AMERICAN UNIVERSITY OF BEIRUT

RELIABILITY OF SPATIALLY VARIABLE UNDRAINED

SLOPES

by
GHINA ALI FAOUR

A thesis
submitted in partial fulfillment of the requirements
for the degree of Master of Engineering
to the Department of Civil and Environmental Engineering
of the Faculty of Engineering and Architecture
at the American University of Beirut

Beirut, Lebanon
September 2014
 DEDICATION
To my late father Ali Faour, I miss him every day.

His words of inspiration and encouragement

in pursuit of excellence, still linger on.

 ACKNOWLEDGMENTS

I would like to sincerely thank my advisor, Dr. Shadi Najjar for his cordiality,
guidance, and support throughout the work of this research. It was a remarkable
experience to work with him. I will forever be grateful for his role in making me what I
am today. I would also like to thank the rest of my thesis committee: Dr. Salah Sadek and
Dr. Ibrahim Alameddine for their encouragement and insightful comments.
Throughout my studies over the years, I have surrounded by friends who provided me
with continuous support and encouragement, I would like to thank them all and wish
them the best of luck.

Finally, I would like to thank my family for their enduring kindness, care and support.

v
 AN ABSTRACT OF THE THESIS OF

Ghina Ali Faour for Master of Engineering

Major: Geotechnical Engineering

Title: Reliability of Spatially Variable Undrained Slopes

Conventionally, the stability of a soil slope is evaluated by adopting a

deterministic approach that is based on a target global factor of safety that is calculated
either through limit equilibrium methods or through numerical analyses. Slope stability
analysis is a branch of geotechnical engineering that is highly amenable to uncertainties.
Spatial variability and model uncertainty are considered the major sources of
geotechnical uncertainties.

To account for such uncertainties in slope stability problems, numerous steps

have been undertaken in recent years to adopt a probabilistic stability analysis that
considers the uncertainties of soil properties in a systematic manner. However, there is
currently an inconsistency in the evaluation of the spatial uncertainty and no accounting
of the model uncertainty in the analysis.

The primary objective of this thesis is to provide slope stability investigators with
a robust reliability analysis that takes into consideration the combined uncertainty of
spatial variability and model uncertainty. To achieve this objective, a thorough
investigation is conducted to evaluate the model uncertainty of common slope stability
models (ex. Bishop, Ordinary Method of Slices, Janbu, and Spencer) by assembling and
analyzing a database of historical failures of slopes. The database is also used to
investigate the possibility of a lower-bound factor of safety for undrained slopes and its
impact on the reliability of slopes. The model uncertainty and the uncertainty due to
spatial variability are then combined within a reliability-based design framework to
recommend design factors of safety that would result in acceptable probabilities of failure
for undrained slopes.

vi
 CONTENTS

ACKNOWLEDGEMENTS……………………………………………… v

ABSTRACT…………………………………………………………………… vi

LIST OF ILLUSTRATIONS……………………………………………. xii

LIST OF TABLES…………………………………………………………... xx

Chapter

1. INTRODUCTION………………………………………………………. 1

1.1. Background ………………………………………………………… 1

1.2. Objective and Approach of Research……………………………….. 4
1.3. Thesis Organization ……………………………………………….... 6

2. LITERATURE REVIEW……………………………………………. 7
2.1. Introduction………………………………………………………….. 7
2.2. Application of Reliability Based Design to Slope Stability Problems 7

2.3. Studies Involving LEM with Random Field Theory………………... 11

2.3.1. Li and Lumb (1987) ……………………………………... 11
2.3.2. Christian et al. (1994) …………………............................. 12
2.3.3. Malkawi et al. (2000)…………………………………….. 13
……………………………………………...
2.3.4. El- Ramly et al. (2002)…………………………………… 14
2.3.5. Low (2003)……………………………………………….. 15
2.3.6. Babu and Mukesh (2004)………………………………… 16
2.3.7. Cho (2007)……………………………………………….. 16
2.3.8. Cho (2010)……………………………………………….. 17

vii
 2.3.9. Wang et al. (2011)……………………………………….. 18

2.4. Studies Involving FEM with Random Field Theory ………………... 19

2.4.1. Griffiths and Fenton (2000)……………………………… 20
2.4.2. Griffiths and Fenton (2004)……………………………… 21
2.4.3. Griffiths et al. (2009)…………………………………….. 22
2.4.4. Griffiths et al. (2010)…………………………………….. 23
2.4.5. Jha and Ching (2013)…………………………………….. 24

2.5. Studies Involving Probabilistic Slope Stability Analyses Based on

Search to Find the Minimum Reliability Index…………………….. 25

2.5.1. Hassan and Wolff (1999)………………………………… 26

…………………………………….…………..
2.5.2. Bhattacharya et al. (2003)………………………………... 27
…………………………………….……….
2.5.3. Xue and Gavin (2007)………………………………........ 27
2.5.4. Deng and Luna (2013)………………………………........ 28

2.6. Studies Involving System Reliability Analysis……………………... 29

2.6.1. Chowdhury and Xu (1995)………………………………. 26

…………………………………….…………..
2.6.2. Hong and Roh (2008)……………………………………. 31
…………………………………….…………..
2.6.3. Huang et al. (2013)……………………………………….. 32
…………………………………….…………..
2.6.4. Zhang et al. (2013)……………………………………….. 32

3. DATABASE COLLECTION………….……………………………. 34

3.1. Introduction………………………………………………………….. 34
3.2. Database Collection…………………………………………………. 35
3.2.1. Slide at Nesset ……………………………………............ 38
3.2.2. Slide at PresterØdbakken………………………………… 41
……………………………………............
3.2.3. Slide at Âs …………………………………….................. 43
3.2.4. Slide at Skjeggerod ……………………………………… 45
3.2.5. Slide at Tjernsmyr………………………………............... 48
3.2.6. Slide at Aulielva …………………….……………............ 50
3.2.7. Slide at Falknstein ……………………..…………............ 52
3.2.8. Slide at Jalsberg…………………..………………............ 54
3.2.9. Slide at Saint- Alban …………………………………….. 57
3.2.10. Slide at Narbonne …………………………………….... 60
3.2.11. Slide at Lanester ……………………………………….. 62

viii
3.2.12. Slide at Cubzac les-Ponts ………………………………. 65
3.2.13. Slide at Lodalen 1 …………………………………….... 67
3.2.14. Slide at Lodalen 2……………………………………..... 69
3.2.15. Slide at Lodaln 3 ……………………………………...... 70
3.2.16. Slide at Rio de janeiro ………………………………….. 70
3.2.17. Slide at New Liskeard ………………………………….. 73
3.2.18. Slide at Bangkok A …………………………………….. 75
3.2.19. Slide at Drammen V ……………………………………. 78
3.2.20. Slide at Drammen VI…………………………………… 79
3.2.21. Slide at Drammen VII …………………………….......... 80
3.2.22. Slide at Pornic…………………………………………... 81
3.2.23. Slide at Saint-Andre…………………………………….. 83
3.2.24. Slide at South of France……………………………........ 85
3.2.25. Slide at NBR Development……………………………... 86
3.2.26. Slide at Portsmouth……………………………………... 89
3.2.27. Slide at Kameda…………………………........................ 90
3.2.28. Slide at Khor Al-Zubair …………………………........... 94
3.2.29. Slide at Lian-Yun-Gang ………………………………... 96
3.2.30. Slide at Congress Street ………………………………... 97
3.2.31. Slide at Daikoku-Cho Dike ……………………………. 100
3.2.32. Slide at Cuyahoga AA ………………………………..... 102
3.2.33. Slide at King's Lynn ………………………………........ 104
3.2.34. Slide at Muar ………………………………................... 106
3.2.35. Slide at North Ridge Dam ……………………………… 107
3.2.36. Slide at Seven Sisters Dike……………………………... 110
3.2.37. Slide at Shellmouth Dam Test Fill……………………… 111
3.2.38. Slide at Juban ………………………………................... 113
3.2.39. Slide at Bradwell ……………………………….............. 115
3.2.40. Slide at Genesee ………………………………............... 117
3.2.41. Slide at Precambrian ………………………………........ 118
3.2.42. Slide at Scrapsgate ………………………………........... 120
3.2.43. Slide at Scottsdale………………………………............. 123
3.2.44. Slide at Iwai ………………………………..................... 124
3.2.45. Slide at Fair Haven…………………………………….... 126
3.2.46. Slide at Boston Marine Excavation …………………….. 128
3.2.47. Slide at Desert View Drive …………………………….. 130
3.2.48. Slide at Siburua ………………………………................ 132
3.2.49. Slide at Tianshenqiao …………………………………... 134
3.2.50. Slide at San Francisco Bay …………………………….. 136
3.2.51. Slide at Carsington ………………………………........... 138
3.2.52. Slide at Atchafalaya ………………………………......... 140

ix
4. QUANTIFICATION OF MODEL UNCERTAINTY AND
INVESTIGATION OF THE PRESENCE OF LOWER-
BOUND FACTORS OF FACTORS OF SAFETY………….. 150
4.1. Introduction………………………………………………………….. 150

4.2. Quantification of Model Uncertainty ……………………………….. 151

4.2.1. The Importance of Quantification of Model Uncertainty... 151

4.2.2. Mean, COV, and Distribution Type of Model Uncertainty 151
[Link]. Evaluation of the Statistical Parameters of the
Model Uncertainty………………………….. 152
[Link]. Probability Distribution of ………………….. 158
4.3. Evidence of a Lower-Bound Factor of Safety of Slopes……………. 160

4.4. Summary ……………………………………………………………. 164

5. IMPACT OF SPATIAL VARIABILITY IN THE

UNDRAINED SHEAR STRENGTH ON THE FACTOR
150
OF SAFETY OF UNDRAINED SLOPES……………………...
5.1. Introduction………………………………………………………….. 165

5.2. Characterization of Soils ………………………………..................... 166

5.2.1. Coefficient of Variation...................................................... 166

5.2.2. Scale of Fluctuation………………………………………. 166

5.3. Brief Summary of Work Done by Jha and Ching (2013)…………… 167

5.4. Method for Combining Uncertainties…………………..…………… 172

6. COMBINATION OF BOTH MODEL UNCERTAINTY

AND SPATIAL VARIABILITY…………………………………... 175

6.1. Introduction…………………………….……………………………. 175

6.2. Conventional Probability Distributions for Factors of Safety………. 175

6.2.1. Normal Distribution……………………………………… 175

x
 6.2.2. Log-normal Distribution……………………………….... 177

6.3. Distribution Types Adopted in the Analysis………………………… 179

6.4. Estimation of the Statistical Parameters of the Factor of Safety….… 180

6.5. Summary……………………………………………………….….… 180

7. RECOMMENDATION OF DESIGN FACTORS OF

SAFETY FOR UNDRAINED SLOPES………………………… 182
………………………….
7.1. Introduction……………………………………………...................... 182
7.2. Reliability – Based Design of Undrained Slopes…………................. 182

7.2.1. Effect of Coefficient of Variation and Scale of…

Fluctuation of Undrained Shear Strength on the…
Probability of Failure……………………………………. 183
7.2.2. Effect of Lower-Bound Factor of Safety on the...
Probability of Failure……………………………………. 191
7.2.3. Recommendation for the Factors of Safety of Undrained
Slopes…………………………………………………. 196
[Link]. Acceptable Probability of Failure…………….. 202
[Link]. Design Graphs for Factors of Safety of …
Slopes………………………………………… 204
7.3. Summary……………………………………….…………................. 212

8. CONCLUSIONS AND RECOMMENDATIONS ………….. 213

8.1. Introduction……………………………………………...................... 213

8.2. Summary of Findings……………………………………………….. 213

8.2. Design Steps………………………………………………………… 218

8.2.1. Estimation of the Probability of Failure of Undrained…

Slopes………………………………………………….. 218
8.2.2. Estimation of the Design Factors of Safety of Undrained
Slopes…………………………………………………… 219
REFERENCES…………………………………………………………. 220

xi
 ILLUSTRATIONS

Figure Page

3.1. Borehole Profile at Nesset (Flaate and Preber 1974)........................... 39

3.2. Critical Slip Surface (Flaate and Preber 1974)………………………. 40

3.3. Critical Slip Surface using SLIDE Software…………......................... 40

3.4. Borehole Profile at PresterØdbakken (Flaate and Preber 1974)........... 42

3.5. Critical Slip Surface at PresterØdbakken (Flaate and Preber 1974)..... 42

3.6. Critical Slip Surface Using SLIDE………………………………....... 43

3.7. Borehole Profile at Âs in Norway (Flaate and Preber 1974)................ 44

3.8. Critical Slip Surface at Âs in Norway (Flaate and Preber 1974).......... 44

3.9. Critical Slip Surface using SLIDE……………………….................... 45

3.10. View of the Slide Occurred at Skjeggerod in Norway (Flaate and

Preber 1974)………………………………………………………….. 45
3.11. Borehole Profile at Skjeggerod in Norway (Flaate and Preber 1974).. 46

3.12. Critical Slip Surface at Skjeggerod (Flaate and Preber 1974)……….. 47

3.13. Critical Slip Surface using SLIDE…………………………………… 47

xii
3.14. Borehole Profile at Tjernsmyr (Flaate and Preber 1974)…………….. 48

3.15. Critical Slip Surface at Tjernsmyr (Flaate and Preber 1974)………… 49

3.16. Critical Slip Surface at Tjernsmyr (Flaate and Preber 1974)………… 49

3.17. Slide at Aulielva (Flaate and Preber 1974)…………………………... 50

3.18. Borehole Profile at Aulielva (Flaate and Preber 1974)………………. 51

3.19. Critical Slip Surface at Aulielva (Flaate and Preber 1974)…………... 51

3.20. Critical Slip Surface using SLIDE…………………………………… 52

3.21. Borehole Profile at Falkenstein (Flaate and Preber 1974)…………… 53

3.22. Critical Slip Surface at Falkenstein (Flaate and Preber 1974)……..… 53

3.23. Critical Slip Surface Using SLIDE ………………………………...... 54

3.24. Borehole Profile at Jalsberg (Flaate and Preber 1974)……………….. 55

3.25. Critical Slip Surface at Jalsberg (Flaate and Preber 1974)…………... 56

3.26. Critical Slip Surface using SLIDE…………………………………… 56

3.27. Soil Profile at Saint-Alban (LA Rochelle et al. 1982)……………...... 58

3.28. Cross- section of the Saint-Alban embankment (LA Rochelle et al.

1982)…………………………………………………………………. 59

3.29. Critical Slip Surface Using SLIDE………………………………....... 59

3.30. Soil Profile at Narbonne (LARochelle et al. 1982)………………....... 60

3.31. Cross-section of the Narbonne embankment (LA Rochelle et al.

1982)…………………………………………………………………. 61

3.32. Critical Slip Surface using SLIDE………………………………….... 62

3.33. Soil Profile at Lanester (LA Rochelle et al. 1982)…………………… 63

3.34. View of the Lanester embankment after Failure (LA Rochelle et al.
1982)………………………………………………………………..... 63

xiii
3.35. Cross-section of the Lanester embankemt ((LA Rochelle et al. 1982). 64

3.36. Critical Slip Surface using SLIDE…………………………………… 64

3.37. View of the Cubzac embankment after failure (LA Rochelle et al.
1982………………………………………………………………....... 65

3.38. Soil Profile at Cubzac- les-Ponts (LA Rochelle et al. 1982)…………. 66

3.39. Cross-Section of the Cubzac embankment (LARochelle et al. 1982).. 66

3.40. Critical Slip Surface using SLIDE…………………………………… 67

3.41. Boring Profile at Lodalen (Sevaldson 1956)…………………………. 68

3.42. Actual and Critical Slip Surfaces at Lodalen (Sevaldson 1956)……... 68

3.43. Critical Slip Surface using SLIDE at Lodalen 1……………………... 69

3.44. Results of Total Stress Analysis using Slide Software at Lodalen 2… 70

3.45. Results of Total Stress Analysis using Slide Software at Lodalen 3… 70

3.46. Summary of Geotechnical Properties, Rio de Janeiro Soft Gray Clay

(Ramalho-Ortigão et al. 1983)……………………………………….. 71

3.47. Results of Total Stress Stability Analysis (Ramalho-Ortigão et al.

1983)…………………………………………………………………. 72

3.48. Results of Total Stress Stability Analysis using SLIDE software…… 73

3.49. Undrained shear strengths used in total stress stability analyses

(Lacasse et al.1977)…………………………………………………... 74

3.50. LEASE-I critical failure arcs at New Liskeard (Lacasse et al. 1977)... 74

3.51. SLIDE critical failure arc at New Liskeard………………………....... 75

3.52. Geotechnical Profile at Bangkok site (Eide and Holmberg 1972)…… 75

3.53. Critical Slip Surface (Eide and Holmberg 1972)…………………….. 77

3.54. Critical Slip Surface using SLIDE software…………………………. 77

xiv
3.55. Soil Properties at Drammen (Kjærnsli and Simons (1962)…………... 78

3.56. Critical Slip Surface (Kjærnsli and Simons 1962)…………………… 79

3.57. Critical Slip Surface Using SLIDE at Drammen V………………….. 79

3.58. Critical Slip Surface using SLIDE software at Drammen VI………... 80

3.59. Critical Slip Surface using SLIDE software at Drammen VII……...... 81

3.60. Embankment Failure at Pornic (Pilot 1972)………………………...... 81

3.61. Embankment Failure at Saint-Andre (Pilot 1972)…………………… 84

3.62. Embankment Failure at Saint-Andre using SLIDE…………………... 84

3.63. Embankment Failure at South of France (Pilot 1972)……………….. 85

3.64. Critical Slip Surface at South of France using SLIDE……………….. 86

3.65. Stratigraphy and Geotechnical Characteristics of the foundation Soil

at NBR Development (Dascal et al. 1972)…………………………… 87

3.66. Results of the total stress stability analysis at NBR Development in

France (Dascal et al. 1972……………………………………………. 88

3.67. Critical Slip Surface using SLIDE………………………………….... 88

3.68. Soil Profile, Index Properties and Field Vane Strengths at…
Portsmouth in USA (Ladd 1972)…………………………………...... 89

3.69. Results of Total Stress Analysis (Ladd 1972)………………………... 90

3.70. Critical Slip Surface using SLIDE…………………………………… 90

3.71. Soil conditions at Kameda site (Hanzawa et al. 1994)……………….. 91

3.72. Undrained Shear Strength at Kameda Site (Hanzawa et al. 1994)…... 92

3.73. Circular Slip Surface (Hanzawa et al. 1994)…………………………. 93

3.74. Circular Slip Surface using SLIDE……………………....................... 93

3.75. Soil Properties at Khor Al-Zubair Site (Hanzawa 1983)…………...... 94

xv
3.76. Circular Slip Surface at Khor Al-Zubair Site (Hanzawa 1983)…….... 95

3.77. Embankment Failure at Khor-Al-Zubair using SLIDE………………. 95

3.78. The Index and the Mechanical Properties of the Subsoil (Chai et al.
2002)………………………………………………………………..... 96

3.79. Failure Surfaces from Field Observations and in Slip Circle Analysis
(Chai et al. 2002)…………………………………………………....... 97

3.80. Failure Surface using SLIDE……………………………………….... 97

3.81. Compressive Strength and Water- Content at Congress Street

(Ireland 1954)……………………………………………………........ 98

3.82. Total Stress Stability Calculations (Ireland 1954)…………………… 99

3.83. Critical Slip Surface using SLIDE………………………………….... 100

3.84. Undrained Shear Strength at Daikoku (Kishida et al. 1983)……....... 101

3.85. Circular Slip Surface at Daikoku Site (Kishida et al. 1983)…………. 101

3.86. Total Stress Stability Analysis using SLIDE………………………… 102

3.87. Soil Properties and Critical Slip Surface at Cuyahoga AA site (Wu et
al. 1975)…………………………………………………………….... 103

3.88. Critical Slip Surface using SLIDE…………………………………… 103

3.89. Soil Properties at King’s Lynn (Wilkes et al. 1972)…………………. 104

3.90. Observed and calculated Slip Surface (Wilkes et al. 1972)………….. 105

3.91. Critical Slip Surface using SLIDE………………………………….... 105

3.92. Variation of Soil Properties with Depth (Indraratna et al. 1992)…….. 106

3.93. Critical Slip Surface at Muar using SLIDE…………………………... 107

3.94. Stability Analyses, North Ridge Dam (Rivard et al. 1978)………….. 109

3.95. Stability Analyses, SLIDE software…………………………………. 109

xvi
3.96. Stability Analyses, Seven Sisters Dike (Rivard et al. 1978)…………. 111

3.97. Stability Analyses using SLIDE for Seven Sisters Dike………...…… 111

3.98. Stability Analyses using SLIDE for Shellmouth Test Fill Dam……... 113

3.99. Schematic Model used in the analysis (Zhang et al. 2005)…………... 114

3.100 Critical Slip Surface using SLIDE…………………………………… 114

3.101. Properties of London Clay at Bradwell (Skempton and LaRochelle

1965)……………………………………………………..................... 115

3.102. Cross Section of Excavated Slope at Bradwell (Skempton and

LaRochelle 1965)…………………………………………………... 116

3.103. Critical Slip Surface at Bradwell using SLIDE………………………. 116

3.104. Stratigraphy and Engineering Properties of the soil at Genesee (Been

et al. 1986)…………………………………………………………. 117

3.105. Slope Geometry at Genesee (Been et al. 1986)………………………. 118

3.106. Critical Slip Surface at Genesee using SLIDE……………………….. 118

3.107. Soil Profile at Precambrian (Dascal et al. 1975)……………………... 119

3.108. Total Stress Analysis (Dascal et al. 1975)…………………………… 120

3.109. Stability Analysis at Precambrian using SLIDE……………………... 120

3.110. Shear Strength Values at Scrapsgate (Golder et al. 1954)…………… 121

3.111. Stability Analysis at Scrapsgate (Golder et al. 1954)………………... 122

3.112. Stability Analysis using SLIDE……………………………………… 122

3.113. Soil Properties at Scottsdale (Parry 1968)…………………………… 123

3.114. Critical Slip Surface using SLIDE…………………………………… 124

3.115. Soil Properties of Organic Clay (Shogaki et al. 2008)……………….. 124

3.116. Soil Properties of Clay1 & Clay2 (Shogaki et al. 2008)……………... 125

xvii
3.117. Soil Properties of Clay1 & Clay2 (Shogaki et al. 2008)……………... 126

3.118. Critical Slip Surface using SLIDE…………………………………… 126

3.119. Stability Analyses (Haupt and Olson 1972)………………………….. 127

3.120. Critical Slip Surface using SLIDE…………………………………… 128

3.121. Undrained Shear Strength using Field Vane (McGinn et al. 1993)….. 129

3.122. Stability Analyses (McGinn et al. 1993)……………………………... 129

3.123. Stability Analyses using SLIDE……………………………………... 130

3.124. Critical Slip Surface at Desert View Drive (Day 1996)……………… 131

3.125. Critical Slip Surface using SLIDE…………………………………… 131

3.126. Total Stress Analysis at Siburua (Wolfskill et al. 1967)……………... 133

3.127. Stability Analyses using SLIDE……………………………………... 133

3.128. Geological Profile at Tianshenqiao (Chen and Shoe 1988)………….. 135

3.129. Critical Slip Surface using SLIDE…………………………………… 135

3.130. Undrained Shear Strength using Field Vane Tests (Duncan and
Buchignani 1973)…………………………………………………….. 136

3.131. Slope Geometry at San Fransisco (Duncan and Buchignani 1973)….. 137

3.132. Critical Slip Surface using SLIDE…………………………………… 137

3.133. Profile at Carsington Dam (Skempton and Coats 1985)……………... 139

3.134. Critical Slip Surface using SLIDE…………………………………… 139

3.135. Soil Properties at Atchafalaya Levees (Kaufman et al.1967)………... 140

3.136. Geometry and Soil conditions at Atchafalaya Site (Kaufman et al.

1967)…………………………………………………………………. 141

3.137. Ctitical Slip Surface using SLIDE…………………………………… 141

xviii
4.1. Values of the ratio of measured to predicted factor of safety for the
52 case histories……………………………………………………… 155
4.2. Values of the model uncertainties for the 52 cases (Shear strength
corrected to UU)……………………………………………………… 156
4.3. Actual and Theoretical best-fit CDFs for the model uncertainty ()… 159

4.4. Relation between sensitivity and liquidity index…………………….. 161

4.5. Evidence of a lower-bound factor of safety of 52 slope failure cases.. 162

4.6. Comparison of measured, predicted, and lower-bound factor of

safety of 52 slope failures……………………………………………. 163

4.7. Predicted Lower-Bound Factors of Safety Excluding Cases with

Highly Sensitive Soils………………………………………………... 164

5.1. Determination of vertical scale of fluctuation(Jha and Ching 2013)… 168

5.2. Relationship between (FSd - FS)/ FSd and V(Jha and Ching 2013) 169

5.3. Relationship between VFS/V and (V, z /Lf , x/ z)(Jha and Ching
2013)…………………………………………………………………. 171
5.4. Relationship between actual VFS/V and VFS/V estimated by Equation
2……………………………………………………………………… 172
6.1. Normal Probability Distribution……………………………………... 176

6.2. Lognormal Probability Distribution………………………………….. 178

6.3. Probability density function of truncated lognormal………………… 179

7.1. Variation of Pf with V and z /Lf for FS =1.5 & FS = 1.7…………… 185

7.2. Variation of Pf with V and z /L for FS =2 & FS = 2.25…………... 186

7.3. Variation of Pf with V and z /Lf for FS =2.5 & FS = 2.75………… 187

7.4. Variation of Pf with V and z /Lf for FS =1.5 by considering spatial

variability only……………………………………………………... 190

7.5. Variation of Pf with V and Sensitivity for z /Lf = 0.1, FS =1.5&

FS=1.7………………………………………………………………... 192

xix
7.6. Variation of Pf with V and Sensitivity for z /Lf = 0.1, FS =2.00&
FS=2.25………………………………………………………………. 193

7.7. Variation of Pf with V and Sensitivity for z /Lf = 0.1, FS =2.50 &
FS=2.75………………………………………………………………. 194

7.8. Recommended factor of safety needed to accomplish a target Pf for

different sensitivities for V =0.1&V=0.2 and z /Lf = 0.1………... 197

7.9. Recommended factor of safety needed to accomplish a target Pf for

different sensitivities for V =0.3&V=0.4 and z /Lf = 0.1………….. 198

7.10. Recommended factor of safety needed to accomplish a target Pf for

different sensitivities for V=0.5 and z /Lf = 0.1…………………… 199

7.11. Recommended factor of safety needed to accomplish a target Pf for

different sensitivities for V=0.1 and V=0.2 and z /Lf = 0.2……….. 200

7.12. Recommended factor of safety needed to accomplish a target Pf for

different sensitivities for V=0.3 and V=0.4 and z /Lf = 0.2……….. 201

7.13. Recommended factor of safety needed to accomplish a target Pf for

different sensitivities for V=0.5 and z /Lf = 0.2………………….. 202

7.14. Recommended factors of safety for different coefficients of variation

for probabilities of failure of 0.001& 0.005 for slopes with ratio z /
Lf = 0.1………………………………………………………………. 205

7.15. Recommended factors of safety for different coefficients of variation

to achieve probabilities of failure of 0.01 & 0.05 for slopes with ratio
z / Lf =0.1……………………………………………………………. 206

7.16. Recommended factors of safety for different coefficients of variation

to achieve probability of failure of 0.1 for slopes with ratio z / Lf
=0.1…………………………………………………………………... 207

7.17. Recommended factors of safety for different coefficients of variation

to achieve probability of failure of 0.001 and 0.005 for slopes with
ratio z/Lf = 0.2……………………………………………………… 208

7.18. Recommended factors of safety for different coefficients of variation

to achieve probabilities of failure of 0.01 & 0.05 for slopes with ratio
of z/Lf = 0.2…………………………………………………. 209

xx
7.19. Recommended factors of safety for different coefficients of variation
to achieve probabilities of failure of 0.1 for slopes with ratio z/ Lf =
0.2…………………………………………………………………….. 210

xxi
 TABLES

Table Page

3.1. Database of undrained slope failure cases.............................................. 36

3.2. Soil properties adopted for the analyses of the 52 cases......................... 142

3.3. Main soil characteristics of soils at Saint- Andre in France (Pilot

1972)…………………………………………………………………... 83

3.4. Soil properties of highly plastic clay North Ridge Dam (Rivard et al.
1978)………………………………………………………………....... 108

3.5. Soil properties of highly plastic clay, Seven Sisters Dike...................... 110

3.6. Stability analyses, Shellmouth Dam test fill........................................... 112

3.7. Geotechnical soil parameters used in the stability analysis for

Tianshenqiao (Chen and Shoe 1988)………………………………….. 134

3.8. Soil stratification ar Carsington Dam (Skempton and Coats 1985)…… 138

4.1. Predicted and lower-bound factors of safety for the slopes in Database 154

4.2. Statistical parameters of  for the four LEM models………………….. 155

4.3. Statistical parameters of the model uncertainty (after shear strength

correction)…………………………………………………………….. 157

xxii
4.4. Statistical parameters of the model uncertainty (after removing the
sensitive cases)………………………………………………………… 158

6.1. Estimation of  and  of the factor of safety………………………….. 181

7.1. Acceptable probability of failure of slopes (Santamarina et al. 1992)… 203

7.2. Recommended factors of safety for different acceptable probability of

failure for V = 0.1 & z/Lf = 0.1………………………………………. 210

7.3. Recommended factors of safety for different acceptable probability of

failure for V = 0.3 & z/Lf = 0.1………………………………………. 211

7.4. Recommended factors of safety for different acceptable probability of

failure for V = 0.5 & z/Lf = 0.1………………………………………. 211

7.5. Recommended factors of safety for different acceptable probability of

failure for V = 0.1 & z/Lf = 0.2………………………………………. 211

7.6. Recommended factors of safety for different acceptable probability of

failure for V = 0.3 & z/Lf = 0.2………………………………………. 212

7.7. Recommended factors of safety for different acceptable probability of

failure for V = 0.5 & z/Lf = 0.2………………………………………. 212

xxiii
 CHAPTER 1

INTRODUCTION

1.1 Background

Traditionally, the stability of a clay slope is evaluated by adopting a deterministic

approach that is based on a global factor of safety obtained either through limit

equilibrium approaches or through numerical analyses. Slope stability analysis is a

branch of geotechnical engineering that is highly amenable to uncertainties. To account

for the different sources of uncertainties and reduce the risk of slope failures, common

practice involves the use of factors of safety that are generally greater than 1.5 (Terzaghi

and Peck 1948). Schweiger et al. (2001) reports that the deterministic approach is simple

and straightforward but does not lead to a realistic mathematical treatment of the

uncertainties involved in the models and parameters affecting the design of slopes. Li and

Lumb (1987) recognized that the factor of safety is not a consistent measure of risk since

slopes with the same safety factor value may exhibit different risk levels depending on

the variability of the soil properties. Accordingly, to account for such variability,

numerous studies have been undertaken in recent years to adopt a probabilistic approach

for slope stability analysis that deals with the uncertainties of soil properties in a

systematic and explicit manner. This probabilistic analysis can facilitate the development

of new perspectives concerning risk and reliability of slopes that are outside the scope of

conventional deterministic models.

1
 Reliability analysis of slope stability has attracted considerable research attention

in the past few decades. Almost all probabilistic methods described in the literature have

at some point been applied to slope stability problems. In general, the probabilistic

procedures differ in assumptions, limitations, capability to handle complex problems, and

mathematical complexity. In the last few decades, some research efforts have targeted

analyzing the effect of spatial variability in the soil properties on the stability of slopes in

the framework of a reliability analysis. Li and Lumb (1987), Christian et al. (1994),

Malkawi et al.(2000), El- Ramly et al. (2002), Low (2003), Babu and Mukesh (2004),

Cho (2007), Cho (2010), and Wang et al. (2011) have targeted analyzing the effect of

spatial variability by using limit equilibirum methods with random field theory. The

above studies differ in both the deterministic and probabilistic methods used in the

analysis and also in the way spatial variability is defined. Malkawi et al. (2000) and Low

(2003) studied the effect of spatial variability by varying the coefficient of variation of

soil properties. On the other hand, Li and Lumb (1987), El- Ramly et al. (2002), Cho

(2007), and Wang et al. (2011) used random fields with an isotropic correlation structure

for defining spatial variability. Conversely, Babu and Mukesh (2004) found that random

fields with an anisotropic correlation structure should be utilized since soil properties

could exhibit a significant degree of anisotropy.

All the above studies have combined the limit equilibrium method (LEM) with

random field theory. However, the inherent nature of LEM is that it leads to a critical

failure surface which could be non-circular in 2-D analysis and the influence of the

random field is only taken into account along the one-dimensional failure line. Thus, to

overcome this limitation, Griffiths and Fenton (2000), Griffiths and Fenton (2004),
2
Griffiths et al. (2009), Griffiths et al. (2010) and Jha and Ching (2013) pursued a more

rigorous method of probabilistic slope stability analysis in which nonlinear finite-

element methods are combined with random field generation techniques. This approach is

currently referred to in the literature as the Random Finite Element Method (RFEM). The

approach captures the effect of soil spatial variability and fully accounts for spatial

correlation and averaging. It is also a powerful slope stability analysis tool that does not

require priori assumptions related to the shape or location of the failure mechanism.

