See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/264276822

Probabilistic and Deterministic Slope Stability Analysis by Random Finite

Elements

Conference Paper · January 2011

DOI: 10.1061/41144(391)9

CITATIONS READS
2 261

3 authors, including:

D. V. Griffiths J. Huang
Colorado School of Mines University of Newcastle
268 PUBLICATIONS   6,593 CITATIONS    121 PUBLICATIONS   1,293 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Infinite and long slopes View project

Backward Erosion Piping: Mechanics and Modeling View project

All content following this page was uploaded by D. V. Griffiths on 12 December 2016.

The user has requested enhancement of the downloaded file.

 Probabilistic and Deterministic Slope Stability Analysis
by Random Finite Elements

Giorgia F. deWolfe1 S.M. ASCE, D.V. Griffiths2 F. ASCE, Jinsong Huang3 M. ASCE
1
Civil Engineer, PhD, Bureau of Reclamation, Denver Federal Center, Denver, Colorado, U.S.A;
email: gdewolfe@usbr.gov
2
Division of Engineering, Colorado School of Mines, Golden, Colorado, 80401-1887, U.S.A; email:
dvgriffiths@mines.edu
3
Division of Engineering, Colorado School of Mines, Golden, Colorado, 80401-1887, U.S.A; email:
jhuang@mines.edu

ABSTRACT: Program PES (Probabilistic Engineered Slopes) provides a repeatable

methodology allowing the user to perform a slope stability analysis on a one-sided
and two-sided sloping structure using a deterministic or probabilistic approach.
Program PES, in contrast with other deterministic or probabilistic classical slope
stability methodologies, is cable of seeking out the critical failure surface without
assigning a pre-defined failure surface geometry. The probabilistic approach of
program PES applies the Random Finite Element Method (RFEM) by Griffiths and
Fenton (1993) taking into account the soil spatial variability and allowing the use of
different random fields to characterize the spatial variation of any material type. The
methodology is compared against the probabilistic approach proposed by the program
SLOPE/W version 7.14 (Geostudio Group, 2007), and demonstrates its potential for
predicting probability of failure (pf) in non-homogeneous soil structures characterized
by phreatic conditions and potential post-earthquake liquefiable conditions. The pf
results obtained by program PES have proved that underestimating the influence that
the soil material variability has on the computation of pf will lead to lower results of
probability and underestimate of the risk of slope instability. Program PES
capabilities could be used by the engineering practice to prioritize intervention
activities within a risk context.

INTRODUCTION

Stability analyses are routinely performed in order to assess the equilibrium

conditions of natural and manmade slopes. The analysis technique chosen depends on
both site conditions and the potential mode of failure, with careful consideration
being given to the varying strengths, weaknesses and limitations inherent in each
methodology.

The motivation driving this study is closely related to the assessment and mitigation
of the hazards caused by the instability processes and the important role that stability
analysis of slopes plays in civil engineering applications and design.

Page 1
 For many years the nature of geotechnical slope stability analysis has been
predominantly deterministic, whether performed using design charts or computers.
It is inherent in this type of approach that the parameters characterizing the soil
materials such as friction angle, cohesion, Young’s modulus, Poisson’s ratio, unit
weight and ground water are also treated as deterministic. Intuitively, it can be
recognized that, where there are materials more homogeneous than others in nature,
there are no perfectly homogeneous natural materials. The deterministic approach,
which does not allow any variation in the soil materials properties, clearly introduces
a high level of approximation to the analysis and characterization of slope stability.
The level of approximation can only be reduced if the natural variation of soil is
taken into account, allowing the soil to be characterized by a range of values for each
parameter instead of a single value.
Soil properties measurements are usually taken over a finite volume, which
represents a local average of the property with respect to the overall size of the site
domain. For this reason, the Local Average Subdivision (LAS) method (Fenton and
Vanmarcke, 1990) has been used to generate the random fields in all the
investigations presented in this work. The random field model provides a useful tool
for the generation of spatially variable soil properties. A random field is characterized
by sets of soil property values, which are randomly generated around their mean
value, and are mapped onto the finite element mesh creating a 2D model of variable
soil. Each set of property values (e.g. cohesion and friction) characterizes an element
within the domain analyzed. The Monte-Carlo method is lastly applied to this model
performing multiple random field realizations. The number of simulation that give a
Factor of Safety (FS)<1 divided by the total number of simulations represents the
probability of failure.
In the computation of slope stability and probability of failure certainly there are
many sources of uncertainty, in addition to those related to soil variability. In current
engineering practice, most slope stability analyses following a deterministic approach
or characterized by a 1D model, do not account for soil variability. The current work
will show that accounting for the influence of soil variability, varying the soil
strength parameters and using a 2D model, leads to more conservative probability of
failure results compared to those computed using classical approaches to geotechnical
problems.

