You are on page 1of 8

57ième CONGRÈS CANADIEN DE GÉOTECHNIQUE 57TH CANADIAN GEOTECHNICAL CONFERENCE

5ième CONGRÈS CONJOINT SCG/AIH-CNN 5TH JOINT CGS/IAH-CNC CONFERENCE
Probabilistic Stability Analysis of an Embankment on Soft Clay
H. El-Ramly, AMEC Earth and Environmental, Edmonton, AB, Canada
N.R. Morgenstern, and D.M. Cruden, Department of Civil and Environmental Engineering, University of
Alberta, Edmonton, AB, Canada
ABSTRACT
A probabilistic slope analysis methodology based on Monte Carlo simulation using Microsoft Excel and @Risk software is
applied to study the performance of the Muar trial embankment in Malaysia. The study demonstrates the techniques used
in quantifying the various sources of uncertainty associated with the undrained shear strengths of clay foundations, and
estimating the probability of unsatisfactory performance. The relationships between embankment height, factor of safety
and probability of unsatisfactory performance at different stages during construction are examined. Sensitivity analyses
show that the uncertainties in Bjerrum’s vane correction factor and the contribution of embankment fill strength to stability
have significant impacts on the reliability of the factor of safety.
RÉSUMÉ
La performance d’un remblais d’essai à Muar, en Malaisie, est évaluée avec une méthode d’analyse probabiliste qui
incorpore la simulation Monte Carlo comme base, et utilise les logiciels Microsoft Excel et @Risk. L’article démontre les
techniques suivis pour quantifier les sources d’incertitudes reliées à l’estimation de résistance au cisaillement des
matériaux de fondations et à l’estimation de la probabilité d’obtenir une performance insatisfaisante. La relation entre
hauteur de remblais, facteur de sécurité et probabilité de performance insatisfaisante durant différents niveaux de
construction est examinée. Des analyses de sensibilité démontrent que la fiabilité du facteur de sécurité est fortement
affecter par certain incertitudes telles que ; le facteur de correction scissométrique de Bjerrum et par la contribution à la
résistance du matériel de remblais.
1. INTRODUCTION
Despite decades of accumulated experience, the ability of
the geotechnical profession to make reliable forecasts of
slope failures remains poor. As an example, in 1989 the
Malaysian Highway Authority built a full-scale trial
embankment in the valley of the Muar River to optimize
the design of a highway embankment on soft marine clay.
A detailed soil investigation program was conducted, and
extensive amounts of data were collected. Prior to
construction, thirty geotechnical consultants were invited
to predict the performance of the trial embankment,
including its height at failure. Brand and Premchitt (1989)
and Poulos et al. (1990) provided a discussion of the
predictions and the results of the field trial. The
embankment failed when the thickness of the placed fill
reached 5.4m. At the time of failure the average
settlement of the embankment was about 0.7 m, and the
height of the embankment above average ground was
4.7 m. Figure 1 is a histogram of predictions of
embankment height at failure. A significant scatter in the
predictions is evident, even though all participants had the
same site characterization data.
The difficulties in predicting field performances in
geotechnical engineering are attributed to the substantial
uncertainties that dominate the profession. Probabilistic
techniques offer an attractive framework for quantifying
and incorporating uncertainty in slope analysis and
design.
The stability of the Muar embankment is analysed using
the probabilistic methodology by El-Ramly et al. (2002a),
and the conventional approach based on the factor of
safety. The relationships between embankment height at
various stages during construction and the probability of
unsatisfactory performance (or failure probabilities) and
factor of safety are examined. The study is one of a series
of case histories (El-Ramly et al., 2002a; 2002b; 2003a)
illustrating the value of probabilistic techniques, and
providing guidelines for acceptable probabilities of
unsatisfactory performance.
Embankment Height (m)
F
r
e
q
u
e
n
c
y
12
10
4.7
0
0
Embankment Failed
6
Figure 1 Histogram of predictions of embankment height
above ground at failure (modified from Kay,
1993).
Session 3F
Page 14
-10
-5
0
5
10
15
0 10 20 30 40 50
Horizontal Distance (m)
E
l
e
v
a
t
i
o
n

(
60
m
)
Deterministic critical
slip surface
Hassan and Wolff
(1999) surface
Embankment
Very Soft
Silty Clay
Soft Silty Clay
Weathered Clay Crust
2
.0
:
1
2
.
