highways are where they cross rock avalanche deposits. Most examples seem to have been stable, relatively long-lived landslide barriers. Topographic blocking has been responsible for putting in place of strong, relatively thick and wide cross-valley barriers [48, 28]. The longer-lived events show the potential for extensive inundation of settled areas and modern infrastructure by large rock avalanche dams. If early or catastrophic failure has been rare, it is not unknown. The largest dam break flood on record for the Upper Indus, that of 1841, is now attributed to failure of an earthquake-triggered rock avalanche, which dammed the river for six months but failed rapidly when overtopped. Only two rock avalanches are known to have engulfed inhabited areas in the past two hundred years, but four massive rockwall failures were identified in the glacier

548 zone in the last fifteen years. This suggests a partitioning of risk among the different major clima-geomorphic zones at different altitudes [25]. Likewise, there is an urgent need to decide among alternative possible models of the time distribution of the many prehistoric rock avalanche deposits. One possibility is that, in the habitable zone, they represent a declining, post-glacial risk, as glacially over steepened and debuttressed sites are 'used up' [28]. Alternatively, groups of events, or the largest events, may record rare, extreme triggering events, especially large earthquakes, perhaps summer storms, or exceptional snowfalls. Each of these has been associated with catastrophic rockslides in the past. Until we have dates for a large fraction of the cases now discovered, it is impossible to determine just what they represent in terms of the contemporary risk of rock avalanches in the zone of permanent settlement. Acknowledgments Thanks to Alexander Strom for useful comments on a draft of the paper. References 1. 2. 3. 4. 5. 6.

7. 8. 9.

10. 11. 12. 13.



Abele, G. (1974) Vom Eis geformte Bergsturzlandschaften: Frage der glazialen Umgestaltung der bergsturze von Sierre, Flims, Ems und vom Fernpass. Zeitschrift fur Geomorphologie 18, 120-147. Abele, G. (1997) Rockslide movement supported by the mobilization of water-saturated valley floor sediments Zeitschrift fur Geomorphologie 41, 1-20. Bovis, M.J. and Evans, S.G. (1996) Extensive deformation of rock slopes in the southern Coast Mountains, southwest British Columbia, Canada, Engineering Geology 44, 163-182. Burgisser, H.M., Gansser, A., and Pika, J. (1982) Late glacial lake sediments of the Indus valley area, northwest Himalaya, Eclogae geologische Helvetica 75, No 1, 51-63. Costa, J.E. and Schuster, R.L. (1987) The formation and failure of natural dams, United States Geological Survey, Open-File Report, 87-392, Vancouver, Washington. Couture, R., Evans, S.G., Locat, J. and Hadjigeorgiou, J. (1999) A proposed methodology for rock avalanche analysis, in Yagi, N., Yamagami, T., and Jiang, J-C. (Eds.) Slope Stability Engineering: Proc. of the International Symposium, Is-Shikoku, November 1999, Rotterdam, Balkema, 1369-1378. Cronin, V.S. (1989) Structural setting of the Skardu intermontane basin, Karakoram Himalaya, Pakistan. Geol. Soc. of America, Special Paper 232, 183-202. Cruden, D.M. (1985) Rock slope movements in the Canadian Cordillera, Can. Geotechnical J. 22, 528540. Dainelli, G. (1922) Studi sul Glaciale, in Dainelli, G., (Ed.) Relazione Scientifiche della Spedizione Italiana de Filippi, nell’Himalaia, Caracorum e Turchestan Cinese (1913-14) Serie II 3, Zanichelli, Bologna. Drew, F. (1875) The Jamoo and Kashmir Territories. Stanford, London. Eisbacher, G.H. and Clague, J.J. (1985) Destructive Mass Movements in High Mountains. Geological Survey of Canada, Paper 84-16, Ottawa. Evans, S.G. (1989) Rock avalanche run-up record, Nature 340, 271. Evans, S.G., Hungr, O., and Engren, E.G. (1994) The Avalanche Lake rock avalanche, Mackenzie Mountains, Northwest Territories, Canada: deposition, Dating and Dynamics. Can. Geotechnical J. 31, 749-768. Evans, S,G, and Clague, J.J. (1988) Catastrophic rock avalanches in glacial environments, in C. Bonnard (Ed.) Landslides. Proc., 5th Intern. Symposium on Landslides, Lausanne, Switzerland, 10-15 July 2, 1153-1158. Fahnestock, R.K. (1978) Little Tahoma Peak rockfalls and avalanches, Mount Rainier, Washington, U.S.A. in B. Voight (Ed.) Rockslides and Avalanches 1, Natural Phenomena. Elsevier, New York, 181196.

549 16. 17. 18. 19.

20. 21. 22. 23. 24. 25.

26. 27. 28. 29.


31. 32. 33. 34.

35. 36. 37. 38. 39. 40. 41. 42.