PROGRAM PES CHARACTERISTICS

Deterministic Theory

Program PES (Probabilistic Engineered Slopes) coded in FORTRAN.95 allows the

user to perform a slope stability analysis on a one-sided and two-sided sloping
structure using a deterministic or probabilistic approach. A brief description of
program PES methodology is given below. For more detailed information on the
elastic- visco-plastic and the strength reduction algorithms used in this study the
reader is referred to Griffiths and Lane (1999) and Smith and Griffiths (2004).
As a first step program PES reads the geometry input parameters from the input data
file generating a finite element mesh of the problem. Subsequently the soils

Page 2
properties recorded in the input data file are assigned to the relative embankment and
foundation deterministic mesh regions and random fields.
The program can allow the analysis of a liquefiable layer ether in the foundation or
in the embankment as well as the partition of a homogeneous embankment into two
materials. Clearly these more complicated components are highly dependent on the
problem analyzed and require modifications of the main program code each time a
different problem is selected.
After the information from the input data files are read Program PES computes the
elastic stress-strain matrix, the shape function at the integrating points, the analytical
version of the stiffness matrix for an 8-node quadrilateral element, and the lower
triangular global matrix kv. Then the program generates the additional loading due
to free-standing water outside of the slope, as well as a pore pressure within the slope.
The water load is equal to the summation of gravity load and pore pressure load, and
is computed before being added to the total load already computed. The program
allows for the analysis of submerged slopes as well as slopes characterized by a
specific water table which can vary in elevation throughout the mesh.
Subsequently program PES computes the strength reduction factor and then
performs a check on whether or not the yield is violated according to the failure
criterion. The theory coded in this section of the program is described in more detail
in the following paragraphs.
Program PES models a 2D plane strain analysis of elastic-perfectly plastic soils
with a Mohr-Coulomb failure criterion using 8-node quadrilateral elements with
reduced integration (4 Gaussian-points per element) in the gravity load generation,
the stiffness matrix generation and the stress redistribution phases of the algorithm.
From the literature, conical failure criteria are the most appropriate to describe the
behavior of soils with both frictional and cohesive components, and the Mohr-
Coulomb criterion is known as the best of this group of failure criteria. Therefore the
program uses the Mohr-Coulomb criterion as failure mechanism in all cases. In terms
of principal stresses and assuming a compression-negative sign convention, the
Mohr-Coulomb criterion can be written as shown in Eq. 1

σ 1′ + σ 3′ σ 1′ − σ 3′
Fmc = sin φ ′ − − c ′ cos φ ′ (1)
2 2

where σ 1′ and σ 3′ are the major and minor principal effective stresses.
In cases where the soil is characterized by a frictionless component (undrained
clays) the Mohr-Coulomb criteria can be simplified into the Tresca criterion
substituting φ = 0 in Eq. 1 and obtaining Eq. 2,

σ (cos θ )
Ft = − cu (2)
3
The failure function F for both criteria can be interpreted as follows:

F<0 stresses inside failure envelope (elastic)

F=0 stresses on failure envelope (yielding)

Page 3
F>0 stresses outside failure envelope (yielding and must be redistributed)

The soil is initially assumed to be elastic and the model generates normal and shear
stresses at all Gauss-points within the mesh. These stresses are then compared with
the Mohr-Coulomb failure criterion.
The elastic parameter E ′ and υ ′ refer to Young’s modulus and Poisson’s ratio of
the soil, respectively. If a value of Poisson’s ratio is assumed (typical drained values
lie in the range 0.2 < υ ′ <0.3), the value of Young’s modulus can be related to the
compressibility of the soil as measured in a 1D oedometer (e.g. Lambe and Whitman
1969) as shown in Eq. 3,
E′ =
(1 + υ ′)(1 − 2υ ′) (3)
mυ (1 − υ ′)

where mv is the coefficient of volume compressibility.

In this study the parameters E ′ and υ ′ have the values of ( E ′ =105 kN/m2 and
υ ′ =0.3) respectively. The total unit weight γ assigned to the soil is proportional to
the nodal self-weight loads generated by gravity. The forces generated by the self
weight of the soil are computed using a gravity procedure which applies a single
gravity increment to the slope. The gravity load vector for a material with unit
weight γ is computed at the element level as shown in Eq. 4, and subsequently
accumulated from each element at the global level by integration of the shape
function [N] as shown in Eq. 5,

gravlo (e ) = γ ∫ N T dV e (4)
Ve

all
gravlo = ∑ γ ∫∫ [N ] dxdy
T
(5)
elemnts

where N represents the shape functions of the element and the superscript e refers to
the element number. This integral evaluates the volume of each element, multiplies
by the total unit weight of the soil and distributes the net vertical force consistently to
all the nodes.