0
:
1
Figure 2 Geometry and stratigraphy of the Muar embankment at failure
2. STRATIGRAPHY AND SOIL CHARACTERISTICS
The Muar Trial embankment had a base area of about
55 m x 90 m and side slopes of 2h:1v; a cross-section of
the embankment is shown in Figure 2. It was constructed
on a crust of weathered clay, about 2.0m thick, underlain
by very soft to soft clays. Two clay layers of different
properties were identified. Immediately below the
weathered crust, 6.0 m of highly plastic, very soft clay
were encountered followed by 9.5 m of soft clay of lower
plasticity and slightly higher strength. The clay sequence
was underlain by 0.7 m of peat on top of dense clayey
sand. Figure 3 shows soil stratigraphy and Table 1
summarizes the physical and mechanical characteristics
of clay layers. Detailed descriptions of the geology and
soil conditions are given in the report prepared by the
Asian Institute of Technology (AIT, 1989) and the
summary paper by Brand and Premchitt (1989).
The undrained shear strengths of subsurface soils were
assessed through 9 field vane soundings. Figure 3 shows
the variation of the undrained shear strength with depth.
The strength of the desiccated crust is much higher than
that of the underlying very soft clay, and decreases from
about 50 kPa at surface to 8 kPa at 2 m depth. The
strengths of the underlying clay layers increase almost
linearly with depth from about 6.5 kPa at 2 m below
surface to 34 kPa at 17 m depth.
3. SOURCES OF UNCERTAINTY
Construction of the embankment took 100 days and,
hence, undrained conditions prevailed at failure. The
stability of the embankment is governed by the undrained
shear strength of foundation soils, and the shear strength
of embankment fill. Uncertainties in shear strength
parameters are attributed to the inherent spatial variability
of the soil layers, and systematic sources of uncertainty
including biases in empirical factors used in deducing
strength values and statistical uncertainties due to limited
amounts of data available. In the following sections,
uncertainties in input parameters whose impacts on the
stability of the embankment could be significant are
quantified.
-17.5
-15.0
-12.5
-10.0
-7.5
-5.0
-2.5
0.0
2.5
0 20 40
Undrained Shear Strength; S (kPa)
u
Clay Crust
Very Soft
Silty Clay
Soft Silty
Clay
60
Dense Clayey Sand
t
C
= -1.77*Elev. + 7.21
t
Cr
= 15.73*Elev. + 2.79
E
l
e
v
a
t
i
o
n
(
m
)
t
C
c
C
Figure 3 Profile of field vane shear strength
Session 3F
Page 15
Table 1. Summary of soil properties of clay layers
Soil Layer LL
(%)
PL
(%)
LI
(%)
e ¸
b
(kN/m
3
)
Clay Crust 90 30 0.6 1.85 15.5
Very Soft Silty
Clay
80 27 1.5 2.60 14.0
Soft Silty Clay 55 23 1.4 1.65 16.0
LL, liquid limit; PL, plastic limit; LI, liquidity, index; e, void
ratio; ¸
b
, bulk unit weight.
3.1 Very Soft and Soft Clay Layers
Figure 3 shows hardly any distinction between the
undrained shear strengths of the very soft and the soft
clay layers in terms of trend and scatter of measurements.
Hence, the data of the two layers are combined and the
undrained shear strengths are treated as one population.
It should be noted that this approximation does not reflect
the geology on site as indicated by the different physical
properties of the two layers (Table 1). It is only a
modelling approach to simplify the statistical treatment of
the data. The validity of this simplification is examined in
the following paragraphs. For illustration purposes, all
parameters pertaining to the subject clay layers are
identified with a “
C
“ subscript.
The undrained shear strength of the clays is considered a
random variable and modelled using Equation 1,
) c + u = u =
C C C M - u C C - u
t S S ( [1]
where S
u-C
is the undrained shear strength at location i, u
C
is Bjerrum’s vane correction factor (Bjerrum, 1972), and
S
u-M
is the measured vane strength at the same location.
The measured vane strength is divided into a trend
component, t
C
, dependent on location, and a residual
component, c
C
, that characterizes the spatial variation of
strength over and above the trend value, as shown in
Figure 3.
Using the method of least squares, a linear trend is fitted
to the vane data (Figure 3). The mean values of the slope,
E[a
1-C
], and intercept, E[a
o-C
], of the trend line are
-1.77 kPa/m and 7.21 kPa, respectively. Since E[a
1-C
] and
E[a
o-C
] are estimated based on a limited amount of data,
they are also uncertain. Assuming a
1-C
and a
o-C
are
normally distributed random variables, statistical theory
(Neter et al. 1990) allows estimating the variances o
2
[a
1-C
]
and o
2
[a
o-C
] as follows;
| |
¯ ÷
=
2
2
C - 1
2
]) E[ (
ı
] [ ı
z z
S
i
M - u
a [2]
| |
)
`
¹
¹
´
¦
¯ ÷
+ =
2
2
2
C - o
2
]) E[ (
] E[
n
1
ı ] [ ı
z z
z
S
i
M - u
a [3]
where E[-] and o
2
[-] are the mean and variance,
respectively, z
i
is the depth, and n is the number of vane
measurements. The estimates a
1-C
and a
o-C
, are usually
correlated. The correlation coefficient p(a
1-C
, a
o-C
) is given
by;
] ı[ ] ı[
] [ ı ] E[
) , ȡ(
1 o
1
2
C - 1 C - o
a a
a
a a
z
÷ = [4]
where o[-] is the standard deviation. Using Equations 2
through 4, the standard deviations o[a
1-C
] and o[a
o-C
] are
0.07 kPa/m and 0.63 kPa, respectively, and the
correlation coefficient p(a
1-C
, a
o-C
) is 0.86.
