Risk Assessment in Geotechnical Engineering:

Stability Analysis of Highly Variable Soils

D.V. Griffiths1 ,F ASCE, Jinsong Huang2 ,M ASCE and Gordon A. Fenton3 ,M ASCE

1
Professor, Colorado School of Mines, 1610 Illinois Street, Golden, CO 80401;
d.v.griffiths@mines.edu
2
Research Professor, Colorado School of Mines, 1610 Illinois Street, Golden, CO
80401; jhuang@mines.edu
3
Professor, Dalhousie University, Halifax, NS, Canada, B3J 1Y9;
gordon.fenton@dal.ca

ABSTRACT: The paper will review the state-of-the-art in the use of finite element
methods for modeling geotechnical engineering problems involving non-typical
geometries and highly variable soil properties. Examples will focus on slope stability
analyses in which traditional limit equilibrium methods, and even well-established
risk assessment methodologies may lead to misleading results.

Keywords: Finite element method, Variable soils, Probability of failure,

Random Fields, Risk assessment.

INTRODUCTION
The finite element method offers a powerful alternative to classical limit
equilibrium methods of slope stability that have remained essentially unchanged for
decades. The method offers the following main advantages:

• No assumption needs to be made in advance about the shape or location of

the failure surface. The failure mechanism “seeks out” the weakest path
through the soil.

• Since there is no concept of slices in the finite element approach there is no

need for assumptions about slice side forces. The finite element method
preserves global equilibrium until “failure” is reached.

• If realistic soil compressibility data is available, the finite element solutions

will give information about deformations at working stress levels.

• The finite element method is able to monitor progressive failure up to and

including overall shear failure.
 Finite element slope stability analysis can hardly be considered a new technique.
The first paper to tackle the subject by Smith & Hobbs (1974) is over 35 years old
followed by an important paper on the topic by Zienkiewicz et al. (1975). Both of
these papers had a very significant influence on the author’s finite element slope
stability software developments over the years. Early publications date back to
Griffiths (1980) and the first ever published source code for finite element slope
stability appeared in the second edition of the text by Smith & Griffiths (1988, 2004).
Readers are also referred to Griffiths & Lane (1999) for a thorough review of how the
methodology works.

This paper will focus initially on demonstrating the use of the finite element
method as applied to slope examples that would not necessarily be amenable to
traditional limit equilibrium methods (LEM). The paper will then go on to discuss
risk assessment methods in geotechnical engineering, particularly for slope stability,
including the most recent developments that combine random fields with finite
element methods in the Random Finite Element Method (RFEM). Example will be
given of system slope reliability, where traditional methods may lead to quite
misleading results.

LONG SLOPES
How long is “infinite”?
It has been noted previously (e.g. Duncan & Wright 2005) that the infinite slope
assumptions can be expected to lead to conservative estimates of the factor of safety.
This is primarily due to support provided at the ends of a finite slope that is not
accounted for in the infinite slope model. Here we present some finite element slope
stability analyses on “long slopes” with uphill and downhill boundary conditions, to
assess the range of validity and conservatism of the “infinite slope” assumptions. The
main question to be addressed is; How long must a slope be for it to be considered
“infinite”?
A typical finite element mesh of 8-noded quadrilateral elements is shown in
Figure 1.

Fig. 1. Mesh of 8-node quadrilateral element for “long slope” analysis.

 For simplicity, we have considered an undrained clay slope for which the infinite
slope equation would give the following factor of safety
cu
FS = (1)
γ sat H cos β sin β

In this example, the properties shown in the caption of Figure 2 were held constant
while L H was gradually increased. As shown, the computed factor of safety
converged on the infinite slope solution of FS = 1.15 from equation (1) for L H
greater than about 16. As expected, the infinite slope solution is always conservative.

Fig. 2. Influence of length ratio on the computed factor of safety for a slope
with H = 2.5 m , β = 30
, cu = 25 kN / m 2 and γ sat = 20 kN / m 3 .

For example, with L H = 2 , the computed factor of safety was FS = 2.86 ; more
than twice the infinite slope value. A typical failure mechanism for a steeper slope is
shown in Figure 3. The figure indicates that as the slope gets longer, the infinite slope
mechanism starts to dominate and the “toe” failure at the downhill end becomes less
important.

Fig. 3. Deformed mesh at failure corresponding to a slope with L H = 2 and

β = 60° indicating a toe mechanism with FS = 1.58 .

Influence of slope angle

A curiosity of the infinite slope equation (1) as shown in Figure 4, is that for
constant H , γ sat and cu , the factor of safety starts to increase as the slope steepens in
the range β > 45° . This result seems counter intuitive since our experience of finite
slopes is that the factor of safety always falls as a slope gets steeper.

Fig 4. Influence slope angle on the factor of safety for an undrained clay slope
with H = 2.5 m , cu = 25 kN / m 2 and γ sat = 20 kN / m 3 .