Others have shown that in nonlinear analyses, the stress paths due to sequential
loading versus the path followed by a single increment to an initially stress-free slope
can be quite different; however the factor of safety appears unaffected when using
elasto-plastic models (e.g. Borja et al 1989, Smith and Griffiths 2004). It is also
important to remember that classical limit equilibrium methods do not account for
loading sequence in their solutions.

Page 4
 In the program the application of gravity loading is followed by a systematic
reduction in soil strength until failure occurs. This is achieved using a strength
reduction factor SRF which is applied to the frictional and cohesive components of
strength in the form of Eq. 6

 tan φ ′  c′
=φ ′f arctan
=   and c′f (6)
 SRF  SRF

The factored soil properties φ ′f and c′f are the properties actually used in each trial
analysis. When slope failure occurs, as indicated by an inability of the algorithm to
find an equilibrium stress field that satisfies the Mohr-Coulomb failure criterion
coupled with significantly increasing nodal displacements, the factor of safety is
given by Eq. 7

FS ≈ SRF (7)

In the literature this method is referred to as the “shear strength reduction technique”
(e.g. Matsui and San 1992).

The reduction of soil strength is followed in the program by the computation of the
total body load vectors. A description of generation of the body loads computed in
the program can be found in deWolfe (2010) and a detailed description of the
algorithm used in the program involving viscoplasticity can be found in Smith and
Griffiths (2004).
After the computation of body load vectors is completed the program generates the
graphic output files respectively a PostScript image of the nodal displacement vectors
and a PostScript image of the deformed mesh. The PostScript plot of the displaced
finite element mesh has an optional grey-scale representation of the material property
random field.

Probabilistic Theory

With regard to the probabilistic analysis computed by program PES, the

probability of failure can be calculated using two different approaches. When the
program is asked to compute the safety factor (FS) for each Monte-Carlo simulation,
the probability of failure is described by the proportion of Monte-Carlo simulations
with FS<1. When the program is asked to compute the probability without
determining the exact value of FS for each simulation, the probability of failure is
described by the proportion of Monte-Carlo slope stability analyses that failed. In this
case the SRF is equal to 1(no strength reduction is actually applied). In this case,
“failure” was said to have occurred if, for any given realization, the algorithm (Mohr-
Coulomb failure criterion) was unable to converge within 500 iterations.
The RFEM code enables a random field of shear strength values to be generated
and mapped onto the finite elements mesh, taking full account of element size in the

Page 5
local averaging process. In a random field, the value assigned to each cell (or finite
elements in this case) is itself a random variable.
The random variables can be correlated to one another by controlling the spatial
correlation length and the cross correlation matrix where the degree of correlation ρ
between each property can be expressed in the range of -1<ρ<1.
More generally the correlation coefficient between two random variables X and Y can
be defined by Eq. 8
COV [ X , Y ]
ρ XY = (8)
σ xσ y
where COV represents the covariance between the two variables X and Y and their
respective standard deviations σx and σy.

Due to the isotropic approach applied throughout this work the following
simplifications can be made with respect to the mean, standard deviation and the
spatial correlation length: µ x = µ y = µ z , σx = σy = σz , and θx= θy= θz .
Using an exponentially decaying (Markovian) correlation function, Eq. 8 can be
rewritten as in Eq. 9 and Eq. 10

2τ
−
θ ln c
ρ =e (9)

 2 
ρ = exp− τ x2 +τ y2  (10)
 θ ln c 

Where ρ is the familiar correlation coefficient, τ is the distance between two points in
the random field and θlnc represent the spatial correlation length.
The spatial correlation length (θ), also referred to in literature as the “scale of
fluctuation”, describes the distance over which the spatially random values will tend
to be significantly correlated in the underlying Gaussian field. Mathematically θ is
defined as the area under the following correlation function (e.g. Fenton and
Griffiths, 2008 from Vanmarcke, 1983);

∞ ∞
θ = ∫ ρ (τ )dτ = 2∫ ρ (τ )dτ (11)
−∞ 0

where τ represents the distance between two positions in the random field. A large
value of θ will imply a smoothly varying field, while a small value will imply a
ragged field.

Another important dimensionless statistical parameter involved in this probabilistic

approach is the coefficient of variation v, which for any soil property can be defined
as

Page 6
 σ
v= (12)
µ
where σ is the standard deviation and μ the mean value of the property.

In brief, the analyses involve the application of gravity loading, and the monitoring
of stresses at all the Gauss points. The program uses the Mohr-Coulomb failure
criterion, which if violated, attempts to redistribute excess stresses to neighboring
elements that still have reserves of strength. This is an iterative process which
continues until the Mohr-Coulomb criterion and global equilibrium are satisfied at all
points within the mesh under quite strict tolerances. Plastic stress redistribution is
accomplished using a visco-plastic algorithm with 8-node quadrilateral elements and
reduced integration in both the stiffness and stress redistribution parts of the
algorithm. For a given set of input shear strength parameters (mean, standard
deviation and spatial correlation length), Monte-Carlo simulations are performed until
the statistics of the output quantities of interest become stable.
A more comprehensive explanation of the random finite elements method,
including local averaging approach and discussion on spatial correlation length can
be found in Fenton and Griffiths (2008).