Fort, M. (2000) Glaciers and mass wasting processes: their influence on the shaping of Kali Gandaki valley, Nepal, Quaternary International 65-66, 101-119. Friedmann, S.J. (1997) Rock-avalanche elements of the Shadow Valley basin, Eastern Majave desert, California: processes and problems, J. of Sedimentary Research 67, 792-804. Gardner, J.G. and Hewitt, K. (1991) A surge of the Bualtar Glacier, Karakoram Ranges, Pakistan: a possible landslide trigger, J. of Glaciology 32, 27-29. Hadley, J.B. (1964) Landslides and related phenomena accompanying the Hebgen Lake earthquake of August 17th, 1959 United States, Geological Survey Professional Paper 435-K Washington, D.C. 107138. Heim, A. (1882) Der Bergsturz von Elm. Zeitschrift Deutschen Geologie Gesellschaft 34, 74-115. Heim, A. (1932) Bergsturz und Menschenleben. Fretz and Wasmuth, Zurich. Hermanns, R.L. and Strecker, M.R. (1999) Structural and lithological controls on large Quaternary rock avalanches (sturzstroms) in arid northwestern Argentina, Geol. Soc. of America Bull. 111, 934-948. Hewitt, K.(1982) Natural dams and outburst floods of the Karakoram Himalaya, Hydrological Aspects of High Mountain Areas, International Association Scientific Hydrology, Publication 138, 259-269. Hewitt, K. (1988) Catastrophic landslide deposits in the Karakoram Himalaya, Science 242, 64-67. Hewitt, K. (1993) The altitudinal organization of Karakoram geomorphic processes and depositional environments, in J.F. Shroder, jr. (Ed.), Himalaya to the Sea: geology, geomorphology and the Quaternary, Routledge, New York, 159-183. Hewitt, K. (1998) Catastrophic landslides and their effects on the Upper Indus streams, Karakoram Himalaya, northern Pakistan, Geomorphology 26, 47-80. Hewitt, K. (1999) Quaternary moraines vs catastrophic rock avalanches in the Karakoram Himalaya, Northern Pakistan, Quaternary Research 51, 220-237. Hewitt, K. (2001) Catastrophic rockslides and the geomorphology of the Hunza and Gilgit River valleys, Karakoram Himalaya. Erdkunde 55, 72-93. Hewitt, K. (2002) Postglacial landform and sediment associations in a landslide-fragmented river system: the transHimalayan Indus streams, Inner Asia, in Hewitt, K., Byrne, M-L., English, M., and Young, G. (Eds.) Landscapes of Transition: Landform Assemblages and transformations in Cold Regions, Kluwer Academic Press, Dordrecht, 63-91. Hewitt, K. (2002) Styles of rock avalanche depositional complexes conditioned by very rugged terrain, Karakoram Himalaya, Pakistan, chapter 15, in Evans, S.G. (Ed.) Catastrophic Landslides: Processes, Events, Environments, Geological Society of America, Reviews in Engineering Geology 15. Hsü, K.J. (1975) Catastrophic debris streams (sturzstroms) generated by rock falls. Geol. Soc. of America, Bull. 86, 129-140. Hsü, K.J. (1978) Albert Heim: observations of landslides and relevance to modern interpretations, in B. Voight, (Ed.) Rockslides and Avalanches 1, Natural Phenomena. Elsevier, New York, 70-93. Hungr, O. (1989) Mobility of rock avalanches. National Research Centre for Disaster Prevention Bulletin. Tsukuba, Japan, 1-13. Hutchinson, J.N., (1988) General report: morphological and geotechnical parameters of landslides in relation to geology and hydrology, in C. Bonnard (Ed.) Landslides. Proceedings, 5th International Symposium on Landslides, Lausanne, Switzerland, 10-15 July 1, 3-35. I.S.D.R. (International Strategy for Disaster Reduction) (2000) Usoi Landslide Dam and Lake Sarez: an assessment of hazard and risk in the Pamir Mountains, Tajikistan, United Nations, Geneva. Iverson, R.M. Reid, M.E. and LaHusen, R.G. (1997) Debris-flow mobilisation from landslides, Annual Review of Earth and Planetary Sciences 25, 85-138 Johnson, B. (1978) Blackhawk landslide, California, U.S.A. in B. Voight (Ed.) Rockslides and Avalanches 1, Natural Phenomena. Elsevier, New York, 418-504. Krieger, M.H. (1977) Large Landslides, composed of Megabreccia, interbedded in Miocene Basin Deposits, Southeastern Arizona. Geol. Survey Professional Paper 1008 , Washington, D.C. Laznicka, P. (1988) Breccias and Coarse Fragmentites: petrology, environments, associations, ores. Developments in Economic Geology 25. Elsevier, New York. Mudge, M.R. (1965) Rock avalanche and rockslide avalanche deposits at Sawtooth Range, Montana, Geol. Soc. of America Bull. 76, 1003-1014. Nicoletti, P.G. and Sorriso-Valvo, M. (1991) Geomorphic controls on the shape and mobility of rock avalanches, Geol. Soc. of America Bull. 103, 1365-1373. Norin, E. (1925) Preliminary notes on the late Quaternary glaciation of the North-Western Himalaya, Geografiska Annaler 7, 165-194.

550 43. 44. 45. 46. 47.

48. 49. 50.


52. 53. 54.



Owen, L.A. (1988) Wet sediment deformation in Quaternary and recent sediments in the Skardu basin, Karakoram Mountains, Pakistan, in D. Croot (Ed.) Glacitectonics, Balkema, Rotterdam, 123-147. Pflaker, G. and Ericksen, G.E. (1978) Nevados Huascaran avalanches, Peru in B. Voight (Ed.) Rockslides and Avalanches, 1 Natural Phenomena. Elsevier, New York, 277-314. Porter, S.C., and Orombelli, G. (1980) Catastrophic rockfall of September 12, 1717 on the Italian flank of the Mont Blanc massif. Zeitschrift für Geomorphologie 24, 200-218. Rouse, W.C. (1984) Flowslides, in D. Brunsden, and D.B. Prior (Eds.) Slope Instability, Wiley, New York, 491-522. Shroder, J.F. jr. (1993) Himalaya to the sea: geomorphology and Quaternary of Pakistan in the regional context, in J.F. Shroder, jr. (Ed.) Himalaya to the Sea: geology, geomorphology and the Quaternary Routledge, London, 1-42. Schuster, R.L. (Ed.) (1986) Landslide Dams: processes, risk and mitigation, Geotechnical Special Publications 3, American Society of Civil Engineers, New York. Searle, M.P. (1991) Geology and Tectonics of the Karakoram Mountains, Wiley, New York. Strom, A.L. (1996) Some morphological types of long-runout rockslides: effects of the relief on their mechanism and of rockslide deposits distribution, in K. Senneset (Ed.) Landslides. Proc. of the Seventh International Symposium on Landslides, 1996, Trondheim, Norway, Rotterdam, Balkema, 1977-1982. Topping, D.J. (1993) Paleogeographic reconstruction of the Death Valley extended region: evidence from Miocene large rock-avalanche deposits in the Amargosa Chaos basin, California. Geol. Soc. of America Bull. 105, 1190-1213. U.S. Army Map Series, U502, Sheet NJ-43-13, Edition 2-AMS, (1962). Whitehouse, I.E. and Griffiths, G.A. (1983) Frequency and hazard of large rock avalanches in the central Southern Alps, New Zealand. Geology 11, 331-334. Yarnold, J.C. (1993) Rock-avalanche characteristics in dry climates and the effect of flow into lakes: insights from mid-Tertiary sedimentary breccias near Artillery Peak, Arizona, Geol. Soc. of America Bull. 105, 345-360. Yarnold, J.C., and Lombard, J.P. (1989) A facies model for large rock-avalanche deposits formed in dry climates. In I.P. Colburn, P.L. Abbott and J. Minch (Eds.) Conglomerates in Basin Analysis: a Symposium dedicated to A.O. Woodford. Pacific Section, S.E.P.M. 62, 9-31. Zanettin, B. (1964) Geology and Petrology of Haramosh-Mango Gusor Area Italian Expeditions to the Karakorum (K2) and Hindu Kush III, Brill, Leiden.