The trend component is removed from all vane
measurements to estimate the statistical population of the
residual component c
C
. The residual component has a
constant mean of zero and a standard deviation, o[c
C
], of
2.45 kPa. Figure 4 shows the histogram of c
C
. It exhibits a
single peak indicating that vane shear strengths of the two
clay layers are consistent in a statistical sense.
E[ ] = 0.00 kPa
[ ] = 2.45 kPa
n = 61
c
c
C
C
o
Figure 4 Histogram of residual component of undrained
shear strength of clay layers; c
C
Based on a review of several embankment and cut slope
failures, Bjerrum (1972, 1973) recommended applying an
empirical correction factor, u, to vane measurements
Session 3F
Page 16
(Equation 1) depending on the plasticity index. Figure 5
shows Bjerrum’s database, as well as additional cases
compiled by Aas et al. (1986). The scatter of the data
around Bjerrum’s recommended relationship is
substantial. Hence, the uncertainty surrounding the vane
correction factor is significant and should be addressed in
any probabilistic assessment.
Figure 5 indicates a relatively uniform scatter of data
points around Bjerrum’s mean curve, irrespective of the
plasticity index. So, the vane correction factor can be
considered a random variable with a mean equal to
Bjerrum’s recommended curve and a constant standard
deviation deduced from the scatter of data around the
mean curve. Assuming u to be normally distributed, the
standard deviation o[u] is 0.15. Based on the plasticity
indices of the subject clay layers, the mean value of
Bjerrum’s vane correction factor, E[u
C
], is estimated to be
0.80.
0.4
0.8
1.2
1.6
0 50 100 150
Plasticity Index (%)
Bjerrum, 1972 & 1973
Compiled by Aas et al., 1986
V
a
n
e

C
o
r
r
e
c
t
i
o
n

F
a
c
t
o
r
;
u
Bjerrum, 1972
Figure 5 Bjerrum’s vane correction factor
3.2 Clay Crust
Published studies (e.g. Lefebvre et al., 1987) suggested
that the field vane overestimates the operational shear
strength of overconsolidated clay crusts due to the
presence of pre-existing fissures. Ferkh and Fell (1994)
recommended the use of a reduced strength for the crust.
Preliminary slope stability analyses of the Muar
embankment using an upper strength threshold of 14 kPa
for the weathered crust showed minimal impact, less than
4 percent, on the factor of safety. Hence, no threshold is
applied to the measured vane strength of the clay crust.
A similar approach to that in Section 3.1 is used to
quantify the uncertainties in the undrained shear strength
of the clay crust. For clarity, all parameters pertaining to
the clay crust are denoted with a “
Cr
” subscript. A linear
trend is fitted to the vane data as shown in Figure 3. The
mean values of the slope and intercept, E[a
1-Cr
] and
E[a
o-Cr
], are 15.73 kPa/m and 2.79 kPa, respectively.
Assuming normal distributions, the standard deviations,
o[a
1-Cr
] and o[a
o-Cr
], are 2.49 kPa/m and 3.62 kPa. The
correlation coefficient p(a
1-Cr
, a
o-Cr
) is -0.92. Removing
trend values from vane measurements, the mean and
standard deviation of the residual component, c
Cr
, are
zero and 4.55 kPa. Figure 6 shows the histogram of the
residual component. The uncertainty in Bjerrum’s vane
correction factor is represented by a normal probability
distribution with a mean of 0.75 (Figure 5) and a standard
deviation of 0.15.
E[ ] = 0.00 kPa
[ ] = 4.55 kPa
n = 11
c
c
Cr
Cr
o
Figure 6 Histogram of residual component of undrained
shear strength of clay crust; c
Cr
3.3 Embankment Fill Material
The embankment was constructed using compacted
clayey sand to sandy clay fill. As part of the site
investigation program, three unconsolidated undrained
triaxial tests were performed on samples of fill material.
The tests yielded undrained shear strength envelopes with
cohesion intercepts ranging between 64 and 19 kPa, and
corresponding friction angles ranging between 12 and 26
degrees. An approximate estimate of fill strength is
obtained by averaging shear strengths corresponding to a
confining stress equivalent to mid height of the
embankment from failure envelopes of three triaxial tests.