Griffiths (2000) studied the effect of the scale of fluctuation and the coefficient of

variation of soil properties. Griffiths (2004) performed a comparison between the simple

and advanced probabilistic approach to study the effect of spatial variability and local

averaging on the probability of failure of slope. Moreover, Griffiths (2009) built on the

work done by Griffiths (2004) and studied the effect of the inclination angle of the slope

on the probability of failure. Additionally, Griffiths (2010) performed a comparison

between limit equilibrium methods and the random finite element method to indicate the

importance and superiority of the RFEM in the stability of slopes.

In a recent study, Jha and Ching (2013) performed a robust and rigorous

probabilistic slope stability analysis using the Random Finite Element Method to study

the effect of slope geometry, mean and coefficient of variation of the soil parameters, and

the scale of fluctuation on the probability of failure of undrained slopes. The authors

conducted the study by collecting a database for 34 real undrained engineered slope

cases. The paper was aimed at quantifying the effect of spatial variability in the undrained

shear strength of clays on the probability of failure of the slopes. An advanced model of

3
spatial variability that takes into account vertical and horizontal spatial variability was

adopted. The vertical scale of fluctuation in the undrained shear strength was back-

calculated for each case in the database using the simplified method presented in Phoon

and Kulhawy (1999). The horizontal scale of fluctuation in the undrained shear strength

was assumed due to the lack of soil data (boreholes) needed to quantify the lateral spatial

variability. One of the major contributions of the study is a relationship between the mean

and the coefficient of variation of the factor of safety from one hand and the slope

geometry, mean and the coefficient of variation of the soil properties, and the scale of

fluctuation in the undrained shear strength from the other hand.

It should be noted that the majority of the published work that is related to

reliability-based design of slopes targets the issue of spatial variability and its effect on

the calculated reliability of the slope. Regarding the evaluation of model uncertainty, only

Malkawi et al. (2000) estimated the model uncertainty by comparing the performance of

the limit equilibrium methods with the Spencer’s method that is considered the most

accurate and rigorous. There are no published studies that aim at characterizing the model

uncertainty of available slope stability methods with published case histories of failed

slopes.

1.2 Objective and approach of research

The primary objective of this thesis is to provide slope stability investigators with

a robust reliability-based design procedure that takes into consideration the combined

uncertainty of spatial variability and model uncertainty. The main goal is to provide

4
designers with a systematic and defendable approach for estimating the probability of

failure of slopes. The main backbone of the proposed study is a simplified and realistic

reliability-based design approach for estimating the probability of failure of slopes by

incorporating the effects of model uncertainty and spatial variability in the probability

distribution of the factor of safety of the slope. What differentiates this tool from other

tools available in the literature is the incorporation of the model uncertainty in the slope

probabilistic analysis. The previous reliability studies found that it is difficult to have

historical observations to compare with results of slope stability methods. Thus, they

don’t take into account the model uncertainty. However, in this study an effort is made to

collect a database that includes historical observations of failed slopes which will allow

for the estimation of the model uncertainty of commonly used slope stability prediction

models. The second objective of the study is to investigate the existence of a physical

lower-bound factor of safety of the slope, which if incorporated in the modeling of the

probability distribution of the factor of safety, could provide a more realistic

quantification of reliability and a more rational basis for design.

These goals will be achieved through the following tasks:

1. Conduct an expanded literature review on slope stability and reliability-based

design of slopes.

2. Collect a database that includes historical failure observations of real slopes.

3. Use the collected database to investigate the existence of a lower-bound factor of

safety for a given slope and to quantify the model uncertainty of the limit

equilibrium methods used for slope stability analyses.

5
 4. Investigate the impact of spatial variability in the undrained shear strength on the

factor of safety of undrained slopes.

5. Combine both the model and the spatial uncertainties to evaluate the statistical

parameters of the factor of safety of an undrained slope.

6. Propose a reliability-based design framework to recommend design factors of

safety that would result in acceptable probabilities of failure for undrained slopes.

1.3 Organization of thesis

In chapter 2, a literature review of some of the recent works done in this field will

be addressed. The literature review targets studies that include historical failure cases of

undrained slopes. It also targets studies where reliability-based design concepts were

employed in the analysis of slope stability. In chapter 3, a database of failure historical

observations of undrained slopes is assembled. In chapter 4, the collected database is used

to analyze biases and uncertainties in current models for predicting the factor of safety of

slopes. Moreover, the database is used to investigate the presence of the lower-bound

factor of safety. In chapter 5, the work done by Jha and Ching (2013) is used to

investigate the impact of spatial variability in the undrained shear strength on the factor

of safety of undrained slopes. Both model uncertainty and spatial variability are

combined in chapter 6 to evaluate the statistical parameters of the factor of safety of

undrained slopes. Finally, a reliability-based design framework is proposed to

recommend design factors of safety that would result in acceptable probabilities of failure

for undrained slopes in chapter 7. Conclusions and contributions of this research are

presented in chapter 8.
6
 CHAPTER 2

LITERATURE REVIEW

2.1 Introduction

Conventionally, the stability of a soil slope is evaluated by adopting a

deterministic approach that is based on a global factor of safety that is generally

evaluated either through limit equilibrium approaches or through numerical analyses. The

input to these slope stability analyses include deterministic soil parameters that are

assigned to the different soil layers. Normally, the selection of design input soil

parameters to be used in the slope stability analyses is based on local experience and

engineering judgment and is most of the time on the conservative side (Schweiger et al.

2001).

To cater for the different sources of uncertainties in the input parameters and in

the predictive models, global factors of safety that generally exceed 1.5 are generally

adopted in slope stability analyses to ensure stafety (Terzaghi and Peck 1948). Schweiger

et al. (2001) indicated that the deterministic approach is simple and straightforward but

does not give realistic and mathematical treatment to the uncertainties involved in the

input soil parameters. Moreover, Li and Lumb (1987) recognized that the factor of safety

is not a consistent measure of risk since slopes with the same safety factor value may

exhibit different risk levels depending on the variability of the soil properties.

Accordingly, to account for such variability, numerous studies have been undertaken in

recent years to adopt a probabilistic stability analysis that deals with the uncertainties of
7
soil properties in a systematic manner. This probabilistic analysis can facilitate the

development of new perspectives concerning risk and reliability that are outside the scope

of conventional deterministic models. According to Kulhawy (1996), a reliability analysis

is the consistent evaluation of design risk using probability theories, and reliability-based

design is any design approach that uses reliability analyses.

2.2 Application of reliability based design to slope stability problems

Slope stability analysis is a branch of geotechnical engineering that is highly

amenable to probabilistic treatment. Almost all of the probabilistic methods described in

the literature have at some point been applied to slope stability problems. Reliability

analysis of slope stability has attracted considerable research attention in the past few

decades.

The reliability of slopes is frequently measured by a ‘‘reliability index,’’, or a

failure probability, Pf, which is defined as the probability that the minimum factor of

safety (FS) is less than unity (i.e., Pf = P(FS < 1)). The ‘‘reliability index,’’ is

evaluated as = --1(Pf) where -1() is the inverse of the standard normal cumulative

distribution function. Various methods have been proposed to estimate and (or) Pf. The

earliest studies appeared in the 1970’s [e.g. Wu and Kraft (1970); Cornell (1971); Matsuo

and Kuroda (1974); Alonso(1976); Tang et al. (1976); and Vanmarcke (1977)] and have

continued steadily [e.g., D’Andrea and Sangrey (1982); Chowdhury and Tang (1987); Li

and Lumb (1987); Oka and Wu (1990) ;Mostyn and Li (1993); Lacasse (1994); Christian

et al. (1994);Chowdhury and Xu (1995); Wolff (1996); Christian (1996); Lacasse and

8
Nadim (1996); Low (1996); Low and Tang (1997a,b); Low et al.(1998); Hassan and

Wolff (1999);Malkawi and Abdulla (2000); Whitman(2000); Duncan (2000); El-Ramly

et al. (2002); Low (2003); Baecher and Cristian (2003); Bhattacharya et al. (2003);

Griffiths and Fenton (2004); Babu and Mukesh (2004); Xu and Low(2006); Low et al.

(2007); Cho (2007); Griffiths et al. (2007); Shinoda (2007); Xue and Gavin (2007); Hong

and Roh (2008); Srivastava and Babu (2008); Griffiths et al .(2009); Cho (2010);

Griffiths et al. (2010); Wang et al. (2010); Kasama and Zen (2011); Christian et al.

(2013); Huang et al. (2013); Zhang et al.(2013); Deng and Luna (2013); Jha and Ching

(2013)].

The above studies differ in the probabilistic procedures for slope stability

analysis. In general, the probabilistic procedures differ in assumptions, limitations,

capability to handle complex problems, and mathematical complexity. These probabilistic

methods are divided into two main categories: (1) approximate methods (traditional

methods) i.e. the First Order Reliability Method (FORM), the First Order Second

Moment Method (FOSM), the Point Estimate Method (PEM), and the Monte Carlo

Simulations Method (MCSM) which are generally used in conjunction with Limit

Equilibrium Slope Stability methods and (2) more advanced methods that use the

Random Finite Element Method. The approximate methods make simplifying

assumptions that limit their application to specific classes of problems. Some studies use

very simple slope models such as the Ordinary Method of Slices (Tang et al. 1976), while

others deal only with frictionless soils (Vanmarcke 1977 and Matsuo and Kuroda 1974).

Moreover, many studies restrict their analyses to a circular slip surface (Vanmarcke 1977

and Alonso 1976). Another limitation of these approximate methods is the inability of

9
these methods to provide any information about the shape of the probability distribution

function where these methods can allow the estimation of the mean and the variance of

the factor of safety only. Finally, the major limitation of these methods is the ignorance of

the spatial variability of soil properties.

Spatial variability is considered one of the major sources of geotechnical

variability in slope stability. Geotechnical variability is a complex attribute that results

from many sources of uncertainties. Spry et al. (1988), Orchant et al. (1988), Filippas et

al. (1988), Kulhawy et al. (1992), Christian et al. (1994), Phoon et al. (1995) and Phoon

et al. (1999) investigated geotechnical variability. Both Christian et al. (1994) and Phoon

et al. (1999) illustrated that the uncertainties in soil properties are comprised from two

sources: scatter in the data and systematic error in the estimate of the properties. Scatter

in the data is due to the real spatial variability within the profile and due to random

testing errors or noise. However, systematic error is due to the statistical error in the mean

value of the property that results from the limited number of tests performed and the bias

in the measurement.

Lacasse and Nadim (1996) stated that variability is attributed to factors such as

variations in mineralogical composition, conditions during deposition, stress history, and

physical and mechanical decomposition processes. To model the spatial variability in

problems involving slope stability, statistical parameters such as the mean and the

variance of the soil properties must be estimated. However, these statistical parameters

are one point statistical parameters and cannot capture the features of the spatial

correlation structure of the soil (El- Ramly et al. 2002). Spatial variations of soil

properties can be effectively described by their correlation structure within the framework

10
of random fields (Vanmarcke 1983). To describe this correlation structure, an

autocorrelation distance is defined which is the distance within which soil properties

show a strong correlation. A large autocorrelation distance value implies that the soil

property is highly correlated over a large spatial extent, resulting in a smooth variation

within the soil profile. On the other hand, a small value indicates that the fluctuation of

the soil property is large (Cho 2010). Some studies in the literature assume isotropic

correlation structure. In contrast, other studies assume anisotropic correlation, when the

investigators recognized that the correlations in the vertical direction tend to have much

shorter distances than those in the horizontal direction.

Below is a summary of the research studies which targeted the probabilistic slope

stability analysis taking into consideration spatial variability.

2.3 Studies involving LEM with random field theory

In spite of the fact that most traditional limit equilibrium methods (LEMs) do not

consider spatial variability, some investigators investigated the impact of spatial

variability of soil properties for slopes by combining the LEM with random field theory.

The theory of random fields (Varmarcke 1977a, 1977b, 1983) is a common approach for

modeling the spatial variability of soil properties. It is also the basis of probabilistic slope

analysis methodology.

2.3.1 Li and Lumb (1987)

Li and Lumb (1987) conducted one of the earliest probabilistic slope stability

analyses that combines LEM with random field theory. The approach presented in the
11
study adopted the Morgenstern and Price method (1965) which is commonly accepted as

one of the accurate and rigorous methods for slope stability analysis that incorporates the

general slip surface. Li and Lumb adopted the First Order Second Moment Method

(FOSM) with some new developments of the technique for analyzing the reliability of

slopes. These developments revolved around defining the reliability index by Hasofer and

Lind (1974) which is considered an invariant risk measure in which all equivalent formats

of the performance function yield the same reliability index. Spatial variability was taken

into account in Li and Lumb study by defining an isotropic correlation structure. The

authors found from their analysis that the probability of failure of the slope is sensitive to

the scale of fluctuation, and recommended that the reliability analyst must pay more

attention to the estimation of the scale of fluctuation. The assumption of perfect

correlation in soil properties was found to lead to an overestimation of the probability of

failure. Nevertheless, Li and Lumb showed that the locations of the deterministic critical

slip surface and the surface with minimum reliability index are very close to each other.

Thus, the authors recommended the use of the deterministic critical slip surface as an

initial trial surface for the general search for the critical slip surface with minimum

reliability index.

2.3.2 Christian et al. (1994)

Christian et al. (1994) conducted a probabilistic analysis using the First Order

Second Moment Method (FOSM) to evaluate the reliability index of slopes. The authors

illustrated the approach by the analysis of a well-known case history (James Bay

12
Embankments). The construction of the dyke followed three scenarios. The first two

alternatives were the construction of the embankment in a single stage either to a height

of 6 or 12m. The second alternative was the multi stage construction. Spatial variability

in addition to systematic and model errors were all taken into consideration in the

analysis. In contrast, errors due to bias were not considered in the analysis due to the

difficulty in determining the magnitude of the errors. Local averaging was considered in

the analysis. Bishop’s method was conducted to evaluate the factor of safety for the dyke

for the first two alternatives. Conversely, the stability analysis for the multistage

construction alternative was done by Morgenstern Price Method. The results showed that

there are difficulties in identifying both the autocorrelation distance and bias. Bias is

considered a significant contributor to the overall uncertainty; however, in general it is

ignored from the analysis. Hence, the authors recommended that the engineers should be

careful when they rely on their judgment to establish the bias contribution.

2.3.3 Malkawi et al. (2000)

Malkawi et al. (2000) investigated the effect of deterministic models [Bishop,

Ordinary Method of Slices (OMS), Janbu, and Spencer] and probabilistic models [First

Order Second Moment Method (FOSM) and Monte Carlo Simulation Method (MCSM)]

on the reliability of homogenous and layered slopes. The authors included the spatial

variability into their analysis and studied the effect of uncertainty of each soil property on

the calculated factor of safety by varying the coefficient of variation of the soil properties.

Furthermore, they conducted a sensitivity analysis to investigate the effect of the seed
13
random number generator and the sample size of soil properties needed for Monte Carlo

Simulation on the reliability index of slopes. They ended up with the following results: In

case of homogenous slopes, both OMS and Bishop result in the same reliability index

regardless of the reliability method used. Conversely, Janbu and Spencer models exhibit

some differences. The FOSM slightly overestimates the reliability index in the case of

Janbu’s model, whereas the MCSM overestimates the reliability index in the case of

Spencer’s model. For the case of layered slopes, only Spencer’s model results in a slight

variation between FOSM and MCSM. The authors also concluded that isn’t sensitive to

the selected random number generator. In contrast, it is sensitive to the sample size of soil

properties where greater than 700 samples are needed in the analysis. Consequently, they

found that FOSM requires fewer calculations and computing time compared to MCSM.

However, with the help of computers in data handling and speed, MCSM proved to be

powerful and effective method for probabilistic reliability analysis.

In spite of the fact that Malkawi et al. (2000) combined LEM with Random Field Theory

to take spatial variability into consideration, they neglected spatial correlation from in

their analysis.

2.3.4 El-Ramly et al. (2002)

El- Ramly et al. (2002) conducted a practical probabilistic slope stability analysis

based on Monte Carlo Simulation by developing a simple spreadsheet using the well-

known software Microsoft Excel 97 and @Risk. The analysis is illustrated by analyzing

14
the dykes of the James Bay hydroelectric project. The authors modeled the geometry, soil

properties, stratigraphy, and slip surface in an excel spreadsheet. The Bishop method is

used for the determination of the deterministic factor of safety. The uncertainties in input

parameters are modeled statistically by representative probability distributions. The

variances of the soil parameters are evaluated using judgment; moreover, the bias in the

vane measurements is adjusted by Bjerrum’s correction factor which also is considered

uncertain in the analysis. The spatial variability of soil parameters was characterized by

an isotropic autocorrelation distance assuming exponential autocovarience functions.

Finally, the authors investigated the efficiency of the analysis by comparing the results

obtained by those obtained using First Order Second Moment Method (FOSM).

The authors concluded that the reliability of a design could be significantly reduced by

the use of empirical factors and correlations and this is proven by the sensitivity analysis

undertaken to study the impact of the uncertainty of Bjerrum’s coefficient on the

reliability of the slope. Additionally, they deduced that ignoring spatial variability of soil

properties and assuming perfect correlation can significantly overestimate the probability

of failure of slopes.

2.3.5 Low (2003)

Low (2003) implemented Spencer’s method both deterministically and

probabilistically in a spreadsheet platform. The author accomplished the search of the

noncircular failure surface by using Spencer’s method involving spatially correlated

15
normal and lognormal variates. Then, he extended the use of the deterministic approach

to the probabilistic analysis (FOSM) in order to calculate the Hasofer-Lind reliability

index simply without involving complex concepts (eigenvalues, eigenvectors…). Both

the probabilities of failures and probability density functions obtained showed a good

agreement with those obtained by the Monte Carlo Simulation method.

2.3.6 Babu and Mukesh (2004)

Babu and Mukesh (2004) investigated the effect of spatial variation of soil

strength on slope reliability for a simple cohesive soil slope. They defined the geometry,

stratigraphy, and soil parameters of the slope. Moreover, they assumed an isotropic

correlation structure. The authors calculated the factor of safety using Bishop’s method.

Next, they calculated the probability of failure of the slope by using the First Order

Second Moment Method. After that, they repeated the same procedure stated above but

by assuming an anisotropic correlation structure by defining both vertical and horizontal

correlation distances. The authors concluded that not only the coefficient of variation of

soil parameters and the correlation distance can affect the probability of failure of the

slope, but also the mean factor of safety can affect the probability of failure. Additionally,

the authors found that there is a significant need to include an anisotropic correlation

structure in the probabilistic slope stability analysis. Performing reliability analysis by

assuming that the correlation distance is the same in both horizontal and vertical

directions leads to an overestimation of probability of failure of slopes.

16
2.3.7 Cho (2007)

Cho (2007) conducted a probabilistic slope stability analysis through a numerical

procedure based on Monte Carlo Simulation (MCS) that considers the spatial variability

of the soil properties based on local averaging. Hassan and Wolff (1999) concluded that

the deterministic critical failure surface is not necessarily the failure surface with the

highest probability of failure. For the sake of that, Cho (2007) adopted the First Order

Reliability Method (FORM) to determine the critical probabilistic failure surface.

Moreover, the author used both FORM to identify the input parameters that have the

greatest impact on the failure probability and Spencer’s method to calculate the reliability

index. The author concluded that the searched critical probabilistic surface showed

somewhat different locations from the critical deterministic surface. Furthermore, the

probability of failure decreases with a decrease in the scale of fluctuation and vise versa.

In addition to that, he deduced that the assumption of the isotropic field is conservative

and the sensitivity of the unit weight is relatively small compared to those of cohesion

and the angle of the internal friction. Finally, Cho (2007) found that in the case of small

scale of fluctuation, a low probability of failure is obtained. Hence, more realizations are

needed to conduct MCSM.

2.3.8 Cho (2010)

All the above case studies used the traditional LEM combined with the random

field theory to calculate the probability of failure by taking spatial variability into

17
consideration. This traditional analysis considers the influence of the random field along

the predetermined critical surface. Cho (2010) proposed his method by using the

Karhunen – Loeve Expansion Method that is independent of the division of slices in the

sliding mass in order to be able to calculate the shear strength at any location along the

trial slip surface. The author considered the strength reduction method in the calculation

of Bishop’s factor of safety. Conversely, he based his probabilistic analysis that accounts

for the spatial variability on a search algorithm that can find the surface with the

minimum reliability index. The author illustrated his approach by analyzing a one layered

slope twice. One time with ø = 0 slope and the other with the c- Ø slope.

Cho (2010) deduced that in the case of ø = 0 slope the critical failure surface identified by

search algorithm always gives smaller factor of safety compared to that obtained from

fixed critical surface. In contrast, the probability of failure that comprises all potential

failure surfaces is greater than that obtained from the fixed critical surface and the

relative difference between the two probabilities decreases when the autocorrelation

distance increases. In contrary, for c- Ø slope, there is no significant difference between

probability of failure obtained from fixed critical surface and that obtained from the

search algorithm methodology. However, the negative correlation between C and Ø has a

significant effect on the variance of the shear strength. This latter affects the probability

of failure significantly.

2.3.9 Wang et al. (2011)

Wang et al. (2011) conducted a probabilistic slope stability analysis based on

MCS by using Simulation Subset in order to improve the efficiency and resolution of the
18
MCS. The analysis was implemented using a spreadsheet package that was used to

explore the effect of spatial variability on the probability of failure of slopes. The

methodology is illustrated through applying it to a cohesive slope and the deterministic

factor of safety was calculated using the Ordinary Method of Slices. The results were

validated by comparing the results with those obtained from other reliability methods.

Wang et al. (2011) modeled the undrained shear strength by a lognormal random field

and by an isotropic correlation structure using an exponential auto covariance function.

The authors found that if the spatial variability is ignored, the probability of failure is

significantly overestimated particularly when the effective correlation length is smaller

than the slope height. Moreover, they concluded that the variance of the factor of safety is

overestimated when the spatial variability is ignored. This variance overestimation may

result in either over conservative or under conservative estimation of the probability of

failure where if the marginal factor of safety (FS=1) occurs at the lower tail of the factor

of safety probability distribution, an overestimation of probability of failure occurs.

However, if the marginal factor of safety (FS =1) is located at the center or approaches

the upper tail of the factor of safety probability distribution, an underestimation of

probability of failure occurs. Further, they deduced that it is appropriate to use only one

given slip surface in the analysis i.e. FOSM or MCS when the spatial variability is

ignored. In contrast, when the spatial variability is considered, the critical slip surface

varies spatially. Thus, the critical probabilistic surface should be investigated by

conducting a search algorithm method to get the surface with the minimum reliability.

19
2.4 Studies involving fem with random field theory

All the above studies have combined the limit equilibrium method (LEM) with

random field theory. However, the inherent nature of LEM is that it leads to a critical

failure surface, which in 2-D analysis appears as a line which could be non-circular and

the influence of the random field is only taken into account along the line and is therefore

one-dimensional. Thus, to overcome this limitation, some investigators pursued a more

rigorous method of probabilistic geotechnical analysis in which nonlinear finite- element

methods are combined with random field generation techniques. This method is called

Random Finite Element Method (RFEM). It captures the effect of soil spatial variability

well where it fully accounts for spatial correlation and averaging. It is also a powerful

slope stability analysis tool that does not require priori assumptions related to the shape

or location of the failure mechanism. The following studies presented the probabilistic

slope stability analysis based on Random Finite Element Method.

2.4.1 Griffiths and Fenton (2000)

Griffiths and Fenton (2000) conducted a Random Finite Element probabilistic

analysis highlighting the influence of the spatial correlation length on the probability of

failure of the slope. Furthermore, the authors performed a parametric study to investigate

the effect of the scale of fluctuation and coefficient of variation of the shear strength on

the stability of the slope. Griffiths and Fenton illustrated the analysis by analyzing an

undrained clay slope. The authors concluded that the probability of failure increases as

20
the coefficient of variation of the undrained shear strength increases. Also they found that

for low values of the coefficient of variation (0< COV< 0.5), the probability of failure

increases as the ratio of the correlation length to the slope height increases. On the

contrary, for high values of the coefficient of variation, the probability of failure

decreases with the increase in the ratio of the correlation length to the slope height. Thus,

perfect correlation overestimates the probability of failure for low values of the

coefficient of variation and for slopes with high factor of safety (FS > 1.4); however, it

underestimates the probability of failure for high values of the coefficient of variation and

for slopes with low factor of safety (FS < 1.40).

2.4.2 Griffiths and Fenton (2004)

In this study the authors performed probabilistic slope stability analysis based on

both simple and advanced methods. In the simple approach, the authors treated the

undrained shear strength of the cohesive slope as a simple random variable and both

spatial correlation and local averaging are ignored. The probability of failure in this

simple methodology was estimated as the probability that the shear strength would fall

below a critical value based on a log-normal probability distribution. The results of the

simple study indicated that the probability of failure increases with the decrease in the

factor of safety. If the factor of safety is greater than one (FS > 1), the probability of

failure increases as the coefficient of variation increases also. However, for FS < 1, lower

values of coefficient of variation tend to give higher values of probability of failure.

Moreover, the results of the simple approach contradicted the practical one; this approach
21
led to a high probability of failure with mean factor of safety = 1.47; however, practical

experience showed that slopes with FS=1.47 rarely fail. To overcome this problem, the

authors proposed two factorization methods that used to reduce the mean value of the

undrained shear strength. Hence, an increase in the strength reduction factor reduces the

probability of failure to an acceptable value. Additionally, the authors conducted a RFEM

to model the slope more realistically. The analysis took into account both spatial

correlation and local averaging. By comparing the results of the simplified probabilistic

analysis and the advanced one, the authors found that the simplified analysis in which

perfect correlation is assumed can lead to unconservative estimates of the probability of

failure and this contradicted all the previous findings of other investigators.

2.4.3 Griffiths et al. (2009)

Griffiths et al. (2009) studied the advantage of Random Finite Element Method

(RFEM) over the traditional probabilistic method (FORM or MCS). The study aimed at

investigating the influence of the spatial correlation length, local averaging and the

coefficient of variation of the strength parameters on the probability of failure of the

slope. The authors found that for a given value of the spatial correlation, there is a critical

value of the coefficient of variation of the strength parameters. Thus, if perfect correlation

is considered, the traditional methods lead to an underestimation of the probability of

failure if the coefficient of variation of the strength parameters exceeds the critical value.

This critical value is influenced by slope inclination, mean factor of safety, and

22
correlation between strength parameters. Its value is lower for steeper slopes with low

factor of safety than less steep slopes with higher factor of safety.

The authors concluded from their analysis using FORM that for a given value of the

coefficient of variation of the undrained shear strength, the probability of failure increases

with the decrease in the factor of safety and for the same value of the factor of safety, the

increase in the coefficient of variation of the undrained shear strength leads to an increase

also in the probability of failure. However, for the same value of the factor of safety,

there is no effect of the slope inclination on the probability of failure. Using RFEM, the

authors varied the coefficient of variation and the correlation length of the undrained

shear strength in order to study the effects of these descriptors on the probability of

failure of the slope. Then, they compare the probability of failure obtained by FORM

with that obtained from RFEM. The authors deduced that ignoring spatial variability

underestimates the probability of failure for high coefficient of variation of the undrained

shear strength and vise versa. Furthermore, the authors deduced that the effect of spatial

variability on steeper slopes is more than that on flatter ones, but the probability of failure

of a steeper slope is higher than that of flatter ones when RFEM is used. When they

studied the effect of the mean factor of safety on the analysis, they found that for low

coefficient of variation of the undrained shear strength, the effect of spatial variability on

slopes of low factor of safety is higher than that of slopes of higher factor of safety.

2.4.4 Griffiths et al. (2010)

In this study the probabilistic slope stability methods in the literature were

reviewed in order to investigate their efficiency in modeling spatial variability of soil

23
properties correctly. Griffiths et al. (2010) found that Point Estimate Method (PEM),

First Order Second Moment Method (FOSM), First Order Reliability Method (FORM),

and Monte Carlo Simulations are all used with the combination of Limit Equilibrium

Method (LEM) for studying the reliability of slopes. The authors performed the analysis

by analyzing a hypothetical slope that has already been analyzed by other authors. The

authors showed that LEM combined with 1D random field can give lower probabilities of

failure than the RFEM and this is due to the fact that RFEM doesn’t require a priori

assumptions related to the shape or location of the failure mechanism and also the failure

mechanism has more freedom to find the weakest path through the random soil, which is

in contrast to the LEM approach, where the failure surface location is fixed before the

random field can be accounted for.

2.4.5 Jha and Ching (2013)

Jha and Ching (2013) performed a probabilistic slope stability analysis using the

advanced method (Random Finite Element Method). The authors conducted the study by

collecting a database for 34 real undrained engineered slope cases. The paper aimed at

studying the effect of slope geometry, mean and coefficient of variation of the soil

parameters, and the scale of fluctuation on the probability of failure. Jha and Ching

(2013) started their analysis by performing a deterministic slope stability analysis for

each case to get the nominal factor of safety. In this deterministic analysis, the authors

transformed the undrained shear strength to the mobilized one and also they used the

strength reduction method to evaluate the factor of safety. Using the RFEM approach, the
24
authors characterized the undrained shear strength of each layer by a random variable

with a mean, coefficient of variation and lognormal distribution. Moreover, spatial

variability was taken into account by defining both vertical and horizontal scales of

fluctuation. The vertical scale of fluctuation was calculated for each case using Phoon

and Kulhawy (1999) approach; however, the horizontal scale of fluctuation was assumed

because it can’t be calculated.

The results of the analysis indicated that statistical parameters of the factor of

safety ( and COV) depend on the coefficient of variation of the undrained shear

strength, scales of fluctuation, and the slope geometry. The authors incorporated the

effect of the slope geometry through the length of the failure surface as obtained from the

deterministic analysis. The authors’ analysis showed that the mean factor of safety is

always less than the deterministic factor of safety. This reduction in the mean doesn’t

depend on the vertical and the horizontal scale of fluctuation; however, it depends on the

coefficient of variation of the undrained shear strength. The reduction in the mean is more

pronounced when the coefficient of variation of the random field is large. Additionally,

the authors found that the coefficient of variation of the factor of safety is always less

than the coefficient of variation of the random field. This variance reduction is more

pronounced when the coefficient of variation of the random field is large and when the

ratio of the vertical scale of fluctuation to the length of the failure surface is small.

Furthermore, the authors reported that the ratio of the horizontal scale of fluctuation to

the vertical one has a minor effect on the mean and the coefficient of variation of the

factor of safety. Finally, the authors proposed a simplified equation to calculate the

25
probability of failure for the undrained engineered slopes that have a spatially variable

shear strengths.

2.5 Studies involving probabilistic slope stability analyses based on search to find
the minimum reliability index

The majority of the studies that conducted either LEM with Random Field or

Random Finite Element Method (RFEM) consider the deterministic critical failure

surface as the surface of the minimum reliability index. Hassan and Wolff (1999)

illustrated that the surface of the minimum factor of safety isn’t necessary the surface that

has the minimum reliability index. Thus, some investigators [Hassan and Wolff (1999),

Bhattacharya et al. (2003), Xue and Gavin (2007) and Deng and Luna (2013)] performed

new methodologies to search for the critical probabilistic surface. The following section

illustrates the work done by these investigators.

2.5.1 Hassan and Wolff (1999)

Hassan and Wolff (1999) proposed a simple and effective method for locating the

critical probabilistic surface. The authors provided an algorithm to search for the surface

of the minimum reliability index using existing deterministic slope stability computer

programs by making a moderate number of multiple runs. Hassan and Wolff applied

offset- values of each of the random variables while keeping the remaining parameters at

their mean values, so they obtained different surfaces with different factors of safety.

26
Among these surfaces, the one that has the minimum factor of safety is considered the

critical probabilistic surface. The authors obtained three values of reliability index; FS

that corresponds to the critical deterministic surface, and f which is the reliability index

of the floating surface. The floating reliability index was used by the authors as an

indicator to check if there is a surface that has a reliability index lower than that obtained

using the deterministic analysis. Furthermore, the authors obtained min that corresponds

to the critical probabilistic surface.

After the application of the technique on case studies (Cannon Dam and Bois

Brule Levee), Hassan and Wolff found that their technique can yield an accurate

estimation of the reliability index. The authors concluded that both the type of the

embankment (layered or homogenous) and the coefficient of variation of soil parameters

have a significant effect on the difference between FS and min. For homogenous slopes

with small coefficient of variation of soil parameters, both the deterministic and the

probabilistic surfaces are close together. Conversely, for stratified slopes and in the case

of high coefficient of variation, considering the critical deterministic surface as the

critical probabilistic surface that has minimum reliability index leads to unconservative

probabilities of failure.