PROGRAM PES APPLICATIONS

Fruitgrowers Dam Deterministic and Probabilistic Slope Stability Analyses

In this section program PES is tested in the analysis of a dam case history.
Fruitgrowers Dam is located in Delta County, Colorado, 6.4 kilometers upstream
from Austin, Colorado on Alfalfa Run, a tributary of the Gunnison River. The dam
was constructed by the Bureau of Reclamation from 1938 to 1939 for the primary
purpose of irrigation. The crest of the dam is at elevation 1674.0 meters (5493 feet).
The dam has a structural height of 16.8 meters (55 feet), hydraulic height of 12.2
meters (40 feet), crest width of 7.6 meters (25 feet), and crest length of 463.3 meters
(1520 feet). An aerial view of Fruitgrowers dam is shown in Figure 1.
The dam is a compacted zoned earthfill structure consisting of a wide central core
protected by a riprap layer on the upstream slope and by a thin gravel shell on the
downstream slope. The embankment core is composed of clay, sand and gravel,
grading to gravel at the outer slopes as shown in Figure 2. A cut-off trench was
excavated to impermeable material. The trench has a bottom width of 2.4 meters (8
feet) and is located 10.7 meter (35 feet) upstream of dam centerline. The surficial
material beneath the dam shell upstream and downstream of the cut-off trench was
stripped to remove top soil and organic material.

Page 7
 FIG. 1. Aerial view of Fruitgrowers dam (Photo courtesy of the BOR)

1690 1690

1675

ELEVATION IN METERS
1675

1675
(GC)-(CL)
ELEVATION IN METERS

Em CL

1660 1660
Qsw Weathered Qsw
Shale Shale

1645 Km 1645

1630
1630

0 30 60 90
DISTANCE IN METERS

EM: Embankment material Qsw: Quaternary slope wash

Geologic label description: GC-CL: gravel with clay alluvium
and sand to lean clay Km: Mancos Shale
CL: Lean Clay Formation

FIG. 2. Cross section G-G’ showing post construction actual dimensions of

Fruitgrowers Dam

Page 8
 The case history of Fruitgrowers Dam was selected because past studies of the site
conducted by the Bureau of Reclamation presented possible post-earthquake
liquefiable conditions in the foundation.
A seismic hazard assessment, conducted by the Bureau of Reclamation (2003),
concluded that the background earthquake sources present in the area will not likely
result in a large liquefaction potential. In August 2004, to address new concerns
created by the presence of silty sand material on the dam abutment, a study was
conducted using data collected from five field explorations performed between 1980
and 1999.
The results of this latest study showed a low likelihood of foundation liquefaction at
the dam site. According to this study, to produce the failure of the embankment a
liquefied continuous lens, longer than 19.5 meters (64 feet), should be present in the
foundation under the right abutment, and from the drill log data collected on each
side of the embankment during the field explorations the presence of such a long
continuous layer is unlikely. As shown in Figure 3, a deterministic post liquefaction
FS of 1.05 was computed for the structure assuming the presence of a 18.3-meter (60-
foot) long liquefiable layer.

66
60

54 FRUITGROWERS DAM
48 DYNAMIC SLOPE STABILITY ANALYSIS
42
ELEVATION IN METERS

36
30
24

18
12
6

0
-6 DISTANCE IN METERS
-12
-15 0 15 30 45 60 75 90 105 120 135

Figure 3: Deterministic post-liquefaction steady state analysis computed in

2004 using the software SLOPE/W version 7.4

From the computer program SLOPE/W version 7.4, the method of analysis
used to compute this result was the Spencer method, coupled with a rigid block
theory technique for the evaluation of the failure surface.

In the “Evaluation of Liquefaction and Post Earthquake Stability” conducted by the

Bureau of Reclamation in August 2004, as well as in previous studies, the dam is
essentially modeled as a homogeneous embankment. Similar to the study conducted
in 2004 the geometry of the current model is based on cross section G-G, Figure 2

Page 9
(post construction actual dimensions) and also represents a homogeneous
embankment.
The phreatic condition characterizing the analysis is also adopted from the model
constructed in 2004 which shows the reservoir elevation at 1672 meters (5485) (top
of active conservation) with 2.44 meters (8 feet) of freeboard, and a downstream toe
water elevation of 1662 meters (5453 feet), 1.22 meter (4 feet) below ground surface.

