You are on page 1of 9

Geotechnical Frontiers 2017 GSP 278 363

Probabilistic Slope Stability Analyses: Effects of the Coefficient of Variation


and the Cross-Correlation of Shear Strength Parameters
Emir Ahmet Oguz1; Yagizer Yalcin2; and Nejan Huvaj3*

1
Graduate Student and Research Assistant, Middle East Technical Univ., Civil Engineering
Department, 06800 Ankara, Turkey. E-mail: eoguz@metu.edu.tr
2
Graduate Student and Research Assistant, Middle East Technical Univ., Civil Engineering
Department, 06800 Ankara, Turkey. E-mail: yyalcin@metu.edu.tr
3*
Assistant Professor, Ph.D., Middle East Technical Univ., Civil Engineering Department, 06800
Ankara, Turkey. E-mail: nejan@metu.edu.tr
Abstract
The assessment of the safety level of natural slopes, road cuts, embankments and levees require
consideration of uncertainties and variability in material properties. In this study, for a number of
slope geometries, including a real-life landslide case, probability of failure (PF) and the most
critical failure surface are investigated with and without cross-correlation of shear strength
properties. Slopes having different traditionally-defined factor of safety (FS) levels are studied.
The uncertainty of soil properties are considered by different levels of coefficient of variation
(COV). Limit equilibrium method is used for slope stability analyses and geotechnical material
properties are considered to have normal statistical distribution. The results of this analyses show
that the PF and the critical failure surface is significantly influenced by the COV level, the
consideration of cross correlation of shear strength parameters, and by the traditional FS level of
the slopes. The inverse relation between FS and PF is demonstrated to be nonlinear and the COV
level has significant effect on this relationship. Results indicate that the deterministic slope
stability analyses resulting in a single FS value is no longer sufficient to evaluate the safety of a
slope in geotechnical engineering, and that the deterministic critical failure surface with
minimum FS value is not always the most critical slip surface. The results presented in this study
could be useful for further understanding of probabilistic slope stability and the effects of soil
variability/uncertainty, with the aim of better geotechnical risk evaluation and communication.

INTRODUCTION

In the practice of geotechnical engineering, it is common to use deterministic analyses methods


to assess the safety of slopes. In deterministic analyses, soil layers are assumed to be
homogeneous and soil properties are selected to be representative values of the natural soil. This
would be an ideal analysis type provided that a thorough site investigation and high quality
extensive laboratory testing is available and they are well-interpreted to select the representative
values. However, this is seldom the case in geotechnical engineering, especially in some
countries where there is lack of strict regulations and quality control. In addition, there is
inherent variability in soils, whereas in practice, based on only a limited number of borehole data
interpolations/extrapolations have to be made to define soil layers at the site. Furthermore, a fair
portion of soil idealization process is carried out based on empirical correlations which adds to
the uncertainties in soil properties. In such circumstances, it would be more suitable to utilize
probabilistic approach which considers the variability of soil properties in terms of a statistical
distribution having mean values and standard deviation.