2.5.2 Bhattacharya et al. (2003)

Bhattacharya et al. (2003) presented a numerical procedure for locating the

surface with minimum reliability index by using a formulation similar to that used to

search for the deterministic critical failure surface. The authors developed a computer
27
program that is an extended version of a deterministic slope program. This program used

Spencer’s method as a deterministic analysis and the Mean First Order Second Method

(MFOSM) as a probabilistic analysis and the Monte Carlo Simulation Method (MCSM)

to search for the critical slip surface. Moreover, this program doesn’t make any priori

assumption regarding the geometry of the slip surface where it can handle any complex

slope geometry and layering. Bhattacharya et al. found that the search of the critical

probabilistic surface is not different from that of the critical deterministic surface in

which the procedure adopted is the same as that used for finding the critical deterministic

slip surface with an additional step for the calculation of the reliability index.

2.5.3 Xue and Gavin (2007)

Xue and Gavin proposed a new method for probabilistic slope stability analysis.

The new approach solved the reliability problem by using a genetic algorithm approach,

which simultaneously locates the critical slip surface and calculates its reliability index.

The authors used Bishop’s method as a deterministic slope stability method. Furthermore,

they illustrated the methodology by analyzing a hypothetical slope which was already

studied by Hassan and Wolff (1999). The proposed approach performed well in

comparison to FOSM and MCSM where it gave a reliability index close to that obtained

by FOSM and MCSM. Conversely, the method has some drawbacks especially due to the

implementation of Bishop’s method only and due to the assumption that the variables are

independent.

28
2.5.4 Deng and Luna (2013)

Deng and Luna (2013) investigated the effect of soil strength parameters on the

probability of failure of the slope for both the deterministic slip surface and the

probabilistic slip surface. They conducted a probabilistic slope stability analysis based on

First Order Reliability Method (FORM) and Mean First Order Second Moment Method

(MFOSM). FORM method was used to search for the surface with the minimum

reliability index. Moreover, they used the Ordinary Method of Slices to calculate the

deterministic factor of safety of the slope. Furthermore, the authors studied the effect of

the distribution type on the reliability index of the slope. The authors found that

compared to FORM, MFOSM tends to provide a relatively higher reliability index;

however, the difference is too small. Hence, it can be used for the analysis. Moreover,

they found that the user should select an appropriate distribution for random variables in

probabilistic analysis. The use of lognormal distribution can result on 6% - 10%

reliability index higher than that obtained using the normal distribution. Further, they

concluded that cohesion is considered the most significant parameter that influences the

probability of failure.

2.6 Studies involving system reliability analysis

For a structural or geotechnical system with several components, the overall

reliability will depend not only on the reliabilities of the individual components but also

on other factors including the correlations between different components or elements of

29
the system (Chowdhury and Xu 1995). Despite the fact that slope stability problems

include many potential slip surfaces each of which has a finite probability of failure

associated with it, the majority of the probabilistic slope stability studies are based on a

predetermined slip surface. Some studies consider that the critical deterministic failure

surface can be considered the surface with the highest probability of failure. Other studies

use the critical deterministic failure surface as a trial surface to find the critical

probabilistic failure surface. In fact, as a soil slope may have many potential slip surfaces,

the failure probability of a slope may be larger than that of sliding along any single slip

surface (Cornell 1967). For the sake of that, many investigators found that the slope

reliability problem may be better solved in the framework of system reliability.

Chowdhury and Xu (1995) stated that there are three scenarios for the calculation of the

system reliability. First scenario, one may simplify the problem by considering it as a

simple series system; in this case failure occurs if any element of the system fails. Second

scenario, considering the problem as a parallel system; in this case the failure of one

element of the system leads to further loading of other elements and consequent decrease

in reliability but the system doesn’t fail unless all elements fail. The final scenario is a

combination of both series and parallel systems. These scenarios are not justified where

the series system can lead to a high failure probabilities; in contrast, parallel system can

lead to a low failure probabilities. Hence, investigators proposed the approximation of the

results by considering the upper and lower bounds of the probability of failure.

The following studies illustrated the use of system reliability for the probabilistic slope

stability analysis.

30
2.6.1 Chowdhury and Xu (1995)

Chowdhury and Xu (1995) performed a system reliability analysis to calculate the

upper and lower bounds within a probabilistic framework. The methodology is based in

the concept of the limit equilibrium method where Bishop’s method is used for the

calculation of the deterministic critical surface that is used later for the comparison with

the lower and the upper bounds of the system. The authors adopted two equations for the

lower and the upper bounds of the probability of failure. The authors concluded from

their study that the difference between the system probability of failure and that of the

critical one is the correlation coefficients between the surfaces. The difference between

the upper and the lower bounds increases as the coefficient of variation of the random

variables increases. Furthermore, the difference between the upper bound and the

probability of failure associated with the critical surface also increases as the coefficient

of variation of the random variables increases.

The authors illustrated the analysis by analyzing two slopes; homogenous and layered

slopes. Chowdhury and Xu (1995) deduced that, in the case of homogenous slopes, the

correlation between the elements of the system is high. Thus, the probability of failure

along different slip surfaces is highly correlated. Therefore, the upper bound of the

probability of failure is highly correlated. Hence, the upper bound of the probability of

failure is very close to the probability of failure of the critical slip surface. Conversely,

for the layered slopes, the correlation is small. Hence, the upper bound probability of

failure is greater than that of the critical surface. Finally, the authors found that the value

of the upper bound probability of failure depends on the coefficient of variation of the

31
random variable. For low values of the coefficient of variation, the upper bound isn’t high

and vise- versa.

2.6.2 Hong and Roh (2008)

Hong and Roh (2008) performed a slope reliability analysis based on the

generalized method of slices as a deterministic analysis and on First Order Reliability

Method as a probabilistic analysis. The authors chose the generalized method of slices

due to its efficiency in modeling slopes with complex geometries. In contrast, the choice

of FORM as a probabilistic analysis is based on its effectiveness in overcoming the

probability distribution tail sensitive problem. The authors considered the slope as a

series system where the failure of any slip surface means failure for all the slope. They

dealt with system reliability by defining a limit state function by getting the minimum of

the ratio of the shear strength to the mobilized shear strength. Moreover, the study aimed

at studying the effect of distribution type and the spatial variability on the probability of

failure. The authors concluded that the probability of failure increases with the increase in

the coefficient of variation of the random variable and as the probability of failure

becomes larger, it is less sensitive to the distribution type. Further, the assumption ththe

soil properties in a soil layer are fully correlated leads to overestimation of the probability

of failure.

32
2.6.3 Huang et al. (2013)

Huang et al. (2013) conducted three types of analyses; deterministic analyses to

investigate the failure regions, probabilistic analyses using Monte Carlo Simulation to

investigate the probability density function of the factor of safety and finally RFEM was

performed to investigate the influence of spatial variability on the probability of failure of

the slope. The authors illustrated the analysis by analyzing a hypothetical slope analyzed

by other investigators Ching et al. (2009) and Low et al. (2011). They dealt with

reliability problem as a system problem where all potential slip surfaces are considered .

The results showed that the probability of failure obtained by FEM is higher than that

obtained by LEM and the probability of failure decreases with increasing spatial

correlation length.

2.6.4 Zhang et al. (2013)

Zhang et al. (2013) extended the work done by Hassan and Wolff (1999) to get a

practical tool for evaluating the system reliability of a soil slope based on computer codes

for deterministic slope stability analysis. Moreover, the authors adopted an equation that

can be used by the user to get bounds of the system reliability. Zhang et al. (2013)

illustrated the effect of the distribution type on the probability of failure. The authors

recommended the use of the method for slopes with relatively simple geometry; however,

the extended method is less accurate for complex slope geometries. The authors deduced

that it is better to use the lognormal distribution for the factor of safety when the

33
coefficient of variation of the factor of safety is greater than about 30% due to larger

uncertainty associated with the basic uncertain variables. Furthermore, they used the

Hassan and Wolff method to judge if the system effect in slope reliability analysis is

obvious or not. They found that the system effect is less obvious if the probability of

failure of the most critical slip surface is greater than those based on other representative

slip surfaces. In this case the system probability of failure is governed by the probability

of failure of the most critical slip surface.

Regardless of the difficulty in estimating the system probability of failure, Zhang et al.

(2013) recommended that its bounds can be estimated based on the probability of failure

of the most critical slip surface using a simple equation.

34
 CHAPTER 3

DATABASE COLLECTION

3.1 Introduction

Conventionally, the stability of a soil slope is evaluated by adopting a

deterministic approach that is based on a target global factor of safety that is calculated

either through limit equilibrium methods or through numerical analyses. Both LEM and

FE slope stability models that are used to predict the factor of safety of a slope may not

be totally accurate, in the sense that the calculated factor of safety may deviate from the

actual factor of safety. The most effective approach that can be used for the evaluation of

this deviation is to rely on actual failure cases when they exist. Despite this, actual well-

documented failure cases for slopes are rare, especially due to the fact that slopes are

typically built to meet high safety requirements. Travis et al. (2010) compiled a large

scale database comprised of 301 actual failure cases. The authors report calculated factors

of safety of the 301 failure cases based on the original publications in which these case

histories were presented. The major drawback of the reported predicted factors of safety

is the lack of consistency in the methods of prediction between the different cases. For

example, some were based on methods assuming circular failure surfaces, but some were

not. Some were based on total stress analyses while others were based on effective stress

analyses. These inconsistencies in the reported factors of safety do not allow for a

systematic and uniform analysis of these published failure case histories of slopes. A

similar effort at compiling a database of slope failures was undertaken by Wu (2009)

35
where eight failure cases were analyzed using various limit equilibrium methods and

finite element methods. Jha and Ching (2013) collected information about 34 idealized

real engineered slope failures (cut and fill slopes). They predicted the factor of safety of

these slopes using finite element method and only total stress analysis was adopted.

Finally, Bahsan et al. (2014) collected 43 case histories, 34 of them were the same as

those collected by Jha and Ching (2013). Bahsan et al. (2014) recalculated the factor of

safety by adopting Limit equilibrium methods (Simplified Bishop’s method and the

Spencer’s method) using a MATLAB code for the LEM calculations.

3.2. Database collection

In this study, 52 case histories are collected from documented failure cases of

undrained slopes and embankments from the year 1956 to 2002 (Table 3.1). These cases

are divided into 43 embankments/fill slopes, 8 cut slopes, and 1 natural slope. Site

locations were spread from Europe, US, South America, Arabian Gulf to Asia. Most of

the cut slopes and fill slopes are parts of road facilities, especially road embankments

located in relatively remote areas such that the failures didn’t have critical consequences

on the surroundings, while others are test embankments that were built to fail. Slope

heights range from 2 to 22 m, and slope angles range from 9 to 69 degrees. The subsoil

natural materials are mostly clays and silty clays with unit weights ranging from 1.1 to 2

t/m3. The embankment fills are typically sandy or silty. The most common tests used to

obtain undrained shear strengths are Unconfined Compression tests (UC) and Field Vane

Tests (FVT). The observed failure surfaces tended to be circular except for two cases.

36
Table 3. 1. Database of Undrained Slope Failure Cases
Slope Slope Slope
No. Site/Country Reference
Type Angle Height
1 Nesset/Norway Flaate and Preber(1974) Fill 27 3m
Presterɸdbakken/
2 Flaate and Preber(1974) Fill 34 3m
Norway
3 Âs/Norway Flaate and Preber(1974) Fill  7m
4 Skjeggerod/Norway Flaate and Preber(1974) Fill  7m
5 Tjernsmyr/Norway Flaate and Preber(1974) Fill  1.5m
6 Aulielva/Norway Flaate and Preber(1974) Fill  2m
7 Falkenstein/Norway Flaate and Preber(1974) Fill  4m
8 Jalsberg/Norway Flaate and Preber(1974) Fill  2.5m
9 Saint Alban/ Canada Pilot et al.(1982) Test Fill  4m
La Rochelle et al. (1974)
Talesnick and Baker (1984)
10 Narbonne/France Pilot (1972) Test Fill  9.6m
Pilot et al (1982)
Talesnick and Baker (1984)
11 Lanester/France Pilot (1972) Test Fill  4m
Pilot et al (1982)
Talesnick and Baker (1984)
Cubzac-les
12 Pilot et al (1982) Test Fill  4.5m
Ponts/France
Talesnick and Baker (1984)
13 Lodalen1/Norway Sevalson(1956) Cut  15.9m
14 Lodalen2/Norway Sevalson(1956) Cut  18.8m
15 Lodalen3/Norway Sevalson(1956) Cut  16.3m
16 Rio de janeiro/Brazil Ramalho-Ortigao et al (1983) Test Fill  2.8m
Ferkh and Fell (1994)
17 New Liskeard/Canada Lacasse et al(1977) Test Fill  6m
18 Bangkok A/Thailand Eide and Holmberg (1972) Test Fill  2m
19 DrammenV/Norway Kjærnsli and Simons (1962) Natural  17.7m
Bjerrum and Kjærnsli (1957)
20 DrammenVI/Norway Kjærnsli and Simons (1962) Natural  15.8m
Bjerrum and Kjærnsli (1957)
21 DrammenVII/Norway Kjærnsli and Simons (1962) Natural  13.2M
Bjerrum and Kjærnsli (1957)
22 Pornic/France Pilot(1972) Fill  3.25m
23 Saint-Andre/France Pilot(1972) Fill  3m
South of
24 Pilot(1972) Fill  6m
France/France

37
 Slope Slope Slope
No. Site/Country Reference
Type Angle Height
NBR Development
25 Dascal et al. (1972) Test Fill  4m
/Canada
26 Portsmouth/USA Ladd(1972) Test Fill  6.5m
27 Kameda/Japan Hanzawa et al.(1994) fill  6.3m
KhorAl - Zubair
28 Hanzawa (1983) Fill  11.5m
no.4/Iraq
Hanzawa et al. (1980)
29 Lian-Yun- Gang/China Chai et al.(2002) Fill  4m
30 Congress Street/USA Ireland(1954) Cut  14.3m
Daikoku-Cho
31 Kishida et al.(1983) fill  14.7m
Dike/Japan
Hanzawa (1983)
32 Cuyahoga AA/USA Wu et al.(1975) fill  18.7m
33 King's Lynn(England) Wilkes(1972) fill  10.5m
34 Muar/Malaysia Indraratna et al.(1992) Test fill  5.5m
North Ridge
35 Rivard et al.(1978) Fill  18.3m
Dam/Canada
Seven Sisters
36 Rivard et al. (1978) Fill  4.3m
Dike/Canada
Peterson et al. (1957)
Shellmouth
37 Rivard et al.(1978) Fill  16.5m
Dam/Canada
38 Juban I/USA Zhang et al.(2005) Fill  6m
Skempton and LaRochelle
39 Bradwell/England Cut  4.60m
(1965)
Duncan and Wright (2005)
40 Genesee/ Canada Been et al.(1986) Fill  7m
41 Precambrian/Canada Dascal et al. (1975) Fill  7.6m
42 Scrapsgate/England Golder et al.(1954) Fill  6m
43 Scottsdale/Australia Parry(1968) Fill  6m
44 Iwai/Japan Shogaki et al.(2008) Fill  4.5
45 Fair Haven/USA Haupt and Olson(1972) Fill  13.8m
Boston Marine
46 McGinn et al. (1993) Cut  14m
Excavation/USA
Desert View
47 Day(1996) Cut  21.7m
Drive/USA
48 Siburua October 5 Wolfskill et al. (1967) Natural  7m

49 Tianshenqiao/China Chen et al. (1988) Natural  14.5m

San Francisco
50 Duncan and Buchignani (1973) Cut  18m
Bay/USA
51 Carsington/England Skempton and Coats (1985) Fill  12.2m
52 Atchafalaya/USA Kaufman et al. (1967) Natural  10.5m

38
 This chapter includes brief descriptions for all the failure cases found in the

literature. The geometries rights before failures were considered to estimate the near

failure condition in the LEM due to the fact that all these cases are failure cases. The 52

cases studied are re-analyzed and the factors of safety are evaluated using four different

limit equilibrium methods (Simplified Bishop method, Ordinary method of slices, Janbu,

and Spencer’s method). Table 3.2 illustrates all the soil properties adopted for the slope

stability analysis of the 52 case histories.

3.2.1. Slide at Nesset

In October 1957 a 50m slide occurred at Nesset in Norway during the

construction of the embankment. The embankment is 3m high and is mostly comprised of

granular soil (sand and gravel) with a probable angle of internal friction of 35° and a unit

weight of 1.9 t/m3. This embankment was built with 2 (horizontal):1 (vertical) side

slopes. The foundation soil at the site consisted of a 2-8m thick deposit of soft clay soil

over a 1m thick layer of silty sand that is placed at 10m below the ground level over

bedrock. The site showed no evidence of the presence of a dry crust; however, quick

clays of the type involved in this slide are noted for their great sensitivity and extremely

brittle failure. This layer of quick clays is present at 8m below the ground surface close to

the permeable sand layer. Soil properties including unit weight, index parameters,

undrained shear strength and sensitivity for the site soils are summarized in the borehole

profile in Fig.3.1. The clay has undrained shear strength of 0.8-2t/m2 that is measured by

unconfined compression tests, fall-cone tests and vane shear tests and a unit weight
39
between 1.50 and 1.80 t/m3. Due to the small movements and disturbances that occurred

after failure, the slip failure was close to cylindrical.

Figure 3. 1 Borehole Profile at Nesset (Flaate and Preber 1974)

Flaate and Preber (1974) computed the minimum factor of safety for the profile shown in

Fig.3.2 by adopting a short term stability analysis using the Swedish slip circle method

and obtained a factor of safety of 0.88. Re-calculation of the factor of safety for the

profile shown in Fig.3.3 was accomplished using the SLIDE software that can evaluate

the factor of safety using different methods (Bishop, Ordinary Method of Slices, Janbu,

and Spencer). The analysis is carried out by adopting the field vane measurements for

assigning the undrained shear strength values for the layers.

40
Figure 3. 2 Critical Slip Surface (Flaate and Preber 1974)

Figure 3. 3 Critical Slip Surface using SLIDE software

41
3.2.2 Slide at Presterɸdbakken

The Presterɸdbakken slide occurred in May 1962 during the construction of a 3m

high embankment. The failure was due to workmanship mistake, when workers noticed

that slight depression occurred in the surface of the embankment they hurried up to solve

the problem by filling the depression with an additional material without seeking an

engineer. That resulted in a slide that began at the centerline of the road with an area of

50x25m. The embankment was constructed of granular fill with an assumed angle of

internal friction of 30° and unit weight of 1.9t/m3 and it was laid at a slope of 1:1.5. The

foundation soil is composed of 1-2m dry crust underlain with 8-10m of clayey silt. Soil

properties are shown in Fig.3.4. The undrained shear strength values are measured using

Unconfined Compression test, field vane test, and Fall-cone tests.

The slide failure was assumed as mere rotational about an axis resulting in a circular slip

surface due to the presence of the same vane shear strength values inside and outside the

sliding zone. Flaate and Preber (1974) analyzed the profile in Fig.3.5 by using the

Swedish slip circular method where FS=0.83 was obtained. The slope is reanalyzed using

SLIDE software to evaluate the minimum factor of safety for the sliding body and the

results are shown in Fig.3.6.

42
Figure 3. 4 Borehole Profile at PresterØdbakken (Flaate and Preber 1974)

Figure 3. 5 Critical Slip Surface at PresterØdbakken (Flaate and Preber 1974)

43
 Conventionally, the stability of a soil slope is evaluated by adopting a

deterministic

Figure 3. 6 Critical Slip Surface using SLIDE

3.2.3 Slide at Âs

The 25x35m area slide took place during the night at the end of September 1962

at Âs in Norway. A depression of about 1-1.5m occurred at the end of construction of the

road embankment when the workers started with the bituminous surfacing. The

embankment height is close to 7m and the slope is inclined at about 27°. The

embankment fill is formed from compacted silt or clay with little amount of crushed

stone. Soil investigation showed that the foundation soil consisted of a 2-3m thick dry

crust underlain by 7m of silty clay. The soil properties of the foundation soil are shown in

Fig.3.7.

44
 Figure 3. 7 Borehole Profile at Âs Iin Norway (Flaate and Preber 1974)

When the depression occurred, no cracks were formed and the vane tests showed no

disturbance. Hence, a circular failure surface was assumed and the factor of safety for the

profile shown in Fig.3.8. was calculated using Swedish slip circle method and FS=0.80

was obtained. The minimum factor of safety is recalculated using SLIDE software and

the results are shown in Fig.3.9.

Figure 3. 8 Critical Slip Surface at Âsin Norway (Flaate and Preber 1974)

45
 Figure 3. 9 Critical Slip Surface using SLIDE

3.2.4 Slide at Skjeggerod

Fig.3.10 shows a slope failure that occurred during the night in the beginning of

September 1963 at Skjeggerod in Norway. The soil mass slid over a distance of 45m

perpendicular to the road forming a slope that is 5-6m high at an inclination of 27°.

Figure 3. 10 View of the Slide Occurred at Skjeggerod in Norway (Flaate and Preber
1974)

46
Soil investigation after the slide resulted in the soil properties shown in Fig.3.11.

Figure 3. 11 Borehole Profile at Skjeggerod in Norway (Flaate and Preber 1974)

The road embankment is mostly of granular material with an angle of internal friction of

30° and unit weight of 2t/m3. According to Flaate and Preber (1974), the slip surface is

assumed to be composite due to the strong disturbance of the soil masses. The authors

computed the factor of safety for the profile shown in Fig.3.12 using the Swedish slip

circle method and a value of 0.73 was obtained. However, in this study the surface is

assumed circular and the factor of safety value is presented in Fig.3.13.

47
Figure 3. 12 Critical Slip Surface at Skjeggerod (Flaate and Preber 1974)

Figure 3. 13 Critical Slip Surface using SLIDE

48
3.2.5 Slide at Tjernsmyr

On November 29, 1957, a slide took place at Tjernsmyr in Norway. At the end of

the road construction, 80m of the embankment slid out. The embankment was between 1-

2m high and was built with 2 (horizontal) : 1 (vertical) side slopes. The fill consisted of

granular material with an angle of internal friction of about 40° and a unit weight of 1.9

t/m3. The foundation soil consisted of 1-2m layer of peat over a 10-20m layer of soft silty

clay with high sensitivity. Soil properties are shown in Fig.3.14.

Figure 3. 14 Borehole Profile at Tjernsmyr (Flaate and Preber 1974)

The movements were small and very few cracks on the surface of the sliding body were

seen so very little disturbance of the soil occurred. Due to the presence of only small

movements, the slip surface was assumed circular and the factor of safety was calculated
49
by Flaate and Preber (1974) using the profile shown in Fig.3.15. and by adopting the

Swedish slip circle method. A factor of safety of 0.87 was obtained by the authors. The

slide at Tjernsmyr is analyzed using SLIDE software and a minimum FS= 0.834 is

obtained as shown in Fig.3.16.

Figure 3. 15 Critical Slip Surface at Tjernsmyr (Flaate and Preber 1974)

Figure 3. 16 Critical Slip Surface using SLIDE

50
3.2.6 Slide at Aulielva

In March 1963 and after the completion of the road construction at Aulielva in

Norway by six months, a slide occurred and the materials slid out perpendicular to the

road towards the Auli river. A photograph of the slide is shown in Fig.3.17.

Figure 3. 17 Slide at Aulielva (Flaate and Preber 1974)

The embankment was 2m high and was laid at an angle of 24° with an angle of

internal friction of 30° and unit weight of 1.9t/m3. The soil investigation indicated the

presence of a 2-4m thick dry crust underlain by a silty clay layer. Fig.3.18. shows the soil

properties of the foundation soil. High disturbance and remolding that occurred within the

sliding body indicated that the slip surface is composite.

51
 Figure 3. 18 Borehole Profile at Aulielva (Flaate and Preber 1974)

Flaate and Preber (1974) analyzed the slope profile shown in Fig.3.19. and obtained a

factor of safety of 0.92 using the Swedish slip circle method by assuming both composite

and circular slip surface. In this study the factor of safety is evaluated using SLIDE

software and the results are shown in Fig.3.20.

Figure 3. 19 Critical Slip Surface at Aulielva (Flaate and Preber 1974)

52
 Figure 3. 20 Critical Slip Surface using SLIDE

3.2.7 Slide at Falkenstein

The slide at Falkenstein in Norway took place in April 1964. The slide occurred

over a 40m width of a recently placed embankment. The fill consisted of granular

material with an assumed angle of internal friction of 35° and unit weight of 1.9t/m3.

Moreover, the embankment was 4m high inclined at 27°.

According to the soil investigation done in the site, no evidence of a dry crust was found

at the site. The soil profile consisted of a 15m soft silty quick clay layer. The soil

properties of the foundation soil material are shown in Fig.3.21.

53
 Figure 3. 21 Borehole Profile at Falkenstein (Flaate and Preber 1974)

Flaate and Preber (1974) evaluated the factor of safety of the slope using the Swedish

Slip Circle method and by assuming a circular slip surface as shown in Fig.3.22. In this

study the factor of safety is evaluated using the profile shown in Fig.3.23.

Figure 3. 22 Critical Slip Surface at Falkenstein (Flaate and Preber 1974)

54
 Figure 3. 23 Critical Slip Surface Using SLIDE

3.2.8 Slide at Jalsberg

In October 1964, a slide took place at Jalsberg in Norway during the construction

of road embankment. The embankment was about 2.5m high and was built at 27°

inclination. The fill consisted of granular soil with an internal friction angle of 35° and a

unit weight of 1.9 t/m3. Soil investigation was carried out and showed the presence of 1-

2m dry crust over a thick deposit of medium to low sensitivity silty clay.

55
 Figure 3. 24 Borehole Profile at Jalsberg (Flaate and Preber 1974)

The soil borehole profile is shown in Fig.3.24. No strong soil deformation occurred due

to the slide and no soil disturbance was detected within the sliding body. Hence, a

circular slip surface was assumed by Flaate and Preber (1974) and the minimum factor of

safety was found to be 1.10 for the profile shown in Fig.3.25. The factor of safety is

recalculated by using SLIDE software and the results are shown in Fig.3.26.

56
Figure 3. 25 Critical Slip Surface at Jalsberg (Flaate and Preber 1974)

Figure 3. 26 Critical Slip Surface using SLIDE

57
3.2.9 Slide at Saint- Alban

In October 13, 1973 a failure of a test embankment occurred in Saint-Alban,

Quebec in Canada. Numerous fissures opened on the top surface of the fill and massive

failure occurred. The purpose of the project was to investigate the failure conditions of

fills built on foundations of soft sensitive clays. The fill was built up until failure

occurred with a height of 4.6m and with a front slope of 1.5 horizontal to 1 vertical.

Moreover, the fill consisted mostly of uniform medium to coarse sand containing about

10% fine sand and 10% gravel. This fill has an internal friction angle of 44° due to the

presence of very pronounced angular-shaped grains. A detailed soil investigation was

conducted in the site indicating the presence of a weathered clay crust extending down to

a depth of 1.8m. This layer is overlain by 0.3m top soil and underlain by a gray to blue

soft silty marine clay with a more silty layer at 5.2m. Below this silty clay layer, a layer

of soft and sensitive clayey silt with sand is found. Finally, a dense fine to medium sand

constitutes the lower part of the deposits from 13.7 to 24.4 m depth. Soil properties

including index parameters, unit weight, and undrained shear strength are shown in

Fig.3.27.

58
 Figure 3. 27 Soil Profile at Saint-Alban (LA Rochelle et al. 1982)

An instrumentation program was conducted and from the observations of the

displacements of the reference points at the toe and of the surface of the fill after failure,

it is evident that the failure surface developed along a circular arc. LA Rochelle et al.

(1974) performed a slope stability analysis using a computer program made by Lefebvre

(1968) that adopt both the simplified Bishop and the Ø = 0 methods. Using the profile

shown in Fig. 3.28., the authors ended up with a factor of safety of 1.2. The slope is

59
reanalyzed with SLIDE software and a FS of 1.27 is obtained. The results are shown in

Fig.3.29

Figure 3. 28 Cross- section of the Saint-Alban embankment (LA Rochelle et al. 1982)

Figure 3. 29 Critical Slip Surface using SLIDE

60
3.2.10 Slide at Narbonne

The embankment failure took place in 1972 in France. The embankment was built

to failure in order to verify the validity of the methods of analysis used in the design of

Narbonne motorway. The foundation soil is relatively heterogeneous and is mainly made

up of a soft low plasticity clay deposit with 12-14m thickness that rests on a layer of

gravel that overlies very sound marls. Fig.3.30. shows the different soil properties of the

foundation material.

Figure 3. 30 Soil Profile at Narbonne (LARochelle et al. 1982)

61
The embankment was constructed in five days and failure occurred at a height of 9.60m

with 37° slope angle. The embankment material is a gravelly and clayey compacted sand

with unit weight = 2.07t/m3 and with the strength parameters c= 53KPa & Ø=26°. When

the large movements took place at failure, the embankment practically moved by rotation

and a circular shape of the failure was formed. Pilot (1972) conducted a total stress

analysis and a factor of safety = 0.83 was obtained using the profile shown in

[Link] using the computer program developed by the Geotechnical Group of

Université Laval, Québec that adopts the simplified Bishop (1955) method for slope

stability analysis. The slope is re-analyzed by using SLIDE software for the evaluation of

the factor of safety. Fig.3.32. shows the results of the analysis.

Figure 3. 31 Cross-section of the Narbonne embankment (LA Rochelle et al. 1982)

62
 Figure 3. 32 Critical Slip Surface using SLIDE

3.2.11 Slide at Lanester

In 1969 in France, engineers constructed the Lanester embankment to provide

them with all the necessary data about the pore pressures generated in the foundations. A

compacted sandy clayey gravel embankment with a density of 1.82 t/m3 and strength

parameters (c =30KPa & Ø=31) was constructed on foundation soil consisting of a layer

of soft, organic sandy clay and silt layer with 8-10m thickness, overlying a layer of gravel

over bedrock. Fig.3.33 shows the different soil parameters of the foundation material.

63
 Figure 3. 33. Soil Profile at Lanester (LA Rochelle et al. 1982)

Before the occurrence of the slide, lateral displacements formed causing the formation of

vertical cracks in the embankment. Failure occurred when the embankment reached a 4m

height. Fig.3.34 shows a photograph for the embankment failure.

Figure 3. 34 View of the Lanester embankment after Failure(LA Rochelle et al. 1982)

64
Pilot (1972) conducted a total stress analysis for the profile shown in Fig.3.35. and a

factor of safety of 1.27 was obtained. The factor of safety is reevaluated using SLIDE

software and the factor of safety shown in Fig.3.36. was obtained.

Figure 3. 35 Cross-section of the Lanester embankemt ((LA Rochelle et al. 1982)

Figure 3. 36 Critical Slip Surface using SLIDE

65
3.2.12 Slide at Cubzac les-ponts

In 1971 a test embankment failure took place at Cubzac-les-Ponts near Bordeaux

in France (Fig.3.37.). The construction of the embankment was part of a research

program aimed at settlement evaluation. The embankment was constructed on soft

alluvial deposits of 9m thickness. An overconsolidated clayey and silty crust (1m in

thickness) overlies a thin layer of strongly organic silty clay of 1m thickness which in

turn overlies slightly organic, soft silty clay 6m in thickness (Fig.3.38.). Failure occurred

when the embankment reached a height of 4.5m. The embankment fill consisted of clean

gravel with unit weight of 2.1t/m3 and an internal friction angle of 35. The shape of the

slip surface appeared to be circular. Stability analysis was carried out by total stress

analysis method using the computer program developed by the Geotechnical Group of

Université Laval, Québec that adopts the simplified Bishop (1955) method for slope

stability analysis and a factor of safety of F =1.44 was obtained (Fig.3.39). The factor of

safety is reevaluated by using SLIDE software as indicated in Fig.3.40.

Figure 3. 37 View of the Cubzac embankment after failure (LA Rochelle et al. 1982)

66
 Figure 3. 38 Soil Profile at Cubzac- les-Ponts (LA Rochelle et al. 1982)

Figure 3. 39 Cross-Section of the Cubzac embankment (LARochelle et al. 1982)

67
 Figure 3. 40 Critical Slip Surface using SLIDE

3.2.13 Slide at Lodalen 1

The slide occurred in Oslo in 1954. The slope was produced by an excavation of a

natural slope. The slope originally had an inclination of approximately 1:2.5 and was

excavated 5-6m and steepened to an inclination of 1:2. The result of a typical boring in

the slide area is shown in Fig.3.41. Underneath the upper layers, the drying crust is found

to be a firm comparatively homogenous marine clay with some thin silt layers. The study

done by Sevaldson ( 1956) adopted the so-called Ø=0 analysis for the evaluation of the

factor of safety. The authors ended up with F = 0.93 (Fig.3.42.). Using SLIDE software,

the factor of safety obtained = 1.012 (Fig.3.43).