This piezometric line was developed during a study also conducted in 2004
investigating the effect of the artesian pressure on the site foundation and
embankment structure (Technical Memorandum No. FW-8312-2, 2004). Figure 4
shows the piezometric line, the geometry and the major units characterizing the
deterministic model created in 2004.
The model representing Fruitgrowers Dam is characterized by the following 3
soil materials.
• The embankment core is composed of clay, sand and gravel, grading
to gravel at the outer slope.
• The foundation material consists of the Mancos Shale Formation (Km)
and is modeled with a thickness of 11 meters (36 feet).
• The Quaternary alluvium (Qal) is characterized by recent alluvial
deposits of the Alfalfa Run and is modeled with a thickness of about
1.83 meters (6 feet).

Before diving into the probabilistic analysis, initial deterministic static

analyses modeling pre- and post-liquefaction conditions were conducted using
program PES.

FRUITGROWERS DAM
DYNAMIC SLOPE STABILITY ANALYSIS

CREST EL. 1674.3 METERS

RESERVOIR EL. 1671.8 METERS

FIG. 4. Representation of the 2004 model used in the deterministic post

liquefaction analysis.

Page 10
 The soil properties used in the 2004 slope stability analysis to characterize the
embankment, foundation, and liquefiable layer are considered generally appropriate
for these two deterministic analyses and are summarized in Table 1.

Table 1. Deterministic soil properties used in the Fruitgrowers Dam pre and post-
liquefaction analyses

Material Unit weight φ ' (º) c ' (kPa)

(kg/m3)
Post liquefaction Embankment 2050 32 20.68
conditions Foundation 2082 30 0.05
Quaternary alluvium 2082 0 14.36
Pre liquefaction Embankment 2050 32 20.68
conditions Foundation 2082 30 0.05
Quaternary alluvium 2082 30 0.05

Subsequently the post liquefaction deterministic model was run using the
probabilistic capability offered in program PES.
The soil properties as probabilistic variables and their statistical parameters
used during the probabilistic analysis are summarized in Table 2.

Table 2. Probabilistic soil properties used in the Fruitgrowers post-liquefaction

analyses

σ σ
µ Distribution
Material characterize by characterize by
Type
lower v higher v
Embankment φ ' (º) 32 3.2 6.4 Lognormal
Embankment c ' (kPa) 20.68 2.07 4.14 Lognormal
Foundation φ ' (º) 30 6 15 Lognormal
Foundation c ' (kPa) 0.05 0.009 0.02 Lognormal
Quaternary alluvium φ ' (º) 0 0.2 0.5 Lognormal
Quaternary alluvium c ' (kPa) 14.36 2.87 7.18 Lognormal

The probabilistic analysis associates one random field with the embankment, one
with the foundation and the liquefiable layer is described by the foundation random
field which is modified to address the new values describing the liquefiable material.
In this probabilistic model only the strength parameters of friction and cohesion are
analyzed in a probabilistic approach; the other parameters, dilation angle, unit weight,
Young’s modulus and Poisson’s ratio are analyzed following a deterministic
approach.
To address the level of uncertainty incorporated with the mean values describing
the properties the same probabilistic model is run one time with a higher Coefficient
of Variation (v) and one time with a lower v. The v values chosen represent suggested

Page 11
values available in the literature for similar soil material. (e.g Phoon and Kulhawy,
1999). The v values used in this analysis for all material types are summarized in
Table 3

Table 3. v values characterizing Fruitgrowers probabilistic runs.

Material lower v higher v

Embankment φ (º) and c ' (kPa)
' 0.1 0.2
Foundation φ ' (º)characterizing
The v values 0.2
and c ' (kPa) the probabilistic analyses 0.5
were chosen evaluating suggested values
Quaternary alluvium φ (º) and c (kPa)
' ' 0.2 0.5

Another critical value in the analysis is the spatial correlation length used to
determine the soil spatial variability. The set of isotropic values chosen to investigate
the spatial correlation length θ for all probabilistic runs is reported in Table 4.

Table 4. Isotropic θ values characterizing Fruitgrowers spatial variation of soil.

θ= 1.22 m
θ= 7.62 m
θ= 18.288 m
θ= 30.48 m
θ= 60.96 m
θ= 91.44 m
θ= 152.4 m

All the probabilistic analyses are run using 1000 Monte-Carlo simulations. It
has been observed during this investigation that the probabilistic model representing
Fruitgrowers dam associated with 1000 Monte-Carlo simulations returns a probability
that can vary up to 2.7% as showed in Figure 5, which represent a repeatable
computation. During all probabilistic and deterministic analyses all soil properties are
considered uncorrelated between each other.
The results of the probabilistic analyses as well as the comparison with the results
generated by the program Slope\W version 7.14 are described in the following
section

Page 12
 Variability in pf results
30

25

pf (%) 20
2.7%
15

10
0 10 20 30 40 50 60
N runs
FIG. 5. Variability in pf results using 1000 Monte-Carlo simulations. To
recognize how much the pf computed by the Fruitgrowers model could vary in a
probabilistic setting the same data file was run 50 times.