ASCE RocNews Spring 2017


Geotechnical Frontiers 2017 GSP 278 364

In this study, safety level of slopes are investigated with deterministic and probabilistic
limit equilibrium methods to compare the results. For different slope angles, in a generic soil,
with and without a water table, the statistical distribution type and parameters of soil shear
strength are changed, with and without cross-correlation of shear strength parameters. The
effects of these on the probabilistic safety of the slope is investigated.
METHODS OF ANALYSES
For the slope stability analyses Rocscience Slide v7, a limit equilibrium analysis tool, is used. In
limit equilibrium method, the sliding soil mass is divided into a number of slices; force and
moment equilibrium of individual slices as well as of the whole slope is calculated. For this
specific study, among the various limit equilibrium methods, Spencer (1967) method is used
since it satisfies all equilibrium conditions. For probabilistic slope stability calculations, material
properties are defined in terms of a statistical distribution and related input parameters such as
the mean and the standard deviation. Then, samples of material properties are randomly selected
from the given statistical distribution and factor of safety calculations are carried out resulting in
probability of the failure (PF) of the slope, determined as the ratio of the number of samples
resulting in F.S. less than 1.00 to the whole F.S. field for all random samples.
Two types of analyses: global minimum (GM) and overall slope (OS) are carried out by
Slide v7. GM-type analysis first performs deterministic slope stability analysis using the mean
values of all parameters and finds the most critical deterministic failure surface having the lowest
FS value and then probabilistic analysis is carried out only for this surface using statistical
distribution of material properties. On the other hand, the OS-type analysis carries out a search N
times, where N is the number of randomly selected soil properties, and carries out probabilistic
analysis for each found critical surface. OS-type analysis takes significantly longer run time as
compared to GM-type analysis. In this study, OS method is used, unless otherwise stated.
Random sampling of material properties from a given statistical distribution can be done
with two sampling methods in Slide v7: Monte-Carlo (MC) and Latin Hypercube (LH). MC
method has been used widely for probabilistic slope stability analyses (Bhattacharya et al. 2003,
El-Ramly et al. 2005, Cho 2010, Le et al. 2014, Akbas and Huvaj 2015). Despite its popularity in
the probabilistic approach, MC method requires long computational time as compared to other
more efficient sampling methods (Le et al. 2014, Cho 2009, Hammah et al. 2009). As part of the
preliminary studies, both LH and MC sampling methods are performed and the PF results are
obtained to be very similar, the only difference being the run time. Therefore, in this study, more
widely used MC sampling method is used, unless otherwise stated.
For soil shear strength, Mohr-Coulomb model is used with cohesion (c) and internal
friction angle (phi) as the random variables, each of which can be defined having normal or
lognormal statistical distributions. Normal distribution has been widely used for statistical
distribution of soil properties (Lumb 1966, Schultze 1971, Harr 1989, Duncan 2000, Baecher and
Christian 2003, Cho 2009). For soil properties (which cannot have a negative value), lognormal
random field is frequently used to model the inherent spatial variability of geotechnical
parameters and has been shown to perform well in the geotechnical literature (Griffiths and
Fenton 2004; Cho 2009, 2010; Tabarroki et al. 2013, Jiang et al. 2014). In this study, unless
otherwise stated, normal (truncated) statistical distribution is utilized for soil parameters c and
phi.
The cohesion and internal friction angle of soil generally have an inversely correlated
relationship (Lumb 1970, Wolff 1985). Therefore, negative cross correlation is considered for
soil strength parameters in the probabilistic analyses in Slide v7. As for the value of this cross

ASCE RocNews Spring 2017


Geotechnical Frontiers 2017 GSP 278 365

correlation coefficient, Wolff (1985) stated that the coefficient is 0.25 based on CU triaxial test
results and -0.47 based on CD test results. Yucemen et al. (1973) and Lumb (1970) reported
coefficients between 0.24 and -0.49, and between -0.37 and -0.7 respectively. Cho (2010) used a
coefficient of -0.5 between cohesion and friction angle. In this study, cross correlation between
cohesion and friction angle has been considered with a coefficient of -0.5, and both cross
correlated and not correlated analyses are investigated.
For this research, probabilistic limit equilibrium slope stability analyses are performed
with circular failure surface search using both global minimum and overall slope methods for a
slope height of 10 m, with different slope angles (ranging from 45 to 70 with a 2.5 increment,
to represent slopes with different deterministic FS levels) and different COV levels (5 to 40%),
with and without cross correlation of shear strength parameters. Slope geometry can be seen in
Figure 4. The results of the analyses include FS, PF and reliability index (RI). According to the
analysis type, the results change such that global minimum analyses give just critical
deterministic surface, mean FS, PF and RI both normal and lognormal, while overall slope
analyses additionally give critical probabilistic surface and different global minimum surfaces.
The soil considered in this study is a generic soil with a cohesion and friction angle, the
values of which are taken from Bhattacharya et al. (2003). The soil is homogenous with mean
properties of cohesion value of 18 kN/m2, friction angle of 30 and unit weight of 18 kN/m3. In
this research, both dry slope and ru = 0.2 cases are investigated, where ru is defined as the average
ratio of pore water pressure to total vertical stress. Unit weight and ru are not considered as
random variables in this study. As for the coefficient of variation, five different values in the
range of 5% to 40% are used to represent different levels of variation in soil shear strength
properties. These values are in the range of reported values in the literature (Harr 1984, Kulhawy
1992, Phoon and Kulhawy 1999, Duncan 2000). Geotechnical and statistical distribution
parameters of generic c-phi soil are given in Table 1. To define the minimum and maximum
values in Slide v7, three times the standard deviation value is added/subtracted from the mean
value, truncating at negative values.
Table 1. Statistical Parameters for the soil used in this study.
Cohesion, c' (kPa) Friction Angle, ' ()
COV %
mean () Standard dev. () mean () Standard dev. ()
Min. 5 18 0.9 30 1.5
Low 13.75 18 2.475 30 4.125
Medium 22.5 18 4.05 30 6.75
High 31.25 18 5.625 30 9.375
Max. 40 18 7.2 30 12