68
 Figure 3. 41 Boring Profile at Lodalen (Sevaldson 1956)

Figure 3. 42 Actual and Critical Slip Surfaces at Lodalen (Sevaldson 1956)

69
 Figure 3. 43 Critical Slip Surface using SLIDE

3.2.14 Slide at Lodalen 2

This slide occurred in another section through the Lodalen site. At this profile, the

slide took place when the cutting reached 18.8m. Slope stability analysis was carried out

by Sevaldson (1956) using the same soil properties for Lodalen 1. A factor of safety of

0.93 was obtained. The results of the analysis conducted using SLIDE are shown in

Fig.3.44.

70
 Figure 3. 44 Results of Total Stress Analysis using Slide Software at Lodalen 2

3.2.15 Slide at Lodalen 3

Another section was analyzed through the Lodalen slide. The slide occurred when

the cutting reached a 16.3m height. Sevaldson (1956) conducted an analysis for the slide

using the Ø=0 analysis method and a factor of safety of 1.35 was obtained. In this study,

the slide is reanalyzed using SLIDE. The results are shown in Fig.3.45.

3.2.16Figure
Slide3.
at45 Results
Rio of Total Stress Analysis using Slide Software at Lodalen 3
de janeiro

71
 To investigate the behavior of embankments founded on soft soils, the Brazilian

Highway Research Institute conducted an extensive research program on an instrumented

trial embankment. The testing site consisted of a clay deposit that is 11m thick and that

overlies sand and gravel layers. Fig. 3.46. illustrates a summary of the geotechnical

properties of the site.

Figure 3. 46 Summary of Geotechnical Properties, Rio de Janeiro Soft Gray Clay

(Ramalho-Ortigão et al. 1983)

The trial embankment was constructed with a steeper slope (1v:2h), a stable slope

(1v:6h), a base length of 80m, and a base width of 40m. Moreover, two triangular berms

were constructed to avoid the possibility of failure occurring outside the instrumented

zone. The embankment fill material was a silty-sand residual soil with unit weight of

1.8t/m3 and strength parameters (c = 10-20 KPa & Ø= 35). The slide occurred when the

72
embankment height was raised to 2.80 m. Ramalho- Ortigao et al. (1983) conducted a

total stress stability analysis employing Bishop’s modified circular arc analysis. The

analysis was made through a program named BISPO and a factor of safety of 1.11 was

obtained (Fig. 3.47.). The slope stability analysis is repeated by using SLIDE and the

results are shown in Fig.3.48.

Figure 3. 47 Results of Total Stress Stability Analysis (Ramalho-Ortigão et al. 1983)

73
 Figure 3. 48 Results of Total Stress Stability Analysis using SLIDE software

3.2.17 Slide at New Liskeard

On June 27, 1963, an embankment failed at New Liskeard in Canada. The

embankment rests on 43m of soft to medium varved clay with a 2.7m thick silty clay

crust. Fig.3.49. shows the undrained shear strength used in the total stress analyses. The

fill material was a gravelly sand with cobbles, boulders and traces of silt and clay with

unit weight of 2.04t/m3 and angle of internal friction of 40. Failure occurred when the

embankment reached 6m. Lacasse et al. (1977) analyzed the stability of New Liskeard

embankment using the ICES-LEASE-I computer program (Bailey and Christian 1969)

with circular arc failure Fig. 3.50. The analysis resulted in a factor of safety of 1.10.

However, using SLIDE software a larger factor of safety is obtained as shown in

Fig.3.51.

74
 Figure 3. 49 Undrained shear strengths used in total stress stability analyses
(Lacasse et al. 1977)

Figure 3. 50 LEASE-I critical failure arcs at New Liskeard (Lacasse et al. 1977)

75
 Figure 3. 51 SLIDE critical failure arc at New Liskeard

3.2.18 Slide at Bangkok A

At noon on the 12th of April in 1972, a slide occurred in Bangkok in Thailand.

The failure took place several days after the critical fill height of an embankment had

been reached. The fill was a test fill constructed to study the bearing capacity of the soft

Bangkok clay and to study the efficiency of sand drains. The embankment height at

failure was 2m with a 2:1 slope and consisted of uniform sand material except for a 20cm
76
layer of gravel material placed at a fill height of 1.6m. The fill material was of unit

weight of 2t/m3 and of internal friction angle of 35. The fill is placed on foundation soil

composed of 1m dry crust. Beneath it, the soft clay extends to a depth of 23m and is

underlain by stiff clay. Fig.3.52. shows the geotechnical profile of the site.

Figure 3.52 Geotechnical Profile at Bangkok Site (Eide and Holmberg 1972)

77
A stability analysis was conducted by Eide and Holmberg (1972) by adopting undrained

total stress analysis. The calculated factor of safety at failure was 1.46 (Fig.3.53). The

slope is reanalyzed using SLIDE software and the results are shown in Fig.3.54.

Figure 3. 53. Critical Slip Surface (Eide and Holmberg 1972)

Figure 3. 54 Critical Slip Surface using SLIDE software

78
3.2.19 Slide at Drammen V

On January 6, 1955, a slide occurred along the river in Drammen, a town 40 km

west of Oslo in Norway. The slide took place in a natural slope made up of normally

consolidated clay to a great depth. The Norwegian Geotechnical Institute investigated the

slide by adopting Ø = 0 analysis for the evaluation of the factor of safety. The stability

analysis was made in the ordinary way, assuming circular sliding surface and using the

soil properties shown in Fig.3.55. Kjærnsli and Simons (1962) reported the analysis and a

factor of safety of 0.74 was obtained (Fig.3.56). The same analysis is carried out by using

SLIDE software and the results are shown in Fig.3.57.

Figure 3. 55. Soil Properties at Drammen (Kjærnsli and Simons (1962)

79
 Figure 3. 56 Critical Slip Surface (Kjærnsli and Simons 1962)

Figure 3. 57 Critical Slip Surface using SLIDE

3.2.20 Slide at Drammen VI

80
(Kjærnsli and Simons 1962) conducted a total stress analsysis for the center profile of the

slide described above for Drammen V. Using the soil properties shown above in Fig.3.55.

the authors ended up with a factor of safety of 0.59. Total stress analysis is carried out

using SLIDE. The results are shown in Fig.3.58.

Figure 3. 58 Critical Slip Surface using SLIDE software at Drammen VI

3.2.21 Slide at Drammen VII

In addition to the analysis carried out for the downstream and the center of the

slide profiles, Kjærnsli and Simons (1962) carried out a total stress analysis for the

upstream profile from the slide. A factor of safety of 0.70 was obtained. However, the

analysis using SLIDE ended up with a factor of safety of 0.87 as shown in Fig.3.59.

81
 Figure 3. 59 Critical Slip Surface using SLIDE software at Drammen VII

3.2.22 Slide at Pornic

In 1963, an embankment failure took place at Pornic in France. The embankment

was built by the highway and motorway construction in France. The embankment height

required was about 8 meters needed to construct a highway by-pass to cross a valley over

about 250m; however, failure occurred when the critical height of the embankment was

4m. This happened four months after the beginning of its construction. Stability analysis

was adopted with the embankment characteristics: unit weight of 2 t/m3, c = 1t/m2, Ø=

40. The embankment placed on a deposit of normally consolidated clay, except for a

dessicated crust of 2m overconsolidated clay. Pilot (1972) conducted stability analyses

under total stress analysis by the Bishop calculation method (circular failure) and a factor

82
of safety of 1.17 was obtained using the profile and soil properties shown in Fig.3.60.

Analysis using SLIDE software is presented in Fig.3.61.

Figure 3. 60 Embankment Failure at Pornic (Pilot 1972)

Figure 3. 61 Embankment Failure at Pornic using SLIDE

83
3.2.23 Slide at Saint-Andre

The slide occurred in 1969 at Saint-Andre (Gironde) in France. Failure happened

at the last stage construction of the embankment. The failure started with a sinking of

about 40cm, then a rotational movement appeared resulting in another sinking of the

platform. Failure occurred when the embankment reached 3m height. Stsability analysis

was performed after failure, taking into account the exact geometry of the embankment at

the instant of sliding as shown in Fig.3.62. and using the geotechnical soil properties

shown in Table 3.3. A factor of safety of F=1.38 was obtained (Pilot 1972). Using

SLIDE, the slope is reanalyzed and the results obtained are shown in Fig.3.63.

Table 3.3 Main soil characteristics of soils at Saint-Andre in France (Pilot 1972)

84
Figure 3. 61 Embankment Failure at Saint-Andre (Pilot 1972)

Figure 3. 62 Embankment Failure at Saint-Andre using SLIDE

85
3.2.24 Slide at South of France

In 1971 during the construction of an embankment on an approach to a bridge in

South of France, a rapid failure occurred in this embankment. The embankment consisted

of slightly muddy loose sand with unit weight of 1.7t/m3 and angle of friction of 35 that

is laid on a 2m-thick river deposit of more or less muddy sands, overlying a 25m thick

layer of lagoonal marine deposits consisting of slightly underconsolidated muddy shelly

clay. Stability analysis was carried out under total stress analysis by the Bishop

calculation method (circular failure) and a factor of safety of 1.30 was obtained using the

profile and soil properties shown in Fig.3.64. The embankment was re-analyzed using the

SLIDE software and the results are shown in Fig.65.

Figure 3. 63 Embankment Failure at South of France (Pilot 1972)

86
 Figure 3. 64 Critical Slip Surface at South of France using SLIDE

3.2.25 Slide at NBR Development

On Friday morning, September 17th, 1971, a failure occurred on a dyke in Canada.

The dyke was erected as a part of the James Bay water resources exploitation project. The

dyke’s dimensions were selected in order to induce failure in a selected direction and at

the same time to involve a large enough soil mass so that the failure could be considered

representative.

Transverse cross section:

 West slope: 1:5

 East slope: 1:4 for the first 15 ft height and 1:2 above 15 ft

Longitudinal cross- section: north and south slope:

 1:8 for the first 15 ft height and 1:4 above 15 ft

87
The dyke which is composed of a medium compacted wet granular material with a unit

weight of 130 pcf and an angle of internal friction of 35 was constructed on a 6 ft thick

layer of organic material underlain by 50 ft clay deposit. The top 8 ft of the clay deposit is

weathered (Fig.3.66). The stability analysis was carried out in terms of a total stress

analysis using the MIT ICES-LEASE computer program (Bishop’s simplified method)

and using the soil properties shown above for the profile shown in Fig.3.67. Dascal et al.

(1972) reported the factor of safety of 1.3.

The dyke is reanalyzed using SLIDE software and the results are shown in Fig.3.68.

Figure 3. 65 Stratigraphy and Geotechnical Characteristics of the foundation Soil at NBR

Development (Dascal et al. 1972)

88
Figure 3. 66 Results of the total stress stability analysis at NBR Development in France
(Dascal et al. 1972)

Figure 3. 67 Critical Slip Surface using SLIDE

89
3.2.26 Slide at Portsmouth

In the spring of 1968, a test embankment was purposely constructed to failure in order to

better define the in situ behavior of the foundation clay. The experimental test

embankment was constructed on a weathered drying crust of several feet underlain by

grey silty clay. Beneath this layer, a non-plastic sand silt layer that overlain a soft clay

soil layer. Soil properties are shown in Fig.3.69. The fill used to construct the test

embankment consisted of fairly clean and well graded sand with an average unit weight

of 115.5 pcf and with an angle of internal friction of 41. Failure occurred when the

embankment reached a height of 21.50 ft. Total stress stability analysis of the failure was

performed by Ladd (1972) using ICES LEASE1 program using the simplified Bishop and

the normal Fellenius method of slices. The factor of safety from this analysis is 0.84

(Fig.3.70). The factor of safety is recalculated using SLIDE and a value of 0.843 is

obtained as shown in Fig.3.71.

Figure 3. 68 Soil Profile, Index Properties and Field Vane Strengths at Portsmouth in
USA (Ladd 1972)

90
Figure 3. 69 Results of Total Stress Analysis (Ladd 1972)

Figure 3. 70 Critical Slip Surface using SLIDE

91
3.2.27 Slide at Kameda

An expressway embankment was constructed on peaty subsoil to investigate the

application of direct shear and cone penetration tests to soil investigation. The

construction site was the Kameda Interchange for the Hokuriku Expressway in Niigata

Prefecture. A detailed stability analysis was made on the embankment that is placed on

peaty soils that vary in thickness from 2m to 6m, and can be classified into two layers;

peat with water content from 100% to 400% and another layer of sandy clay with peaty

material with water content from 45% to 75%. Fig.3.72. shows the soil conditions at the

proposed site and the physical properties of the peaty soils. Fig.3.73. shows the undrained

shear strength of the soil. The fill material used for the embankment consists of upland

sand containing about 10% of fine particles smaller than 0.074mm and Tertiary mudstone

with more fine particles up to 20% maximum. The unit weight of the sand was 1.9

t/m3and the angle of internal friction was 35.

Figure 3. 71 Soil conditions at Kameda site (Hanzawa et al. 1994)

92
 Figure 3. 72 Undrained Shear Strength at Kameda Site (Hanzawa et al. 1994)

Failure occurred when the embankment height reached 6.3m where large deformations

took place together with settlement causing tension cracks and heave. The failure

occurred during the stage where an excavation for an irrigation canal was under

construction. Stability analysis was carried out and a factor of safety of 0.98 was obtained

(Fig.3.74).The factor of safety is re-evaluated again using SLIDE software and the results

are shown in Fig.3.75.

93
Figure 3. 73 Circular Slip Surface (Hanzawa et al. 1994)

Figure 3. 74 Circular Slip Surface using SLIDE

94
3.2.28 Slide at KhorAl - Zubair no.4

Five earth fills were constructed for the aim of preloading of an alluvial marine

clay with an area of 500mx500m. The unit weight of the fill was suggested to be 2 t/m3

with an angle of internal friction of 35. The fill was placed on a 17m layer of Alluvial

marine clay underlain by 2m silt and fine sand and finally with 1m dilluvial hard clay.

Fig.3.76. shows the soil properties of the site.

Figure 3. 75 Soil Properties at Khor Al-Zubair Site (Hanzawa 1983)

Failure occurred when the fill reached 11.5m height. Stability analysis was accomplished

using circular arc method as shown in Fig.3.77 and the factor of safety obtained was 1.03.

Moreover, the slope is analyzed by using SLIDE and a factor of safety of 1.387 is

obtained Fig.3.78.
95
Figure 3. 76 Circular Slip Surface at Khor Al-Zubair Site (Hanzawa 1983)

Figure 3. 77 Embankment Failure at Khor-Al-Zubair using SLIDE

96
3.2.29 Slide at Lian-Yun-Gang

The test site is located in an alluvial plain in the Lian-Yun-Gang area, Jiangsu

province, China (Chai et al. 2002). The built-to-failure embankment had a length of 45m

with a base width of 42m. The embankment has a 1V:1.75H slope and comprised of

compacted sandy clay with unit weight of 1.9 t/m3 and an angle of internal friction of

35. The soil profile where the embankment was placed consisted of a 2m thick clay crust

underlain by an 8.5m thick soft clay layer. Below the soft layer there were medium-to-

stiff sandy clay and silt sand layers. The soil properties of the soft deposit are

summarized in Fig.3.79. Slope stability analysis was carried out using the finite element

method and the subsoil and embankment fill material were represented by 8-node

quadrilateral and 6-node triangular elements. Failure occurred when the embankment

height reached a height of 4.04m. The factor of safety obtained from Chai et al.(2002)

analysis equals to 1.01 with a circular slip surface as shown in Fig.3.80. The slope is re-

analyzed by adopting Limit equilibirum method not the finite element method using

SLIDE software and the results are shown in Fig.3.81.

Figure 3.78 The Index and the Mechanical Properties of the Subsoil (Chai et al. 2002)

97
Figure 3. 79 Failure Surfaces from Field Observations and in Slip Circle Analysis
(Chai et al. 2002)

Figure 3.80 Failure Surface using SLIDE

98
 3.2.30 Slide at Congress Street

During the spring and summer of 1952, a portion of the Congress Street

“superhighway”, just east of Halsted Street, in Chicago, was built in an open cut. The cut

was for the most part in glacial clay and failed when the excavation reached a depth of 47

feet. The soil profile consisted of a deposit of sand and miscellaneous fill underlain by a

gritty blue clay as shown in Fig.3.82.

Figure 3. 81 Compressive Strength and Water- Content at Congress Street (Ireland 1954)

99
The approximate location of the slip surface is known from field evidence. A nearly

vertical escarpment formed at the top of the slope and a crack formed in the bottom of the

cut near the center line. It was assumed that the slip surface must be tangential to the stiff

layer. Slope stability analysis was conducted using the Ø=0 analysis and a factor of safety

of 1.11 was obtained (Ireland 1954) as shown in Fig. 3.83. The slope failure is analyzed

using SLIDE and a factor of safety of 1.486 is obtained Fig.3.84.

Figure 3. 82 Total Stress Stability Calculations (Ireland 1954)

100
 Figure 3. 83 Critical Slip Surface using SLIDE

3.2.31 Slide at Daikoku-Cho Dike

On March 19, 1981, a dike failed in Japan. The dike was constructed for a

reclaimed land for the urban waste material on a highly plastic marine clay which is

considered a normally consolidated clay. The soil profile consisted of a highly plastic

marine clay 10m to 15m thick underlain by a sandy gravel with N values more than 50.

Below the sandy gravel, a stiff clay and a fine sand with some gravels were found. The

undrained shear strength profile used for analysis is shown in Fig.3.85. The embankment

consisted of mudstone with unit weight of 1.5t/m3 and with an angle of friction of 35.

Kishida et al. (1983) and Hanzawa (1983) carried out slope stability analysis by adopting

both the simplified and the advanced Ø=0 method. The difference between the simplified

and advanced methods is the undrained shear strength values either from unconfined

compression test or SHANSEP. The authors ended up with a factor of safety of 0.91

Fig.3.86. Slope stability analysis is carried out using SLIDE and a factor of safety of

1.042 is obtained Fig.3.87.

101
Figure 3. 84 Undrained Shear Strength at Daikoku (Kishida et al. 1983)

Figure 3. 85 Circular Slip Surface at Daikoku Site (Kishida et al. 1983)

102
 Figure 3. 86 Total Stress Stability Analysis using SLIDE

3.2.32 Slide at Cuyahoga AA

In March 1971, an embankment failure occurred in the Cuyahoga River valley

about 5 miles south of Cleveland in Ohio in USA. The embankment fill consisted of

clayey silt with an undrained shear strength of 4.3t/m2. The subsoil conditions at the site

consisted of thick layer of lacustrine clay made up mainly of varved clay which is a silty

clay with silt and fine sand laminations. This layer overlies a layer of glacial till. An

organic silty sand layer is found below the glacial till (Fig.3.88).

103
 Figure 3. 87 Soil Properties and Critical Slip Surface at Cuyahoga AA site (Wu et al.
1975)

Wu et al. (1975) analyzed the stability of the embankment by the Morgenstern and Price

method using the circular arc method. The authors assumed a total stress analysis

assuming that the loading takes place in the undrained condition. A factor of safety of

1.25 was obtained. A slope stability analysis is conducted using SLIDE and a factor of

safety of 0.815 is obtained as shown in Fig.3.89.

Figure 3. 88 Critical Slip Surface using SLIDE

104
3.2.33 Slide at King's Lynn

In 1972, an induced failure occurred in a trial embankment at King’s Lynn in

England. The trial aimed at getting data regarding the installation of the sand drains (type,

spacing, diameter) and investigating the types of instruments needed for monitoring the

final work. The fill consisted of a light weight material called Carstone. The embankment

was laid on an upper alluvial layer composed of firm blue silty clay, soft brown clay and

soft blue peaty clay. The upper alluvial layer is underlain by a layer of peat of 2m thick.

Below the peat layer, a soft blue clay with traces of peat was found. Finally, a blue brown

clayey sand was present that overlyies a weathered Kimmeridge layer (Fig.3.90.).Wilkes

et al. (1972) conducted a total stress stability analysis based on a circular slip surface and

a factor of safety of 1 was resulted as shown in Fig.3.91.

Figure 3. 89 Soil Properties at King’s Lynn (Wilkes et al. 1972)

105
 Figure 3. 90 Observed and calculated Slip Surface (Wilkes et al. 1972)

When failure occurred, the final slip zone was non-circular and was deeper than

anticipated. Thus, total stress analysis was adopted for analyzing the slope using SLIDE

software by assuming an irregular slip surface and the results are shown in Fig.3.92.

Figure 3. 91 Critical Slip Surface using SLIDE

106
3.2.34 Slide at Muar

The Malaysian Highway Authority selected an appropriate site on the Muar

plain to construct full-scale test embankment built to failure to investigate in detail the

behavior of Muar clay deposits. The subsurface geology at the site revealed the existence

of a 2m thick weathered crust overlying a 16.50 thick layer of soft silty clay. This soft

silty clay is composed of an upper very soft and a lower soft silty clay. Below this lower

clay layer a thick peaty soil layer was found followed by a stiff sandy clay layer

(Fig.3.93).

Figure 3. 92 Variation of Soil Properties with Depth (Indraratna et al. 1992)

The embankment fill consisted of soil material with unit weight of 2.05t/m3 and strength

parameters of c=1.9t/m2 and an angle of internal friction of 26. The embankment failed

by the development of a quasi-slip circle type of rotational failure at a critical height of

107
5.5m with a tension crack propagating vertically through the crust and the fill. Indraratna

et al. (1992) conducted both finite element analysis and limit equilibrium method to

evaluate the stability of the slope. The slope is reanalyzed using SLIDE software and the

factor of safety is evaluated. The result is shown in Fig.3.94.

Figure 3. 93 Critical Slip Surface at Muar using SLIDE

3.2.35 Slide at North Ridge Dam

The North Ridge Dam is a 21.30 m high homogenous earth dam located

southeast of Raymond, Alberta. The fill material consisted of lean to medium plastic clay

of glacial origin with an average liquid limit of 31%, and an average plastic limit of 15%.

The fill was compacted to an average dry density of 17.9 t/m3 at an average water content

of 14.6%. The fill had a c= 41.4 KPa and Ø= 27 and was placed on a foundation soil

108
consisting of 3-5.8m of sand overlying 11-15m of soft highly plastic clay. A lower sand

layer underlies the clay to an unknown depth. The sand has an angle of internal friction of

29. The highly plastic clay has the properties shown in Table 3.4.

. Table 3.4. Soil Properties of Highly Plastic Clay, North Ridge Dam (Rivard et al. 1978)

Construction of the earth dam started in June 1953 and failure occurred at a fill height of

18.30m in mid-September. Failure of the embankment was indicated by cracks and slope

bulging. Rivard et al. (1978) carried out an effective stress analysis. However, Peterson

et al. (1957) conducted a total stress analysis as shown in Fig.3.95. and a factor of safety

of 1.23 was obtained. Using SLIDE software, a total stress analysis is conducted using

the soil properties illustrated above by assuming a circular slip surface. The results are

shown in Fig.3.96.

109
Figure 3. 94 Stability Analyses, North Ridge Dam(Rivard et al. 1978)

Figure 3. 95 Stability Analyses, SLIDE software

110
3.2.36 Slide at Seven Sisters Dike

In 1953, a slide took place at Seven Sisters on the Winnipeg River, Manitoba.

The dike failed when the embankment reached a critical height of 4.3m. The dike was

constructed of medium to highly plastic clay compacted to greater than 90% standard

proctor density at a water content of 5-10% above the Standard Proctor optimum water

content. The foundation conditions of the site consisted of 4.6m of highly plastic clay

underlain by a low to medium plastic glacial clay. Table 3.5 summarizes the physical

characteristics of the highly plastic clay.

Table 3.5 Soil Properties of the Highly Plastic Clay, Seven Sisters Dike (Peterson et al.
1957)

A total stress analysis was conducted by Peterson et al. (1957) using the profile shown in

Fig.3.97. and a factor of safety of 1.4 was obtained. Another total stress analysis was

conducted using SLIDE software and the results are shown in Fig.3.98.

111
 Figure 3. 96 Stability Analyses, Seven Sisters Dike (Rivard et al. 1978)

Figure 3. 97 Stability Analyses using SLIDE

3.2.37 Slide at Shellmouth Dam Test Fill

To evaluate the shear strength and the development of pore pressures for the

design of Shellmouth Dam, the Prairie Farm Rehabilitation Adminstration (PFRA)

112
constructed the Shellmouth Dam Test Fill. The fill was constructed of a well-graded

mixture of sand and gravel to a height of 16.8m with slopes of 1.75:1, 2:1, and 5:1. The

foundation of the test fill consisted of two clay layers, each about 6-7.6m thick, separated

by a 3m continuous sand layer. The two clay layers possessed similar soil properties as

shown in Table 3.6.

Table 3.6. Stability Analyses, Shellmouth Dam Test Fill (Rivard et al. 1978)

The sand layer between the clay layers was a mixture of poorly-graded and silty sands

with an angle of internal friction of 38. At a fill height of 16.80m, failure occurred and

construction stopped. An effective stress analysis was carried out by Rivard et al. (1978)

by assuming a circular arc surface. A total stress analysis was conducted using SLIDE

software and a factor of safety of 0.94 was obtaioned as shown in Fig.3.99.

113
 Figure 3. 98 Stability Analyses using SLIDE

3.2.38 Slide at Juban

In late 2002, the south approach embankment of the overpass of the Juban Road

experienced slope surface failures. The Juban Road is classified as a local collector that

crosses Interstate Highway 12 in Livingston Parish, Louisiana (Zhang et al. 2005). Based

on the field observations and laboratory results, the failure occurred due to the shrinkage

cracks that formed during the dry season. During the wet season, the water infiltrates into

the soil mass through the shrinkage cracks causing swelling that lead to the decrease in

the shear strength of the soil. The slope consisted of cohesive soil with a specific gravity

of 2.72 and contained 30.6% silt, 41.5% sand, and 27.9% clay. Its index parameters are as

follows (PL=15, LL= 37, PI=22). The soil has a maximum unit weight of 1.868t/m3 and

an unconfined compressive strength of 10.6 KPa. Both the embankment soil and the

foundation soil have the same soil properties. Zhang et al. (2005) conducted a total stress

114
analysis of the slope profile shown in Fig.3.100. The authors reported a factor of safety of

0.991 for an assumed circular slip surface.

Figure 3. 99 Schematic Model used in the analysis (Zhang et al. 2005)

Another total stress analysis is conducted by using SLIDE software and the factor of

safety is calculated by adopting different methods (Bishop, Ordinary method of slices,

Janbu, and Spencer). The results are shown in Fig.3.101.

Figure 3. 100 Critical Slip Surface using SLIDE

115
3.2.39 Slide at Bradwell

Five days after the excavation was completed in the London clay at Bradwell,

the slope failed. The excavation is 48.5 ft deep. The lower 28 ft of the excavation is in

London Clay and is inclined at 0.5(Horizontal) : 1(vertical). The London Clay is overlain

by 9ft of Marsh clay where the excavation slope was inclined at 1:1 (45). About 11.5 ft

of clay from the excavation was placed at the top of the excavation, over the marsh clay.

The clay fill was also inclined at 1:1. Fig.3.102. shows the soil properties of the site. A

representative unit weight for the London clay at the site was 120 pcf. The clay fill was

assumed to crack to the full depth of the fill and thus its strength was ignored. The marsh

clay was reported to have a total unit weight of 105 pcf. The analysis and the cross-

section adopted by Skempton and La Rochelle (1965) in the analysis of the slope is

shown in Fig.3.103. The SLIDE analysis (Fig.3.104) resulted in a factor of safety of 1.76.

Figure 3. 101 Properties of London Clay at Bradwell (Skempton and LaRochelle

1965)

116
Figure 3. 102 Cross Section of Excavated Slope at Bradwell (Skempton and LaRochelle
1965)

Figure 3. 103 Critical Slip Surface at Bradwell using SLIDE

117
3.2.40 Slide at Genesee

On July 28, 1982, longitudinal cracking was observed along the top of an

embankment located in Genesee, 60 km southwest of Edmonton. The fill was constructed

using a weathered clay shale compacted to 95% of standard Proctor maximum dry

density. The Genesee site is underlain by bedrock of the Late Cretaceous Paskapoo

Formation consisting of a sequence of shale, siltstone, sandstone, and coal deposits.

Moreover, the site is located on the side of a pre glacial valley in the bedrock surface and

was covered by a till deposit during glaciations. Fig.3.105. shows the stratigraphy of the

soil deposit. The failure occurred on a clay foundation with a thick crust, so the failure

surface is more likely to be circular. The failure took place when the embankment

reached a critical height of 12m. The slope was analyzed by Been et al. (1986)

(Fig.3.106) by adopting a total stress analysis and a factor of safety of 1 was obtained.

Another total stress analysis of the slope is conducted using SLIDE and the same factor

of safety is obtained as indicated in Fig.3.107.

Figure 3. 104 Stratigraphy and Engineering Properties of the soil at Genesee (Been et al.
1986)
118
Figure 3. 105 Slope Geometry at Genesee (Been et al. 1986)

Figure 3. 106 Critical Slip Surface at Genesee using SLIDE

119
3.2.41 Slide at Precambrian

On the morning of August 22, 1973, an embankment failure occurred at

Precambrian in Canada. The failure happened when the height of the embankment

reached 30 ft (7.6m). The fill material was a well-graded granular material, with particles

ranging from fine sand to cobbles. The density of the embankment material was

considered to be 130 pcf and the angle of internal friction was 35. The soil profile of the

foundation soil consisted of a 3-ft thick organic soil (muskeg) overlying a marine clay

formation with a 45 ft thickness. The marine formation consisted of an irregular sequence

of silty clay layers and thin seams of silt or fine sand. Below the marine formation, a

layer of fine to medium sand containing some gravel, cobbles and rocks having a

thickness of about 50 ft underlies the clay formation and rests directly on the bedrock

(Fig.3.108). Dascal et al. (1975) carried out the stability analysis by adopting a total stress

analysis and a circular slip surface was assumed (Fig.3.109) resulting in a factor of safety

of 1.1. The factor of safety is re-evaluated using SLIDE and the results are shown in

Fig.3.110.

Figure 3. 107 Soil Profile at Precambrian (Dascal et al. 1975)

120
 Figure 3. 108 Total Stress Analysis (Dascal et al. 1975)

Figure 3. 109 Stability Analysis at Precambrian using SLIDE

3.2.42 Slide at Scrapsgate

121
 At the end of January 1953, disastrous floods occurred on the east coast of

England. These floods led to the failure of many banks of earth constructed on saltings.

The slide at Scrapsgate is one of these banks failures. The failure occurred towards the

landward side and extended from a vertical tension crack near the top of the seaward

slope to the toe of the slip. The bank fill consisted of brown London Clay. The fill was

placed on a foundation site consisting of a 22-ft deep layer of soft grey organic peaty silty

clay. A layer of firm to stiff grey-brown London Clay underlies the soft clay (Fig.3.111).

Golder et al. (1954) conducted a short term analysis for the bank fill using the profile

shown in Fig.3.112. and a factor of safety of 1.3 was obtained. Fig.3.113. shows the

results of analyzing the slope using SLIDE.

Figure 3. 110 Shear Strength Values at Scrapsgate (Golder et al. 1954)

122
Figure 3. 111 Stability Analysis at Scrapsgate (Golder et al. 1954)

Figure 3. 112 Stability Analysis using SLIDE

123
3.2.43 Slide at Scottsdale

In February 1965, a major slip failure occurred along the Scottsdale Railway

Levée in Launceston, Australia. The Scottsdale Railway portion of the levee was

designed to be built in three stages. The first stage was constructed as an extension to the

existing railway embankment and was completed in July 1962. The second stage began in

January 1965 and was completed on 5th February, 1965. One day after the construction of

the second stage, the slip occurred. The levees consisted of an imported sandy clay. It

was constructed with a total length of 150 ft and net height of 7 ft above ground level.

The foundation soil consisted of a soft black clay layer extending to a depth of 50 ft. A

sand layer underlies the clay. Fig.3.114. shows the soil properties of the site.

Parry (1968) conducted a total stress analysis for the Scottsdale levee and a minimum

factor of safety of 1.6 was obtained. The factor of safety is recalculated by adopting the

total stress analysis and using SLIDE software (Fig.3.115).

Figure 3. 113 Soil Properties at Scottsdale (Parry 1968)

124
 Figure 3. 114 Critical Slip Surface using SLIDE

3.2.44 Slide at Iwai

The site is located on the Holocene lowland in Ibaraki Prefacture in Japan. The

foundation soil consisted of an organic clay layer bounded between two clay layers (C1

&C2). The soil properties of the layers are shown in Fig.3.116. and Fig.3.117.

Figure 3. 115 Soil Properties of Organic Clay (Shogaki et al. 2008)

125
 Figure 3. 116 Soil Properties of Clay1 & Clay2 (Shogaki et al. 2008)

The failure embankment was studied by Shogaki et al. (2008) to examine the effect of

soil variability and plasticity index on the inherent strength anisotropy of the soft clay

layers. Shogaki et al. (2008) conducted a total stress analysis by using the strength

measured by Unconfined Compression tests for the slope geometry shown in Fig.3.118.