An explanation of this effect for infinite slopes comes from the fact that as the
slope becomes steeper, the length of the potential failure surface available to resist
sliding is increasing at a faster rate than the down-slope component of soil weight
trying to cause sliding. Even a short slope analysis with L H = 2 demonstrates this
effect as shown in Fig. 4 (Griffiths et al. 2011a).

STRATIFIED SLOPES
James Bay Dike
The next example to be considered here is the James Bay Dike slope shown in Figure
5. The geometry has a terraced cross-section with four different soil types consisting
of cohesionless soil in the embankment and undrained clays in the foundation. This
profile has attracted considerable interest (see e.g., El Ramly et al. 2002, Duncan et al.
2003) because published LEM solutions that assumed circular failure mechanisms
(e.g. Bishop’s method), led to unconservative estimates of the factor of safety.
Although limit equilibrium procedures are available for estimating the factor of safety
associated with non-circular surfaces (e.g. Spencer’s method), it is still hard to
guarantee that the critical surface corresponding to the minimum factor of safety has
been found.

Fig. 5. FE geometry and soil properties assigned to the James Bay dike.

The benefits of the FE slope stability approach are even more striking in an
example such as this in which the factor of safety can be accurately estimated, and the
corresponding failure mechanism observed. The sudden displacement increase shown
in Figure 6 indicates that FS ≈ 1.27 and the deformed mesh at failure given in
Figure 7 clearly shows the anticipated non-circular critical failure mechanism.

Fig. 6. FE solution of James Bay Dike by strength reduction indicating

FS = 1.27 .

Fig. 7. Deformed mesh at failure demonstrating a non-circular failure

mechanism.
 Multiple failure mechanisms
As mentioned in the Introduction, the finite element method “seeks out” the most
critical failure path through the soil, and unlike many LEM approaches, does not
require the user to anticipate in advance where the critical failure mechanism might
lie. The example shown in Figures 8 makes this point quite clearly by demonstrating
multiple mechanisms, which all have the same factor of safety of FS = 1.38 . A
traditional approach could easily miss one or more of these surfaces, which could
lead to an unsafe design if the goal of the analysis, for example, was to identify
locations for possible soil reinforcement.

Fig. 8. Multiple failure mechanisms of an undrained slope.

Checkerboard slope stability analysis.

Soils and rocks are the most variable of all engineering materials, so when an
engineer chooses “characteristic values” of the soil shear strength for a slope analysis,
it is very likely that some parts of slope consist of soil that is stronger than the
characteristic values, and other parts that are weaker.

In this section we take a simple 2D undrained clay slope and assign the slope two
different properties arranged in a checkerboard pattern as shown in Figure 9.
Fig. 9. Slope stability analysis with checkerboard strength pattern. The darker
zones are stronger.

The 45° undrained clay slope has a height of H = 10 m a unit weight of

γ sat = 20 kN/m3 and a foundation depth ratio of D = 1.5 . The mean strength of
cu = 50 kPa was held constant, while the stronger soil was made stronger and the
weaker soil was made weaker. The results of the factor of safety analysis by strength
reduction are shown in Figure 10. Clearly the weaker soil “wins”! This trend will be
repeated when wider ranges of strength values are incorporated into an analysis, such
as later in this paper when we discuss random field modeling of soils.

Fig. 10. Influence of strength ratio in checkerboard slope analysis.

3D SLOPE STABILITY ANALYSIS

3D slope stability analysis has received considerable attention in the literature (e.g.
Hungr 1987, Seed et al. 1990, Duncan 1996, Stark and Eid 1998, Chen et al. 2005,
Griffiths and Marquez 2007, Michalowski 2010), yet the vast majority of slope
stability analyses in research and practice, are still performed in 2D under the
assumption of plane strain conditions. Even when 2D conditions are not appropriate,
3D analysis is rarely performed. There are a number of reasons for this. The majority
of work on this subject has shown that the 2D factor of safety is conservative (e.g.
lower than the “true” 3D factor of safety), and existing methods of 3D slope stability
analysis are often complex, and not well established in practice. A further
disadvantage of some 3D LEM approaches, is that being based on extrapolations of
2D “methods of slices” to 3D “methods of columns”, they are complex, and not
readily modified to account for realistic boundary conditions in the third dimension.
The advantages of FE slope stability methods become even more attractive in 3D.
Here we demonstrate when 3D may be justified, and also show that great care must
be taken in subscribing to the received wisdom that “2D is always conservative”.

When is plane strain a reasonable approximation?

The first issue addressed for a homogeneous slope, is to consider the question
“how long does a slope need to be in the third dimension for a 2D analysis to be
justified?” Figure 11 shows a simple mesh that might be used for a 3D slope analysis.

Fig. 11. FE mesh for 3D slope stability analysis using 20-node hexahedral
elements.

The boundary conditions are such that one side ( z = 0) is fully fixed and the other
( z = 0) (typo) allows vertical movement only implying a plane of symmetry. The
bottom ( y = D) of the slope is fully fixed, while the back ( x = 0) and front-side
( z = L 2) (typo) of the slope allow vertical movement only. The results from a series
of FE analyses with different depth ratios ( L H ) while keeping all other parameters
constant are shown in Figure 12. It can be seen that the factor of safety in 3D is
always higher than in 2D, but tends to the plane strain solution of FS = 1.25 for depth
ratios of the order L H > 10 . It is shown that results of the same slope with a coarser
mesh gave slightly higher values of FS .