DISSECTED ROCKSLIDE AND ROCK AVALANCHE DEPOSITS; TIEN SHAN, KYRGYZSTAN K. ABDRAKHMATOV1 Institute of Seismology, Asanbay 52/1, Bishkek, 720060, Kyrghyzstan A. STROM Institute of the Geospheres Dynamics, Russian Academy of Sciences, Leninskiy Avenue, 38-1, 119334, Moscow, Russia Abstract Rockslides and rock avalanches in Northern and Central Tien Shan, that have been deeply dissected by subsequent erosion, or which internal structure was studied in trenches and road cuts, are described. Presence of intensively comminuted debris overlaid by coarse blocky material is found out in most of the case studies. At those cases, where different types of parent rocks outcrop in the rockslide scars, the resultant deposits are composed of the unmixed 'layers' of debris originated from these rock types. Such grain size distribution and unmixing of debris can be considered as typical features of large-scale massive rock slope failures. Basal sliding surfaces with different relationships between rockslide debris and underlying soil can be observed at several sites. Comprehensive study of deeply dissected deposits of rockslides and rock avalanches can shed light on mechanism of their rapid motion and of 'rock' ĺ 'debris' transformation. 1.


The Tien Shan – one of the largest mountain systems of Central Asia – has many large (106-108 m3) and giant (•109 m3) massive rock slope failures that form high natural dams and vast aprons of rock avalanches. Numerous rockslides occurred on slopes composed of several types of sedimentary, metamorphic and igneous rocks of Palaeozoic and Precambrian, rarely of Mesozoic and Neogene ages. Sometimes different lithologic units very in colours and can be easily recognised both in the scars and in the resultant deposits. It provides good opportunity to compare mutual position of various types of rocks before and after the emplacement. Some of such rockslides are of especial interest as far as they have been deeply dissected by subsequent erosion. These gorges form giant 'natural trenches', which allow detail study of the internal structure of rockslide deposits. In such cross-sections we can obtain comprehensive information 1

E-mail of corresponding author: [email protected]

551 S.G. Evans et al. (eds.), Landslides from Massive Rock Slope Failure, 551–570. © 2006 Springer. Printed in the Netherlands.

552 about debris grain size composition and its distribution in the rockslide bodies as well as on the structure of basal sliding surface and other boundaries inside rockslide debris. Data that can be obtained by comprehensive study of deeply dissected deposits of rockslides and rock avalanches can shed light on mechanism of their rapid motion and of 'rock' ĺ 'debris' transformation. 2.

Case Studies

Hereafter data on the Kokomeren and the Inylchek dissected rockslides (the latter can be classified as rock avalanche), first described few years ago [28, 29] and of several other deeply eroded rockslides and rock avalanches (Figure 1) some of which still await detail research are presented.

Figure 1. Location of rockslides and rock avalanches in the Kyrgyz Tien Shan described in this paper. Black triangles – rockslides and rock avalanches: 1 – Kokomeren, 2 – Aksu, 3 – Djashilkul, 4 – Bashi-Djaya, 5 – 1911 Ananievo, 6 – Djusumdy-Bulak, 7 – Karakul, 8 – Inylchek and Atdjailau, 9 – "Ancient".

2.1. THE KOKOMEREN ROCKSLIDE The Pleistocene Kokomeren rockslide (Figure 1, No 1; 41.92º N, 74.23º E) about 1.0 km3 in volume descended from the left bank of the Kokomeren River 5-7 kilometres downstream from the Kyzyloi intermountain depression and formed a dam up to 400 m high (Figure 2). Subsequently it was dissected by the Kokomeren River that almost completely removed its distal part. It allow to conclude that the lowermost part of rockslide crest was at or near its front, contrary to many other high Tien Shan rockslide blockages, which have been dissected at their proximal parts due to lowering of dam’s crests at the feet of the scars from which they descended. Such morphology when distal part is higher, corresponds to the 'primary' morphological type proposed in [30]. Similar 'primary' debris distribution characterises numerous high blockages in Karakoram Hi-

553 malaya [18, 20], and other well-known rockslide dams such as Köfels in the Tyrolean Alps [16] and Usoi in Pamirs [12].

Figure 2. The Kokomeren rockslide (after [28]). Key to legend: 1a - sandstone, 1b - sandstone fragments in rockslide debris; 2a - granite, 2b, c - granite in rockslide debris, forming a huge massif (b) and composed of angular boulders (c); 3a – alternating granite and sandstone, 3b-d - shattered granite and sandstone in rockslide deposits (b - on the map, c, d - on the cross-sections, c - shattered granite, d - shattered sandstone); 4 limestone, schist and diabase alternation; 5 - alluvium; 6 - rockslide scar.

554 Thus, the Kokomeren rockslide which lowest part was nearer to the opposite bank of the valley could be considered as rather rare exception. Remaining part of rockslide debris rests on the left bank of the river on a terrace 150m above the riverbed. Frontal part of the dam filled the ancient valley, which was 40m above the present-day riverbed and now only its minor portion remains intact on the right bank of the river just above the road. This portion of debris is easily attainable, while study of the left-bank part of the dam requires crossing of the powerful Kokomeren River. This rockslide is of particular interest because its internal structure can be investigated in detail due to different colours of rocks involved in the failure [28, 29, 31]. Its 1.6km-high scar exposes (from its top to foot) dark-grey sandstone, reddish granite and granite and sandstone alternation. The same sequence can be observed in the rockslide deposits on both banks of the river. On its left bank the upper part of the dams’ body, up to 250 m thick, is composed of tremendous blocks and huge angular boulders of red granite, overlain, in turn, by a 'layer' of blocks and big fragments of dark-grey sandstone. The lower part, up to 150 m thick, consists of 'layers' of intensively comminuted alternating granite and sandstone (see section A-B on figure 2). The same succession occurs in the isolated segment on the right bank of the river (see section C-D on figure 2). The only difference is that at this part upper granite unit is composed of angular boulders up to 2 - 3 meters across while on the left bank – of large block hundreds meters in size (in the lateral direction). It should be pointed out that the boundary between the slightly and the intensively crushed rock units is abrupt, without any transitional zone. Correspondence of the succession of debris varieties in the rockslide deposit to the lithostratigraphy in the scar, and retention of the bedrock structure show that lower part of the sliding mass, up to 150 m thick, moved as a single unit without internal mixing. Intensively comminuted granite sampled from the lower unit at the right bank of the river has grain-size composition typical of sand (Table 1) with specific gravity of 2.67 2.68 g/cm3 and extremely low coefficient of permeability, comparable with that of clay. The latter was measured in the laboratory after compaction of small amount of debris. Separate particles are angular quartz and feldspar crystal fragments. The smallest particles are mainly the quartz scales, and some bigger quartz grains have conchoidal micro-texture that indicates high local stress in the process of rock crushing. Table 1. Grain size composition of two samples of shattered granite from the lower zone of the Kokomeren blockage (right bank of the river). Fractions (mm)




Sample 1 (M-A) 1(M-D) 2(M-A) 2(M-D)