The failure surface was assumed circular with a tension crack appearing in the

embankment height. The factor of safety is recalculated using SLIDE software and the

results are shown in Fig.3.119.

126
 Figure 3. 117 Total Stress Stability Analyses (Shogaki et al. 2008)

Figure 3. 118 Critical Slip Surface using SLIDE

3.2.45 Slide at Fair Haven

During the construction of a highway beginning at the New York State Border in

Fair Haven, Vermont and ending at a section open to traffic in Castleton, Vermont, an

embankment failure occurred in November 1971. The subsurface conditions consisted of

a layer of brown and grey varved silt extending from the surface to a depth of 15 to 20 ft
127
(layer A). A layer of grey and blue varved silt (layer B) 10 to 15 ft in thickness was

placed beneath layer A. Underlying layer B, a stratum of grey varved silty clay 10 ft in

thickness (layer C) existed. Below the varved clay a 5 to 10 ft thick layer of grey silty

sand resting on a shale bedrock was found. Atterberg Limits tests indicate that the clay

had a LL = 37, PL = 21, and PI = 16 while the silt has a LL = 30 and is non-plastic. The

fill has a unit weight of 130 pcf and an angle of internal friction of 35. The properties of

the different layers in the soil profile are:

 First layer: pcf, c = 0, Ø = 30

 Second layer: pcf, c = 0, Ø = 30
 Third layer: pcf, c = 700, Ø = 0
 Fourth layer: pcf, c = 20000, Ø = 45

Haupt and Olson (1972) performed a total stress analysis by assuming a circular failure

surface (Fig.3.120.) a factor of safety of 1.66 was obtained. The factor of safety is re-

evaluated by using SLIDE and a factor of safety of 1.88 resulted as shown in Fig.3.121.

Figure 3. 119 Stability Analyses (Haupt and Olson 1972)

128
 Figure 3. 120 Critical Slip Surface using SLIDE

3.2.46 Slide at Boston Marine Excavation

In September 1993, an excavation failure occurred in Boston Marine Clay. The

excavation is for the construction of 915m of I-90 adjacent to Logan Airport, Boston,

Massachusetts. The failure took place when the excavation reached a depth of 13.4m. The

stratigraphy from the ground surface downwards consisted of granular and cohesive fill,

organic silt, and a thick layer of marine clay formed from a grey clay with inter layered

silt seams of fine sands. The marine clay overlying glacial deposits (Fig.3.122) shows the

undrained shear strength used for the analysis.

129
Limit equilibrium analyses were performed by McGinn et al. (1993) and a factor of safety

of 0.97 was obtained as shown in Fig.3.123. Analysis using SLIDE is conducted and the

results are shown in Fig.3.124.

Figure 3. 121 Undrained Shear Strength using Field Vane (McGinn et al. 1993)

Figure 3. 122 Stability Analyses (McGinn et al. 1993)

130
 Figure 3. 123 Stability Analyses using SLIDE

3.2.47 Slide at Desert View Drive

In 1990, the failure of the Desert View Drive fill embankment occurred in La

Jolla. The failure took place when an excavation group cut into the embankment to

construct a building pad and house. The failure started as a circular mode and eventually

increasing to a wedge type movement of the entire fill embankment. Total stress analysis

was conducted by Day (1996) using the section shown in Fig.3.125. and the shear

strength parameters Ø = 18 and c = 24 KPa. The analysis was carried out using the

STABL computer program. The computer program, using the modified Janbu method of

slices, calculated a minimum factor of safety of 1.15 for the fill embankment. The

location of the critical failure surface is shown in Fig.3.125. Another analysis is

conducted using SLIDE software. The results are shown in Fig.3.126.

131
Figure 3. 124 Critical Slip Surface at Desert View Drive (Day 1996)

Figure 3. 125 Critical Slip Surface using SLIDE

132
3.2.48 Slide at Siburua

In the late evening of August 12, 1964, a slide took place in the downstream

slope of a dam composed of shaley clay. The foundation soil consisted of a thin layer of

sand and gravel underlain by a stratum of compacted clay core. Below the clay core, a

layer of cored sandstone overlies a stiff shaley fat clay with brown and red weathering

planes. Finally, an extremely hard red clay is found at the bottom. Wolfskill et al. (1967)

conducted total stress analysis using different methods. The first method requires the

equilibrium of a rigid free body of circular failure surface. The second method was the

ordinary method of slices. Moreover, the authors used Bishop method for comparing the

results of the factor of safety and finally they used the Morgenstern –Price general

method of slices.

The authors conducted the analysis for the slides that occurred during different periods.

They did the analysis for the slides that occurred on July 15, August 12, and October 5.

The following table illustrates the results they obtained using different slope stability

analyses methods. Fig.3.127. shows the geometry of the slope and the slip surface that

was obtained.

Total stress analysis is conducted for the slide that occurred on October 5 which is

considered the major slide in the dam. The analysis conducted using SLIDE software and

the results are shown in Fig.3.128.

133
Figure 3. 126 Total Stress Analysis at Siburua (Wolfskill et al. 1967)

Figure 3. 127 Stability Analyses using SLIDE

134
3.2.49 Slide at Tianshenqiao

On December 24, 1985, a landslide occurred at the Tianshenqiao Hydroelectric

Power Project in Guangxi Province, China. The landslide killed 48 people. The slide area

is on the right bank of the Nanpanjiang River. The slope consisted mainly of Quaternary

alluvium and talus covered by road fill and underlain by middle Tertiary bedrock,

composed of shales and sandstones. Table 3.7. shows the soil properties of the slope soil

materials.

Table 3. 7 Geotechnical Soil Parameters used in the Stability Analysis for Tianshenqiao
(Chen and Shoe 1988)

135
 Chen and Shoe (1988) conducted a total stress analysis for the profile shown in

Fig.3.129. and using the soil properties shown in Table.3.6. A factor of safety of 1.03 was

obtained. The factor of safety is re-calculated using SLIDE (Fig.3.130).

Figure 3. 128 Geological Profile at Tianshenqiao (Chen and Shoe 1988)

Figure 3. 129 Critical Slip Surface using SLIDE

136
3.2.50 Slide at San Francisco Bay

On August 20, 1970, a failure of a slope excavated underwater took place during

construction of a new shipping terminal at the port of San Francisco. The soil conditions

at the site were found to be quite uniform over the entire area. The profile consisted of

about 80 ft to 100 ft of San Francisco bay mud underlain by firmer clays and sands. The

San Francisco bay mud is a normally consolidated, slightly organic clayey silt or silty

clay of marine origin. The clay has moderate plasticity, with a liquid limit of about 50%

and a plastic limit of about 30%. Fig.3.131. shows the undrained shear strength profile

determined by the Field Vane shear tests. Duncan and Buchignani (1973) conducted a

total stress analysis of the site based on the undrained shear strengths and using the slope

geometry shown in Fig.3.132. The authors obtained a factor of safety of 1.17. Moreover,

the authors showed that the effect of sustained loading (creep) under undrained

conditions was probably the reason to reduce the shear strength and cause the failure.

Figure 3. 130 Undrained Shear Strength using Field Vane Tests (Duncan and
Buchignani 1973)

137
 Figure 3. 131 Slope Geometry at San Fransisco (Duncan and Buchignani 1973)

Duncan and Wright (2003) performed new slope stability calculations using a program

with Spencer’s procedure of slices. The minimum factor of safety calculated was 1.17.

The factor of safety is re-calculated by using SLIDE software and a factor of safety of 1.2

is obtained as shown in Fig.3.133.

Figure 3. 132 Critical Slip Surface using SLIDE

138
3.2.51Slide at Carsington

In June 1984, a failure occurred in Carsington Dam. The dam is located

near the village of Hognaston in Derbyshire. The failure occurred due to the heavy

rainfall that caused cracks in the dam followed by total failure. The foundation strata

consisted of the materials shown in Table 3.8.

Table 3.8 Soil Stratification at Carsington Dam (Skempton and Coats 1985)

A stability analysis was carried out by Skempton and Coats (1985) by dividing the sliding

mass above the slip surface into a number of vertical slices and by assuming that the

forces between the slices are inclined at an angle of 10 to the horizontal. The authors

used the profile shown in Fig.3.134. to evaluate the factor of safety. A factor of safety of

139
1.1 was obtained. Another total stress analyses is carried out in this study by taking the

same profile and soil properties. The results are shown in Fig.3.135.

Figure 3. 133 Profile at Carsington Dam (Skempton and Coats 1985)

Figure 3. 134 Critical Slip Surface using SLIDE

140
3.2.52 Slide at Atchafalaya

In December 1964, the construction of the Atchafalaya test sections was

commenced. Six months after construction of the test sections, two test sections had

moved laterally to the extent that cracks developed in the surface of the fills. In this study

the test section that exhibited more cracks was considered for analysis. The fill consisted

of fat clay and the soil properties of the site including the index parameters, undrained

shear strength, unit weight and the preconsolidation pressures are shown in Fig.3.136.

Figure 3. 135 Soil Properties at Atchafalaya Levees (Kaufman et al.1967)

141
Kaufman et al. (1967) conducted a total stress analysis for the test section using the soil

properties shown in Fig.3.136 and using the geometry shown in Fig.3.137. The factor of

safety was evaluated and a value of 1.10 was obtained. The factor of safety is recalculated

using the SLIDE software as shown in Fig.3.138.

Figure 3. 136 Geometry and Soil conditions at Atchafalaya Site (Kaufman et al. 1967)

Figure 3. 137 Ctitical Slip Surface using SLIDE

142
Table 3. 2. Soil Properties Adopted for the Analyses of the 52 Cases

Case Number/ Slope Type/ ØFill Layer Name H (m)  (t/m3) SuU SuR Su test type LL (%) PI LI S Notes
2 2
Slope Name Fill Type (t/m ) (t/m )

Fill 3 1.9 ˉ ˉ ˉ ˉ ˉ ˉ
Clayey silt 5 1.72 0.9 2.71 40 40 1 3.5
(1)
Fill/ granular 35 Clay I 2.5 1.77 1.5 2.2 FC,UC 44 50 1.27 7 a,c,m
Nesset
Clay II 2.5 1.7 1.5 0.36 39 50 1.69 42
Sand (Ø = 30 ) 7 2 ˉ ˉ ˉ ˉ ˉ ˉ
Fill 3 1.9 ˉ ˉ ˉ ˉ ˉ ˉ
(2)
Fill/ granular 30 Dry Crust 7 1.8 2.2 0.48 FC,UC,VT 43 45 1.111 4.5 a,c,m
Presterɸdbakken
Clayey Silt 13 1.8 1 0.333 35 45 1.667 30
Fill 2 1.9 2 2 ˉ ˉ ˉ ˉ
Dry Crust 2 2 4.5 2.25 38 30 0.53 2
(3)
Fill/ clayey ˉ Silty Clay 1 1 1.84 2.5 0.833 FC,UC,VT 42 21 0.14 3 a,c,m
As
Silty Clay 2 3 1.85 1.5 0.5 38 18 0.38 3
Silty Clay 3 14 1.84 1 0.045 38 18 1.6 22
Fill 2 2 ˉ ˉ ˉ ˉ ˉ ˉ
Dry Crust 1 2 2.1 8 4 45 15 0.167 2
(4)
Fill/ granular 30 Dry Crust 2 2 1.6 2 0.667 FC,UC,VT 78 36 0.916 3 a,b
Skjeggerod
Clayey silt 6 2.1 1 0.016 30 10 1.5 62
Silty Clay 10 1.9 1 0.034 28 10 1.22 29
Fill 1.5 1.9 ˉ ˉ ˉ ˉ ˉ ˉ
Peat 2 1.4 0.6 ˉ ˉ ˉ ˉ ˉ
(5) Silty Clay 1 1 1.88 1.2 0.0315 35 16 2 38
Fill granular 40 FC,UC a,c,n,m
Tjernsmyr Silty Clay 2 4 1.74 1.1 0.073 40 15 1.2 15
Silty Clay 3 1 1.79 1.5 0.021 32 10 2.5 70
Silty Clay 4 12 1.86 1.1 0.078 38 13 1.5 14
Fill 2 1.9 ˉ ˉ ˉ ˉ ˉ ˉ
Dry Crust 2.5 1.76 4 1 40 18 0.8 4
(6)
Fill/ granular 30 Silty Clay 1 4.5 1.75 2 0.4 FC,UC,VT 52 24 0.85 5 a,b
Aulielva
Silty Clay 2 2 1.8 1.85 0.617 58 33 0.5 3
Silty Clay 3 11 1.85 3 1 58 33 0.5 3
Fill 4 1.9 ˉ ˉ ˉ ˉ ˉ ˉ
Organic Silt with
2 1.74 3.2 3.2 ˉ ˉ ˉ ˉ
Sand
(7)
Fill/ granular 35 Clayey silt with FC,UC a,b
Falkenstein 5 1.64 0.8 0.011 25 5 3 70
gravel and shells

Silty Clay 1 2 1.92 1.2 0.022 21 9 2.2 50

Silty Clay 2 11 2 5.25 0.04 22 10 2.9 127
Fill 2.5 1.9 ˉ ˉ ˉ ˉ ˉ ˉ
Dry Cryst 1 1 1.94 1.5 2 ˉ ˉ ˉ 1
Dry crust 2 1 1.8 1.5 0.75 ˉ ˉ ˉ 2
Silty Clay 1 1 1.8 0.9 0.456 ˉ ˉ ˉ 2
(8)
Fill/ granular 35 Silty Clay 2 3 1.78 0.9 0.225 FC,UC,VT ˉ ˉ ˉ 4 a,b
Jalsberg
Silty Clay 3 2 1.77 1 0.167 ˉ ˉ ˉ 6
Silty Clay 4 2 1.72 2.5 0.5 ˉ ˉ ˉ 5
Silty Clay 5 5 1.76 1.25 0.125 ˉ ˉ ˉ 10
Silty Clay 6 5 1.8 1 0.111 ˉ ˉ ˉ 9

143
Table 3.2.(Continued.)
(Case
Slope
Number) H  SuU SuR Su test LL
Type/ Layer Name 3 2 2 PI LI S Notes
Slope ØFill (m) (t/m ) (t/m ) (t/m ) type (%)
Fill Type
Name
Fill 4 1.88 ˉ ˉ ˉ ˉ ˉ ˉ
Top Soil 0.4 1.92 4 1.333 ˉ ˉ ˉ ˉ
(9)
Fill/ Clay Crust 1 1.2 1.68 3 0.2 55 25 1.4 15
Saint 35 CIU,UU,VT a,b,d
granular Clay Crust 2 3 1.68 3 0.2 60 100 1.4 15
Alban
More Silty 1 0.5 1.76 1.6 0.025 42 22 2.5 64
More silty 2 4.7 1.73 3 0.077 40 20 2 39
Fill 9.6 2.07 ˉ ˉ ˉ ˉ ˉ ˉ
Soft Clay and
2 1.96 2.8 1.12 42 21 0.4 2.5
Silt
Soft, organic
2 1.9 3 0.375 42 21 0.95 8
clay and silt
Peat 1 1.73 2.9 0.045 40 19 2.31 64
(10) Fill/ Silty Sand with
26 1.5 2.08 2.5 0.312 LVS,VS 30 10 0.91 8 a,b,d
Narbonne granular some clay
Sand, silt and
clay of low 1.5 2.06 1.9 0.585 30 10 0.6 3.25
plasticity 1
Sand, silt and
clay of low 3 1.98 2 0.002 30 10 12 1000
plasticity 2
Fill 4 1.82 ˉ ˉ ˉ ˉ ˉ
Dry Crust 1 1 1.6 3.6 3.6 ˉ ˉ ˉ 1
Dry Crust 2 1 1.4 1.4 0.175 100 58 1.12 8
Dry crust 3 1.5 1.35 1.3 0.108 120 70 1.2 12
Soft, organic
sandy clay and 1.5 1.33 1.533 0.191 120 70 1.12 8
silt 1
(11) Fill/ Soft, organic
30 VS a,b,d
Lanester granular sandy clay and 1 1.36 1.8 0.075 100 50 2 24
silt 2
Soft, organic
sandy clay and 1 1.39 1.95 0.244 120 70 1.12 8
silt 3
Soft, organic
sandy clay and 1 1.41 2.05 0.293 100 50 0.83 7
silt 4
Fill 4.5 2.21 ˉ ˉ ˉ ˉ ˉ ˉ
Dry Crust 1 1 1.3 4.5 1.56 80 60 0.42 2.88
Dry Crust 2 1 1.8 3 0.43 80 60 0.9 7
Dry Crust 3 1 1.22 2.2 0.667 140 90 0.52 3.3
Dry Crust 4 1 1.5 2.2 0.667 110 70 0.52 3.3
Soft, organic
1 1.48 2.2 0.55 90 50 0.75 4
(12) silty clay 1
Fill/
Cubzac- 35 Soft, organic VS a,b,d
granular 1 1.71 2.2 0.275 90 50 0.7 8
les ports silty clay 2
Soft, organic
1 1.43 2.2 0.55 110 70 0.75 4
silty clay 3
Soft, organic
1 1.6 5 0.125 114 75 0.75 4
silty clay 4
Soft, organic
1 1.48 5 0.172 110 65 0.41 2.9
silty clay 5

144
Table 3.2. (Continued.)
Slope
Case
Type/ H  SuU SuR LL
Number) Layer Name 3 2 2 Su test type PI LI S Notes
Fill ØFill (m) (t/m ) (t/m ) (t/m ) (%)
Slope Name
Type
Clay with shells
4 2 4 2 38 18 0.3 2
and sand 1
Clay with shells
2 1.9 4 1 30 10 0.9 4
and sand 2
(13) Clay with shells
Lodalen 1 2 2 4 1.2 30 15 0.7 3.33
and sand 3
(14) Cut/
ˉ Clay with shells UC c,e
Lodalen 2 - 3 2 4 1.7 40 20 0.1 2.35
and sand 4
(15)
Lodalen 3 Clay with shells
2 2 4 2 35 15 0.2 2
and sand 5
Clay with shells
3 2 4 1.5 35 15 0.4 2.67
and sand 6
Clay with shells
and sand 7
3 2 5 2.5 35 15 0.2 2
Fill 2.8 1.8 ˉ ˉ ˉ ˉ ˉ ˉ
Crust 2.5 1.5 1.5 1.5 160 100 1
(16) Soft Grey Clay 1 2.5 1.3 0.75 0.21 130 80 0.8 3.6
Fill/
Rio de 35 VS, UU b,e
granular Soft Grey Clay 2 2.5 1.3 0.96 0.31 120 80 0.75 3.1
janeiro
Soft Grey Clay 3 2.5 1.3 1.2 0.41 100 65 0.65 2.9
Soft Grey Clay 4 2.5 1.3 1.4 0.51 100 65 0.4 2.74
Fill 6 2.04 ˉ ˉ ˉ ˉ ˉ ˉ
Crust 5 2 4.8 0.1 70 10 2.5 48
(17) Soft clay 1 7 1.3 2 0.067 30 10 4 200
Fill/
New 40 UC,UU,VS,SHANSEP a,b
granular Soft clay 2 10 1.3 1.2 0.06 30 10 4 200
Liskeard
Soft clay 3 2 1.3 2 0.06 30 10 4 200
Soft clay 4 8 1.3 3.2 0.16 30 10 4 200
Fill 2 2 ˉ ˉ ˉ ˉ ˉ ˉ
Crust 1 1.5 2.9 ˉ 90 50 1.2 13
Clay 1 2 1.33 0.991 0.123 150 90 1 8
(18) Fill/ Clay 2 2 1.32 0.838 0.105 150 90 1 8
20 VS a,b
Bangkok A granular Clay 3 2 1.34 1.117 0.14 150 90 1 8
Clay 4 1.5 1.35 1.33 0.166 150 90 1 8
Clay 5 1 1.35 1.62 0.2 150 90 1 8
Clay 6 1 1.37 1.755 0.22 150 90 1 8
Very soft to soft
5 1.85 1 0.12 38 18 1 8
(19) grey silty clay 1
DrammenV Very soft to soft
(20) 5 0.85 1 0.123 35 15 1 8
Cut/ grey silty clay 2
DrammenVI ˉ CIU,CID, UC, VS b,e
ˉ Very soft to soft
(21) 3 0.9 2 0.215 32 12 1.08 9
Drammen grey silty clay 3
VII Very soft to soft
4 0.95 2.5 0.4 32 12 0.83 6
grey silty clay 4

145
Table 3.2. (Continued.)

Slope
(Case Number) H  SuU SuR Su test LL
Type/ Layer Name 3 2 2 PI LI S Notes
Slope Name ØFill (m) (t/m ) (t/m ) (t/m ) type (%)
Fill Type
Fill 4 2 1 1 ˉ ˉ ˉ ˉ
Crust 2 2 2.5 2.5 ˉ ˉ ˉ 1
Blue and brown
very plastic Clay 3 1.5 1.4 0.14 80 45 1.22 10
1
Blue and brown
(22) Fill/ very plastic Clay 5 1.5 4.2 0.42 80 45 1.22 10
40 VS a,b
Pornic clayey 2
Blue and brown
very plastic Clay 4 1.5 5 0.5 80 45 1.22 10
3
Blue and brown
very plastic Clay 3 1.5 4.5 0.45 80 45 1.22 10
4
Fill 3 2 ˉ ˉ ˉ ˉ ˉ
Organic Clay 1 1.15 1.5 0.6 83 37 0.08 2.5
(23) Fill/
35 Very Peaty Clay 2 0.27 1.2 0.004 VS ˉ ˉ ˉ 300 a,b
Saint -Andre granular
Peat 0.5 0.15 1.3 0.0048 ˉ ˉ ˉ 300
Organic Mud 5 0.7 1.6 0.16 102 47 1.17 10
Fill 6.5 1.7 ˉ ˉ ˉ ˉ ˉ ˉ
(24) Soil 1 5 1 2.8 0.7 64 32 1 4
Fill/
South of 35 Soil 2 5 1 2.2 0.55 VS 64 32 1 4 a,b
granular
France Soil 3 5 1.7 3 0.75 64 32 1 4
Soil 4 5 1.7 3 0.75 64 32 1 4
Fill 1.5 2.08 ˉ ˉ ˉ ˉ ˉ
Medium plastic
3 1.75 3.8 0.475 58 33 1.2 8
grey clay
Grey Clay of high
3 1.4 3 0.333 70 40 1.21 9
(25) plasticity 1
Fill/
NBR 35 Grey Clay of high VS a,b
granular 3 1.48 2 0.286 62 32 1.1 7
Development plasticity 2
Grey Clay of high
3 1.48 2.4 0.267 75 40 1.21 9
plasticity 3
Light Grey Clay 3 1.7 3.2 0.32 42 12 1.22 10
Silt and Fine Sand 3 1.96 5 0.72 42 12 0.8 7
Fill ˉ 17.28 ˉ ˉ ˉ ˉ ˉ ˉ
Medium Clay 1.5 18.54 48 3.83 45 20 1.2 12.5
(26) Fill/ Soft Clay 1 1.5 18.54 14.4 3.83 35 15 0.5 3.75
30 VS b,e
Portsmouth granular Soft Clay 2 4.6 17.12 14.4 3.83 35 15 0.5 3.75
Soft Clay 3 3 18.85 14.4 3.83 35 15 0.5 3.75
Sand Silt (Ø =30) 1.5 20.4 ˉ 3.83 35 15 0.5 3.75
Fill 6.3 1.9 ˉ ˉ ˉ ˉ ˉ ˉ
Peat 5.5 1.15 2.4 0.8 ˉ ˉ ˉ 3
(27) Fill/
35 Sandy Clay with UC,VS,qc a,b,l
Kameda granular 2.2 1.6 3.35 1.116 ˉ ˉ ˉ 3
Peat
Peat 2 3 1.212 4.025 1.35 ˉ ˉ ˉ 3
Fill 11 1.85 ˉ ˉ ˉ ˉ ˉ ˉ
(28)
Fill/ Clay 5 1.8 2.5 0.156 55 27 1.48 16
KhorAl - 35 VS b,f,g
granular Hard Failure 1 1.8 5000 1428 45 20 0.55 3.5
Zubair no.4
Clay 14 1.75 2.5 0.714 55 30 0.567 3.5

146
 Table 3.2. (Continued.)
Case
Slope
Number) H  SuU SuR Su test LL
Type/ Layer Name 3 2 2 PI LI S Notes
Slope ØFill (m) (t/m ) (t/m ) (t/m ) type (%)
Fill Type
Name
Fill 2 1.9 0.5 ˉ 20 1.25 4
Top Crust 1 0.5 1.7 2.4 1.6 20 1 1.5
Top Crust 2 1 1.7 1.6 1.067 21 1 1.5
Clay 1 1 1.5 0.76 0.304 25 1.2 2.5
(29) Clay 2 1 1.5 0.8 0.08 22 1 10
Fill/
Lian-Yun- 25 Clay 3 1 1.6 0.92 0.23 VS 30 1.25 4 a,b, h
clayey
Gang Clay 4 1 1.6 1.06 0.424 30 1.5 2.5
Clay 5 1 1.6 1.2 0.6 30 1.75 2
Clay 6 1 1.7 1.7 0.68 30 2 2.5
Clay 7 1 1.7 1.8 0.72 30 2 2.5
Clay 8 4 1.7 1.9 0.76 30 2 2.5
sand and
miscellanous fill (Ø 1.8 17.3 ˉ ˉ 28 13 0.385 3.25
=30)
(30) Medium gritty blue
Cut/ 4.3 20.74 74.8 23 31 13 0.538 3.63
Congress ˉ clay 1 UC, VS a,b
-
Street Medium gritty blue
6.1 20.11 43.18 11.87 32 14 0.538 3.63
clay 2
Stiff to very stiff
2.7 20.11 43.18 11.87 31 14 0.643 4
gritty blue clay
Fill 15 1.5 ˉ ˉ
Marine Clay 1 2.5 1.9 3 0.43 100 60 0.833 7
(31) Marine Clay 2 2.5 1.9 4 1.334 110 55 0.64 3
Fill/
Daikoku- 30 Marine Clay 3 2.5 1.9 4 1.6 VS 120 61 0.426 2.5 a,b
granular
Cho Dike Sand 1(Ø =40) 2.5 2 3.5 ˉ ˉ ˉ ˉ ˉ
Sand 2(Ø =30) ˉ 2 ˉ ˉ ˉ ˉ ˉ ˉ
Sand 3(Ø =40) ˉ 2 ˉ ˉ ˉ ˉ ˉ ˉ
Fill 20 22 ˉ ˉ ˉ ˉ ˉ ˉ
Glacial Till 3 18.9 191.5 83.3 30 20 1 2.3
(32)
Fill/ Varved Clay 6 19.6 96 41.6 30 20 1 2.3
Cuyahoga 35 UC a,c
granular Silty Clay 4.5 19.6 43 18.7 30 20 1 2.3
AA
Organic Silty Clay 1.5 19.6 33.5 14.55 30 20 1 2.3
Varved Clay 18 19.6 96 41.6 30 20 1 2.3
Fill 4.5 2.02 ˉ ˉ ˉ ˉ ˉ ˉ
Upper Alluvial 1 1.7 3 1.2 62 22 0.091 2.5
Peat 1 1.2 1.8 0.1125 45 25 1.6 16
Lowe Alluvial 1 1.7 2.2 0.0346 62 41 2.2 52
(33) Fill/
35 Lower Alluvial 2 1 1.7 3 0.0236 VS 60 32 2.72 93 a,b
King's Lynn granular
Lower Alluvial 3 1 1.7 3 0.107 20 10 1.8 28
Sand(Ø = 30) 0.6 2 ˉ ˉ ˉ ˉ ˉ ˉ
Weathered
1.2 1.7 4.2 1.3 60 42 0.476 3.25
Kimmeridge
Fill 5.5 2.05 1.9 ˉ ˉ ˉ ˉ ˉ
Crust 2 1.65 2.5 0.5 70 40 1.25 5
Very Soft Clay 6.1 1.55 1.1 0.3 80 45 0.888 3.67
(34) Fill/
26 Soft Clay 9.7 1.55 2 0.5 VS 70 40 1.125 4 b,e
Muar clayey
Peat 0.5 1.5 1.1 0.5 75 45 0.67 2.2
Sandy Clay 4.6 1.6 3.5 0.75 68 43 1.2 4.67
Sand(Ø = 30°) 3 1.6 3.5 ˉ ˉ ˉ ˉ ˉ

147
 Table 3.2. (Continued.)
(Case Slope
H  SuU SuR Su test LL
Number) Type/ Layer Name 3 2 2 PI LI S Notes
ØFill (m) (t/m ) (t/m ) (t/m ) type (%)
Slope Name Fill Type

(35) Fill 19 2.01 4.14 ˉ ˉ ˉ ˉ ˉ

Fill/
North Ridge 27 Sand(Ø = 29°) 4 1.81 ˉ ˉ UC ˉ ˉ ˉ ˉ a,c
clayey
Dam Highly Plastic Clay 13 1.81 4.825 1.93 72 51 0.313 2.5
Fill/Clay
(36) 13 1.94 2.54 ˉ ˉ ˉ ˉ ˉ
Fill Fill 4.2
Seven
Fill/Rock 35 1.85 ˉ ˉ UC ˉ ˉ ˉ ˉ
Sisters Dike
ˉ Highly Plastic Clay 4.6 1.686 1.5 0.6 97 67 0.268 2.5
Fill 17 1.98 ˉ ˉ ˉ ˉ ˉ ˉ
(37) Highly Plastic Clay
6.1 1.74 3.445 0.984 58 37 0.486 3.5
Shellmouth Fill/ (upper clay)
43.5 UC a,c
Dam Test granular Sand(Ø = 38) 2.8 1.96 ˉ ˉ ˉ ˉ ˉ ˉ
Fill Highly Plastic Clay
6.4 1.74 3.445 0.984 58 37 0.486 3.5
(Lower clay)
Fill 1 1.55 0.53 0.212 ˉ ˉ ˉ 2.5
(38) Fill/ 6
ˉ Fill 2 1.6 1.89 0.756 UC ˉ ˉ ˉ 2.5 c,i
Juban I Clayey
Foundation Soil 10 1.65 2.9 1.16 ˉ ˉ ˉ 2.5
Clay Fill 3.5 17.3 0.1 0.1 ˉ ˉ ˉ ˉ
(39) Cut/
ˉ Marsh Clay 2.8 16.5 14.4 14.4 VS ˉ ˉ ˉ ˉ a,b
Bradwell ˉ
London Clay 9.8 18.9 86.2 43.1 92 67 0.104 2
Fill, weathered
22 1.8 7 7 ˉ ˉ ˉ ˉ
clay shale
Clay Crust 5 1.8 7 2.3 60 40 0.25 3
(40) Fill/ Un-weathered
ˉ 10 1.8 2.3 2.3 VS, UU 60 30 0 1 a,b,g
Genesee clayey Clay
Firm to stiff grey
10 1.8 5 2.3 25 7 0.2 2
clay
Bedrock 5 2 500 500 ˉ ˉ ˉ ˉ
Fill 9 20.5 ˉ ˉ ˉ ˉ ˉ ˉ
Clay with low
2.3 17.75 72 8.95 40 25 8
plasticity
(41) Fill/ Clay with medium
35 2.3 18.06 47.9 5.3 VS 30 15 9 a,b
Precambrian granular plasticity
Low plastic clay
4.6 17.3 38.3 2.54 30 2 15
with silt
Clay 15 18.9 24 1.484 30 2 16
Bank Fill 6.1 16.88 47.87 16.75 ˉ ˉ ˉ ˉ
(42) Fill/
ˉ Sogt Organic silty UU,VS a,b
scrapsgate clayey 7.6 15.71 16.75 3.35 67 47 ˉ 3
clay with peat
Fill 1.8 17.3 ˉ ˉ
Brown Clay 0.9 17.3 10.53 4.25 145 100 0.4 2.5
(43) Fill/
25 Dark Grey Clay 1 2.1 17.3 10.05 4.02 VS ˉ ˉ ˉ 2.5 a,b
scottsdale granular
Dark Grey Clay 2 2.1 17.3 14.36 5.75 ˉ ˉ ˉ 2.5
Soft Clay 4.9 17.3 24 9.6 ˉ ˉ ˉ 2.5