Figure 12: Comparison of 3D and 2D solutions for a φu = 0 slope with

o

cu / (γ H ) = 0.20 .

Is plane strain conservative?

The assumption that 2D analyses lead to conservative factors of safety needs
some qualification. Firstly, a conservative result will only be obtained if the “most
pessimistic” section in the 3D problem is selected for 2D analysis (see e.g., Duncan
1996). In a slope that contains layering and strength variability in the third dimension,
this conservative 2D section may not be intuitively obvious. Secondly, the corollary
of a conservative 2D slope stability analysis is that back analysis of a failed slope will
lead to an unconservative overestimation of the soil shear strength (e.g. Arellano &
Stark 2000). Bromhead & Martin (2004) argued that some landslide configurations
with highly variable cross-sections could lead to failure modes in which the 3D
mechanism was the most critical. Other investigators have also indicated situations
where more critical 3D factors of safety were observed (e.g., Chen & Chameau 1982
and Seed et al. 1990).

Finite element slope stability analysis offers us the opportunity to perform

objective comparisons in which 2D and 3D factors of safety are compared for
variable soil conditions. This point is highlighted in the 3D example shown in Figure
13 which represents a 2:1 slope of height 10 m , foundation depth 5 m and a length
in the out-of-plane direction of 60 m with smooth boundary conditions.

Fig. 13. Three-dimensional slope mesh and at failure including an oblique layer
of weak soil. (bad figure quality: cannot see the numbers)

An oblique zone of weak soil (shaded black) with undrained strength

cu = 20 kN/m 2 has been introduced into the slope with the surrounding soil four
times stronger with cu = 80 kN/m 2 . The 3D factor of safety was found to be
approximately 1.5 and the mechanism clearly follows the weak zone as also shown
in Figure 13.

When 2D stability analyses are then performed on successive slices in the x − z

plane moving from y = 0 m to y = 60 m , the result shown in Figure 14 is obtained.
As a check, the 2D analyses were performed both by finite elements and by a
standard LEM. It can be seen that towards the boundaries of the 3D slope
( y < 21 m and y > 34 m ) where the majority of soil in the sections is strong, the 2D
results led to higher and therefore unconservative estimates of the factor of safety. On
the other hand, at sections towards the middle of the slope ( 21 m < y < 34 m ) where
there is a greater volume of weak soil, the 2D results led to lower, and therefore
conservative estimates of the factors of safety. The 2D factor of safety closely
approached unity at y = 29 m . An even more critical 2D plane however, is the oblique
one that runs right down the middle of the weak soil. This 2D plane has a 2.5:1 slope
and is flatter than the x − z planes considered previously. A 2D slope stability
analysis on this plane gives an even lower factor of safety of about 0.7 . This result,
also shown on Figure 14, is less than half of the factor of safety given by the 3D
analysis, and would be considered excessively conservative, even by geotechnical
design standards.

Figure 14: Factors of safety from 3D analysis and various 2D sections.

Even in the rather simple problem considered here, the results have shown a
quite complex relationship between 2D and 3D factors of safety. The results confirm
that 2D analysis will deliver conservative results, but only if the most pessimistic
plane in the 3D problem is selected. Even so, this result may lie well below the “true”
3D factor of safety. More importantly however, it has also been shown that selection
of the “wrong” 2D plane could lead to an unconservative result.

RISK ASSESSMENT IN GEOTECHNICAL ENGINEERING

Soils and rocks are the most variable of all engineering materials, yet this is often
coupled with inadequate site data. These factors combine to make geotechnical
engineering one of the most appropriate areas for the application of probabilistic
tools.
Risk assessment and probabilistic analyses in geotechnical engineering are
rapidly growing areas of importance and activity for practitioners and academics (e.g.
Baecher and Christian 2003, Fenton and Griffiths 2008). At a recent G-I specialty
conference called Georisk 2011 for example, several important state of practice
papers were presented (e.g. Christian and Baecher 2011, Lacasse and Nadim 2011,
Scott 2011) and in this GeoCongress 2012, Lacasse et al. 2012 have presented a
comprehensive review of the state of risk assessment and mitigation in geo-practice.
It is now commonplace for major geotechnical conferences to include sessions on risk
assessment in geotechnical engineering.

Of all areas of geotechnical engineering, slope stability analysis has received

greater attention using risk assessment tools than any other, since the concept of
replacing a “factor of safety” by a “probability of failure” is immediately appealing to
many engineers (see e.g. Alonso 1976, Catalan and Cornell 1976, Li and Lumb 1987,
Oka and Wu 1990, Chowdhury and Xu 1992, Mostyn and Soo 1992, Juang et al.
1992, Mostyn and Li 1993, Lacasse 1994, Lacasse and Nadim 1996, Liang et al 1999,
Malkawi et al. 2000, Griffiths and Fenton 2000,2004, Duncan 2000, El Ramly et al.
2002, Bhattacharya et al. 2003, Babu and Mukesh 2004, Jiminez-Rodriguez et al.
2006, Low and Tang 2007, Hong and Roh 2009, Griffiths et al. 2009a, Huang et al.
2010, Ching et al. 2010, Mbarka et al. 2010, Wang et al. 2011).