11,90 6,90 6,40 7,80

27,90 31,60 27,50 24,00

27,60 27,60 24,70 22,50


0.050.01 Content (%) 8,40 17,10 10,50 14,10 11,20 21,10 15,30 17,10




10 km of undissected plateau. The least developed montane topography is found in the northeastern range flank where large tracts of gently arched carbonates are separated by deep slot canyons in limestone. Further west, several such catchments have developed dendritic drainage networks in which volcaniclastic rocks are exposed. In order to understand this topographic evolution, we briefly consider fluvial incision of bedrock. Streams erode bedrock through abrasion by bedload and suspended load, and joint block plucking [15, 20, 22, 49]. The efficacy of these processes is a function of water discharge and channel bed slope [26] and sediment flux [27]. If a stream is under capacity, its sediment load is limited by supply from its valley sides. Thus, there is a feedback between fluvial incision and hillslope erosion. An evolving mountain belt may initially have modest local relief, and thus small sediment loads derived from

576 shallow valley sides. These sustain only low fluvial incision rates, which in turn do not promote effective catchment expansion and hillslope mass wasting. Erosion is thus outpaced by tectonic uplift resulting in growth of a regional topographic high. This is the state of the easternmost Finisterre Mountains whose form is broadly that of a large growth anticline, the crest and north flank of which have a contiguous limestone envelope. Subdued local relief is associated with internal, karstic drainage, while shallow, elongate catchments drain with the regional topographic dip to the north. As regional topography steepens because of continued rock uplift, rates of mass wasting increase. Debris is supplied faster to fluvial systems of increasing stream power, accelerating valley lowering. However, it is generally by hillslope mass wasting alone that catchment expansion may occur. In the Finisterre Mountains, this phase of landscape evolution is represented along the edges of the limestone plateau of the Huon Peninsula. On the plateau, limestones have become progressively karstified, with high rates of infiltration and subsurface flow. Seepage concentrates along subsurface permeability contrasts and emerges where such interfaces are exposed. Sapping [4] then results in undercutting of valley heads and side walls, and slope failure may ensue. On the north flank of the eastern Finisterre Mountains, several drainage networks branch out from single points along canyon-like valleys. Upstream of these points, lowpermeability volcaniclastic rocks are exposed below steep headwalls, suggesting that catchment expansion initiated when their contact with the overlying limestones was exhumed. In this area, headwaters consist of large (some >5 km), amphitheatre-shaped concavities, filled with debris lobes. These deposits have a chaotic topography, sometimes with pressure ridges and ponded drainage, implying catastrophic failure of the plateau edge. Analogous features have been described for historic failures of the limestone cap of the nearby island of New Britain. In the eastern Finisterre Mountains, valley-head landslide scars form semicircular clusters. Large transverse catchments have several such aggregates. Some are located on major, active faults, and downstream of such clusters, river valleys follow the same structures. Evidently, large faults have guided and facilitated drainage propagation, possibly by focusing seepage. More generally, we conclude that the pattern of landslidedriven drainage network expansion in the Finisterre Mountains reflects the organization of seepage in the antecedent, undissected topography. Downstream of headwalls, valley widths are constant and equal to the diameter of the corresponding landslide clusters. Valley sides are poorly dissected, and mass wasting occurs principally through slope-clearing landslides. Away from the plateau margin, most mountain ridges and peaks are defined by coalescing, multiple-kilometre scale landslide scars, whose pattern responds to the local gradients associated with the established valley network. Generally, headscarps are steep and arcuate; hummocky debris deposits fill the lower parts of scars and spill into adjacent valleys. The 1988 Kaiapit landslide [40] is a recent example of such a failure. Without obvious trigger, this landslide collapsed the entire south face of a spur descending from the main divide of the Finisterre Mountains, involving displacement of ~1.5 km3 rock mass. The failure had a height of 1.5 km, a base width of 2.5 km, and a concave shape with slope gradients as steep as 60°. Landslide debris traveled 6.5 km down two adjoining valleys, leaving 150-200-m-thick deposits, and killing 75 people. Slope clearing landslides such as the Kaiapit example generate concavities that concentrate runoff on a scale sufficient

577 to initiate watersheds. On the south flank of the Finisterre Mountains, most lower-order catchments were apparently created by this mechanism. Once the sub-escarpment drainage pattern is established, development of a complex ridge-and-peak topography proceeds rapidly. Runoff concentration causes fluvial incision of landslide scars and deposits, resulting in the formation of steep inner valleys, bounded by debris terraces. Such terraces are found throughout the southern and western Finisterre Mountains. In most valleys, fluvial incision has progressed beyond the base of landslide deposits into bedrock. Thus, the drainage network is entrenched in the uplifting rock mass, consolidating runoff into trunk streams. Continued fluvial incision reduces the upper length scale of the local topography, replacing the initial landslide-induced concavities with entire drainage networks. Consequently, the potential for slope failure on a multiple-kilometre scale, which currently dominates erosion of the eastern Finisterre Mountains, is eventually removed from the landscape. In the ensuing phase of orogenic evolution, erosion occurs primarily through local slope failure in response to fluvial incision. The upper length scale of such landslides is constrained by the local drainage density. In this scenario, large, drainage altering landslides are extremely rare and the topographic template is essentially fixed. Major rearrangements of this mature montane landscape can only be caused by progressive, lateral motion of channels, for example as a result of anisotropic substrate resistance or structural entrainment of drainage, renewal of large-scale landsliding due to changes in tectonic and/or climatic boundary conditions, or horizontal advection of topographic elements through an orogenic system. This last point should be amplified. In our discussion of the topographic evolution of the Finisterre Mountains we have assumed that surface motion was approximately vertically upward. This may be appropriate in the case of the Finisterre Mountains, especially if we consider the establishment of drainage networks in a passively advecting limestone cap. However, in many orogens, horizontal rock displacement rates are an order of magnitude greater than tectonic uplift rates [53]. This implies that rock displacement paths are largely horizontal through active orogens [50], and that topographic elements are carried into zones of enhanced denudation where the deformation field has a larger vertical component. Thus, major drainage elements can be added to a catchment by horizontal advection across the crest line of a mountain belt. One example of this is the Landsborough River in the western Southern Alps of New Zealand. Other possible examples have been identified in this and other mountain belts. Moreover, horizontal advection of rock mass makes range divides prone to frequent, large landslides of the ‘Kaiapit’ type, especially in headwaters of relatively wet catchments facing with the direction of rock advection. Preliminary observations in the Central Mountains of Taiwan indicate that kilometre-scale landslides cluster along the east side of the main divide, as predicted from the eastward motion of rock mass through the orogen and toward its typhoon-prone east flank. We anticipate a similar prevalence of very large landslides in the headwaters of west-draining catchments in the Southern Alps of New Zealand. To conclude this section we ask whether the key geomorphic features of the Finisterre Mountains, including the prominence of very large landslides, are shared with other presteady-state orogens. It would be easy to attribute these features to the special geology of the Finisterres, notably their thick limestone cap and strong permeability contrast