148
Table 3.2. (Continued)

Slope
(Case
Type/ H  SuU SuR Su test LL
Number) Layer Name 3 2 2 PI LI S Notes
Fill ØFill (m) (t/m ) (t/m ) (t/m ) type (%)
Slope Name
Type
Fill 4.5 1.5 ˉ ˉ ˉ ˉ ˉ ˉ
Clay 1 2 1.36 1.21 0.4033 67 34 0.5 3
(44) Fill/
25 Organic 4 1.01 1.55 0.484 UC 655 370 0.6 3.2 a,c
Iwai granular
Clay 2 4 1.36 1.23 0.41 67 34 0.5 3
Sand(Ø =30°) 1.5 1.6 ˉ ˉ ˉ ˉ ˉ ˉ
Fill 14 20.5 ˉ ˉ ˉ ˉ ˉ ˉ
Berm(Ø =30) 5.8 22 ˉ ˉ ˉ ˉ ˉ ˉ
Brown and grey
2.6 18.7 ˉ ˉ ˉ ˉ ˉ ˉ
varved silt(Ø =30°)
(45) Fill/
35 Grey and blue LVS a
Fair Haven granular 4.7 9 ˉ ˉ ˉ ˉ ˉ ˉ
varves silt(Ø =30°)
Grey varved silty
6.4 9 33.5 2.6 37 16 1.31 13
clay
Grey Silty(Ø =45°) 3 26 960 960 ˉ ˉ ˉ ˉ
Cohesive Fill 8.1 1.8 6 6 ˉ ˉ ˉ ˉ
Organic Silt 1.9 1.75 6.5 1.86 35 5 0.6 3.5
(46)
Upper Marine Clay 4 1.6 8 0.62 30 12 1.42 13
Boston Cut/
ˉ Middle Marine Clay 9 1.6 5 0.385 VS 30 12 1.42 13 a,b
Marine ˉ
Excavation Lower Marine Clay 4 1.6 6 0.462 30 12 1.42 13
Glaciomarine
8 1.78 7 0.54 26 7 1.29 11
Deposits
(47) Fill V 2 2.4 0.96 ˉ ˉ ˉ 2.5
Fill/
Desert View 18 ˉ a,b,j
clayey
Drive Bedrock V 2 500 ˉ ˉ ˉ ˉ ˉ
(48)
Compacted Clay
Siburua ˉ 9 2.04 0.8 0.8 UU 45 21 1 a,k
Core
October 5
New Fill V 1.85 1.96 0.784 ˉ ˉ ˉ ˉ
Old Fill V 1.85 1.96 0.784 ˉ ˉ ˉ ˉ
clay with rock
V 1.85 0 0 ˉ ˉ ˉ ˉ
(Ø =21.8°)
fine sand and
medium sand V 1.85 2.94 1.176 ˉ ˉ ˉ ˉ
(49) Fill/ (Ø =20.8°)
21.8 ˉ a,i
Tianshenqiao clayey
grey and dark silty
V 1.81 3.43 1.372 ˉ ˉ ˉ ˉ
clay(Ø =10.2°)
gravels and sands
V 1.9 0 0 ˉ ˉ ˉ ˉ
(Ø =24.2°)
Tertaairy Bedrock
V 2.4 3.92 1.568 ˉ ˉ ˉ ˉ
(Ø =45°)
Debris V 0.39 4 ˉ ˉ ˉ ˉ ˉ
(50) Sand(Ø =30) V 2 ˉ ˉ ˉ ˉ ˉ ˉ
San Cut/ Mud 10 0.6 1.6 0.04 50 20 2 40
ˉ UU a,b
Francisco - Clay 1 5 0.6 2 0.05 50 20 2 40
Bay Clay 2 5 0.6 3 0.075 50 20 2 40
Clay 3 5 0.6 4.2 0.105 50 20 2 40

149
 (Case Slope
Number) Type/ H  SuU SuR Su test LL
Layer Name 3 2 2 PI LI S Notes
Slope Fill ØFill (m) (t/m ) (t/m ) (t/m ) type (%)
Name Type
Core V 1.85 6.5 4.333 74 42 0.048 1.5
(51) Fill/ Zone I V 2.05 6.5 2.167 79 45 0.2 3
ˉ ˉ a
Carsington clayey Zone II V 2.1 6.5 6.5 ˉ ˉ ˉ 1
Protection V 1.85 6.5 6.5 ˉ ˉ ˉ ˉ
Peat and soft
12 15.71 31.6 1.975 100 70 1.5 16
organic Clay
(52) Fill/
ˉ Soft to Medium UC,UU a,c
Atchafalaya Clayet 15 17.28 42 14 90 65 0.492 3
Clay
Medium Clay 12 18.06 52.66 21 80 50 0.2 2.5

Abbreviations
H: Layer Thickness
: Unit Weight
SuU: Undisturbed Undrained Shear Strength
SuR: Remolded Shear Strength
S: Sensitivity
FC: Fall Cone
UC: Unconfined Compression
VT: Vane Test
UU: Unconsolidated Undrained
CIU: Consolidated Isotropic, Undrained
LVS: Laboratory Vane Test
SHANSEP: Stress History and Normalized soils Engineering Parameters
CID: Consolidated Isotropic, Drained
qc: Point Resistance measured in CPT
V : Variable

Notes
a : Remolded shear strength is calculated from sensitivity
b : Analysis based on Uncorrected Field Vane test
c : Analysis based on Unconfined compression test
d : Liquidity index is given
e : Remolded shear strength is given
f : Assume Atterberg limits
g : Assume high shear strength (500 t/m2) to indicate the presence of a hard layer
h : PI is estimated from figure
i : Correlation between dry unit weight and water content to get Unconfined compression strength
LVS : laboratory field vane
j : Sensitivity is assumed
k : Analysis based on Unconsolidated Undrained Testing
l: sensitivity is assumed
m: Field vane shear strength values are not clear in the paper
n: Peat properties are assumed
o: Fill properties are assumed
p: Lodalen (1), (2), and (3) have same properties but differ in the slope angle and the slope height
q:Drammen V, Drammen VI, and Drammen VII have same soil properties but different slope angle and height

150
 CHAPTER 4

Quantification of Model Uncertainty and Investigation of the

lower-bound factor of safety for slopes

4.1 Introduction

In slope stability analysis and design, one frequently uses models to evaluate the

stability of slopes. These models represent the physical phenomenon of slope stability by

mathematical or numerical solutions for the factor of safety of slopes. For instance, the

Limit Equilibrium Methods of Bishop, Ordinary Method of slices, Janbu, and Spencer are

popular methods among engineers for studying the stability of slopes. The effectiveness

of these models has never been thoroughly tested due to the lack of databases of

historical published records of slope failures. Databases are needed for evaluating biases

and uncertainties in these models for predicting the factor of safety of slopes. In this

chapter, the database discussed in chapter 3 and summarized in Table 3.1. is used to

accomplish the following objectives: (1) quantify the model uncertainty of these slope

stability models by evaluating the statistics {mean and coefficient of variation (COV)}

and the probability distribution of the ratio of measured to predicted factor of safety for

each method, and (2) investigate the presence of a lower-bound factor of safety that can

be calculated using information on the slope geometry and site-specific soil properties.

151
4.2 Quantification of model uncertainty

4.2.1 The Importance of Quantification of Model Uncertainty

Model uncertainty is the uncertainty associated with the geotechnical model due

to the inability of the model to fully represent the true physical behavior of a geotechnical

system. Model uncertainty arises from unavoidable idealizations in analytical or

numerical models for predicting engineering behavior. Mathematical modeling of any

physical process generally requires approximations to create a usable model.

Unavoidably, the resulting models are simplifications of complex real world

phenomenon. Consequently there is uncertainty in the model prediction even if the model

inputs are known with certainty.

The magnitude of model uncertainty is important for geotechnical decision making. If

model uncertainty is not considered, the geotechnical predictions and hence the decisions

based on the geotechnical predictions might be biased. Tang and Gilbert (1993) and

Lacasse and Nadim (1994) noted that the calculated probability of failure without

considering model uncertainty was not the actual failure probability of geotechnical

systems.

4.2.2 Mean, COV, and Distribution Type of Model Uncertainty

The model uncertainty in a slope stability calculation method could be quantified

from a database of slope failure case histories in terms of a mean (bias), a standard

deviation (and/or Coefficient of variation) and a probabilistic density distribution of the

152
ratio of the measured to predicted factor of safety of the failed slopes. In this study, a

framework for characterizing model uncertainty using observation data is proposed. The

proposed framework is illustrated by characterizing the model uncertainty of four limit

equilibrium methods for slope stability analysis (Bishop, Ordinary method of slices,

Janbu, and Spencer.

[Link] Evaluation of the Statistical Parameters of the Model Uncertainty

In an initial analysis, predicted factors of safety for the 52 case histories presented

in Table 3.1. are calculated using the SLIDE software for the four Limit Equilibrium

Methods. In this analysis, values of undrained shear strength as reported in the original

references (see Table 3.2) are used as input to the model. An investigation of the original

references for the different case studies indicates that different types of tests were

conducted for each case history to evaluate the undrained shear strength. In this initial

analysis, only the results of field vane tests are adopted. For cases that do not include

field vane records, the unconfined compression test results are adopted instead.

The 52 case histories include cohesive soils in the upper portion of the slope. Slope

stability calculations usually show tension at the interfaces between slices as well as on

the bottom of the slices. When tension develops, numerical problems in the slope stability

calculations could occur (Duncan and Wright 2005). To overcome these problems the

tension forces should be eliminated. Introducing tension cracks to the analysis can

eliminate these tensile stresses. In the analysis conducted in this research study, an

automatic tension crack search procedure using SLIDE software is applied.

153
 To quantify the model uncertainty, the ratio of the measured factor of safety to

that of the predicted factor of safety (is calculated using SLIDE for the four Limit

Equilibrium Methods for all the cases in the database. The predicted factors of safety

from the different methods are presented in Table 4.1 for the 52 cases. Since the cases are

actual historical failed slopes, the measured factor of safety could be realistically

assumed to be approximately equal to 1. Based on this assumption, the ratio of measured

to predicted factors of safety  was calculated for all the cases in the database and for the

different slope stability methods considered.

The calculated values of  for the four models are presented in Fig.4.1 for the 52

cases analyzed in this study. Results on Fig. 4.1 indicate that the ratio of the measured to

predicted factor of safety varies significantly between the different cases with minimum

and maximum values of about 0.5 and 1.8, respectively. The mean value of the ratio of

measured to predicted factor of safety  is found to vary from 0.964 (for the Spenser

method) to 1.036 (for the Janbu method). The mean of  is a direct measure of the “bias”

in the prediction model. Based on the statistics of  as reflected in Table 4.2, it could be

concluded that the 4 limit equilibrium methods could be considered to be relatively

unbiased with mean values of  that are very close to 1.0. The coefficient of variation

(COV) of  is an indication of model uncertainty. Results on Fig. 4.1 and Table 4.2

indicate that the predictions of the different models show considerable scatter with COVs

ranging from 0.256 (for the ordinary method of slices) to 0.285 (for the method of Janbu).

These COV values could be considered to be significant and in line with model

uncertainties that are generally encountered in other areas of geotechnical engineering.

154
Table 4. 1 Predicted and Lower-Bound Factors of Safety for the Slopes in Database
FS FS LB
N0 FS FS LB LB LB
Slope Name OM Janb Janb
. Bishop Spencer Bishop OMS Spencer
S u u
1 Nesset 1.205 1.1 1.080 1.198 0.271 0.274 0.266 0.267
0.987 3
0.9 0.940 0.986 0.089 0.089 0.087 0.089
2 Presterɸdbakke
3 As
n 0.810 0.8
7 0.674 0.808 0.525 0.504 0.489 0.519
4 Skjeggerod 0.700 0.7
1 0.642 0.702 0.01 0.011 0.01 0.012
5 Tjernsmyr 0.834 0.8
0 0.740 0.832 0.078 0.078 0.071 0.073
6 Aulielva 1.137 1.1
3 1.078 1.134 0.228 0.228 0.214 0.228
7 Falkenstein 1.071 1.0
3 1.067 1.067 0.318 0.322 0.616 0.546
8 Jalsberg 1.145 1.1
1 1.094 1.141 0.384 0.381 0.603 0.602
9 Saint Alban 1.276 1.1
1 1.197 1.272 0.047 0.047 0.048 0.05
10 Narbonne 0.698 0.7
7 0.657 0.697 0.18 0.18 0.142 0.148
11 Lanester 1.273 1.2
3 1.136 1.267 0.529 0.636 0.614 0.624
12 Cubzac- les 1.438 1.3
7 1.333 1.433 0.388 0.383 0.388 0.386
13 Lodalen1
ports 1.012 1.0
5 0.926 1.010 0.399 0.399 0.371 0.398
14 Lodalen2 0.879 0.8
1 0.836 0.881 0.409 0.409 0.38 0.407
15 Lodalen3 1.153 1.1
8 1.115 1.156 0.496 0.496 0.463 0.493
16 Rio de janeiro 1.148 1.1
5 1.139 1.150 0.586 0.591 0.635 0.626
17 New Liskeard 1.662 1.2
7 1.668 1.670 0.005 0.01 0.005 0.007
18 Bangkok A 1.803 1.7
3 1.727 1.800 0.057 0.057 0.058 0.057
19 DrammenV 0.602 0.6
8 0.547 0.600 0.052 0.052 0.047 0.053
20 Drammen VI 0.746 0.7
0 0.694 0.750 0.1 0.1 0.085 0.094
21 DrammenVII 0.870 0.8
5 0.819 0.873 0.109 0.109 0.103 0.111
22 Pornic 1.133 1.1
7 1.040 1.128 0.43 0.465 0.486 0.447
23 Saint -Andre 1.346 1.2
2 1.207 1.330 0.145 0.154 0.146 0.148
24 South of France 1.574 1.5
4 1.394 1.569 0.412 0.407 0.373 0.412
25 NBR 1.525 1.4
3 1.487 1.560 0.195 0.194 0.191 0.196
26 Portsmouth
Development 0.843 0.8
9 0.805 0.839 0.2 0.198 0.177 0.198
27 Kameda 1.082 0.9
0 0.941 1.058 0.41 0.393 0.35 0.413
28 KhorAl - Zubair 1.387 1.1
9 1.319 1.396 0.421 0.417 0.409 0.448
29 Lian-Yun- no.4 Gang 0.987 0.8
3 0.958 0.987 0.608 0.459 0.613 0.605
30 Congress Street 1.457 1.4
9 1.349 1.456 0.624 0.622 0.621 0.624
31 Daikoku-Cho 1.042 0.9
5 0.890 1.029 0.365 0.358 0.3 0.361
32 Cuyahoga
Dike AA 0.900 0.9
4 0.803 0.901 0.391 0.391 0.353 0.392
33 King's Lynn 1.059 1.0
0 1.013 1.140 0.105 0.01 0.114 0.197
34 Muar 0.660 0.6
4 0.583 0.652 0.161 0.162 0.137 0.157
35 North Ridge 1.515 1.4
6 1.409 1.508 0.868 0.794 0.834 0.872
36 Seven Sisters
Dam 1.646 1.6
0 1.490 1.642 0.723 0.729 0.664 0.72
37 Shellmouth
Dike 1.125 1.0
3 1.051 1.113 0.323 0.442 0.272 0.302
38 Dam Juban
TestIFill 0.818 0.8
2 0.799 0.813 0.35 0.35 0.343 0.35
39 Bradwell 1.761 1.7
2 1.624 1.680 0.911 0.914 0.85 0.907
40 Genesee 1.001 1.0
8 0.935 1.001 0.557 0.557 0.532 0.555
41 Precambrian 1.022 0.9
0 1.041 1.046 0.081 0.085 0.097 0.089
42 scrapsgate 0.944 0.9
9 0.821 0.943 0.273 0.273 0.25 0.271
43 scottsdale 1.484 1.4
4 1.396 1.481 0.846 0.771 0.787 0.841
44 Iwai 1.732 1.7
2 1.732 1.757 0.682 0.68 0.498 0.755
45 Fair Haven 1.883 1.8
3 1.731 1.865 0.798 0.661 0.795 0.796
46 Boston Marine 1.259 1.2
8 1.147 1.254 0.251 0.251 0.251 0.251
47 Desert View
Excavation 1.134 1.0
6 1.063 1.132 0.83 0.798 0.788 0.826
48 Siburua
Drive 1.084 1.0
9 0.992 1.083 0.881 0.881 0.866 0.882
49 Tianshenqiao
October 5 0.944 0.9
8 0.912 0.943 0.711 0.648 0.64 0.712
50 San Francisco 1.201 1.2
1 1.067 1.181 0.147 0.147 0.15 0.147
51 Carsington
Bay 0.830 0.7
0 0.751 0.822 0.466 0.484 0.443 0.477
52 Atchafalaya 1.084 1.0
9 0.992 1.083 0.303 0.303 0.282 0.299
FS : Factor of Safety, LB: Lower-Bound
8

155
 2.4 Bishop
2.2 OMS
2.0 Janbu
1.8
(FS (measured) / FS(predicted))

Spencer
1.6
1.4
1.2
1.0
0.8
0.6
0.4
0.2
0.0
0 5 10 15 20 25 30 35 40 45 50 55

Case Number

Figure 4. 1 Values of the ratio of measured to predicted factor of safety for the 52 case
histories
The above results indicate that uncertainty in slope stability models for undrained slopes

is considerable and needs to be incorporated in any reliability-based design analysis that

aims at characterizing the risk of failure of undrained slopes. The model uncertainty as

reflected by the ratio of measured to predicted factor of safety from 52 documented case

histories is not sensitive to the slope stability method utilized (see Fig. 4.2). Predictions

from all methods were found to be relatively unbiased but are associated with a degree of

uncertainty that could be statistically reflected through a COV of about 0.27 in the ratio

of measured to predicted factor of safety.

Table 4. 2 Statistical Parameters of  for the Four LEM Slope Models

All Cases Bishop OMS Janbu Spencer

mean 0.966 0.990 1.036 0.964
Standard deviation 0.264 0.254 0.295 0.263
Coefficient of variation 0.273 0.256 0.285 0.273

156
 To investigate the sensitivity of the model uncertainty to the choice of the test

method used to measure the undrained shear strength of the soil, the undrained shear

strength values that were reported in the original case histories were corrected to make

them equivalent to the shear strength obtained from the Unconcolidated Undrained (UU)

triaxial test, which is considered as the most representative technique for measuring the

undrained strength. To this end, the undrained shear strength that was measured using

field vane test procedure was considered to be equivalent to that of the UU test, while the

strength that was measured using unconfined compression tests (UC) was multiplied by a

factor of 1.3 to make it equivalent to a UU strength as indicated by Olson and Dennis

(1982). The number of cases with UC tests are 13 out of the 52 cases (Tjernsmyr,

Skejeggerod, Shellmouth Dam test fill, Seven Sisters Dike, PresterØdbakken, North

Ridge Dam, Nesset, Juban, Iwai, Falkenstein, Cuyahoga, Congress Street, and As).

2.4 Bishop
2.2 OMS
2.0 Janbu
1.8
(FS (measured) / FS(predicted))

Spencer
1.6
1.4
1.2
1.0
0.8
0.6
0.4
0.2
0.0
0 5 10 15 20 25 30 35 40 45 50 55

Case Number

Figure 4. 2 Values of the model uncertainties for the 52 cases (Shear strength corrected
to UU)

157
Table 4. 3. Statistical Parameters of the model uncertainty (After shear strength
Correction)

All Cases (Bishop) (OMS) (Janbu) (Spencer)

mean 0.940 0.967 1.012 0.940
sd 0.262 0.252 0.295 0.262
COV 0.279 0.261 0.291 0.279

The ratio of the measured to the predicted factor of safety was reevaluated for these 13

cases with UC tests and plotted on Fig. 4.3. The updated statistics for  are presented in

Table 4.2 and indicate very small differences between the mean and COV of  with and

without the correction.

The 52 case histories that are assembled in the database are located in different countries.

Some of these slopes are found on soils with high sensitivity. Sensitivity is defined as the

ratio of the undrained shear strength of an undisturbed sample of soil to the undrained

shear strength of a remolded sample of the same soil tested at the same water content. A

more detailed analysis of the data is conducted to find if cases that have soil with high

sensitivity could have a certain effect on the bias and uncertainty of the models. 9 cases

out of the 52 cases have soil with high sensitivities (as high as 65). The statistical analysis

that was conducted on  was repeated without taking the cases with high sensitivities into

consideration. The calculated values of the mean and the COV of  are summarized in

Table 4.3. Results in Table 4.3. indicate that removing the sensitive cases doesn’t have a

significant effect on the model uncertainty of the four models. There is a small decrease

in the COV and the mean of . For instance, the COV of Bishop decreases from 0.279 to

0.273 and the mean decreases from 0.94 to 0.928. This decrease is also applicable for the

other three models (OMS, Janbu, and Spencer).

158
Table 4. 4 Statistical parameters of model uncertainty (after removing the sensitive
cases)

Non-Sensitive (Bishop) (OMS) (Janbu) (Spencer)

mean 0.928 0.957 1.003 0.930
sd 0.254 0.247 0.288 0.254
COV 0.273 0.258 0.287 0.273

In the balance of this thesis, the statistical parameters of  as presented in Table 4.2. are

adopted in any analysis related to reliability-based design of undrained slopes.

[Link] Probability Distribution of 

Due to the limited number of research studies that target the model uncertainty for

slope stability design problem, there is a lack of information on the distribution type

needed to model this uncertainty for slopes. To investigate the applicability of commonly

used probability distributions that could be used to model the uncertainty in , the

cumulative distribution function (CDF) of  was determined for the four models in

Figs.4.4 (a,b,c, and d) and tested against theoretical normal and lognormal CDFs that

could be used to model the data. Results in Fig.4.4 (a,b,c, and d) indicate that the

lognormal distribution could provide a realistic representation of the actual data more

than the normal distribution particularly at the left hand tail of the distribution. To

validate the hypothesis that the lognormal distribution is considered the best fit of the

data, the p-values associated with the Kolmogrov-Smirnov test for the lognormal

distribution was computed using R software. The p-values for the four models were

found to be greater than 0.05 indicating that there is no sufficient evidence to reject the

lognormal distribution hypothesis.

159
 1.0
Cumulative Distribution Function (CDF) Bishop OMS (b)
(.
0.8 (.
COV( = 0.279
COV( = 0.262

0.6

0.4 Normal
Distribution Normal
Distribution
Lognormal
Distribution Lognormal
0.2 Distribution
Actual Data
(N=52) Actual Data
(N=52)
0.0
0.0 0.5 1.0 1.5 2.0 0.0 0.5 1.0 1.5 2.0
(FS(measured)/FS(predicted)) (FS(measured)/FS(predicted))

1.0
Janbu
Cumulative Distribution Function (CDF)

( c) Spencer (d)
0.8 (.
(.
COV( = 0.291
COV( = 0.261
0.6

0.4 Normal
Normal
Distribution
Distribution
Lognormal Lognormal
0.2 Distribution Distribution
Actual Data Actual Data
(N=52) (N=52)
0.0
0.0 0.5 1.0 1.5 2.0 0.0 0.5 1.0 1.5 2.0

(FS(measured)/FS(predicted)) (FS(measured)/FS(predicted))

Figure 4. 3 Actual and Theoretical best-fit CDFs for the model uncertainty ()

160
4.3 Evidence of a lower-bound factor of safety of slopes

Results from 52 case histories of slope failures show significant scatter in the ratio

of measured to predicted factor of safety. Part of this scatter results from the uncertainties

in the values of the undrained shear strength. However, this uncertainty in the predicted

factor of safety can be reduced by introducing a physical lower-bound factor of safety.

The lower-bound factor of safety represents the minimum, possible factor of safety for

the undrained slope and can be calculated by assuming that the shear strength of soil

reduces to the fully remolded undrained shear strength. To validate the hypothesis of a

lower-bound factor of safety, an analysis is presented for the slope cases available in the

database. A predicted lower-bound factor of safety can be calculated using SLIDE by

replacing the undisturbed undrained shear strength with the remolded shear strength. The

remolded undrained shear strength represents the lowest possible strength for a clay.

Measurements of the undrained remolded shear strength are available for 6 out of

52 cases. These remolded shear strengths are typically measured using unconfined

compression tests or unconsolidated-undrained Triaxial tests on soil samples that have

been remolded at constant water content. For 46 out of 52 cases, information about the

sensitivity of the clay was used to calculate the remolded shear strength . Sensitivity is

defined as the ratio of the undisturbed strength to the remolded strength measured at the

same water content. Some cases have sensitivity given in the reference that discussed the

case while others have either liquidity indices as given or have index parameters that

could be used to estimate the liquidity indices. The well-known correlation between

liquidity index and sensitivity as presented by Bjerrum (1954) was used for this purpose.

This correlation is presented in Fig. 4.5.

161
 Figure 4. 4 Relation between sensitivity and liquidity
index

A predicted lower-bound factor of safety was calculated using Slide for the 52 cases in

the database using the 4 Limit Equilibrium methods. The lower-bound factor of safety

was calculated by replacing the undisturbed undrained shear strength with the remolded

undrained shear strength as mentioned above. The predicted lower-bound factor of safety

are presented in Table 4.1 and plotted on Fig. 4.6. The data on Fig. 4.6 support the

hypothesis of a lower-bound factor of safety because none of the data points fall above

the measured factor of safety (assumed to be equal to 1.0 for a failed slope) for the four

models. Each case of the database has a different calculated lower-bound factor of safety

since the calculated lower-bound depends on the properties of the soil and the geometry

of the slope. The predicted lower-bound factors of safety were found to range from

minimum values that are almost equal to zero (for highly sensitive quick clays) to

maximum values of about 0.9, with a mean value ranging from 0.37 to 0.39, depending

on the method used to predict the lower-bound factor of safety. A summary of measured,

predicted and lower-bound factors of safety for all cases in the database is shown in Fig.

4.7.

162
 2.4
Bishop
2.2 LB(BISHOP).
OMS
2.0  LB(OMS) = 0.372
Janbu
1.8  LB(Janbu) = 0.371
 LB(Spencer)= 0.393 Spencer
1.6

FS(lower Bound)

1.4
1.2
1.0
0.8
0.6
0.4
0.2
0.0
0 5 10 15 20 25 30 35 40 45 50 55

Case Number

Figure 4. 5 Evidence of a lower-bound factor of safety of 52 slope failure cases

A more detailed analysis is carried out to investigate the effect of the sensitive

cases on the calculated lower-bound factor of safety. The case histories that are

characterized with high sensitive soils are removed from the database (Skejggerod,

PresterØdbakken, Saint Alban, Narbonne, Bangkok A, NBR Development, Portsmouth,

King’s Lynn, New Liskeard and Precambrian). The elimination of the cases with soils

with very high sensitivities resulted in a significant increase in the mean of lower-bound

factor of safety for all models. The range of the mean of the lower-bound factor of safety

increased from 0.37 to 0.39 to a higher range of 0.45 to 0.47 (see Fig. 4.8). This

observation is important since the lower-bound factor of safety is expected to have a

more considerable effect on the design of a slope as the magnitude of the lower bound

increases.

163
 2.4 FS(Predicted) FS(Predicted)
2.2 Bishop OMS
FS(measured) FS(measured)
FS (Lower Bound),FS(Predicted,FS(measufred)

2.0 FS(Remolded) FS(Remolded)

1.8
1.6
1.4
1.2
1.0
0.8
0.6
0.4
0.2
0.0
0 5 10 15 20 25 30 35 40 45 50 55 0 5 10 15 20 25 30 35 40 45 50 55

Case Number Case Number

2.4
Janbu FS(Predicted) FS(Predicted)
2.2 Spencer
FS (Lower Bound),FS(Predicted,FS(measufred)

FS(measured) FS(measured)
2.0 FS(Remolded)
FS(Remolded)
1.8
1.6
1.4
1.2
1.0
0.8
0.6
0.4
0.2
0.0
0 5 10 15 20 25 30 35 40 45 50 55 0 5 10 15 20 25 30 35 40 45 50 55

Case Number Case Number

Figure 4. 6 Comparison of measured, predicted, and lower-bound factor of safety of 52 slope

failures

164
 2.0
Bishop
1.8  LB(BISHOP).
OMS
1.6
 LB(OMS) = 0.449 Janbu
 LB(Janbu) = 0.447 Spencer
1.4  LB(Spencer)= 0.471
FS (Lower Bound)

1.2
1.0
0.8
0.6
0.4
0.2
0.0
0 5 10 15 20 25 30 35 40 45 50 55
Case Number

Figure 4. 7 Predicted Lower-Bound Factors of Safety Excluding Cases with Highly

Sensitive Soils

4.4 Summary

Analysis of a database containing results from 52 undrained failed slopes

indicates that the factor of safety of slopes can be predicted using four models (Bishop,

OMS, Janbu, and Spencer) without introducing significant bias to the predicted factor of

safety. The coefficient of variation in the ratio of measured to predicted factor of safety

(model uncertainty ranges between 0.261 to 0.291. Results from 52 slope failures

provide evidence of the existence of the lower-bound factor of safety that can be

calculated using the undrained remolded shear strength of the soil and information about

the geometry of the slope.

165
 CHAPTER 5

Investigation of the impact of spatial variability in the undrained

shear strength on the factor of safety of undrained slopes

5.1 Introduction

Soil is a natural material. It exhibits considerable variation in space due to

depositional and post depositional processes and therefore it brings unavoidable

uncertainties in the estimation of input soil parameters used for defining the strength and

stiffness characteristic of the in situ soil deposit. Uncertainty in the input parameters

results in uncertainty in the output. In geotechnical engineering analysis and design,

various sources of uncertainties are encountered and well recognized. Several features

contribute to such uncertainties, like: geological details missed in the exploration

program and estimation of soil properties that are difficult to quantify.

Spatial variability of soils contributes to the total uncertainty in civil engineering designs.

Reliability Based Designs provide a consistent framework to quantify the uncertainties.

One of the primary steps in the reliability analyses of geotechnical systems is to

characterize the in-situ spatial variability of soil properties.

166
5.2 Characterization of soils

To model spatial variability of soils, generally two parameters are used as

measures of variability and correlation in soils, namely, the coefficient of variation and

the scale of fluctuation.

5.2.1 Coefficient of Variation (COV)

The coefficient of variation is one way of normalizing the variance and is a

widely used measure of variability. Due to the fact that this parameter is simple to

interpret, many soil statistical studies are based on it. The geotechnical literature has

considerable information on the estimates of COV for almost all soil properties. Both

Spry et al. (1988) and Phoon et al. (1999) reported typical estimates for the coefficient of

variation for various geotechnical properties.

5.2.2 Scale of Fluctuation

While COV is used as a parameter to describe how variable a process is, the scale

of fluctuation is used to describe the spatial correlation in a random process. Spatial

variability can be effectively described by the correlation structure (Vanmarcke 1983). To

describe this correlation structure, an autocorrelation distance is defined which is the

distance within which soil properties show a strong correlation. A large autocorrelation

distance value implies that the soil property is highly correlated over a large spatial

167
extent, resulting in a smooth variation within the soil profile. On the other hand, a small

value indicates that the fluctuation of the soil property is large (Cho 2010).

As illustrated in Chapter 2, many studies adopted the coefficient of variation in

characterizing spatial variability. The coefficient of variation (COV) does not reflect the

spatial correlation of soils which is an indispensible descriptor in the geotechnical site

characterization. The complexity of the problem reduces to a very simplistic level if the

spatial correlation isn’t taken into consideration. However, this simplicity is not realistic.

In order to conduct a probabilistic geotechnical analysis in a rational framework, the

method has to take the effect of soil correlation into account. According to the literature,

Random Finite Element Method (RFEM) is able to combine local averaging theory, soil

spatial correlation, in addition to the COV that describes how variable the soil is.

Numerous reliability analyses in the literature adopted the RFEM for incorporating

spatial variability in the analysis (Chapter 2).

In a recent study, Jha and Ching (2013) performed a robust and rigorous probabilistic

slope stability analysis using the Random Finite Element Method to study the effect of

spatial variability of soil properties on the probability of failure of undrained slopes. The

following section provides a brief summary of the work done by Jha and Ching (2013)

that is adopted later in the analysis.

168
5.3 Brief Summary of Work done by Jha and Ching (2013)

Jha and Ching (2013) performed a robust and rigorous probabilistic slope stability

analysis using the Random Finite Element Method (RFEM) to study the effect of slope

geometry, mean and coefficient of variation of the soil parameters, and the scale of

fluctuations on the probability of failure of undrained slopes. The authors conducted the

study by collecting a database for 34 real undrained engineered slope cases. The paper

was aimed at quantifying the effect of spatial variability in the undrained shear strength

of clays on the probability of failure of the slopes. An advanced model of spatial

variability was adopted. This model took into account both vertical and horizontal spatial

variability of the undrained shear strength in addition to the COV of the undrained shear

strength. The vertical scale of fluctuation (dz) in the undrained shear strength was back-

calculated for each case in the database using an approximate method proposed by

Vanmarcke (1977) as follows:

dz  0.8 d ( d = (d1 + d2 + d3 + d4 + d5 )/5) Equation 5-1

where d is the average vertical interval of the intersection points between the Su profile

and its trend (t). Fig.5.1. shows a typical example of estimating dz. There are cases where

the detailed Su borehole data are not given but the trends are known. For these cases the

authors assumed dz =2.5m.