The Random Finite Element Method (RFEM)

The goal of a probabilistic slope stability analysis is to estimate the probability of
slope failure as opposed to the ubiquitous factor of safety used in conventional
analysis. Several relatively simple tools exist for performing this calculation that
include the First Order Second Moment (FOSM) method and the First Order
Reliability Method (FORM). The FORM method in particular has now been
developed to a quite significant level of sophistication to tackle correlation and
system slope reliability (e.g. Low et al. 2007, Low et al. 2011).

A legitimate criticism of these first order methods however, is that they are unable to
properly account for spatial correlation in the 2D or 3D random materials, and are
inextricably linking with “old fashioned” slope stability methods that involve simple
shapes for the failure surfaces (typically circular).

To overcome these deficiencies, a method called the Random Finite Element

Method (RFEM) that combines random field theory with deterministic finite element
analysis was developed by the authors in the early 1990’s and has been applied to a
wide range of geotechnical applications (e.g. Griffiths and Fenton 2007, Fenton and
Griffiths 2008). In a stability analysis, input to RFEM is provided in the form of the
mean, standard deviation and spatial correlation length of the soil strength parameters
which may consist of several layers with different statistical input parameters. In the
absence of site specific information, there is an increasing number of publications
presenting typical ranges for the standard deviation of familiar soil properties (e.g.
Lee et al. 1983).
In RFEM, local averaging is fully accounted for at the element level indicating
that the mean and standard deviation of the soil properties are statistically consistent
with the mesh density. Since the finite element method of slope stability allows
mechanisms to “seek out” the most critical path through the soil, the method offers
great promise for more realistic reliability assessment of slopes and other
geotechnical applications. The flow chart for a typical RFEM slope stability analysis
is shown in Figure 15.

Fig. 15. Flow chart for a typical RFEM slope stability analysis.

The RFEM codes developed by Griffiths and Fenton for a range of geotechnical
applications are freely available in source code from the authors’ web site at
www.mines.edu/~vgriffit/rfem. The 2D slope stability program is called
rslope2d. A couple of failure mechanisms computed using this program for slopes
with quite different spatial correlation lengths but with the same mean and standard
deviation of strength parameters are shown in Figure 16. The spatial correlation
length is expressed in dimensionless form relative to the height of the embankment,
e.g. ΘC = 0.5 means the spatial correlation length is 0.5H etc. It is seen that the
slope with the higher spatial correlation length in the lower figure gives a quite
smooth failure mechanism more like the classical “mid-point” circle. The soil with a
lower spatial correlation length in the upper figure however, displays a quite complex
system of interacting mechanisms which would defy analysis by any traditional LEM.
Fig. 16. Typical failure mechanisms from an RFEM analysis with two different
spatial correlation lengths.

Following the results of Griffiths and Fenton (2004), the RFEM results for an
undrained clay slope with a spatially random, lognormally distributed dimensionless
undrained strength given by C = cu ( γ sat H ) is shown in Figure 17. The computed
probability of failure by RFEM ( p f ) is given as a function of the spatial correlation
length ( ΘC = θ ln C H ) and the coefficient of variation (VC = σ C µC ) . It can be seen
that an increasing correlation length may either increase or decrease the slope failure
probability depending on the input coefficient of variation VC .

Fig. 17. Influence of the spatial correlation length and coefficient of variation on
the probability of failure of an undrained slope ( µC = 0.25 ) .
 In order to interpret these results, a couple of key deterministic solutions
considering a homogeneous soil, should be kept in mind. (i) if µC = 0.25 , FS = 1.47
and (ii) if µC = 0.17 , FS = 1.0 . The diverging results in Figure 17 can then be
explained by considering the limiting cases of ΘC → 0 and ΘC → ∞ . As ΘC → 0 ,
the local averaging is maximized, and the slope becomes essentially homogeneous
with a uniform strength given by the median of the strength. If the median falls below
0.17 , all simulations fail and p f → 1 . On the other hand, if the median is greater
than 0.17 , none of the simulations fail and p f → 0 . On the other hand, as
ΘC → ∞ , each simulation involves a uniform soil with the property varying from one
simulation to the next, so p f → P [C < 0.17 ] .

For example, in the case of µC = 0.25, VC = 0.5 , the parameters of the underlying
normal distribution of ln C are given as

µln C = ln µ X − ln {1 + VC 2 } = −1.498
1
2 (2)
σ ln C = ln {1 + VC 2
} = 0.472
hence

 ln 0.17 − µln C 
pf = Φ  = 0.28 (3)
 σ ln C 

which is shown as the asymptotic trend of the line corresponding to VC = 0.5 as

ΘC → ∞ in Figure 17.