578 with underlying rocks. However, we have observed similar geomorphic trends in the eastern Greater Caucasus of Azerbaijan, where very large landslides dominate the headwaters of catchments propagating into a thick pile of muddy, clastic and calcareous sediments. The eastern tip of this mountain belt is arid and in the absence of considerable erosion, weak rocks have built several kilometers of relief. Fluvial dissection of this topography commences where orographically-forced precipitation first generates significant runoff. In this transitional area rapid catchment expansion occurs, as in the eastern Finisterre Mountains, by the propagation of large, deep-seated landslides into elevated topography with low, undulating relief. However, great escarpments are absent, as a result of the lesser competence of the sedimentary rocks constituting the tip of the Azeri Caucasus. Both the eastern Greater Caucasus and the eastern Finisterre Mountains contrast with the emerging, southern tip of the Taiwan mountain belt where rapid orogen growth occurs from below sea level to elevations of up to 4 km. In south Taiwan, structurally controlled drainage rapidly makes way for regularly spaced streams traversing the structural and topographic grain of the mountain belt. These catchments appear to grow self-similarly, with the expanding mountain belt, and without known evidence of catastrophic, landslide-induced changes. In summary, we have found that drainage initiation in growing mountain belts is commonly retarded due to lack of orographic forcing of precipitation, high infiltration rates in sedimentary and possibly karstified cover rocks, and low sediment yields from subdued relief. However, once fluvial incision has started, rapid drainage network propagation may be driven by multiple-kilometer-scale landslides, the location of which is strongly linked with upslope seepage patterns. In such cases, the mode and rate of drainage network expansion are governed not by fluvial incision but by hillslope mass wasting at valley heads. Slope-clearing landslides initiate the formation of tributary catchments in the wake of retreating headwaters. Stream entrenchment then follows from runoff concentration in trunk streams and tributaries, promoting valley floor lowering through landslide debris and into bedrock. With increasing dissection of the landscape, the potential for catchment altering landslides is reduced, although it remains high around the main divide of orogens with strong, but opposing tectonic and climatic asymmetries. This course of events is shared by some, but apparently not all pre-steadystate mountain belts. It is likely that in mountain belts where significant surface runoff occurs in newly emerging topography, for example due to location within the monsoon belt (e.g., Taiwan), drainage networks are established before topographic growth permits multiple-kilometre-scale slope failure. In such cases, the maximum length scale of hillslopes is limited throughout the orogen by effective fluvial dissection. We attribute the fact that the Finisterre Mountains do not appear to follow this evolutionary path despite their position in the humid tropics to an inferred high seepage loss and subdued surface runoff over the limestone cap.

3. Landslides Limit Relief and Balance Tectonic Fluxes Having addressed the role of landslides in pre-steady-state mountain belts, we now turn to mass wasting in common, ridge-and-valley topography. The aim of this section is to demonstrate that landsliding is the dominant mode of hillslope mass wasting where

579 creation of relief, by the combined effects of rock uplift and fluvial and/or glacial valley lowering, occurs faster than regolith production by weathering of newly exhumed rocks. Landslides effectively limit relief in such landscapes; and, as a consequence, landslides balance the tectonic rock mass flux where valley floor long profiles are in steady state. Hillslope erosion is often represented as a diffusion process, in which the hillslope sediment transport rate Qs is proportional to the local topographic slope, and its spatial variation is proportional to the vertical erosion or aggradation rate of the substrate such that wz wt


w2z , wx 2


where x is distance from the divide, z is elevation, t is time, and N is a diffusion coefficient. This expression implies that the steady-state profile of hillslopes dominated by diffusion processes, and underlain by a homogeneous substrate, is parabolic. Thus, the topographic fingerprint of diffusion is a unique, positive correlation of local topographic gradient and upslope area. Convex-up hillslopes are common in upland landscapes with low erosion rates, where erosion occurs by splash, wash, and creep. In tectonically active mountain belts, they tend to be limited to sections of drainage divides not recently affected by slope failure. Splash, wash and creep are limited by the rate of production of regolith by weathering of intact rock mass. This is a slow process, limited by the kinetics of the chemical reactions involved. Where the rate of valley floor lowering is greater than the rate of regolith production, weathering-limited mass wasting cannot keep pace with local base level lowering, and valley sides become progressively undercut. Eventually, this will give rise to landslides involving not only weathered material, but also unweathered rock mass. In active mountain belts, rates of rock uplift and fluvial incision are commonly greater than 1 mm yr-1, and significantly faster than most weathering processes [18]. Therefore, bedrock landslides dominate hillslope mass wasting in tectonically active mountain belts. This is supported by several lines of evidence. A power law relation exists between the rates of erosion and silicate weathering across a range of climates and catchment sizes [34], implying that the two are intimately linked, through weathering-limited mass wasting, and the associated refreshing of the weathering front. But, in a number of active mountain belts, erosion rates are up to an order of magnitude higher than would be expected from measured silicate weathering rates (J. West, Personal Communication, 2003). In such areas, weathering rates may be at the kinetic limit for a given substrate and climate, a limit that is subdued by the absence of continuous, organic-rich soils, due to relentless mass wasting: active orogens have bedrock landscapes. Fundamentally, the stability of a hillslope is determined by its surface geometry, the density, cohesion and frictional properties of its substrate, the depth of potential failure plains, and the gravitational acceleration. A change in any of these parameters might cause the destabilization and failure of a slope. For example, the topographic gradient may increase due to undercutting by river erosion at the base of the slope. Similarly, the frictional or cohesive strengths may decrease by weathering of material, seismic shaking, or wetting of the rock mass, which also increases the weight of the slide block. In the absence of any external landslide triggers, substrate properties determine the

580 maximum stabile gradient of a hillslope. Rock mass strength decreases with increasing spatial scale because of the influence of spatially distributed discontinuities. The mountain-scale strength of the rock mass therefore limits the steepness of bedrock landscapes [43]. In such landscapes the maximum hillslope height is determined by the spacing between higher-order streams and the bulk mass strength of the interfluves. Given effective fluvial bedrock incision, it may therefore be expected that dry mountain belts have greater relief than their wetter equivalents. The rock mass control on topographic development was illustrated in a terrain analysis of the northwestern Himalaya [9]. There, the frequency distributions of slopes were found to be essentially indistinguishable among different mountain regions, despite differences in denudation rates of up to an order of magnitude (Figure 2). In each region, most slopes fell between 20° and 45°, the mean slope was 32° ± 2°, and the modal slope was only marginally steeper. This similarity of slope distributions suggests homogenous topographic characteristics, largely independent of denudation variations, and set by rock mass strength. The rapid decrease in the frequency of slopes steeper than 35° implies that such slopes are prone to collapse. They do not, in general, survive for geomorphologically significant amounts of time. Interestingly, this cut off value is only slightly higher than the maximum stable slope in loose, dry sand, implying that the rock mass strength in the northwest Himalayas, and probably most other mountain belts, is determined by through-going discontinuities rather than the properties of the intact rocks: to first order, mountains are built of low-cohesion material. Another, frequently used method of terrain analysis considers the relation of local slope and upslope (drainage) area across a landscape [36]. It has been demonstrated that the principal upland erosion processes have significantly different area-slope fingerprints. We have mentioned the positive correlation of area and slope for ‘diffusion’-dominated topography. Similarly, bedrock rivers have a power-law relation between area and slope with a negative scaling exponent whose normal value is between -0.3 and -0.6 [44, 49]. Bedrock landslides commonly have straight failure plains and are therefore characterized by a constant local slope for a range of upslope areas. Although they exist, it is rare to find examples of the flat area-slope relation