169
 Figure 5. 1 Determination of vertical scale of fluctuation(Jha and Ching 2013)

The authors studied the effect of Su spatial variability on the statistics of the factor of

safety by quantifying the effect of the coefficient of variation (V) by varying the values

of V to be 0.1, 0.3, and 0.5. Moreover, they investigated the effect of the horizontal scale

of fluctuation (dx) by taking different values of (dx) to be 10dz, 20dz, and 30dz

according to Phoon and Kulhawy (1999). In addition to that, they studied the effect of

both the vertical scale of fluctuation (dz) and the geometry of the slope by quantifying the

ratio of (dz /Lf ) where Lf is the length of the failure surface. By analyzing the 34 cases,

the authors estimated the mean of the factor of safety (FS) and the coefficient of variation

of the factor of safety (VFS). They concluded that FS is always less than the deterministic

factor of safety (FSd) and VFS is always less than COV of Su. In the aim of understanding

170
the reasons behind this reduction, the authors studied the correlation between this

reduction and V, dx, dz, and Lf. They investigated the change of the ratio (FSd - FS)/ FSd

versus the dimensionless factors (V, dx/dz, and dz /Lf). Their analysis showed that (FSd -

FS)/ FSd is strongly correlated to V only and this is shown in Fig.5.2.

It is shown in Fig.5.2. that there is deviation of the ratio of (FSd - FS)/ FSd from the trend

line and this variability increases as V increases. The authors took into account this

variability by denoting an error term ε1 with a standard deviation of s1. Finally, they

ended up with an equation for the estimation of FS as follows:

FS = (1- 0.115( ) - s1* Z1) * FSd Equation 5-2

Where: V is the coefficient of variation of Su

s1 = 0.06 x V0.85

171
 Z1 is modeled as the standard normal random variable N(0,1). As a result:

FS = (1- 0.115( ) - 0.06* V0.85* Z1) * FSd Equation 5-3

The same analysis illustrated above was repeated by the authors to dictate the reason

behind the reduction of the coefficient of variation of the factor of safety (V FS ). The

analysis resulted in the conclusion that VFS is strongly correlated with both V and dz /Lf

and this is shown in Fig.5.3. The authors conducted a regression analysis to get an

equation for the estimation of the coefficient of variation of the factor of safety (VFS):

VFS = (0.2606 * bdz/ Lf * bdx/ dz * bv + 0.0466*Z2) *V Equation 5- 4

Where bdz/ Lf = exp[3.1226+1.5027* ln(dz /Lf ) +0.1655*ln(dz /Lf)2]

bdx/ dz = exp[-0.4999+0.1668*ln(dx/ dz )]
bv = exp[-0.6349- 0.731*ln(V) - 0.1691*ln(V)2]
Z2 is modelled as the standard normal variable N(0,1).

The value 0.0466 that is placed in the equation is related to the standard deviation of the

error between the estimated and the actual ratio (VFS/V) as shown in Fig.5.4.

172
Figure 5. 3 Relationship between VFS/V and (V, z /Lf , x/ z)(Jha and Ching 2013)

173
 Estimated VFS/V based on Equation 2

Actual VFS/V
Figure 5. 4 Relationship between actual VFS/V and VFS/V estimated by Equation 2

5.4 Method for Combining Uncertainties

The work done by Jha and Ching (2013) is implemented to evaluate the

expected value of FS, E(FS) and the variance of FS, Var(FS).

Both the expected value and the variance of the factor of safety could be evaluated using

a first order approximation as follows (Gilbert 1999):

174
 E (FS   EZ FS Z (  (5-5)

Var (FS   EZ  FS Z (   Var (  Var (E 

2
Z FS Z Z FS Z
(5-6)

Where:

  FS Z and  FS Z are the mean value and standard deviation of FS (both are

random variables since they are function of the uncertain model

parameters Z (see Equation 5-3 and 5-4).

 E (FS  and Var (FS  are the expected mean and variance of FS as obtained

from the first order approximation. It should be noted that

It is worthwhile, mentioning that Equation (5-6) indicates that the uncertainty in

FS arises from two sources:

1. Random Variability that is modeled by the model parameters, E Z ( FS Z 2

2. Uncertainty in the model parameters themselves, VarZ ( FS Z  and VarZ (E FS Z 

The first and second moments for  FS Z and  FS Z themselves can be

approximated as functions of the first and second moments of the model parameters Z ,

using first order Taylor series expansion such that:

( 
E Z  FS Z  h ( Z 

(5-7)

T
h  h 
(
VarZ E FS Z   


 CZ   (5-8)
  Z i Z 
   Z Z 

175
 (
E Z  FS Z
= h  ( Z  (5-9)

T
h  h 
(
VarZ  FS Z  
 
  
 CZ 

 (5-10)
  Z i Z 
   Z i Z 


Where:

h   
  



and h 

are vectors containing the partial derivatives of
 
 Zi Z 
 Z
 i Z 


h (Z  and h (Z  , respectively, evaluated at the mean values of the model

parameters

h (Z  and h (Z  are the expressions of the probabilistic model of the  FS and  FS ,

respectively.

Equations 5-5 and 5-6 allow for estimating the mean and variance of FS given

information about the spatial variability of the undrained shear strength as reflected by

the coefficient of variation of the undrained shear strength (V) and the vertical and

horizontal correlation distances (dz and dx). Other important input to these equations are

the predicted factor of safety (FSd) and the length of the failure surface (Lf).

176
 CHAPTER 6

Combination of Both Model Uncertainty and Spatial Variability

6.1 Introduction

In this chapter, a mathematical framework is provided to study the reliability and

evaluate the statistical parameters (mean and variance) of the factor of safety of

undrained slopes. This framework models the factor of safety that combines both model

uncertainty and spatial variability. Moreover, a practical approach for incorporating a

lower-bound factor of safety in the reliability analysis is investigated and a probability

distribution that can accommodate a lower-bound factor of safety is recommended.

6.2 Conventional Probability Distributions for Factor of Safety

Traditionally, normal and lognormal probability distributions have been used to

model the uncertainty in the factor of safety. Parameters and mathematical forms of the

normal and lognormal distributions in addition to the advantages and disadvantages of

both distributions are described in the following sections.

6.2.1 Normal Distribution

The normal distribution is the most widely known and used distribution in

engineering due to the simplicity of its mathematical form and to the physical
177
significance of the parameters describing it. The normal distribution which is also

referred to as the Gaussian distribution has a probability density function (PDF) that is

defined over a range of values that extend from - to +. The normal distribution is

symmetrical in shape and is defined by two parameters, the mean and the standard

deviation .Because of symmetry, the mean of the normal distribution is equal to the

median value (50th percentile value). The PDF of the normal distribution is shown in the

following equation:

1 x 2
1  
f X (x)  e 2  
x 
 2 Equation 6-1

The probability density function (PDF) of examples of normal distributions are shown in

Fig. 6.1.

Figure 6. 1 Normal Probability Distribution

When the normal distribution is used to model uncertainty in the capacity or factor of

safety of engineering systems, the distribution has a shortcoming in that the left-hand tail

178
of the distribution can extend to values that are less than zero (see Figure 6.1.). Negative

values of capacity and factor of safety are not physically possible in engineering design.

6.2.2 Lognormal Distribution

The lognormal distribution has a probability density function (PDF) that is

defined over a range of values that extends from zero to +. The probability density

function (PDF) of the lognormal distribution is given by the mathematical expression

shown below:

2
1 ln (x)  

1 2   
f X (x)  e 0x
x 2 Equation 6-2

The lognormal distribution is skewed to the right and is defined by two parameters,  and

.These two parameters represent the mean and the standard deviation of the natural

logarithm of the variable.  and  can be evaluated using the following equations:

2 Equation 6-3
  E ln ( X    ln (  X  
2

(
 2  Var ln ( X    ln 1  2 

Note: xmedian e
is approximately equal to  for  < 0.3 Equation 6-4

Where is the coefficient of variation of the random variable x that is defined as the ratio

of the standard deviation to the mean of random variable. The probability density

function of examples of lognormal distributions are shown in Fig.6.2.

179
 Figure 6. 2 Lognormal Probability distribution

The lognormal distribution has been used widely to model the uncertainty in the

load and capacity in conventional reliability analyses in civil engineering in general and

in geotechnical engineering in particular (Tang 1988 and 1990; Hamilton and Murff

1992; Tang and Gilbert 1993; API 1993; Hornsell and Toolan 1996; Bea et al. 1999;

McVay 2000; 2002; and 2003; Kulhawy and Phoon 2002; Phoon et al. 2003; AASHTO

2004).

The main reasons for the wide-spread use of the lognormal distribution are related

to the fact that it is skewed to the right and has a lower bound of zero. However, the

lognormal distribution, with a lower tail that extends to zero, does not capture the realistic

possibility that ther is a physical minimum or lower bound for the capacity or factor of

safety of geotechnical engineering systems. This lower-bound factor of safety could be

greater than zero and is not modeled properly by conventional distributions.

180
6.3 Distribution Types Adopted in the Analysis

In this analysis, the lognormal distribution is assumed to model the uncertainty in

the factor of safety. To incorporate the lower-bound factor of safety into reliability

assessments, a simple approach is adopted through the use of a truncated lognormal

probability distribution (Fjeld 1977, Rodriguez et al. 1988). A Lognormal distribution

that is truncated at a lower-bound factor of safety (FSLB) can be used to accomplish this

purpose. The use of truncated lognormal distribution is convenient because the

parameters describing the distribution are the same as those of the non-truncated

distribution with the addition of one extra parameter, the lower-bound factor of safety

(FSLB). However, the mean and the coefficient of variation of truncated lognormal

distribution can be quite different than the mean and the coefficient of variation of non-

truncated distribution, especially as the lower-bound factor of safety increases and

becomes close to the mean or median factor of safety. The probability density function of

an example truncated lognormal distribution is illustrated on Fig.6.3.

Truncated Lognormal

COV FS

FS(LB) FS

Figure 6. 3 Probability density function of truncated lognormal

181
6.4 Estimation of the Statistical Parameters of the Factor of Safety

The statistical parameters of the factor of safety are estimated by combining the

uncertainties in the factor of safety (FS) due to model uncertainty and due to spatial

variability. To accomplish this objective, the factor of safety is assumed to be equal to the

product of two random variables as indicated in Equation 6-5.

FS = model.FSspatial Equation 6-5

The first random variable (FSspatial) models the effect of spatial variability in the

undrained shear strength on FS. The mean and the coefficient of variation of FSspatial are

estimated as indicated in chapter 5 using Equations 5-5 and 5-6. The second random

variable (modelrepresents the model uncertainty as reflected in the ratio of the

measured to predicted factor of safety of the slope. The mean and coefficient of variation

of (model are evaluated from the analysis of the database which includes the real case

histories of failed slopes as illustrated in chapter 4.

The distributions of the two random variables are assumed lognormal. Thus, exact

solutions that allow for combining the uncertainties in both parameters to calculate the

parameters  and of the total factor of safety are available and result in a total factor of

safety that is also lognormally distributed. The mathematical expressions shown in Table

6.1 can be used for this purpose.

182
Table 6. 1. Estimation of and of the factor of safety

Probability Distribution Probability Distribution

Function, for Xi for Y

Y  g(X)

n
Y  b  Xi
ai Lognormal(Xi,Xi) Lognormal(Y,Y)
i 1
n
 Y   a i  Xi  ln(b)
i1
COV[ln(Xi ),ln(Xj)]  i,j XiXj n n
Y    a i a jCOV[ln(Xi ), ln(X j)]
i 1 j1

6.5 Summary

Most reliability analyses focus on the mean, variance, and an assumed

mathematically convenient distribution to model the left-hand tail of the factor of safety

distribution. In this analysis, the lognormal distribution is assumed to model the

uncertainty in the factor of safety and this is due to the fact that it is skewed to the right

and has a lower-bound of zero (does not allow negative values). For analysis in which the

effect of the lower-bound factor of safety is included in the analyses, a truncated

lognormal distribution is used instead of the conventional lognormal distribution.

183
 CHAPTER 7

Recommendations for design factors of safety for undrained slopes

7.1 Introduction

In this chapter, a reliability-based design framework is proposed to recommend

design factors of safety that would result in target probabilities of failure for undrained

slopes. As a first step, uncertainties due to spatial variability and model uncertainty are

combined to evaluate the probability of failure of undrained slopes that are designed with

different factors of safety. In the second step, the effect of incorporating a lower-bound

factor of safety on the probability of failure is investigated. The third and final step

consists of recommending factors of safety to be used for different design scenarios to

achieve target levels of risk.

7.2 Reliability-based design of undrained slopes

The sensitivity of the probability of failure (Pf) to variations in the deterministic

factor of safety, coefficient of variation of the undrained shear strength, vertical

correlation distance, and lower-bound factor of safety is investigated in this section. A

number of spatially variable clay slopes that cover the typical range of design slope

conditions are considered. For each analyzed slope, a reliability analysis is conducted to

quantify the probability of failure (Pf) of the slope with and without the inclusion of the

estimated lower-bound factor of safety. For the two cases, the probability of failure (Pf) is

defined as the probability that the factor of safety is less than one as is the convention.

184
Equation 7-1 is used to calculate the probability of failure (Pf) and the reliability index,

without the inclusion of the lower bound factor of safety:

( )
= Φ( )= Φ(- Equation 7-1

Where Φ() is the standard normal cumulative distribution function, andare statistical

parameters (lognormal distribution) that are related to the mean and coefficient of

variation of the factor of safety, and is the reliability index.

In the presence of the lower-bound factor of safety, the lognormal distribution,

with a left tail that extends to zero, doesn’t capture the realistic possibility that there is a

physical minimum or lower-bound for the factor of safety of the slope. This lower-bound

is expected to be greater than zero. Therefore, a truncated lognormal distribution is used

to model the uncertainty in the factor of safety of the slope in the presence of the lower-

bound factor of safety. The mathematical expression of the probability of failure (Pf)

including the lower-bound is presented in Equation 7-2.

( ) ( )
( ) ( )
( ( ) ) Equation 7-2
( )

Where LB is the lower-bound factor of safety of the slope.

7.2.1 Effect of coefficient of variation and scale of fluctuation of undrained shear

strength on the probability of failure

As illustrated in chapter 5, both the ratio of dz /Lf and the coefficient of variation

(V) of the undrained shear strength have a significant effect on the statistical parameters

185
of the factor of safety. To study the effect of dz /Lf and the coefficient of variation of the

undrained shear strength (V) on the probability of failure (Pf) of slopes, a number of

spatially variable slopes with different slope geometries and different soil properties were

considered. The coefficient of variation (V) was varied between 0.1, 0.2, 0.3, 0.4, and 0.5

and dz /Lf was taken as 0.001, 0.005, 0.01, 0.05, 0.1, 0.15, 0.2, and 0.3. Due to the fact

that dx doesn’t have a significant effect on the probability of failure (Pf) of slopes, the

ratio of dx/dz is taken to be 20 (Phoon and Kulhawy 1999). The probability of failure of

each case is calculated using Equation 7-1 where the statistical parameters andare

estimated using the procedure shown in chapter 6 that takes into account both model

uncertainty and spatial variability. In this analysis, the statistical parameters of the model

uncertainty are those evaluated for Spencer’s method knowing that there is no large

difference in the statistical parameters between the four methods.

Figures 7.1.,7.2., and 7.3. show typical results for the variation of probability of

failure (Pf) with the ratio of dz /Lf for different deterministic design factors of safety and

different coefficients of variation. The deterministic factors of safety are estimated based

on the mean of the undrained shear strength that is measured using Undrained

Unconsolidated tests (UU) as an input. As a result, these design factors of safety could be

different from actual design factors of safety that are conventionally used in slopes,

whereby conservative estimates (rather than the mean) and other test methods (such as

the unconfined comporession tests) of the undrained shear strength are generally adopted.

For such studies, the design factors of safety are expected to be lower than the values

adopted in Figs. 7.1 to 7.3.

186
 z/Lf
0.00 0.05 0.10 0.15 0.20 0.25 0.30
1.00
Probability of Failure

0.10
V=0.1
V=0.2
V=0.3
V=0.4 FS=1.5
V=0.5

0.01

z/Lf
0.00 0.05 0.10 0.15 0.20 0.25 0.30
1.00
Probability of Failure

0.10

V=0.1
V=0.2 FS=1.7
V=0.3
V=0.4
V=0.5
0.01

Figure 7. 1Variation of Pf with V and z /Lf for FS =1.5 & FS = 1.7

187
 z/Lf
0.00 0.05 0.10 0.15 0.20 0.25 0.30
1.00

V=0.1
V=0.2 FS=2
Probability of Failure

V=0.3
V=0.4
V=0.5

0.10

0.01

z/Lf
0.00 0.05 0.10 0.15 0.20 0.25 0.30
1.000
V=0.1
V=0.2
FS=2.25
V=0.3
Probability of Failure

V=0.4
0.100 V=0.5

0.010

0.001

Figure 7. 2 Variation of Pf with V and z /Lf for FS =2 & FS = 2.25

188
 z/Lf
0.00 0.05 0.10 0.15 0.20 0.25 0.30
1.000
V=0.1
V=0.2
Probability of Failure

FS=2.5
V=0.3
V=0.4
0.100 V=0.5

0.010

0.001

z/Lf
0.00 0.05 0.10 0.15 0.20 0.25 0.30
1.0000
V=0.1
V=0.2
V=0.3 FS=2.75
Probability of Failure

0.1000 V=0.4
V=0.5

0.0100

0.0010

0.0001
Figure 7. 3 Variation of Pf with V and z /Lf for FS =2.5 & FS = 2.75

189
The results show that the probability of failure (Pf) depends on the coefficient of

variation, the factor of safety, and the scale of fluctuation. The primary conclusion from

Figures 7.1.,7.2., and 7.3. is that the spatial variability has a significant effect on the

probability of failure of slopes. The probability of failure increases significantly with the

increase in the coefficient of variation. As an example, consider the case where the factor

of safety is equal to 1.5 and dz /Lf = 0.15. For this case, Pf is found to increase from 15%

for the case where the undrained shear strength is the least variable (V=0.1) to 48% for

the case where the undrained strength is highly variable (V=0.5). This increase in Pf

diminishes with the increase in the factor of safety due to the fact that the magnitude of Pf

decreases at any value of the ratio dz /Lf with the increase in the factor of safety. Along

the same lines, it is worth noting that the calculated probabilities of failure for the case

with FS = 1.5 (which is a common design case) are relatively high (range from 15% to

50% depending on the coefficient of variation of the undrained shear strength) compared

to typical probabilities of failure that are considered as acceptable in engineering practice.

The second conclusion is that there is a significant decrease in (Pf) with the

increase of FS. For example, consider the case where the ratio z /Lf = 0.15 and V = 0.5.

For this case, the probability of failure (Pf) decreases from 48% for FS = 1.5 to 2% for FS

= 2.75. For the case with the lowest spatial variability (V = 0.1) and z /Lf = 0.15, the

probability of failure (Pf) decreases from 14% for FS = 1.5 to 0.07% for FS = 2.75.

The third conclusion from Figures 7.1.,7.2., and 7.3. is that there is a threshold

value for the ratio of dz /Lf (dz /Lf = 0.1), above which the probability of failure (Pf)

slightly increases until it reaches another threshold (dz /Lf =0.2) beyond which the

190
probability of failure (Pf) remains constant with the increase of the ratio dz /Lf. This

increase in the probability of failure between dz /Lf = 0.1 and dz /Lf = 0.2 is related to the

variance reduction in the undrained shear strength due to averaging along the failure

surface. As dz /Lf decreases (either due to small scale of fluctuation or long failure

surface), there is more averaging in the undrained shear strength that occurs along the

failure surface leading to variance reduction which ultimately translated into a reduction

in the probability of failure. The effect of this averaging seems to be minor at lower

factors of safety but increases slightly when the factor of safety increases. The reason

why this effect is minor is related to the fact that model uncertainty masks the uncertainty

due to spatial variability when the two sources of uncertainty are combined.

To portray the effect of dz /Lf on the calculated probability of failure, an analysis

is conducted without including the model uncertainty in the analysis. The results of the

analysis are presented in Fig.7.4 which shows the variation of the probability of failure

(Pf) with dz /Lf for a design factor of safety of 1.5 and for different coefficients of

variation of the undrained shear strength. Results on Fig.7.4 indicate that the effect of

variance reduction in the undrained shear strength becomes visible at ratios of z /Lf that

around 0.2. This effect is shown as a decrease in the probability of failure. The maximum

benefit from variance reduction is achieved at ratios of z /Lf that are around 0.1, since

no considerable further reduction in the variance occurs below that level. This significant

effect of variance reduction on the probability of failure doesn’t appear clearly in Figures

7.1, 7.2, and 7.3 since the model uncertainty dominates the probability of failure in the

reliability analysis.

191
 0.35
V=0.1
0.30 V=0.2 FS=1.5
V=0.3
Probability of Failure 0.25 V=0.4
V=0.5
0.20

0.15

0.10

0.05

0.00
0.00 0.05 0.10 0.15 0.20 0.25 0.30

z/Lf

Figure 7. 4Variation of Pf with V and z /Lf for FS =1.5 by considering spatial variability
only

Fig.7.4. shows the significant effect of spatial variability on the probability of

failure of slopes. For instance, the probability of failure of a slope that has a ratio of z /Lf

= 0.2 increases from about 0 when V=0.1 to a high Pf of about 27% when V= 0.5. To

explain this result, one should look at the increase in the coefficient of variation of the

factor of safety (VFS) that enters in the calculation of the probability of failure. VFS is

calculated using Equations 5-1& 5-2 in chapter 5. It is found that VFS increases from

0.061 for V=0.1 to 0.174 for V=0.5. This increase in VFS doesn’t explain the large

increase in the probability of failure from 0 to 27%. There is another factor that leads to

the increase in the probability of failure which is the mean of the factor of safety (FS)

which is found to decrease from 1.467 for V =0.1 to 1.129 for V=0.5. Hence, spatial

variability has two effects on the probability of failure; (1) increasing the total uncertainty

of the factor of safety, (2) decreasing the mean of the factor of safety.

192
7.2.2. Effect of lower-bound factor of safety on the probability of failure

To illustrate the effect of the lower-bound factor of safety on the probability of

failure of slopes, a number of homogenous slopes with spatially variable undrained shear

strength are considered. The probability of failure is evaluated using Equation 7-2 that

takes into consideration the lower-bound factor of safety. This lower-bound factor of

safety is calculated by replacing the undisturbed undrained shear strength with the

remolded shear strength that is evaluated using the sensitivity equation. The sensitivity of

clays is defined as the ratio of the undisturbed undrained shear strength to the remolded

undrained shear strength of the clay. The analysis is conducted for different conditions

for the lower-bound factor of safety (clays with different sensitivities). Sensitivities of

1.5, 1.75, 2, 2.25, 2.5, and 3 are considered in the analysis for the calculation of the

lower-bound factor of safety, and for showing the effect of these lower-bound values on

the probability of failure of slopes.

Curves showing the variation of the probability of failure as a function of the sensitivity

of clays and the coefficient of variation of the undrained shear strength for slopes with

ratio of z /Lf = 0.1 are shown in Figures 7.5, 7.6,and 7.7. The results are illustrated

according to different values of deterministic factors of safety. The curves on Figures 7.5,

7.6,and 7.7. represent the cases were the uncertainty in the undrained shear strength takes

different values (V =0.1, 0.2, 0.3, 0.4, and 0.5).

193
 FS lower-bound / FS predicted
0.67 0.57 0.5 0.45 0.4 0.36 0.33
0.5
V=0.1
V=0.2
0.4
Probability of Failure
V=0.3
V=0.4
V=0.5
0.3

0.2

0.1
FS=1.5
0.0
1.50 1.75 2.00 2.25 2.50 2.75 3.00

Sensitivity of Homogenous Slopes

FS lower-bound / FS predicted
0.57 0.5 0.45 0.4 0.36 0.34
0.5
V=0.1
V=0.2
FS=1.7
0.4 V=0.3
Probability of Failure

V=0.4
V=0.5
0.3

0.2

0.1

0.0
1.75 2.00 2.25 2.50 2.75 3.00

Sensitivity of Homogenous Slopes

Figure 7. 5 Variation of Pf with V and Sensitivity for z /Lf = 0.1, FS =1.5& FS=1.7

194
 FS lower-bound / FS predicted
0.5 0.45 0.40 0.36 0.33 0.30 0.28 0.26 0.25
0.2

V=0.1 FS = 2.00
V=0.2
Probability of Failure

0.2 V=0.3
V=0.4
V=0.5
0.1

0.1

0.0
2.00 2.25 2.50 2.75 3.00 3.25 3.50 3.75 4.00
Sensitivity of Homogenous Slopes

FS lower-bound /FS predicted

0.44 0.40 0.36 0.33 0.30 0.28 0.26 0.25 0.24 0.22
0.10
V=0.1
FS = 2.25
V=0.2
0.08 V=0.3
Probability of Failure

V=0.4
V=0.5
0.06

0.04

0.02

0.00
2.25 2.50 2.75 3.00 3.25 3.50 3.75 4.00 4.25 4.50

Sensitivity of Homogenous Slopes

Figure 7. 6 Variation of Pf with V and Sensitivity for z /Lf = 0.1, FS =2.00& FS=2.25

195
 FS lower-bound / FS predicted
0.40 0.36 0.33 0.30 0.28 0.26 0.25 0.23 0.22
0.05
V=0.1
FS = 2.50
V=0.2
0.04
Probability of Failure
V=0.3
V=0.4
0.03 V=0.5

0.02

0.01

0.00
2.50 2.75 3.00 3.25 3.50 3.75 4.00 4.25 4.50
Sensitivity of Homogenous Slopes

FS lower-bound / FS predicted
0.36 0.33 0.30 0.28 0.27 0.2 0.24 0.22 0.21 0.20
0.015
V=0.1
V=0.2
FS = 2.75
V=0.3
V=0.4
V=0.5
0.010
Pf

0.005

0.000
2.75 3.00 3.25 3.50 3.75 4.00 4.25 4.50 4.75 5.00

Sensitivity of Homogenous
Figure 7. 7 Variation of Pf with V and Sensitivity for z /Lf = 0.1, FS =2.50 & FS=2.75

196
The results shown on Figs. 7.4 to 7.7 correspond to a ratio of z /Lf = 0.1 which is a ratio

that is typical for undrained slopes. The primary conclusion from Figures 7.5, 7.6, and

7.7 is that a lower-bound factor of safety can have a significant effect on the calculated

probability of failure. For example, consider a typical case where the factor of safety of

the slope is 1.5. If the sensitivity of the soil is 1.75 (ratio of lower-bound to predicted

factor of safety of about 0.57), the probability of failure decreases to half of its magnitude

compared to the case where there is no lower-bound factor of safety (Fig.7.5).

The second conclusion from Figures 7.5, 7.6, and 7.7 is that there is a threshold value for

the sensitivity (sensitivity of about 3), below which the lower-bound factor of safety

affects the probability of failure. Above this threshold, the lower-bound factor of safety

has essentially no effect on the probability of failure.

The effect of the lower-bound factor of safety on the probability of failure is influenced

by the magnitude of the deterministic factor of safety and by the value of the coefficient

of variation; as the deterministic factor of safety increases, the lower-bound becomes

more effective in reducing the probability of failure. Also, as the uncertainty in the

coefficient of variation (V) increases, the probability of failure becomes more sensitive to

the lower-bound factor of safety where the importance of the lower-bound factor of

safety increases more. This is related to the fact that as the coefficient of variation

increases, the mean factor of safety decreases. Thus, the magnitude of the ratio of the

lower-bound factor of safety to the mean factor of safety becomes larger making the

probability of failure more sensitive to the lower-bound factor of safety. It should be

noted that the same conclusions are applicable for a ratio of z /Lf = 0.2.

197
7.2.3 Recommendation for the factors of safety of undrained slopes

In this section, recommendations for the factor of safety of undrained slopes are

presented to achieve target levels of acceptable risk. This analysis is conducted for

different levels of the spatial variability of the undrained shear strength, and different

conditions for the lower-bound factor of safety (clays with different sensitivities).

Relationships between the factor of safety and the probability of failure were

established for cases with different lower-bound factors of safety (as indicated by the

sensitivity) and different coefficients of variation of the undrained shear strength. Plots

showing these relationships for coefficients of variation of 0.1, 0.2, 0.3, 0.4, and 0.5 are

shown in Figures 7.8., 7.9., and 7.10 for the case with z/Lf = 0.1 while similar

relationships are shown in Figures 7.11., 7.12., and 7.13 for the case with z/Lf = 0.2.

The relationships between the factor of safety and the probability of failure illustrate the

effect of the lower-bound factor of safety on the probability of failure by comparing the

probability of failure of the slope without the inclusion of the lower-bound with that with

the inclusion of the lower-bound factor of safety.

For the case with z/Lf = 0.1, the relationships shown in Figs. 7.8., 7.9., and 7.10 indicate

that as the factor of safety increases, the probability of failure decreases as expected. For

relatively small coefficients of variation of the undrained shear strength (V =0.1 & 0.2),

when the lower-bound is incorporated in the analysis, the relationship between FS and pf

seems to be unaffected by the lower-bound for cases with sensitivities ranging from 2 to

3. For sensitivities smaller than 2, the lower-bound factor of safety starts to play a role in

decreasing the probability of failure for a given factor of safety. The importance of the

198
 lower-bound becomes more significant for cases involving higher coefficients of

variation of the undrained shear strengths (V = 0.3, 0.4, and 0.5), where considerable

effects of lower-bound on the probability of failure are noticed from sensitivities as high

2.5 and 2.75.

0.200

V=0.1 NO LB
S=3
Probability of Failure

S=2.75
S=2.5
S=2.25
S=2
0.100 S=1.75

0.000
1.50 1.75 2.00 2.25 2.50 2.75 3.00
FS
0.200
NO LB
V=0.2 S=3
S=2.75
Probability of Failure

S=2.5
S=2.25
S=2
S=1.75
0.100 S=1.5

0.000
1.50 1.75 2.00 2.25 2.50 2.75 3.00

FS

Figure 7. 8 Recommended factor of safety needed to accomplish a target Pf for different

sensitivities for V =0.1&V=0.2 and z /Lf = 0.1

199
 0.300
NO LB
V=0.3 S=3
S=2.75
Probability of Failure

S=2.5
S=2.25
0.200 S=2
S=1.75
S=1.5

0.100

0.000
1.50 1.75 2.00 2.25 2.50 2.75 3.00

FS
0.400
NO LB
V=0.4
S=3
S=2.75
Probability of Failure

0.300 S=2.5
S=2.25
S=2
S=1.75
S=1.5
0.200

0.100

0.000
1.50 1.75 2.00 2.25 2.50 2.75 3.00
FS

Figure 7. 9 Recommended factor of safety needed to accomplish a target Pf for different

sensitivities for V =0.3&V=0.4 and z /Lf = 0.1

200
 0.500

NO LB
V=0.5 S=3
0.400 S=2.75
Probability of Failure

S=2.5
S=2.25
S=2
0.300 S=1.75
S=1.5

0.200

0.100

0.000
1.50 1.75 2.00 2.25 2.50 2.75 3.00

FS

Figure 7. 10 Recommended factor of safety needed to accomplish a target Pf for different

sensitivities for V=0.5 and z /Lf = 0.1

Similar results are obtained for the cases with a ratio z /Lf = 0.2 as indicated in Figures

7.11., 7.12., and 7.13., except that the calculated probabilities of failure for a given

coefficient of variation, a given factor of safety, and a given lower-bound factor of safety

(sensitivity) are slightly higher than the probabilities of failure of slopes with ratio z /Lf

= 0.1. This is expected given that the effect of variance reduction diminishes for the

case of z /Lf = 0.2.

201
 0.200

V=0.1 NO LB
S=3
Probability of Failure
0.150
S=2.75
S=2.5
S=2.25
S=2
0.100 S=1.75
S=1.5

0.050

0.000
1.50 1.75 2.00 2.25 2.50 2.75 3.00
FS

0.300
NO LB
V=0.2 S=3
0.250 S=2.75
Probability of Failure

S=2.5
S=2.25
0.200 S=2
S=1.75
0.150

0.100

0.050

0.000
1.50 1.75 2.00 2.25 2.50 2.75 3.00
FS

Figure 7. 11Recommended factor of safety needed to accomplish a target Pf for different

sensitivities for V=0.1 and V=0.2 and z /Lf = 0.2

202
 0.300

0.250 V=0.3 NO LB
S=3
Probability of Failure

S=2.75
0.200 S=2.5
S=2.25
S=2
0.150 S=1.75

0.100

0.050

0.000
1.50 1.75 2.00 2.25 2.50 2.75 3.00

FS

0.500
NO LB
S=3
V=0.4 S=2.75
0.400 S=2.5
Probability of Failure

S=2.25
S=2
S=1.75
0.300 S=1.5

0.200

0.100

0.000
1.50 1.75 2.00 2.25 2.50 2.75 3.00
FS

Figure 7. 12 Recommended factor of safety needed to accomplish a target Pf for different

sensitivities for V=0.3 and V=0.4 and z /Lf = 0.2

203
 0.500
NO LB
V=0.5 S=3
S=2.75
0.400 S=2.5
Probability of Failure

S=2.25
S=2
0.300 S=1.75
S=1.5

0.200

0.100

0.000
1.50 1.75 2.00 2.25 2.50 2.75 3.00

FS

Figure 7. 13 Recommended factor of safety needed to accomplish a target Pf for different

sensitivities for V=0.5 and z /Lf = 0.2

[Link] Acceptable probability of failure

A reliability analysis calculates the probability of failure of a given system. In

geotechnical engineering, acceptable probabilities of failure vary according to how

important the structure is and how dangerous the consequences of failure are. Thus, the

concept of risk may be understood in terms not only of the likelihood that certain events

will occur but also of what these events consist of and what they would lead to in terms

of environmental damage, casualties, financial losses, and other undesirable outcomes

(Salgado et al. 2014). Chowdhury and Flentje (2003) suggested maximum values for the

probability of failure of slopes that have different functions and which have different

consequences of failure. The acceptable probabilities of failure range could be as low as

204
0.001 for cases involving slopes that might results in loss of lives upon failure to about

0.15 for slopes and temporary slopes where failure would not result in potential loss of

lives. Christian et al. (1994) suggested that Pf = 0.001 would be a reasonable number to

use in design. Loehr et al. (2005) set the range of Pf from 0.001 to 0.01 for slopes: 0.01

for relatively low potential risk and 0.001 for high potential risk. Santamarina et al.