On the other hand, as ΘC → 0 , the median of the shear strength is given by

Median C = exp( µln C ) = exp(−1.498) = 0.22 > 0.17 (4)

hence p f → 0 .

First order methods and single random variable Monte-Carlo methodologies that
treat each simulation as a homogeneous material can be considered special cases of
RFEM with ΘC → ∞ but cannot be guaranteed to deliver conservative results.
 Influence of Mesh Refinement
A commonly asked question of any finite element analysis, including RFEM, is
the extent to which mesh refinement and discretization errors affect the results. As
mentioned previously, the statistics of the random field mapped onto the finite
element mesh are adjusted in a consistent way to account for element size.

Fig. 18. Influence of mesh density on p f for an undrained slope with

ΘC = 1 and µC = 0.25

This is an integral part of the Local Average Subdivision method (Fenton and
Vanmarcke 2000). As for the overall discretization issue, Figure 18 shows the
influence of mesh refinement for two different cases. It can be seen that the finer
mesh gives somewhat higher values of p f , which is to be expected, since more paths
are available for failure to occur.

IMPORTANCE OF SPATIAL VARIABILITY

In the following section we present two short examples that emphasize the
importance of proper modeling of spatial variability in slope risk assessment.

Infinite Slope Example

Fig. 19. Geometry and parameters of an infinite slope

 This is one of the oldest and simplest types of slope problem in which the failure
mechanism is assumed to be purely translational with the failure plane at the base of
the layer. In the absence of pore pressures ( u = 0 ) , the factor of safety can be
expressed explicitly by the equation
c′ tan φ ′
FS = + (2)
γ H sin β cos β tan β
In this example (Griffiths et al. 2011b) the cohesion is defined
by µc′ = 10 kN/m 2 and σ c′ = 3.0 kN/m 2 and the tangent of the friction angle by
µ tan φ ′ = 0.5774 and σ tan φ ′ = 0.1732 . The remaining parameters are assumed to be
deterministic with values given by H = 5.0 m , β = 30° , and γ = 17.0 kN/m3 .
Substitution of these deterministic parameters and the mean values of the random
variables into Eq. (2) leads to a deterministic factor of safety of FS = 1.27 .
From Eq. (2), and assuming c′ and tan φ ′ are uncorrelated, we can estimate
the mean and standard deviation of FS by the FOSM as

µ c′ µ ′
µ FS ≈ + tan φ (3)
γ H sin β cos β tan β
2 2 (4)
 1  2  1  2
σ FS ≈   σ c′ +   σ tan φ ′
 γ H sin β cos β   tan β 

which gives µ FS = 1.27 and σ FS = 0.311

Assuming that FS is lognormal, the probability of failure is then given by

 µ 
p f =P [ FS <1] = P [ ln( FS ) < ln(1) ] = Φ  − ln FS  (5)
 σ ln FS 
where the mean and standard deviation of the underlying normal distribution of
ln( FS ) are given by µln(FS) = 0.2113 and σ ln(FS) = 0.2409 . After substitution

 0.2113 
p f = Φ − = Φ [ −0.8772] = 1 − Φ [ 0.8772] = 1 − 0.810 = 0.19 (6)
 0.2409 
hence the probability of failure is approximately 19.0%. It should be noted that
this result, being based on the deterministic Eq.(2), assumes failure always occurs at
the base of the layer.
The same problem was then solved using RFEM by including lognormal and
uncorrelated c′ and tan φ ′ and a range of spatial correlation lengths defined in
dimensionless form as Θ = θ H (assumed in this example to be the same for both
c′ and tan φ ′ ) . The results shown in Figure 20 indicate that the FORM (FOSM?)
results are consistently unconservative, but less so as Θ → ∞ . This is because in
RFEM, failure takes place along the weakest path, which doesn’t necessarily occur at
the base of the layer. For shorter values of Θ , the critical plane is more likely to
occur above the base and p f is higher. The figure also shows a typical random field
and failure plane from the RFEM Monte-Carlo analyses.

Fig. 20. Comparison of RFEM and FORM results for an infinite slope analysis.

Three dimensional slope reliability

Since the 2-d factor of safety is generally considered to be conservative,
practitioners are reluctant to invest in the more time-consuming 3-d approaches. A
key question to be addressed is, under what circumstances will the probability of
failure of a slope predicted by a full 3-d analysis be higher than that obtained from an
equivalent 2-d analysis?
In all the RFEM analyses that follow (Griffiths et al. 2009b), the bottom of the
mesh ( y = − H ) is fully fixed and the back of the mesh ( x = 0 ) is allowed to move
only in a vertical plane. Both “rough” and “smooth” boundary conditions have been
considered at the ends in the out-of-plane direction ( z = 0 and L ) . In the rough
cases the ends are fully fixed and in the smooth case, they are allowed to move only
in a vertical plane. It is noted that unlike the deterministic study shown previously,
there is no symmetry in the RFEM analyses due to the spatial varying soil properties.
In this study, it was determined that 2000 realizations of the Monte-Carlo process for
each parametric group, was sufficient to give reliable and reproducible estimates of
the probability of failure p f .
The undrained clay slope shown in Figure 21 demonstrates an important
characteristic in 3D slope analysis called the “preferred” failure mechanism width W .
This is the width of the failure mechanism in the z − direction that the finite element
analysis “seeks out”. Over a suite of Monte-Carlo simulations the average preferred
failure mechanism width is called Wcrit . It will be shown that this dimension has a
significant influence on 3D slope reliability depending on whether the length of the
slope L is greater than or less than Wcrit .