Fraction of area

0.05 0.04




5 4



0.01 0 0




Slope (degrees) Figure 2. Slope distributions from subregions in the northwestern Himalaya. Slopes were calculated as best-fit planes to a 4 × 4 grid cell matrix in a ~90 m DEM. Areas 1-3 have apatite fission track ages of 0-1 my; Areas 4 and 5 have apatite fission track ages of 1-6 my. Regardless of the ten-fold contrast in denudation rates, implied by these fission track ages, there are few significant differences in slope statistics among them. After: Burbank et al. [9].

581 associated with this geometry in large topographic data sets of active mountain belts, probably due to the mixing with other process signals over the characteristic length scale range of bedrock landslides (101 m – 103 m). For example, debris flows typically dominate channels with upstream drainage areas of less than 1 km2, or 103 m equivalent length scale. Such ‘colluvial’ channels have a power law relation between area and slope with a small, negative scaling exponent [35]. This mixing detracts from the overwhelming importance of bedrock landslides in many active mountain belts. A scenario for the erosion of mountain belts has now emerged in which the rate of bedrock uplift is matched by the rate of valley lowering (steady state longitudinal river profiles) but surpasses the rate of weathering. Then interfluves grow until topographic elements become unstable and collapse, producing bedrock landslides. Given sufficient transport capacity of the rivers, this type of landscape yields sediment, principally by landsliding, at a rate that is solely determined by the rate of rock uplift, and independent of local relief. Confirmation comes from a study of local relief and erosion rates by Montgomery and Brandon [37]. They found that a well defined, linear relation exists between erosion rates and local relief, calculated over 10 km, for catchments outside areas of active mountain building [3]. However, erosion rates in active orogens vary by an order of magnitude whereas mean local relief over 10 km is fairly constant, between 1.0 km and 1.5 km (Figure 3). This implies that topographic relief is not a first order control on the rate of hillslope mass wasting in active mountain belts, that erosion rates are set, instead, by external, tectonic forcing, and that there is a limit to local relief, imposed by bedrock landslides. 10 9


Erosion Rate (mm y-1 )

8 7






4 3 H






0 0




Mean Local Relief (m) Figure 3. Plot of erosion rate versus mean local relief (measured over 10 km) from mostly tectonically inactive areas (open circles) and tectonically active, convergent areas (solid squares). NZ is Southern Alps, New Zealand, NP is Nanga Parbat region, western Himalaya, T is Taiwan, H is central Himalaya, D is Denali portion of the Alaska Range, A is European Alps, BC is British Columbia. After: Montgomery and Brandon [37].

582 4. Landslide Magnitude and Frequency If landslides dominate the erosion of active mountain belts, it is important to quantify their long-term impact. Extrapolating short-term geomorphic observations to time scales pertinent to landscape evolution and orogen dynamics requires an understanding of the scaling behaviour of the processes involved, in particular the magnitude and frequency with which they occur [52]. The magnitude-frequency distribution of landslides is characterized by a maximum at small to intermediate size events (103 m2) and a broad, negative power law tail for larger landslides (Figure 4). This power law scaling holds true whether the landslide size is defined as the scar area [25], or the total area disturbed [41], and whether landslides are triggered over a long period of time [19, 23], or almost instantaneously [19, 21]; it also holds true if landslide volume is considered instead of area [8], although volume is typically much more difficult to measure (both in the field and in air photographs). For an idealized landslide size distribution to be power-law distributed across the size range x  [c, f) , the size probability density is defined as p( x) { DcD x D 1 , c>0, D >0


where D is the power law scaling exponent [45], and x is usually defined as planform area. The scaling exponent explicitly determines the impact of large versus small landslides on integrated measures such as the total area disturbed, or the volume of material yielded. Power law scaling is typically observed for areas greater than 10005000 m2 up to the largest landslide areas for which a distribution can be reliably estimated (of the order of 105 m2). The power law property of the landslide size distribution introduces several complications. First, the disturbance area and eroded volume of a landslide are highly variable. Second, there is no characteristic landslide scale that dominates the erosion budget: a power-law distribution indicates that events at many scales play an important role. This makes it hard to quantify the pattern and rate at which a mountain landscape evolves by landsliding. At present there exists no mathematical means of assessing the flux of sediment from a zone dominated by landslide mass-wasting. In other words, no differential operator or partial differential equation (analogous to the diffusion equation) yet exists to formalize the relationship between mountain relief and landslide sediment flux (quite apart from the difficulty in calibrating such an equation were it to exist). Power laws pose further technical problems that have impeded conceptual progress in several respects. It is well known in the statistics community that heavy-tailed distributions are difficult to characterize reliably [2]. The steepness of a negative powerlaw tail, which represents the relative frequency of small versus large events, cannot be estimated with confidence unless the sample size (the number of landslides) is very large. If the underlying distribution is only asymptotically a power law, as is probably the case with landslides, then the frequency of small to medium events can strongly distort any estimate of the power-law scaling [45]. The practical consequence of

583 10

Probability density, p(x)





Southern Alps West Flank











$D 10









Area, x [m ]

Figure 4. Examples of landslide size distributions, from the western Southern Alps, of New Zealand, plotted as a probability density function p(x) plotted in log [p(x)] versus log(x) form. Solid squares show the probability density of landslides in the Whataroa catchment, mapped at 1:25,000, N = 3986; open circles show the probability-density of landslides in a larger part of the western Southern Alps, mapped at 1:50,000, N = 5086. The data sets show similar scaling of landslide magnitude and frequency. Above a cut-off size, related to the resolution of the mapping and/or a break in the failure mechanism, the data scale as a power-law. This portion of the data is the tail end of the distribution and represents about a quarter of the observed landslides. D is the slope of the best fit power-law, and values are almost identical at D = 1.45 for both data sets. After: Stark and Hovius [45].