Programs PES and SLOPE/W: Deterministic and Probabilistic Slope Stability

Results Comparison.

The result from the deterministic pre-liquefaction model run using program PES
shows a FS=1.66 (Figure 6) while the SLOPE/W result according to Spencer’s
Method returns a FS=1.746 (Figure 7). The deterministic post-liquefaction model
computed by PES returned a value of FS=1.09 (Figure 8) when the SLOPE/W result
on the same model according to Spencer’s Method returned a FS=1.06 (Figure 9).

Estimated FS=1.66

FIG. 6. Displacement file showing displacement associated with the deterministic

pre-liquefaction conditions at Fruitgrowers Dam.

Page 13
 FRUIT GROWERS DAM
STATIC SLOPE STABILITY ANALYSIS 2009
pre liquef action material properties

1.746

Non liquefiable Km Liquefiable weathered Km Embankment

Unit Weight: 2082 kg/m3 Unit Weight: 2082 kg/m3 Unit Weight: 2050 kg/m3
Cohesion: 0.0478 kPa Cohesion: 0.0478 kPa Cohesion: 20.68 kPa
Phi: 30° Phi: 30° Phi: 32°
FIG. 7. Graphic representation according to Spencer’s Method of the SLOPE/W
results describing the deterministic pre-liquefaction conditions at Fruitgrowers
Dam.

The loading applied in the post liquefaction analysis are vertical gravity load only.
The post liquefaction analysis results from both programs assumes the presence of a
liquefiable layer, 84.12 meter (276 feet) long (16.45 meter or 54 feet downstream
from the centerline of the dam), while the post liquefaction deterministic analysis
computed in the 2004 obtained a FS=1.05 assuming the presence of a continuous
liquefiable layer 18.29 meter (60 feet) downstream of the centerline of the dam.

In the probabilistic analysis computed by PES the deterministic variables are

characterized by the same values used in the post-liquefaction analysis and the
probabilistic values are described by the statistical parameters summarized in the
previous section. In the probabilistic analysis computed using SLOPE/W, the failure
surface associated with the FS of 1.06 (Figure 9) was chosen as the critical one to test
with the probabilistic approach offered by SLOPE/W.
Estimated FS=1.09

FIG. 8. Representation of the displacement associated with the deterministic

post-liquefaction conditions at Fruitgrowers Dam (program PES).
Page 14
 The soil properties statistical parameters and soil spatial variation parameters used
in this analysis are the same as those used in the analysis run with program PES, and
are summarized in Tables 2, 3, and 4.

3-Non liquefiable Km
1-Embankment 2-Liquefiable weathered Km
Unit Weight: 2082
Unit Weight: 2050 kg/m3 Unit Weight: 2082 kg/m3
kg/m3
Cohesion: 20.68 kPa Cohesion: 13.36 kPa
Cohesion: 0.0478kPa
Phi: 32° Phi: 0°
Phi: 30°

FIG. 9: Graphic representation according to Spencer’s Method of the SLOPE/W

3
results describing the deterministic post-liquefaction conditions at Fruitgrowers
Dam.

Tables 5 and 6 respectively summarize the results from the SLOPE/W analyses and
the analyses run with PES. Figure 10 shows a direct comparison of the results from
the two programs for both lower and higher v.

The results showed in Figure 10 outlines fundamental differences between the two
programs. A detailed effort has been made during this study to comprehend the
differences among the two programs, but while for the program PES a full version of
the program’s code is available, for the program SLOPE/W the author of this research
has to solely rely upon the program manual, published by Geostudio, which does not
provide detailed information on the program code.

Page 15
Table 5. Results from the Fruitgrowers probabilistic analyses run
with the program SLOPE/W.

(θ) m Low v High v

pf % pf %
1.22 3.8 20.12
3.05 12.53 34.23
4.57 19.37 39.02
6.09 23.37 43.48
7.62 26.41 45.48
9.14 28.47 45.95
10.67 28.21 46.39
12.19 28.69 46.4
15.24 29.14 46.35
152.4 29.28 46.72

Table 6. Results from the Fruitgrowers probabilistic analyses run

with the program PES.

(θ) m Low v High v

pf % pf %
1.22 94.7 98.6
7.62 72.9 95.8
18.29 70.3 89
30.48 66.7 82.7
60.96 66.4 78.6
91.44 65.9 77.9
152.40 67.3 73.5

The pf trend shown in Figure 10, corroborated by the trend results between program
PES and the program SLOPE/W compared in the probabilistic validation presented in
deWolfe (2010). The results presented in Figure 10 confirm that the probability of
failure computed by SLOPE/W is unconservative with respect to the probability of
failure estimated by program PES.
Figure 10 shows that for high values of spatial correlation the pf results from both
programs will show very little variation which is expected because high values of
spatial correlation correspond to a virtually homogeneous soil material at each
simulation. Lower values of spatial correlation instead emphasize a very different
trend between the two programs.