scale of cu
W
(kNm −2 )

z
L
x

Fig. 21. Slope failure with (isotropic) Θ = 2.0 and rough boundary condition
 The length ratio (see Fig. 11) is varied in the range 0.2 < L H < 16 to investigate
the influence of three-dimensionality, with results presented in Figure 22.

0.30
3-d RFEM (rough)
3-d RFEM (smooth)
0.25
2-d RFEM
0.20
pf

0.15

0.10

0.05

0.00
0 2 4 6 8 10 12 14 16
L/H

Fig. 22. Probability of failure versus slope length ratio

,

(Vcu = 0.5 Θ = 1.0, FS = 1.39 slope angle 2h : 1v ) (bad quality. What is theta?)

In the case of smooth boundary conditions, the p f of one slice ( L H = 0.2 )

in the 3-d analysis is equivalent to that given by a 2D RFEM analysis since the 3D
analysis is essentially replicating plane strain. It is also shown in the smooth case that
as L H is increased, p f initially decreases, reaching a minimum before rising to
eventually exceed the 2D value. In the rough case, p f is close to zero for a narrow
slice and increases steadily as L H is increased due to a gradual reduction in the
supporting influence of the rough boundaries in the 3D case. As the length ratio is
increased in both the rough and smooth cases, the 3-d p f eventually exceeds the 2D
value, indicating that 2D analysis will be always give unconservative results if the
slope is long enough. It may also be speculated that p f → 1 as L H → ∞
regardless of boundary conditions.

For the case of smooth boundary conditions, let us define the critical slope length
Lcrit and the critical slope length ratio ( L H )crit as being that value of L H for
which the slope is safest and its probability of failure p f a minimum. It will be
shown that this minimum probability of failure in the smooth case occurs when
Lcrit ≈ Wcrit . If we reduce the slope length ratio below this critical value ( L < Lcrit ) ,
the slope finds it easier to form a global mechanism spanning the entire width of the
mesh with smooth end conditions, so the value of p f increases, tending eventually to
the plane strain value. However, if we increase the slope length ratio above this
critical value ( L > Lcrit ) , the slope finds it easier to form a local mechanism. Since
L > Wcrit the mechanism has more opportunities to develop somewhere in
the z − direction hence p f again increases.

CONCLUDING REMARKS
The paper has demonstrated the power and advantages of the finite element method
for both deterministic and probabilistic slope stability analysis in highly variable soils.
Results were presented indicating the limitations of 2D analysis in infinite slope and
3D slope analysis. It was shown that 2D slope analysis is only conservative if the
most pessimistic plane in the 3D geometry is chosen. Even then, the result may be
excessively conservative. More seriously however, poor selection of the 2D plane for
analysis could lead to unconservative results.
Examples of slope risk analysis were presented using the random finite
element method (RFEM) developed by the authors. It was shown that single random
variable approaches can give unconservative results compared with RFEM using 2D
random fields. The key benefit of RFEM is that it does not require any a priori
assumptions related to the shape or location of the failure mechanism. In an RFEM
analysis, the failure mechanism has freedom to “seek out” the weakest path through
the random soil, which generally leads to more simulations reaching failure. The
importance of spatial variability was further demonstrated in two examples involving
an infinite slope and a 3D slope. In both cases, failure to account for spatial
variability could lead to unconservative results.