RESULTS OF ANALYSES
The first series of analyses are carried out without cross correlation of shear strength properties.
For a 10-m high slope having ru = 0.2, with different slope angles in the range of 45 to 75
degrees, for a given slope angle, it is observed that as COV increases, PF versus COV graph
shows different behavior depending on the deterministic safety level (FS) of the slope (Figure 1).
For slopes having deterministic FS > 1.00 (slope angles in the range of 45 to 60 degrees), with
the increase of COV, PF increases. For slopes having deterministic FS < 1.00 (slope angles in the
range of 62.5 to 70 degrees) as COV increases, PF decreases (Figure 1). The results indicate that
COV effect reverses at the boundary of safe-unsafe slope (FS=1). Similar result has been

ASCE RocNews Spring 2017


Geotechnical Frontiers 2017 GSP 278 366

obtained by Griffiths and Fenton (2004) and Le et al. (2014) where they report that, an increase
of COV causes an increase of PF for FS > 1.00 slopes, and this trend reverses for FS < 1.00
slopes.

Figure 1. Probability of failure versus COV % for different slope angles.


When we study Figure 1, among relatively safer slopes (slope angles of 45 to 60 degrees),
the slope of PF-COV relation increases dramatically after a certain value of COV. For a slope at
60 degree angle, having a COV value of 0% or 5% will significantly influence PF, such that PF
will dramatically increase from 0% to 32%. Whereas for a relatively safer slope (slope angle of
45 degree), COV values of 0% and 20% do not have significant influence on PF, such that PF
will only change from 0% to 4%. This would indicate that, for slopes that are critical/close to
failure, reducing variability in soil shear strength with extensive site investigation etc (therefore
having less COV) could dramatically improve the probabilistic safety level of the slope, for
example in a slope stabilization project.
For a 10-m high slope having ru = 0.2, with different slope angles in the range of 45 to 75
degrees, Figure 2 presents the results on effects of COV for different FS values. It can be seen in
Figure 2 that, the relation between traditional deterministic FS and PF is nonlinear and the
nonlinearity of this relationship is dependent on COV level. For relatively low levels of COV
(such as COV=5%) the relation between FS and PF resembles a deterministic type relation, i.e.
when FS>1.00 then PF<50% (close to 0%), and when FS<1.00 then PF>50% (close to 100%). In
other words, theoretically, if COV=0%, deterministic and probabilistic slope stability analyses
will give same result and there will not be a need to carry out a probabilistic analyses. For
relatively high levels of COV (such as COV=40%) the FS-PF relation becomes more of a linear
relationship, and deterministically safe slopes having FS values of 1.06 and 1.36 correspond to
41% and 19% PF, respectively. This demonstrates that, large values of COV, i.e. significant
variation in material properties, can cause large values of PF, and deterministic FS alone will not
be sufficient to indicate this danger in slope stability.