erroneous inference is a faulty emphasis on either small or large events. In several recent studies [23, 25, 41], the steepness of the power-law scaling was underestimated, largely as a result of unsophisticated statistical analysis. This has resulted in the inference that large landslides dominate the erosion budget, since integration of the power-law magnitude-frequency distribution indicated a strong dependence on the largest events. Recent work [19, 45] has shown that the power-law distribution of landslide size-frequency is steep, and reasonably consistently so for a variety of data sets (with some exceptions). The scaling exponent D expresses this steepness and generally varies between 1.3 and 1.5. In light of these recent studies it is clear that the area disturbed by landslides, over the long term, is dominated by small to medium scale failures (up to an area of around 103104 m2). Most of the landslide data sets assembled over the years are unreliable in their representation of the magnitude-frequency distribution of small to medium scale failures. Only those data sets acquired with great care, high quality air photography, and detailed field verification can be regarded as having counted accurately the smaller landslides [8, 19, 21]. For these very rare data sets, there is convincing evidence for a rollover, or break in scaling, typically at around 1000-5000 m2. The mean and most common (modal) size landslides are approximately of this scale, but the strong asymmetry of the landslide size-frequency distribution means that these averages are not equal.

584 Some major challenges remain. First, most data sets undercount the smaller failures and misrepresent the frequency of the dominant events. The rollover in the magnitudefrequency distribution in these cases is unreliable, and the estimate of the power-law component of the distribution is distorted. Fortunately, this distortion is quantifiable [45], and a more reliable value of Dcan be elicited if a censoring model is applied. However, no reliable estimate can be made of the area disturbed for such data sets. Given the importance of such estimates, for example in the evaluation of soil loss and mobilization of particulate organic matter, there is a clear need for high fidelity, regional landslide maps. Second, the volume eroded by landsliding also remains difficult to quantify, particularly where the power-law scaling is steeper than previously thought, since the smaller, poorly enumerated failures are now seen to play a stronger role. This is due, in part, to the fact that the scaling relationship between landslide thickness and landslide planform area remains unclear. This scaling is important because it sets the transformation between the area-frequency and the volume-frequency distributions. For strictly soil/regolith failures, it could be argued that the depth of landslide failure is approximately constant. For failures that involve bedrock, however, it has been argued that the depth of failure likely correlates with landslide length scale, giving a volume to area relation of V ~ A3/2 [23]. At present, neither model has been vindicated with field data, but must be so before any reliable estimate of total landslide sediment flux can be made. If the constant thickness model applies, then the volume eroded by landslides is set by the frequency of the average area landslide and is weakly dependent on the power-law scaling. In contrast, if the scaling thickness model applies, then the total erosion volume is a more equal function of small and large landslides, with a weighting that is a sensitive function of the power-law scaling exponent D 5. Landslide-driven Sediment Flux Finally, we shift our focus to the output of sediment from active mountain belts. This output is set to first order by the tectonic mass flux. Systematic, long-lived trends in sediment delivery to the mountain front may result from changes in tectonic and/or climatic boundary conditions [10, 32, 54]. Superimposed on these long-term trends are shorter term (300 mm precipitation have triggered significant numbers of landslides [39]. During the largest storm on record, in March 1988, landslides accounted for 89% of the sediment mobilized and 87% of the sediment delivered to Lake Tutira [38], suggesting that a relationship exists between landslide intensity in the catchment and sedimentation in the lake (Figure 6). This, together with the observation that the sediment delivery ratio of the catchment scales linearly with storm magnitude, has provided a context for the interpretation of the stratigraphic record of the lake. Using historic data only, Trustrum et al. [47] have shown that above the precipitation threshold for landsliding, the impact of storm events increases, seemingly in exponential

Figure 5. Schematic showing changing relationship between channel erosion and hillslope response. In (A), frequent low to moderate discharge/wear events mainly lower the central channel thalweg, cutting through the parabolic channel shape, and leaving hillslopes untouched. A rare, intense flood fills the channel (B), and high sediment flux and water levels work to widen the channel out, restoring a wider parabolic shape consistent with the previous lowering. This wider parabola undercuts and oversteepens adjacent hillslopes, and landslides result (C), restoring stability in the hillslope-channel relationship.

fashion, with their size (Figure 6), such that the two largest storms have generated about half the sediment supplied to Lake Tutira between 1895 and 1988. From longer (2 kyr) lake records, it appears that the magnitude-frequency distribution of sediment layers attributed to storm-triggered landsliding, and the duration of time intervals between landslide episodes can be described by power laws, with scaling exponents of approximately -2.1 and -1.4, respectively [17] (Figure 6). Thus, the Lake Tutira record suggests that landslide intensity closely tracks local, meteorological conditions. This

586 applies globally, although the relation between landsliding and storm size is likely to be obscured by other local factors such as geology, vegetation, land use, and the history of landscape perturbation. Similar observations have been made for earthquake-triggered landslides. Notably, the area affected by slope failure, the epicentral landslide intensity, and the total mobilized sediment volume scale with earthquake magnitude [29]. Such observations provide a semi-quantitative basis for natural hazard risk prediction, and modelling of erosional landscape evolution and the associated sediment flux [6]. There are some important differences between storm-triggered landslides and earthquake triggered landslides. First, storm-triggered landslides result primarily from local pore water pressure gradients and changes of pore water pressure that are likely to be most pronounced in the shallow subsurface. Storm-triggered slope failure is therefore likely to be located at the soil/regolith-rock interface or above it, although deeper, bedrock failures may occur. In contrast, seismic ground motion affects local stress fields well below the topographic surface and may trigger a relatively large number of deepseated, bedrock-involved landslides. Such landslides are likely to produce coarser debris than their shallow counterparts, and onward transport may be more difficult as a result. Second, storm-triggered landslides occur at a time when the transport capacity of rivers is enhanced, and surface runoff on hillslopes ensures effective downslope translation of

Figure 6. Statistics of landslide-driven sediment supply from a 32 km2, rainfall-dominated catchment to landslide-dammed Lake Tutira, North Island, New Zealand. (A) Relation of sediment layer thickness in Lake Tutira to storm rainfall in the catchment. (B) Average and cumulative sediment layer thickness in Lake Tutira for specified storm magnitudes. (C) Frequency distribution of 316 storm-related sediment layers that accumulated in Lake Tutira over a 2250 yr period. The best-fit regression line computed from all points has a slope of -2.06. (D) Distribution of intervals between storm-related sediment layers (•3 mm) in Lake Tutira. The slope of the regression line is -1.4. After: Trustrum et al. [47] and Gomez et al. [17].