Page 16
 FIG. 10: Comparison of the results from programs PES and SLOPE/W for both lower and higher v.
PES results are based on the deterministic F.S of 1.09 and the SLOPE/W results on the deterministic F.S of 1.06
To gain a better prospective on the comparison between the element size and the spatial correlation length in this

Page 17
model, it is important to remember that a single square element size is equal to 0.91 meter (3 feet), and the total
dimensions of the problem are approximately, 21.95 meters (72 feet) in height and 146.3 meters (480 feet) in length.
 The trend showed in Figure 10, by SLOPE/W results, associates lower pf to a
highly spatially variable soil (low spatial correlation) and a higher pf with a more
homogeneous soil (high spatial correlation). In the other hand, program PES show
results which associate higher pf with more variable soils and lower pf to a more
homogeneous soil. As mentioned in program SLOPE/W manual, the program does
not apply any reduction to the standard deviation or the mean values of a random
property unless the length between two sections, ∆Z , is equal to or greater than the
scale of fluctuation or spatial variation length.
In the specific case of the model representing Fruitgrowers Dam the average
distance between two slices is approximately 1.22 meters and therefore no reduction
was ever applied to the standard deviation or the mean values of a random property
through all analyses. In general in the case of a deterministic FS>1 a random field
characterized by a reduced mean and variance values will lead to higher probability
of failure, and that could explain why the SLOPE/W results are consistently
unconservative with respect to the results computed by program PES. Instability in
the results produced by program PES can be observed when the spatial correlation
length value is equal to or smaller than the element size. In general, cases where the
element size is greater than the spatial correlation length do not represents a very
meaningful model, when instead, if many elements are able to define the variability
inside the spatial correlation length, this can be considerate a representative model.
Even for the cases when this may apply, one unstable result certainly cannot
in anyway change the overall interpretation of the analysis results trend.
Without a doubt it is quite difficult to determine the correct value of a soil variability
and this parameter represents a key component of this probabilistic analysis. Only
expert engineering judgment supported by exploration can truly lead to the
understanding of what that meaningful range of soil variability is for a specific
material. The results computed by the program PES and shown in Figure 10 clearly
emphasizes that not accounting properly for soil variability will lead to
unconservative results of pf or non-convergence and underestimate the probability of
slope instability. It needs to be remembered that the high probability of failure
computed by program PES associated with Fruitgrowers dam is strictly dependent on
the liquefaction of a continuous layer approximately 1.5 to 2 times the height of the
embankment. Even though the presence of potentially liquefiable material has been
corroborated by field testing in the area, the absolute continuity of the potentially
liquefiable layer still remains uncertain. Furthermore, based on the blow counts
values describing the strength of the weathered shale characterizing the potentially
liquefiable layer, liquefaction can occur only for an event associated with a high
seismic return period, such as the 50,000-year return period characterized by an
acceleration value of 0.27g. The probability of such event occurring in this area is
highly unlikely. For further information on the seismicity associated with
Fruitgrowers Dam the reader is referred to the Bureau of Reclamation seismic study
conducted in 2004 (Bureau of reclamation 2004).

Page 18
CONCLUDING REMARKS

Program PES provides a repeatable methodology able to improve the confidence

associated with the computation of probability of slope instability, which is a key
component of risk assessment for an engineering structure.
The probabilistic approach used in program PES applies a combination of the
random field technique and the finite element method.
At the core of the RFEM approach is the capability of accounting for spatially
random shear strength parameters and spatial correlation. This methodology
combines a non-linear elasto-plastic finite element analysis with random field theory
generated using the Local Average Subdivision Method (Griffiths and Fenton, 2004).
More specifically the spatially variable soil properties are correlated through the
parameter spatial correlation length or scale of fluctuation (θ), which indicates the
distance within which the values of a property show a relatively strong correlation,
and the parameter correlation coefficient (ρ). The main advantage of the RFEM over
traditional probabilistic slope stability techniques is that RFEM enables slope failure
to develop naturally by “seeking out” the most critical mechanism.
The methodology utilized in program PES is compared against the probabilistic
approach proposed by the program SLOPE/W version 7.14, and demonstrates its
potential for predicting probability of failure in a non-homogeneous soil structure
characterized by phreatic conditions and a possible liquefiable layer. While the results
computed from the deterministic analyses using programs PES and SLOPE/W show
a very close agreement, the results from the probabilistic analyses from the two
programs are generally in disagreement, and the SLOPE/W results consistently show
lower values of pf than obtained using program PES.