ASCE RocNews Spring 2017


Geotechnical Frontiers 2017 GSP 278 367

Figure 2. The nonlinear relation between FS vs PF and the effect of COV%.


In reality, most soil parameters are not independent random variables, they are dependent
on each other. For example, an increase in unit weight most probably will indicate a denser soil
and slightly higher shear strength. Another example could be that, since Mohr-Coulomb failure
envelope is frequently simplified as a linear line, as cohesion increases, friction angle can be
expected to decrease. In this study, a negative cross correlation between cohesion and friction
angle is utilized and influence of cross correlation on PF is investigated, for a 10-m high slope
having ru = 0.2. The coefficient of negative cross correlation between cohesion and friction angle
is reported to change between -0.2 and -0.7 in the literature. In this research, the correlation
coefficient is taken as -0.5. The results presented in Figure 3 show that considering cross
correlation decreases PF for slopes having deterministic FS > 1.00, but slightly increases PF for
slopes having deterministic FS < 1.00. Le et al. (2014) and Jiang et al. (2014) stated that more
conservative results can be obtained without cross correlation of soil parameters.

(a) (b)
Figure 3. Effects of cross correlation on PF-COV relation (a) 45 degree slope, (b) 65 degree
slope.
In a deterministic limit equilibrium slope stability analysis, typically a search is carried
out and the most critical failure surface is found to be the one corresponding to the minimum FS

ASCE RocNews Spring 2017


Geotechnical Frontiers 2017 GSP 278 368

value. When probabilistic approach is used, it is seen that the failure surface having the lowest
deterministic FS value is not always the critical probabilistic surface which has the highest PF
value (Figure 4). Therefore, OS analysis is performed for this study, because it is more likely to
have several global minimum surfaces with different set of random variable samples and OS
analysis searches the whole slope for global minimum surface. In Figure 4, differences between
the most critical deterministic and probabilistic failure surfaces (obtained by Overall Slope
method) is shown for a 10-m high, 65 degree slope with ru=0.2 and COV=40% for cohesion and
friction angle, without cross correlation. It can be seen that although critical probabilistic surface
has slightly larger mean FS value (FS=0.915) than deterministic failure surface (FS=0.911), the
PF value is slightly greater. Cho (2009) and Cho (2010) reported similar results and stated that
generally probabilistic analyses are carried out on critical deterministic surface, but that surface
may not have the highest PF value. Bhattacharya et al. (2003), Griffiths and Fenton (2004), Cho
(2009) and Cho (2010) also state that the minimum reliability index which corresponds to the
highest PF value, occurs on critical probabilistic surface and deterministic slope stability
analyses do not seek out the surface with the highest PF value.

Figure 4. Critical deterministic and probabilistic failure surfaces.

CASE STUDY: SUGAR CREEK EMBANKMENT IN IOWA, U.S.A

In this section, the effects of COV levels and cross correlation are illustrated through a real life
case, Sugar Creek Embankment in Iowa, U.S.A. This case has been studied by Cho (2007) and
Akbas (2015). Cho (2007) focuses on the spatial correlation effects on PF and Akbas (2015)
considers this case for comparison of limit equilibrium, finite element and random finite element
methods. The soil parameters are taken from the study of Cho (2009). Figure 5 shows both the
geometry and soil parameters of this case. The COV levels are considered to be five different
values between 5% and 40% and assigned to the upper three layers at the same time. The soil
parameters of lower two layers are considered to be constant because the sliding surface does not
pass through these layers (failure surface passes through the highly weathered shale layer at its
lowest elevation).