(1992) made an effort to determine an acceptable probability of failure of slopes. The

results are summarized in Table 7.1.

Table 7. 1. Acceptable probability of failure of slopes (Santamarina et al. 1992)

Acceptable
probability of
Conditions failure
Temporary structures: no potential life loss, low repair cost 0.1
Minimal consequences of failure: high cost to reduce the probability of failure
0.1-0.2
(bench slope or open pit mine)

Minimal consequences of failure: repairs can be done when time permits (repair 0.01
cost is less than cost of reducing probability of failure)
Existing large cut in interstate highway 0.01-0.02
Large cut on interstate highway to be constructed < 0.01
Lives may be lost when slopes fail 0.001

The typical acceptable probabilities of failure range from as high as 20% for cases with

minimal consequences of failure to values of 0.001 for extreme cases where lives may be

lost when slopes fail. In this study, probabilities of 0.001, 0.005, 0.01, 0.05, and 0.1 are

considered acceptable probabilities of failure for the recommendation of the factors of

safety.

205
[Link] Design graphs for factors of safety of undrained slopes

In this section, required factors of safety are recommended to achieve different

target probabilities of failure for different cases of spatial variability and soil sensitivity.

The factors of safety recommended to achieve probabilities of 0.001, 0.005,0.01, 0.05,

and 0.10 according to different coefficients of variations ( V= 0.1, 0.3, and 0.5) for slopes

with ratio z /Lf = 0.1 and z /Lf = 0.2 are presented in Figures 7.14, 7.15., 7.16., 7.17.,

7.18., and 7.19.

For the case where the lower-bound factor of safety is not included in the

analysis, results on Figs. 7.14 to 7.19 indicate that the required factor of safety decreases

significantly as the target reliability level of the slope increases. For example, for the case

of intermediate spatial variability (V = 0.3) in the undrained shear strength, the required

factor of safety decreases from a high value of 2.8 for the case with a target probability of

failure 0.001 (slopes that have loss of lives) to a low value of 1.76 for the case where the

target probability of failure is 0.1. These required factors of safety decrease when the

lower-bound factor of safety is incorporated in the analysis. For example, when a lower-

bound factor of safety that is consistent with a sensitivity of 2 is incorporated, the

required factor of safety decreases from 2.8 to 2.0 (for the case with a target probability

of failure 0.001) and from 1.76 to 1.66 (for the case with a target probability of failure

0.1). These results are significant because they indicate that the lower-bound factor of

safety could play a significant role in reducing conservatism in the design, particularly for

slopes designed for a higher reliability level.

206
 3.0

2.8
Required Factor of Safety Pf = 0.001
2.6

2.4

2.2

2.0

1.8
V =0.1
1.6
V = 0.3
1.4
V = 0.5
1.2
1.50 1.75 2.00 2.25 2.50 2.75 3.00

Sensitivity
3.0

2.8
Pf = 0.005
Required Factor of Safety

2.6

2.4

2.2

2.0

1.8 V =0.1
1.6 V = 0.3
V = 0.5
1.4

1.2
1.50 1.75 2.00 2.25 2.50 2.75 3.00
Sensitivity
Figure 7. 14 Reccommended factors of safety for different coefficients of variation for
probabilities of failure of 0.001& 0.005 for slopes with ratio z / Lf = 0.1

207
 3.0

2.8
Pf = 0.01
Required Factor of Safety

2.6

2.4

2.2

2.0

1.8
V =0.1
1.6
V =0.3
1.4
V = 0.5
1.2
1.50 1.75 2.00 2.25 2.50 2.75 3.00
Sensitivity

3.0

2.8 V =0.1
Required Factor of Safety

Pf = 0.05
V = 0.3
2.6
V = 0.5
2.4

2.2

2.0

1.8

1.6

1.4

1.2
1.50 1.75 2.00 2.25 2.50 2.75 3.00

Sensitivity
Figure 7. 15 Recommended factors of safety for different coefficients of variation to
achieve probabilities of failure of 0.01 & 0.05 for slopes with ratio z / Lf =0.1

208
 3.000

2.800 V = 0.1
Required Factor of Safety Pf = 0.1
V = 0.3
2.600
V = 0.5
2.400

2.200

2.000

1.800

1.600

1.400

1.200
1.50 1.75 2.00 2.25 2.50 2.75 3.00
Sensitivity
Figure 7. 16 Recommended factors of safety for different coefficients of variation to
achieve probability of failure of 0.1 for slopes with ratio z / Lf =0.1

The cases that were analyzed above represent the case with a z / Lf =0.1. To check the

sensitivity of the results to the ratio of z / Lf , the analysis is repeated for a ratio of z / Lf

=0.2 and the results are included in Figs. 7.17 to 7.19. For the case where the lower-

bound factor of safety is not included in the analysis and for the case of intermediate

spatial variability (V = 0.3) in the undrained shear strength, the required factor of safety

decreases from a high value of 2.82 (instead of 2.8 for z / Lf =0.1) for the case with a

target probability of failure 0.001 to a low value of 1.83 (instead of 1.76 for z / Lf =0.1)

for the case where the target probability of failure is 0.1. These results indicate that the

resulting required factor of safety is not highly sensitive to the ratio of z / Lf.

209
 3.0

2.8
Required Factor of Safety

Pf = 0.001
2.6

2.4

2.2

2.0

1.8
V = 0.1
1.6
V = 0.3
1.4 V = 0.5

1.2
1.50 1.75 2.00 2.25 2.50 2.75 3.00
Sensitivity

3.0

2.8
Required Factor of Safety

Pf = 0.005
2.6

2.4

2.2

2.0

1.8
V = 0.1
1.6
V = 0.3
1.4 V = 0.5

1.2
1.50 1.75 2.00 2.25 2.50 2.75 3.00
Sensitivity
Figure 7. 17 Recommended factors of safety for different coefficients of variation to
achieve probability of failure of 0.001 and 0.005 for slopes with ratio z/Lf = 0.2

210
 3.0

2.8
Pf = 0.01
Required Factor of Safety

2.6

2.4

2.2

2.0

1.8
V =0.1
1.6
V =0.3
1.4 V = 0.5

1.2
1.50 1.75 2.00 2.25 2.50 2.75 3.00
Sensitivity

3.0

2.8 V =0.1
Pf = 0.05
Required Factor of Safety

V = 0.3
2.6
V = 0.5
2.4

2.2

2.0

1.8

1.6

1.4

1.2
1.50 1.75 2.00 2.25 2.50 2.75 3.00
Sensitivity
Figure 7. 18 Recommended factors of safety for different coefficients of variations to
achieve probabilities of failure of 0.01 & 0.05 for slopes with ration of z/Lf = 0.2

211
 3.0

2.8 V = 0.1
Pf = 0.1
Required Factor of Safety
V = 0.3
2.6
V = 0.5
2.4

2.2

2.0

1.8

1.6

1.4

1.2
1.50 1.75 2.00 2.25 2.50 2.75 3.00
Sensitivity

Figure 7. 19 Recommended factors of safety for different coefficients of variation to

achieve probabilities of failure of 0.1 for slopes with ratio z/ Lf = 0.2

Design tables are also shown in the following section to illustrate the required factors of

safety needed to achieve a target level of the probability of failure for the ratios of z / Lf

= 0.1 & 0.2 for different coefficients of variations.

Table 7.2 Recommended Factors of Safety needed to accomplish a target Pf for different
sensitivities for V =0.1 & z / Lf = 0.10

Sensitivity 1.5 1.75 2 2.25 2.5 2.75 3

Pf
0.001 1.498 1.745 1.989 2.218 2.41 2.55 2.62
0.005 1.492 1.73 1.946 2.12 2.23 2.26 2.3
0.01 1.48 1.71 1.9 2 2.1 2.145 2.145
0.05 1.43 1.6 1.7 1.75 1.77 1.78 1.79
0.1 1.375 1.5 1.57 1.59 1.6 1.61 1.61

212
Table 7.3 Recommended Factors of Safety needed to accomplish a target Pf for different
sensitivities for V =0.3 & z / Lf = 0.10

Sensitivity
pf 1.5 1.75 2 2.25 2.5 2.75 3

0.001 1.499 1.747 1.990 2.235 2.450 2.650 2.800

0.005 1.495 1.738 1.970 2.180 2.350 2.450 2.500
0.01 1.490 1.782 1.945 2.135 2.260 2.330 2.350
0.05 1.450 1.650 1.780 1.870 1.940 1.960 1.970
0.1 1.410 1.570 1.680 1.740 1.760 1.770 1.770

Table 7.4 Recommended Factors of Safety needed to accomplish a target Pf for different
sensitivities for V =0.5 & z / Lf = 0.10

Sensitivity
1.5 1.75 2 2.25 2.5 2.75 3
Pf
0.001 1.499 1.748 1.996 2.245 2.48 2.71 2.93
0.005 1.497 1.744 1.987 2.225 2.45 2.64 2.8
0.01 1.494 1.739 1.979 2.205 2.405 2.57 2.68
0.05 1.47 1.69 1.89 2.05 2.16 2.25 2.29
0.1 1.443 1.645 1.81 1.93 2.02 2.045 2.06

Table 7.5 Recommended Factors of Safety needed to accomplish a target Pf for different
sensitivities for V =0.1 & z / Lf = 0.20

Sensitivity
Pf 1.5 1.75 2 2.25 2.5 2.75 3

0.001 1.498 1.745 1.98 2.21 2.4 2.55 2.6

0.005 1.492 1.73 1.946 2.12 2.23 2.26 2.3
0.01 1.485 1.71 1.905 2.035 2.105 2.145 2.145
0.05 1.43 1.6 1.7 1.75 1.77 1.78 1.79
0.1 1.375 1.5 1.57 1.59 1.6 1.61 1.61

213
Table 7.6 Recommended Factors of Safety needed to accomplish a target Pf for different
sensitivities for V =0.3 & z / Lf = 0.20

Sensitivity
1.5 1.75 2 2.25 2.5 2.75 3
Pf
0.001 1.499 1.747 1.992 2.24 2.47 2.7 2.85
0.005 1.495 1.738 1.978 2.2 2.4 2.55 2.66
0.01 1.49 1.728 1.916 2.16 2.33 2.476 2.53
0.05 1.45 1.65 1.82 1.93 2 2.08 2.1
0.1 1.41 1.58 1.7 1.78 1.83 1.85 1.85

Table 7.7 Recommended Factors of Safety needed to accomplish a target Pf for different
sensitivities for V =0.5& z / Lf = 0.20

Sensitivity
Pf 1.5 1.75 2 2.25 2.5 2.75 3

0.001 1.499 1.748 1.997 2.245 2.49 2.73 2.97

0.005 1.497 1.744 1.989 2.23 2.466 2.68 2.88
0.01 1.494 1.738 1.979 2.211 2.43 2.6 2.8
0.05 1.468 1.69 1.9 2.09 2.25 2.32 2.38
0.1 1.438 1.645 1.825 1.97 2.08 2.14 2.18

7.3 Summary

The major conclusion from this chapter is that a lower-bound factor of safety can

cause a significant increase in the calculated reliability for slope design. The effect of the

lower-bound factor of safety on the reliability is more pronounced when the uncertainty

in the undrained shear strength is large. Moreover, reliability analyses will provide more

robust, realistic, and useful information for decision making purposes if they include

information about lower-bound factor of safety. Finally, it is found that the probability of

failure of slopes is affected mostly by the ratio of the lower-bound to mean factor of

safety.

214
 CHAPTER 8

Conclusions and Recommendations

8.1 Introduction

In this Chapter the conclusions and recommendations that resulted from this

research work besides the design steps that may be followed to evaluate the probability of

failure and the design factors of safety of slopes are presented. The conclusions and

recommendations are based on a reliability-based design analysis that was conducted for

undrained slopes while incorporating the effects of both model uncertainty and

uncertainty due to spatial variability. The model uncertainty was evaluated based on a

database of 52 case histories of failure slopes, while the uncertainty due to spatial

variability was evaluated from the findings of the study by Jha and Ching (2013) who

used a random finite element analysis to characterize this uncertainty for undrained

slopes.

8.2 Summary of Findings

The following findings, conclusions, and recommendations emerged from the

study conducted in this thesis:

1. Based on the analysis of 52 case histories of failed slopes, it was concluded that

the model uncertainty as reflected in the ratio of measured to predicted factor of

safety, , for 4 commonly used Limit Equilibrium Methods (Bishop, Janbu, OMS,

and Spencer) has a mean of about 1.0 and a coefficient of variation that is in the
215
 order of 0.27 to 0.29. The parameter  was found to be properly modeled by a

lognormal probability distribution.

2. Based on the same database, it was found that there is strong evidence of the

existence of a lower-bound factor of safety for undrained slopes that could be

estimated from information regarding the remolded undrained shear strength of

the soil. The lower-bound factor of safety has a mean value of that ranges from

0.45 to 0.47 when cases of very high soil sensitivity were excluded from the

analysis. For such values of lower-bound factors of safety, the lower-bound is

expected to have a considerable effect on the design of a slope. This effect

increases as the magnitude of the lower bound factor of safety increases.

3. The spatial variability in the undrained shear strength of clays as reflected in the

coefficient of variation and the scale of fluctuation has a direct effect on the mean

of the factor of safety and its coefficient of variation. As the coefficient of

variation in the undrained shear strength increases, the mean of the factor of

safety decreases compared to its deterministic design value and the coefficient of

variation in the factor of safety increases. The effect of the scale of fluctuation is

in reducing the uncertainty in the factor of safety since it results in a variance

reduction due to spatial averaging along the failure surface. The ratio of the scale

of fluctuation to the length of the failure surface dictates the magnitude of the

variance reduction due to averaging.

4. The probability of failure that was calculated for the cases were the uncertainties

in the spatial variability and model uncertainty are combined indicated that the

probability of failure decreases as (1) the design factor of safety increases, (2) the

216
 coefficient of variation in the undrained shear strength decreases, (3) the ratio of

z / Lf decreases below a threshold value of 0.2, and (4) the lower-bound factor of

safety increases.

5. The effect of the lower-bound factor of safety on the reliability of the slope was

found to be significant and depends on the magnitude of the lower-bound factor

of safety relative to the design factor of safety, the coefficient of variation of the

undrained shear strength, and on the magnitude of the design factor of safety. For

relatively small values of the coefficient of variation of the undrained shear

strength (V = 0.1 and 0.2), the lower-bound factor of safety was found to have an

effect on the reliability only for cases with relatively small sensitivities (less than

2.0). For cases with higher coefficients of variation of the undrained shear

strength (V > 0.3), the effect of the lower-bound factor of safety on the reliability

is evident at higher sensitivities (2.75 to 3.0).

6. The probability of failure could be reduced by more than half for cases were a

lower bound factor of safety is included in the analysis. This reduction in the

probability of failure due to the lower bound translates into a reduction in the

required factor of safety that could be used in a design with a target level of

reliability.

7. Relationships between the target probability of failure and the required factor of

safety were established and presented for cases with different conditions of spatial

variability (different V and different z/Lf ) and different lower-bound factors of

safety as reflected in the sensitivity of the soils. Based on these relationships,

recommendations were made for the factors of safety to be used in the design of

217
 undrained slopes for different target reliability levels. The range of the different

reliability levels was chosen to be between a probability of failure of 10%

(applicable for temporary slopes where no loss of lives is envisaged) and a

probability of failure of 0.1% (applicable to slopes with potential loss of lives

upon failure). As an example, for the case of intermediate spatial variability (V =

0.3), the required factor of safety decreases from a high value of 2.8 for the case

with a target probability of failure 0.1% to a low value of 1.76 for the case where

the target probability of failure is 10%. These required factors of safety decrease

to 2.0 (for pf = 0.1%) and 1.66 (for pf = 10%) when the lower-bound factor of

safety is incorporated in the analysis.

8.3 Design Steps

This section illustrates the steps that may be followed to evaluate the probability

of failure and the design factors of safety of undrained slopes.

8.3.1 Estimation of the probability of failure of undrained slopes

1. A deterministic slope stability analysis should be adopted for the evaluation of

the deterministic factor of safety of the undrained slope.

2. Using the results of the deterministic analysis, the length of the failure surface

(Lf) should be measured.

3. Based on the undrained shear strength borehole, both the vertical scale of

fluctuation (z) and the coefficient of variation (V) can be evaluated. For

218
 estimating z, either equation 5-1 shown in chapter 5 can be used or an

assumed value between 1-2.5m can be adopted as recommended by Phoon

and Kulhawy (1999).

4. After estimating both z and Lf, the ratio of z / Lf can be evaluated simply.

5. Suppose the ratio of z / Lf, deterministic factor of safety, and the coefficient

of variation of undrained shear strength are evaluated using the above steps;

the probability of failure of the slope can be evaluated using graphs 7.1, 7.2, &

7.3.

8.3.2 Estimation of the design factor of safety of undrained slopes

The following steps are needed to investigate the required factor of safety needed to

accomplish a target level of probability of failure.

1. The ratio of z / Lf is evaluated as illustrated before in section 8.3.1.

2. The sensitivity of the soil of each slope can be evaluated.

3. Using Tables 7.2, 7.3, 7.4, 7.5, 7.6, & 7.7 and by knowing the sensitivity of

the soil the required factor of safety needed to achieve a target level of

probability of failure can be estimated.

219
 REFERENCES

Alonso, E. E. 1976. “Risk analysis of slopes and its application to slopes in Canadian
sensitive clays.” Geotechnique, 26, 453–472.
American Association of State Highway and Transportation Officials (AASHTO). (2004)
LRFD Bridge Design Specifications, Washington, D.C.
Bea, R.G., Jin, Z., Valle, C., and Ramos, R. (1999) “ Evaluation of Reliability of
Platform Pile Founations”, Journal of Geotechnical and Geoenvironmental Engineering,
ASCE, Vol. 125, No.8, pp. 696-704.
Bhattacharya, G., Jana, D., Ojha, S., and Chakraborty, S. 2003. “Direct search for
minimum reliability index of earth slopes.” Comput. Geotech., 306, 455–462.
Babu, G. L. S., and Mukesh, M. D. 2004. “Effect of soil variability on reliability of soil
slopes.” Geotechnique, 545, 335–337.
Crooks, J.H.A., Been, K., and Mickleborough, B.W. 1986. “ An Embankment Failure on
Soft Fissured Clay.” Can. Geotech. J., Vol. 23, pp. 528-540.
Chowdhury, R. N., and Tang, W. H. 1987. “Comparison of risk models for slopes.”
Proc., 5th Int. Conf. on Applications of Statistics and Probability in Soil and Structural
Engineering, Vol. 2, Univ. of British Columbia, Vancouver, BC, Canada, 863–869.
Chen, Z., and Shao, C. (1988). “Evaluation of minimum factor of safety inslope stability
analysis.” Can. Geotech. J., 25(4), 735–748.
Christian, J. T., Ladd, C. C., and Baecher, G. B. (1994). “Reliability applied to slope
stability analysis.” J. Geotech. Eng., 120(12), 2180–2207.
Chowdhury, R. N., and Xu, D. W. 1995. “Geotechnical system reliability of slopes.”
Reliab. Eng. Syst. Saf., 47, 141–151.
Christian, J. T. 1996. “Reliability methods for stability of existing slopes.” Proc.,
Uncertainty in the Geologic Environment: From Theory to Practice, C. D. Shackelford,
P. P. Nelson, and M. J. S. Rotheds., ASCE, Reston, Va., 409–419.
Chai, J.-C., Miura, N., and Shen, S.-L. (2002). “Performance of embankments with and
without reinforcement on soft subsoil.” Canadian Gesotechnical Journal, 39(4), 838–
848.
Cho, S. E. 2007. “Effects of spatial variability of soil properties on slope stability.” Eng.
Geol., 923–4, 97–109.

220
Dascal, O., Tournier, J. P., Tavenas, F., and La Rochelle, P. (1972). “Failure of a test
embankment on sensitive clay.” Proceedings of the Specialty Conference on
Performance of Earth and Earth-Supported Structures, ASCE, Lafayette, IN, 129– 158.
Dascal, O., and Rournier, J. P. (1975). “Embankment on soft and sensitive clay
foundation.” J. Geotech. Eng., 101(3), 297–314.
D’Andrea, R. A., and Sangrey, D. A. 1982. “Safety factors for probabilistic slope
design.” J. Geotech. Engrg. Div., 1089, 1108–1118.
Day, R. W. (1996). “Failure of Desert View Drive embankment.” J. Perform. Constr.
Facil., 10(1), 11–1.
Duncan, J. M. 2000. “Factors of safety and reliability in geotechnical engineering.” J.
Geotech. Geoenviron. Eng., 1264, 307–316.
Duncan, J. M., and Wright, S. G. (2005). Soil strength and slope stability, Wiley,
Hoboken, NJ.
Eide, O., and Holmberg, S. (1972). “Test fills to failure on the soft Bangkok Clay.”
Proceedings of the Specialty Conference on Performance of Earth and Earth- Supported
Structures, ASCE, Lafayette, IN, 159–180.
El-Ramly, H., Morgenstern, N. R., and Cruden, D. M. 2002. “Probabilistic slope stability
analysis for practice.” Can. Geotech. J., 39, 665– 683.
Flaate, K., and Preber, T. (1974). “Stability of road embankments in soft clay.” Canadian
Geotechnical Journal, 11(1), 72–88.
Fjeld, S. (1977) “ Reliability of Offshore Structures”, Proceeings, 9th Ofshore Technology
Conference. OTC 3027, PP459-471.
Ferkh, Z., and Fell, R. (1994). “ Design of embankments on soft clay.” Proceedings of
Thirteenth International Conference on Soil Mechanics and Foundation Engineering,
New Delhi, India, 733-738.

Golder, H.Q., and Palmer, D.J. (1954). “ Investigation of a Bank Failure at Scrapsgate,
ISLE of Sheppey, Kent.” Géotechnique, 5, 55-73.

Griffiths, D. V., and Fenton, G. A. 2004. “Probabilistic slope stability analysis by finite
elements.” J. Geotech. Geoenviron. Eng., 1305, 507–518.
Griffiths, D. V., Fenton, G. A., and Denavit, M. D. 2007. “Traditionaland advanced
probabilistic slope stability analysis.” Proc., GeoDenver 2007 Symp., K. K. Phoon, G. A.
Fenton, E. F. Glynn, C. H. Juang, D. V. Griffiths, T. F. Wolff, and L. Zhang, eds., ASCE,
Reston, Va., 1–10.
Griffiths, D. V., Huang, J., and Fenton, G., (2010). “ Comparison of slope reliability
methods of analysis.” GeoFlorida, 1952-1961.

221
Haupt, R., and Olsen, J. (1972). “Case history—Embankment failure on soft varved silt.”
Proc. Spec. Conf. Perf. of Earth and Earth Supported Structures, 1(1), 1–28.
Hasofer, A. M., and Lind, N. C. 1974. “Exact and invariant second moment code
format.” J. Engrg. Mech. Div., 1001, 111–121.
Hanzawa, H., Kishida, T., Fukusawa, T., and Hideyuki, A. (1994). “A case study of the
application of direct shear and cone penetration tests to soil investigation: design and
quality control for peaty soils.” Soils and Foundations, 34(4), 13–22.
Hassan, A. M., and Wolff, T. F. 1999. “Search algorithm for minimum reliability index
of earth slopes.” J. Geotech. Geoenviron. Eng., 1254, 301–308.
Hong, H. P., and Roh, G., (2008). “ Reliability evaluation of earth slopes.” J. Geotech.
Engrg., 134, 1700-1705.

Huang, J., Griffiths, D. V., and Fenton, G. A., (2013). “ A benchmark slope for system
reliability analysis.” Geo- Congress, 962-971.

Indraratna, B., Balasubramaniam, A.S., and Balachandran, S. (1992). “Performance of

Test Embankment Constructed to Failure on Soft Marine Clay.” J. Geotech. Engrg.,
118(1), pp. 12-33.

Jha, S. and Ching, J. (2013). “Simplified reliability method for spatially variable
undrained engineered slopes.” Soils and Foundations, 53(5), 708-719.
Kjærnsli, B., and Simons, N. (1962). “Stability investigations of the North Bank of the
Drammen River.” Géotechnique, 12(2), 147–167.
Kulhawy, F. (1992). On Evaluation of Static Soil Properties. In R. Seed, & R. Boulanger,
In stability and Performance of Slopes and Embankments II (pp. 95-115). New York:
American Society of Civil Engineers.
Kulhawy, F. H. (1996). From Casagrande's "Calculated Risk" to Reliability Based Design
in Foundation Engineering. Civil Engineering Practice, Boston Society of Civil Engineers
, 43-56.
Kulhawy, F.H., and Phoon, K.K. (2002) “ Observations on Geotechnical Reliability-
Based Design Development in North America”, Proc. Int. Workshop on Foundation
Design Codes and Soil Investigation in view of International Harmonization and
Performance Based Design, Tokyo, Japan, April, pp.31-48.
Lacasse, S. M., Ladd, C. C., and Barsvary, A. K. (1977). “Undrained behavior of
embankments on New Liskeard varved clay.” Canadian Geotechnical Journal, 14(3),
367–388.
Li, K. S., and Lumb, P. 1987. “Probabilistic design of slopes.” Can. Geotech. J., 24, 520–
531.

222
Lacasse, S. 1994. “Reliability and probabilistic methods.” Proc., 13th Int. Conf. Soil
Mechanics and Foundation Engineering, New Delhi, India, 225–227.
Lacasse, S., and Nadim, F. 1996. “Uncertainties in characterizing soil properties.” Proc.,
Uncertainty 1996, C. D. Shackelford, P. P. Nelson, and M. J. S. Roth, eds., ASCE,
Reston, Va., 49–75.
Low, B. K. 1996. “Practical probabilistic approach using spreadsheet.” Proc.,
Uncertainty in the Geologic Environment: From Theory to Practice, Vol. 2, Reston, Va.,
1284–1302.
Low, B. K., and Tang, W. H. 1997a. “Reliability analysis of reinforced embankments on
soft ground.” Can. Geotech. J., 345, 672–685.
Low, B. K., and Tang, W. H. 1997b. “Efficient reliability evaluation using spreadsheet.”
J. Geotech. Geoenviron. Eng., 1237, 749–752.
Low, B. K., Gilbert, R. B., and Wright, S. G. 1998. “Slope reliability analysis using
generalized method of slices.” J. Geotech. Geoenviron. Eng., 1244, 350–362.
Low, B. K. 2003. “Practical probabilistic slope stability analysis.” Proc., Soil and Rock
America 2003, 12th Panamerican Conf. on Soil Mechanics and Geotechnical
Engineering, and 39th U.S. Rock Mechanics Symp., MIT, Cambridge, Mass., Vol. 2,
Verlag Glückauf GmbH, Essen, Germany, 2777–2784.
Low, B. K., Lacasse, S., and Nadim, F. 2007. “Slope reliability analysis accounting for
spatial variation.” Georisk, Vol. 1, Taylor and Francis, London, 177–189.
Matsuo, M., and Kuroda, K. 1974. “Probabilistic approach to the design of
embankments.” Soils Found., 141, 1–17.
Mostyn, G. R., and Li, K. S. 1993. “Probabilistic slope stability— State of play.” Proc.,
Conf. Probabilistic Meth. Geotech. Eng., K. S. Li and S.-C. R. Lo, eds., Balkema,
Rotterdam, The Netherlands, 89– 110.
Malkawi, A. I., Hassan, W. F., and Abdulla, F. A., (2000). “ Uncertainty and reliability
analysis applied to slope stability.” Structural Safety 22, 161-187.

McGinn, M., and O’Rourke, T.D., Performance and Evaluation of Tunnels, Excavations,
and Retaining Structures, ASCE, pp. 480-495.

Oka, Y., and Wu, T. H. 1990. “System reliability of slope stability.” J. Geotech. Engrg.,
116, 1185–1189.

Peterson, R., Iverson, N.L., and Rivard, P.J.1957. “Studies of Several Dam Failures on
Clay Foundations.” Proceedings of the 4th International Conference on Soil Mechanics
and Foundation Engineering, London, England, Vol. 11, pp. 348-352.

223
Parry, R. (1968). “Field and laboratory behavior of a lightly overconsolidated clay.”
Géotechnique, 18, 151–171
Pilot, G. (1972). “Study of five embankment failures on soft soils.” Proceedings of the
Specialty Conference on Performance of Earth and Earth-Supported Structures, ASCE,
Lafayette, IN, 81–100.
Pilot, G., Trak, B., and La Rochelle, P. (1982). “Effective stress analysis of the stability
of embankments on soft soils.” Canadian Geotechnical Journal, 19(4), 433–450.
Rivard, P., and Lu, Y. (1978). “Shear strength of soft-fissured clays.” [Link]. J.,
15(3), 382–390.
Ramalho-Ortigão, J. A., Werneck, M. L. G., and Lacerda, W. A. (1983). “Embankment
failure on clay near Rio de Janeiro.” Journal of Geotechnical Engineering, 109(11),
1460–1479.
Rodriguez, S.G.H., Henriques, C.C.D., Ellwanger,G.B., and de Lima, E.C.P. (1998)
“Structural Reliability with Truncated Probability Distributions”, Proceedings of the
International Conference on Offshore Mechanics and Arctic Engineering, OMAE, July,
5-9, 1998,Lisbon,Portugal, 6 pp.
Skempton, A.W., and P. La Rocheelle (1965). “ The Bradwell Slip: A Short-term Failure
in London Clay.” Géotechnique, 15, No.3, pp. 221-242.

Skempton, A.W., and Coats, D. J. (1985). “Carsington dam failure.” Failures in

Earthworks, Thomas Telford, London, 203–220.
Shinoda, M. 2007. “Quasi-Monte Carlo simulation with low discrepancy sequence for
reinforced soil slopes.” J. Geotech. Geoenviron. Eng., 1334, 393–404.
Shogaki, T., and Kumagai, N. (2008). “ A slope Stability analysis Considering Undrained
Strength Anisotropy of Natural Clay Deposits.” Soils and Foundations, Vol.48, No.6,
805-819.
Tang, W. H., Yucemen, M. S., and Ang, A. H. S. 1976. “Probability based short-term
design of slopes.” Can. Geotech. J., 13, 201–215.
Travis, Q.B., Schmeeckle, M.W., and Sebert, D.M. (2010). “ Meta-analysis of 301 slope
failure calculations. I: Database description.” Journal of Geotechnical and
Geoenvironmental Engineering, 1(1), 264.
Vanmarcke, E. H. 1977. “Reliability of earth slopes.” J. Geotech. Engrg. Div., 10311,
1247–1265.
Wolfskill, L. A., and Lambe, T. W (1967). “Slide in the Siburua Dam.” [Link] Mech. and
Found. Div., 93(SM4), 107–13.
Wu, T. H., Thayer, W. B., and Lin, S. S. (1975). “Stability of embankmenton clay.” J.
Geotech. Engrg. Div., 101(9), 913–932.

224
Wolff, T. F. 1996. “Probabilistic slope stability in theory and practice.” Uncertainty in the
geologic environment: From theory to practice, C. D. Shackelford, P. P. Nelson, and M.
J. S. Roth, eds., ASCE, Reston, Va., 419–433.
Whitman, R. V. 2000. “Organizing and evaluating in geotechnical engineering.” J.
Geotech. Geoenviron. Eng., 1267, 583–593.
Wu, T.H. (2009). “Reliability of geotechnical predictions.” Geotechnical Risk and Safety,
Proceedings of the 2nd International Symposium on Geotechnical Safety and Risk, CRC
Press, Taylor & Francis Group, Gifu, Japan, 3-10.
Xu, B., and Low, B. K. 2006. “Probabilistic stability analyses of embankments based on
finite-element method.” J. Geotech. Geoenviron. Eng., 13211, 1444–1454
Zhang, Z., Tao, M., and Morvant, M. (2005). “Cohesive slope surface failure and
evaluation.” J. Geotech. Geoenviron. Eng., 131(7), 898–906.
Zhang, J., Huang, H. W., Juang, C. H., and Li, D. Q., (2013). “ Extension of Hassan and
Wolff method for system reliability analysis of soil slopes.” Engineering Geology 160,
81-88.

225
226