Acknowledgement
The authors wish to acknowledge the support of NSF grant CMMI-0970122 on
“GOALI: Probabilistic Geomechanical Analysis in the Exploitation
of Unconventional Resources" and KGHM Cuprum, Wrocław, Poland through the
Framework 7 EU project on “Industrial Risk Reduction”.
References
Alonso, E. E., (1976). “Risk analysis of slopes and its application to slopes in
Canadian sensitive clays.” Géotechnique, 26:453-472.
Arellano, D. & Stark, T.D. (2000). “Importance of three-dimensional slope stability
analysis in practice.” Slope Stability 2000, GSP no. 101, D.V. Griffiths et al.
(eds.), ASCE: 18-32.
Babu, G. L. S. and Mukesh M. D., (2004), “Effect of soil variability on reliability of
soil slopes.” Géotechnique, 54(5):335–337.
Baecher G.B. and Christian, J.T. (2003) “Reliability and statistics in geotechnical
engineering.” John Wiley & Sons, New York.
Bhattacharya, G. Jana, D. Ojha, S. and Chakraborty, S. (2003). “Direct search for
minimum reliability index of earth slopes.” Comput Geotech, 30(6): 455–462.
Bromhead, E.N. & Martin, P.L. (2004). “Three-dimensional limit equilibrium
analysis of the Taren landslide.” In Advances in Geotechnical Engineering
(Skempton Conference), Thomas Telford, vol. 2: 789-802.
Catalan, J. M. and Cornell, C. A. (1976). “Earth slope reliability by a level-crossing
method.”, ASCE J Geotech Eng Div, 102( GT6):691-604
Chen, R.H. & Chameau, J.L. (1985). “Three-dimensional limit equilibrium analysis
of slopes.” Géotechnique, 33(1): 31-40.
Ching, J.Y., Phoon K.K. and Hu Y.G. (2010) “Observations on Limit
Equilibrium-Based Slope Reliability Problems with Inclined Weak Seams.” J
Eng Mech, ASCE, 136 (10), pp.1220-1233.
Cho, S. E. (2007). “Effects of spatial variability of soil properties on slope stability.”
Eng Geol, 92 (3-4): 97–109.
Chowdhury R.N. and Xu D.W. (1992). “Reliability index for slope stability
assessment - 2 methods compared” Reliability Engineering & System Safety,
37(2): 99-108
Christian J.T. and Ladd C.C., Baecher G.B. (1994). “Reliability applied to slope
stability analysis.” J Geotech Eng ASCE, 120(12): 2180-2207.
Christian J.T. and Baecher G.B. (2011). “Unresolved Problems in Geotechnical Risk
and Reliability.”GeoRisk 2011, C.H. Juang et al. eds., GSP No. 224, ASCE CD
pp.50-63.
Duncan, J.M (1996). “State of the art: Limit equilibrium and finite-element analysis
of slopes.” J Geotech Geoenv, 122(7): 577-596.
Duncan, J.M (2000). “Factors of safety and reliability in geotechnical engineering.”
J Geotech Geoenv Eng, ASCE, 126(4): 307-316.
Duncan, J.M., Navin, M. & Wolff, T.F. (2003). “Discussion on Probabilistic slope
stability analysis for practice”. Can Geotech J, 40(4): 848–850.
Duncan, J.M. & Wright, S.G. (2005). “Soil strength and slope stability”. John Wiley
& Sons, Hoboken , New Jersey.
El-Ramly, H., Morgenstern, N. R., and Cruden, D. M. (2002). “Probabilistic slope
stability analysis for practice.” Can Geotech J, 39:665–683.
Fenton, G. A., and Griffiths, D. V., (2008). “Risk Assessment in Geotechnical
Engineering.” John Wiley & Sons, Hoboken, New Jersey.
Fenton G.A. and Vanmarcke E.H. (2000) “Simulation of random-fields via local
average subdivision.” J Geotech Eng, ASCE, 116(8):1733-1749.
Griffiths, D.V. (1980). “Finite element analyses of walls, footings and slopes”. Proc
Symp Comp Applic Geotech Probs Highway Eng, M.F. Randolph (ed.), PM
Geotechnical Analysts Ltd, Cambridge, UK: 122-146.
Griffiths, D. V., and Fenton, G. A. (2000). “Influence of soil strength spatial
variability on the stability of an undrained clay slope by finite elements.”
Proc GeoDenver 2000 Symposium. (eds. D.V. Griffiths et al.), Slope Stability
2000, GSP No. 101, ASCE, pp.184-193.
Griffiths, D. V., and Fenton, G. A. (2004). “Probabilistic slope stability analysis by
finite elements.” J Geotech Eng, 130(5): 507-518.
Griffiths, D. V. and Marquez, R.M. (2007). “Three-dimensional slope stability
analysis by elasto-plastic finite elements.” Géotechnique, 57(6): pp.537-546.
Griffiths, D.V., Huang, J. and Fenton G.A. (2009a) “Influence of spatial variability on
slope reliability using 2D random fields.” J Geotech Geoenv Eng, vol.135, no.10,
pp.1367-1378.
Griffiths, D.V., Huang, J. and Fenton, G.A. (2009b) “On the reliability of earth slopes
in three dimensions.” Proc R Soc A, vol.465, issue 2110, pp.3145-3164.
Griffiths, D.V., Huang, J. and deWolfe, G.F. (2011a) “Numerical and analytical
observations on long and infinite slopes.” Int J Numer Anal Methods Geomech,
35(5): 569-585.