ASCE RocNews Spring 2017


Geotechnical Frontiers 2017 GSP 278 369

Figure 5. Model Geometry and Soil Parameters


The deterministic FS value using the soil properties given in Figure 5, is found as 1.63
and FS(mean) value is found to be between 1.63 and 1.53 for different COV values, using OS-
type probabilistic slope stability analyses. Similar to the trends reported in the preceding sections
of this study, the results of this embankment case show that, increasing COV causes an increase
in PF value. When the cross correlation coefficient of -0.5 is used for the correlation of cohesion
and friction angle, the safety level of this slope increases and the probability of failure decreases
as expected.
It is seen that critical probabilistic surface does not coincide with the critical deterministic
surface (Figure 6). Although the critical probabilistic surface has higher PF, the FS value is much
greater than critical deterministic surface. The critical deterministic and probabilistic surfaces are
illustrated in Figure 6.

Figure 6. Critical deterministic and probabilistic failure surfaces.

ASCE RocNews Spring 2017


Geotechnical Frontiers 2017 GSP 278 370

CONCLUSIONS
This research shows the effects of different COV levels and effect of including cross correlation
of shear strength parameters in probabilistic slope stability analyses. Thousands of analyses were
performed by considering the following cases;
- Different slope angles: from 45 to 70 with a 2.5 increment, to represent different
deterministic safety levels
- Different COV levels: five levels between 5% and 40%
- Cases with dry soil and soils having a water table (ru=0.2)
- Not cross correlated and cross-correlated cohesion and friction angle
- Overall slope and global minimum type analyses
- Using Monte Carlo sampling method and normal statistical distribution
At the end of this study, the following conclusions are obtained;
- Deterministic limit equilibrium analysis is not always sufficient for determination of the
location of the most critical failure surface. The obtained surface is not always the critical
probabilistic surface having the highest PF value.
- PF and the critical failure surface is significantly influenced by COV level, the
consideration of cross correlation, and by deterministic safety (FS) level of the slopes.
- The inverse relation between FS and PF has been demonstrated to be nonlinear and the
COV level has significant effect on this relationship.
- Slopes traditionally considered as safe (deterministic FS>1.00) may have higher PF
values depending on the COV level.
- Increase of COV level causes increase in PF value for slopes with FS (deterministic)
value greater than 1.00. For slopes having FS (deterministic) value smaller than 1.00, the
effect is reverse.
- Considering cross correlation between cohesion and friction angle significantly decreases
PF for slopes having FS(deterministic)>1.00, but increases for others.
The results presented in this study could be useful for further understanding of probabilistic slope
stability and the effects of soil variability/uncertainty, with the aim of better geotechnical risk
evaluation and management.
REFERENCES
Akbas, B. (2015). Probabilistic slope stability analysis using limit equilibrium, finite element and
random finite element methods. MS. Thesis, Middle East Technical University,
Akbas, B. and Huvaj, N. (2015). Probabilistic Slope Stability Analyses Using Limit Equilibrium
and Finite Element Methods, Proc. 5th International Conference on Geotechnical Safety and
Risk, Eds. Schweckendiek et al., Rotterdam, the Netherlands
Baecher G. B. and Christian J. T. (2003). Reliability and Statistics in Geotechnical Engineering. John
Wiley & Sons Inc., Chichester, West Sussex, England.
Bhattacharya G., Jana D., Ojha S. and Chakraborty S. (2003). Direct search for minimum
reliability of earth slopes. Computers and Geotechnics, Vol. 30, Issue 6, pp. 455-462.
Cho S. E. (2007). Effects of spatial variability of soil properties on slope stability. Engineering
Geology, Vol. 92, Issue 3-4, pp. 97-109.
Cho S. E. (2009). Probabilistic stability analyses of slopes using the ANN-based response surface.
Computers and Geotechnics, Vol. 36, No. Issue 5, pp. 787-797.
Cho S. E. (2010). Probabilistic Assessment of Slope Stability That Considers the Spatial Variability
of Soil Properties. Journal of Geotechnical and Geoenvironmental Engineering, Vol. 136,
Issue 7, pp. 975-984.