In the author’s opinion the difference in pf computed by the two programs can be
explained by the following three observations:
1. Both programs PES and SLOPE/W produce results of deterministic FS, pf , mean
and standard deviation of FS, but it is important to remember that, for both
probabilistic and deterministic analyses, program SLOPE/W represents a 1D
model of the soil property correlations along the potential failure surface, while
PES characterizes the soil property correlations using a 2D model. In the
probabilistic approach, the program PES investigates the soil variability through
the spatial correlation length over the entire foundation and embankment zones
while SLOPE/W investigates the soil variability only along the line characterizing
the critical slip surface.
2. Another major difference between the two programs is that SLOPE/W will
perform the probabilistic analysis on a failure surface found using traditional
slope stability methods (Jambu, Spencer, Bishop etc.) that require a subdivision of
the slope into columns, while the program PES based on a strength reduction
allows the modeled slope to fail naturally by “seeking out” the path of least
resistance of each Monte-Carlo simulations. In the author’s opinion, the number
of columns initially selected by the user in program SLOPE/W not only

Page 19
 influences the precision of the deterministic FS, but also influences the
computation of the probability of failure.
3. Another component that may lead to the low values of probability by SLOPE/W,
especially at lower values of the spatial correlation length (θ), is the difference in
the way local averaging is implemented in the two programs.

The establishment of a robust methodology provided by this research will not

only allow testing of the stability conditions of dams during modification phases, but
will also help estimate the probability of failure in cases involving post-earthquake
liquefaction. Although in the current study interest was concentrated on a classical
two-sided embankment geometry, the methodology can be applied to a wide range of
geotechnical engineering problems, taking into account the soil spatial variability and
its capability of “seeking out” the critical failure surface without assigning a pre-
defined failure surface geometry.

The current work has proven that not accounting for spatial variability can lead to
unconservative results with respect to more classical approaches computing
probability of failure in geotechnical problems.

Page 20
REFERENCES

Borja A.W., Lee S. R., and Seed R.B. (1989) “Numerical simulation and Excavation
in Elast-plastic soils”. Int J Numer Anal Methods Geomech, 13930;231-249.

Bureau of Reclamation (BOR), November 2003. “Seismic hazard Assessment,

Fruitgrowers Dam, Grand Mesa Project, Colorado” Technical Memorandum No. D-
8330-2003-29, Bureau of Reclamation, Internal publication, Denver, Colorado.

Bureau of Reclamation (BOR), August 2004. “Fruitgrowers Dam Issue Evaluation:

Evaluation of Liquefaction and Post Earthquake Stability” Fruitgrowers Dam Project,
Colorado, Upper Colorado Region.Technical Memorandum No.FW-8312-3 Bureau
of Reclamation, Internal publication, Denver, Colorado.

deWolfe, G. F. (2010). “ Probabilistic and Deterministic Slope Stability Analysis by

Finite Element. Unpublished dissertation. Colorado School of Mines, department of
Engineering.

Fenton, G.A. and Vanmarcke, E. H. (1990). “Simulation of random fields via local
average subdivision.” J Eng Mech, ASCE, Vol. 116, No. 8, pp. 1733-1749.

Fenton, G.A. and Griffiths, D.V. (2008). “Probabilistic Methods in geotechnical

Engineering” Springer Wien NewYork.

Geostudio Group, 2007. “SLOPE/W 2007 software for Slope Stability Analysis”
Geo-SLOPE International,

Griffiths, D.V., and Fenton, G.A. (1993). “Seepage beneath water retaining structures
founded on spatially random soil”, Géotechnique, 43(6), pp. 77-587.

Griffiths, D.V. and Lane, P.A., (1999) “Slope stability analysis by finite elements”.
Géotechnique, vol. 49, no. 3, pp. 387-403.

Griffiths, D.V. and Fenton, G.A., (2004). “Probabilistic slope stability analysis by
finite elements.” J Geotechnical and Geoenvironmental Engrg., vol.130, no.5,
pp.507-518.

Lambe, T. W., Whitman R. V., (1969). “Soil Mechanics” Wiley, New York, pp. 553.

Matsui, T. and San, K-C., (1992). “Finite element slope stability analysis by shear
strength reduction technique”. Soils and Foundations, vol. 32, no. 1, pp. 59-70.

Phoon, K.-K., and Kulhawy, F.H., (1999). “Characterization of geotechnical

variability”. Canadian Geotechnical Journal, 36, pp.612–624.

Page 21
 Smith I. M. and Griffiths D.V., (2004). “Programming the Finite Element Method, 4th
edn”, Wiley & Sons, West Sussex, England.

Vanmarcke E.H., (1983). “Random Fields: Analysis and Synthesis”. MIT Press,
Cambridge, Mass.

Page 22

View publication stats