587 debris. This reduces the potential residence time of sediment in the montane catchment. Earthquakes, on the other hand, do not appear to correlate with specific meteorological conditions. They can generate very large volumes of landslide debris when the potential for onward transport is low. Third, storm-triggered slope failure appears to affect all steep locations in ridge-and-valley landscapes, possibly with a bias towards slope toes [28] where onward transport is guaranteed. Seismic strong ground motion is strongest at ridge crests [16]. As a result, co-seismic landslides often cluster around high points and deposit debris on hillslopes rather than on channel floors. The combined effect of these differences is a potentially very significant difference in the residence times of stormgenerated and co-seismic landslide debris in montane catchments. We reinforce this point, briefly, with an example from Taiwan, using data assembled by the Taiwan Water Resources Agency [48]. In 1999, central west Taiwan was struck by a Mw 7.6 earthquake (return time 50-70 yr) that triggered more than 22,000 landslides in the epicentral area. The year following the earthquake was relatively dry and without major storms. Although the sediment concentration in rivers draining the epicentral area was elevated, the sediment loads of most Taiwanese rivers remained below the 30-yr average [11]. However, when several big typhoons hit Taiwan in 2001, a disproportionately large number of landslides occurred throughout the central Taiwan mountains, and the average sediment concentration in affected rivers increased by up to 8,000 ppm. As a result the sediment yield of epicentral catchments increased by a factor 3-11, potentially making the Choshui river (drainage area 3,000 km2), albeit temporarily, the third most important river (globally) in terms of suspended sediment supply to the ocean. This was primarily due to the remobilization of the debris of co-seismic landslides that remained in the landscape, and the preparation of other slopes, by seismic cracking and shattering of the substrate, for failure during subsequent storms. This example indicates that the sediment cascade from valley side to mountain front may have many steps, and that sediment production and transfer together determine the supply to nearby basins [6, 25]. Importantly, it implies that the distant sedimentary record of mountain belt erosion is likely to be dominated by storm-driven input, even though most sediment may have a co-seismic origin. Moreover, the recent events in Taiwan have shown that the probability of slope failure remains elevated in epicentral areas for years, and possibly decades, after a major earthquake. Taiwan offers a unique opportunity to study the geomorphic response to a large, seismic perturbation in full. In closing we reemphasize the crucial role of landslides in the erosion and topographic evolution of active mountain belts. Landslides drive the expansion of drainage networks in uplifting rock mass, and counter the tectonic mass flux into orogenic systems. Moreover, they are the source of most sediment eroded from the continents, and the probability distributions of landslides and their triggers are a firstorder control on the variability of the sediment flux from active mountain belts. Finally, landslides are the primary cause of material damage and loss of life associated with earthquakes and rainstorms in upland areas [46]. This is, without doubt, the strongest motivation for further investigations into the mechanisms, patterns and rates of landsliding.

588 Acknowledgements We thank the many organizations and people who have supported our work on landslides by providing access to imagery and data, assistance in the field, and help with data processing. Supported by EC Framework 5 grant EVG1-2001-0003.

References 1.

2. 3. 4.

5. 6. 7. 8. 9.

10. 11. 12. 13. 14.

15. 16. 17. 18. 19. 20.


Abbott, L. D., Silver, E. A., Anderson, R. S., Smith, R., Ingle, J. C., Kling, S. A., Haig, D., Small, E., Galewsky, J., and Sliter, W. (1997). Measurement of tectonic surface uplift in a young collisional mountain belt, Nature, 385, 501-507. Adler, R.J., Feldman, R.E., and Taqqu, M.S., 1998: A Practical Guide to Heavy Tails: Statistical Techniques and Applications, Birkhauser, Boston. Ahnert, F. (1970). Functional relationship between denudation, relief, and uplift in large, mid-latitude drainage basins, American Journal of Science, 268, 243-263. Baker, V. R. (1990). Spring sapping and valley network development, in Higgins, C. G., and Coates, D. R. (eds.), Groundwater Geomorphology: The Role of Subsurface Water in Earth-Surface Processes and Landforms, Geological Society of America Special Paper 252, pp. 235-265. Beaumont, C., Fulsack, P., and Hamilton, J. (1991). Erosional control of active compressional orogens, in McClay, K. R. (ed.), Thrust Tectonics, Chapman and Hall, New York, pp. 1-18. Benda, L., and Dunne, T. (1997). Stochastic forcing of sediment supply to channel networks from landsliding and debris flow, Water Resources Research, 33, 2,849-2,863. Berner, R.A. (1993). Weathering and its effect on atmospheric CO2 over Phanerozoic time, Chemical Geology, 107, 373-374. Bradinoni, F., and Church, M. (submitted). Representing the landslide magnitude-frequency relation: Capilano River Basin, British Columbia, Earth Surface Processes and Landforms. Burbank, D. W., Leland, J., Fielding, E., Anderson, R. S., Brozovic, N., Reid, M. R., and Duncan, C. (1996). Bedrock incision, rock uplift and threshold hillslopes in the northwestern Himalayas. Nature, 379, 505-510. Clift, P., and Gaedicke, C. (2002). Accelerated mass flux to the Arabian Sea during the middle to late Miocene, Geology, 30, 207-210. Dadson, S. J., Hovius, N., Chen, H., Dade, W. B., Hsieh, M. L., Willett, S. D., Hu, J. C., Horng, M. J., Chen, M. C., Stark, C. P., Lague, D., and Lin, J. C. (submitted). Erosion of the Taiwan orogen, Nature. Davies, H. L., Lock, J., Tiffin, J. L., Okuda, Y., Murakami, F., and Kisimoto, K. (1987). Convergent tectonics in the Huon Peninsula region, Papua New Guinea, Geo-Marine Letters, 7, 143-152 Densmore, A. L., Anderson, R. S., McAdoo, B. G., and Ellis, M. A. (1997). Hillslope evolution by bedrock landslides, Science, 275, 369-372. Findlay, R. H., Arumba, J., Kagl, J., Nekitel, S., Ninkama, J., Kopi, G., Tekeve, B., and Abbott, L. D. (1998). The stratigraphy and tectonics of Markham, Geological Survey of Papua New Guinea Memoir 16. Foley, M. G. (1980). Bed-rock incision by streams. Geological Society of America Bulletin, Part II, 91, 2,189-2,213. Geli, L., Bard, P.Y., and Jullien, B. (1988) The effect of topography on earthquake ground motion: a review and new results, Bulletin of the Seismological Society of America, 78, 42-63. Gomez, B., Page, M., Bak, P., and Trustrum, N. (2002). Self-organized criticality in layered, lacustrine sediments formed by landsliding, Geology, 30, 519-522. Goudie, A. (1995). The Changing Earth, Blackwell, Oxford. Guzzetti, F., Malamud, B.D., Turcotte, D.L., and Reichenbach, P. (2002). Power-law correlations in landslide areas in central Italy, Earth and Planetary Science Letters, 195, 1