ASCE RocNews Spring 2017


Geotechnical Frontiers 2017 GSP 278 371

deWolfe G. F., Griffiths D. V. and Huang J. (2010). Probabilistic and Deterministic Slope Stability
Analysis by Random Finite Element. GeoTrends 2010: The Progress of Geological and
Geotechnical Engineering in Colorado at the Cusp of a New Decade Conference, Colorado,
U.S.A., Section: Earth Retention and Slope Stability, pp. 99-111.
Duncan J. M. (2000). Factors of Safety and Reliability in Geotechnical Engineering. Journal of
Geotechnical Engineering, Vol. 126, Issue 4, pp. 307-316.
El-Ramly H., Morgenstren N. R. and Cruden D. M. (2005). Probabilistic assessment of stability of a
cut slope in residual soil. Geotechnique, Vol. 5, Issue 1, pp. 77-84.
Griffiths D. V. and Fenton G. A. (2004). Probabilistic Slope Stability Analysis by Finite Elements.
Journal of Geotechnical and Geoenvironmental Engineering, Vol. 130, Issue 5, pp. 507-518.
Hammah R. E., Yacoub T. E. and Curran J. H. (2009). Probabilistic Slope Analysis with the Finite
Element Method. Asheville 2009, North Carolina, U.S.A.
Harr M. E. (1984). Reliability-based design in civil engineering. 1984 Henry M. Shaw Lecture,
Department of Civil Engineering, North Carolina State University, Raleigh, N.C.
Harr M. E. (1989). Probabilistic estimates of multivariate analyses. Applied Mathematical
Modelling, Vol. 13, Issue 5, pp. 313-318.
Jiang S. H., Li D. Q., Cao Z. J., Zhou C. B. and Phoon K. K. (2014). Efficient System Reliability
Analysis of Slope Stability in Spatially Variable Soils Using Monte Carlo Simulation.
Journal of Geotechnical and Geoenvironmental Engineering, Vol. 141, Issue 2, 04014096.
Kulhawy F. H. (1992). On the evaluation of soil properties. ASCE Geotechnical Special
Publications, Vol. 31, pp. 95-115.
Le T. M. H., Sanchez M., Gallipoli D. and Wheeler S. (2014). Probabilistic modelling of auto-
correlation characteristics of heterogeneous slopes. Geomechanics and Geoengineering,
Vol. 10, Issue 2, pp. 95-108.
Lumb P. (1966). The variability of natural soils. Canadian Geotechnical Journal, Vol. 3, Issue 2,
pp. 74-97.
Lumb, P. (1970). Safety factors and the probability distribution of soil strength. Canadian
Geotechnical Journal, 7(3), 225242.
Phoon K. K. and Kulhawy F. H. (1999). Evaluation of geotechnical property variability. Canadian
Geotechnical Journal, Vol. 36, Issue 4, pp. 625-639.
Schultze E. (1971). Frequency distributions and correlations of soil properties. First International
Conference on Applications of Statistics and Probability to Soil and Structural Engineering
Proceedings, Hong Kong University Press, pp. 372-387.
Suchomel R. and Masin D. (2010). Spatial variability of soil parameters in an analysis of a strip
footing using hypoplastic model. Seventh European Conference on Numerical Methods in
Geomechanics Proceeding, Trondheim, Norway, Taylor & Francis Group, London, pp. 383-
388.
Tabarroki M., Ahmad F., Banaki R., Jha S. and Ching J. (2013). Determining the factors of safety of
spatially variable slopes modeled by random fields. Journal of Geotechnical and
Geoenvironmental Engineering, Vol. 139, Issue 12, pp. 2082-2095.
Wolff, T. H. (1985). Analysis and design of embankment dam slopes: A probabilistic approach.
Ph.D. thesis, Purdue University, Lafayette, Ind.
Yucemen, M. S., Tang, W. H., and Ang, A. H. S. (1973). A probabilistic study of safety and design
of earth slopes, Structural Research Series Vol. 402, University of Illinois, Urbana, Ill.

ASCE RocNews Spring 2017

You might also like