Griffiths, D.V., Huang, J. and Fenton, G.A. (2011b) “Probabilistic infinite slope
stability analysis.” Comput Geotech, 38(4): 577-584.
Griffiths, D. V. and Lane, P. A. (1999). “Slope stability analysis by finite elements.”
Géotechnique, 49(3):387–403.
Hassan, A. M., and Wolff, T. F. (1999). “Search algorithm for minimum reliability
index of earth slopes.” J Geotech Geoenv Eng, 125(4):301–308
Hong, H. P. and Roh, G. (2008). “Reliability Evaluation of Earth Slopes.” J Geotech
Geoenv Eng, 134(12):1700-1705
Huang, J., Griffiths, D.V. and Fenton, G.A. (2010). “System reliability of slopes by
RFEM.” Soils Found, vol.50, no.3, pp.343-353.
Hungr, O. (1987). “An extension of Bishops simplified method of slope stability
analysis to 3 dimensions.” Géotechnique, 37(1):113-117.
Jimenez-Rodriguez R., Sitar N. and Chacon J. (2006) “System reliability approach to
rock slope stability.” Int J Rock Mech Min Sci. 43(6):847-859.
Juang C.H., Lee D.H. and Sheu C, (1992) “Mapping slope failure potential using
fuzzy-sets.” J Geotech Eng, ASCE, 118(3): 475-494.
Lacasse, S. (1994). “Reliability and probabilistic methods.” In Proc 13th Int Conf
Soil Mech Found Eng, New Delhi, India, pp. 225–227.
Lacasse, S., and Nadim, F. (2011). “Learning to Live with Geohazards: From
Research to Practice.” GeoRisk 2011, C.H. Juang et al., eds., GSP No. 224, ASCE
CD, pp.64-116.
Lacasse, S., Nadim, F., Høeg, K and Gregersen,O. (2012). “Risk assessment and
mitigation in geo-practice.” GeoCongress 2012, K. Rollins al., eds., ASCE CD.
Lacasse, S., and Nadim, F. (1996). “Uncertainties in characterizing soil properties.”
In C.D. Shackelford et al, editor, GSP No 58, Proceedings of Uncertainty ’96, pp.
49–75.
Lee, I. K., White, W., and Ingles., O. G. (1983). “Geotechnical Engineering.” Pitman,
London
Li, K. S., and Lumb, P. (1987). “Probabilistic design of slopes.” Can Geotech J,
24:520–531.
Liang R.Y., Nusier O.K. and Malkawi A.H. (1999) “A reliability based approach for
evaluating the slope stability of embankment dams.” Eng Geol, 54(3-4):271-285.
Low, B.K., Zhang, J. Tang, W.H. (2011). “Efficient system reliability analysis
illustrated for a retaining wall and a soil slope.”, Comput Geotech, 38(2),
196-204.
Low, B. K., Lacasse, S. and Nadim, F. (2007). “Slope reliability analysis accounting
for spatial variation.”, Georisk , 1(4), 177-189.
Low, B.K. and Tang W.H. (2007). “Efficient spreadsheet algorithm for first-order
reliability method.”, J Eng Mech, ASCE , 133(12), 1378-1387.
Mbarka S., Baroth J., Ltifi M., Hassis, H. and Darve, F. (2010) “Reliability analyses
of slope stability Homogeneous slope with circular failure.” European J Env Civ
Eng 14(10):1227-1257.
Malkawi A.I.H., Hassan W.F. and Abdulla F.A. (2000) “Uncertainty and reliability
analysis applied to slope stability.” Struc Safety, 22(2):161-187.
Michalowski, R.L. (2010) “Limit analysis and stability charts for 3D slope failures.”
J Geotech Geoenv Eng, 136(4): 583-593.
Mostyn, G. R., and Soo, S. (1992). “The effect of autocorrelation on the probability of
failure of slopes.” In 6th Australia, New Zealand Conference on Geomechanics:
Geotechnical Risk, pp. 542–546.
Mostyn, G. R., and Li, K.S. (1993). “Probabilistic slope stability -- State of play.”
Proc Conf Probabilistic Meth Geotech Eng, eds. K.S. Li and S-C.R. Lo, pub. A.A.
Balkema, pp. 89-110.
Oka Y. and Wu T.H. (1990). “System reliability of slope stability.” J Geotech Eng
ASCE, 116(8): 1185-1189.
Scott, G. (2011) “The practical application of risk assessment to dam safety.” GeoRisk
2011, C.H. Juang et al. eds., GSP No. 224, ASCE CD pp.129-168.
Seed, R.B., Mitchell, J.K. & Seed, H.B. (1990). “Kettleman Hills waste landfill slope
failure. II Stability Analysis.” J Geotech Eng ASCE, 116(4): 669-690.
Smith, I.M. & Hobbs, R. (1974). “Finite element analysis of centrifuged and built-up
slopes.” Géotechnique, 24(4): 531-559.
Smith, I.M. & Griffiths, D.V. (1988), (2004). “Programming the Finite Element
Method”. 2nd ed., 4th ed., John Wiley & Sons, Chichester, U.K.
Stark T.D. and Eid H.T. (1998). “Performance of three-dimensional slope stability
methods in practice.” J Geotech Geoenv Eng, 124(11): 1049-1060.
Wang Y., Cao Z.J. and Au S.K. (2011) “Practical reliability analysis of slope stability
by advanced Monte Carlo simulations in a spreadsheet.” Can Geotech J
48(1):162-172.
Zienkiewicz, O.C., Humpheson, C. & Lewis, R.W. (1975). “Associated and
non-associated viscoplasticity and plasticity in soil mechanics.” Géotechnique,
25(4): 671-689.