It is estimated to be 60 kPa.
The uncertainty in the operational shear strength of
embankment fill is attributed to two factors. First, the
procedure used above to estimate the average shear
strength is approximate, particularly since the initial
stresses within the embankment are not geostatic and
depend largely on the compaction energy. Second, the
strains experienced by the compacted fill, due to the
Session 3F
Page 17
distortion of the underlying soft clay, exceeds the strain
levels at peak strength. Hence, the operational shear
strength of the fill at failure is an unknown value between
the peak and residual strengths. In the stability analyses
in the following sections, the average shear strength of
embankment fill along the slip surface is considered a
random variable and modelled using a normal probability
distribution with a mean equal to 60 kPa. The standard
deviation is assigned, judgementally, a value of 12 kPa; a
coefficient of variation of 0.2.
4. PROBABILISTIC SLOPE ANALYSIS
4.1 Probabilistic Methodology and Input Variables
The stability of the Muar embankment at failure is
analysed using the probabilistic methodology developed
by El-Ramly et al. (2002a). Only brief descriptions of the
methodology are provided here and the reader is referred
to the original paper for more details.
The probabilistic methodology is spreadsheet-based and
makes use of the Microsoft Excel (Microsoft, 1997) and
@Risk (Palisade, 1996) software. The embankment
geometry, stratigraphy, soil properties, critical slip
surface(s), and selected method of slope analysis are
modelled in an Excel spreadsheet. The Bishop method of
slices (Bishop, 1955) is used in the spreadsheet. Three
input parameters are considered random variables; the
shear strength of the embankment fill, and the undrained
shear strengths of the clay crust and the underlying soft
clays. Using @Risk functions, each of these input
parameters is assigned appropriate probability
distributions. To account for the various sources of
uncertainty addressed in Section 3, each parameter is
represented by a number of variables. The uncertainties
due to the spatial variability of clay layers are represented
by the observed probability distribution functions of the
residual components, whose histograms are shown in
Figures 4 and 6. The biases in Bjerrum’s vane correction
factors and the statistical uncertainties in regression
coefficients of the linear trends of the undrained shear
strengths are represented by normal probability
distributions with the means and standard deviations
estimated in Section 3. The correlations between slopes
and intercepts of trend lines are established in the
spreadsheet using @Risk tools. Table 2 summarizes the
statistical parameters of all input variables.
4.2 Autocorrelation Distance and Spatial Variability
To account for the spatial variability of undrained shear
strengths along the slip surface, an estimate of the
autocorrelation of in-situ soils is required. In the absence
of adequate data to conduct a site specific assessment,
empirical estimates of horizontal and vertical
autocorrelations distances are made based on typical
values in the literature (El-Ramly et al., 2003). A range of
30-40 m is postulated as a possible range for the
horizontal autocorrelation distance and 1-3 m as a
possible range for the vertical autocorrelation distance. To
simplify the analysis of spatial variability, an equivalent
isotropic autocorrelation distance, rather than different
horizontal and vertical distances, is adopted. Using the
approximate procedure proposed by El-Ramly (2001) and
the typical ranges above, the equivalent isotropic
autocorrelation distance, r
o
, of the marine clays in the
foundation is in the range of 5-15 m. The probabilistic
analyses in the following sections are based on an
intermediate value of 10 m. The sensitivity of the outputs
to the autocorrelation distance is investigated in a latter
section.
The spatial variability of the residual components of
undrained shear strengths along the slip surface is
modelled using one-dimensional, stationary random fields
(Vanmarcke, 1983). At any given location along the slip
Table 2. Statistical parameters of input variables
Input variable Soil spatial variability Systematic uncertainty
E[--] o[--] PDF E[--] o[--] PDF
Embankment fill; S
u-Fill
(kPa) 60.0 12.0 N. -- -- --
Trend
a
1-Cr
(kPa/m)
-- -- -- 15.73 2.49 N.
a
o-Cr
(kPa)
-- -- -- 2.79 3.62 N.
Residuals; c
Cr
(kPa)
0.00 4.55 Exper.
(Figure 6)
-- -- --
Weathered clay
Crust
Bjerrum factor; u
Cr
-- -- -- 0.75 0.15 N.
Trend
a
1-C
(kPa/m)
-- -- -- -1.77 0.07 N.
a
o-C
(kPa)
-- -- -- 7.21 0.63 N.
Residuals; c
C
(kPa)
0.00 2.45 Exper.
(Figure 4)
-- -- --
Very soft/soft
clay layers
Bjerrum factor; u
C
-- -- -- 0.80 0.15 N.
Note: E[-], mean; o[-], standard deviation; PDF, probability density function; a
1
and a
o
, slope and
intercept of linear trends of undrained shear strengths; c, residual component; Exper., experimental
probability distribution; N, normal probability distribution.
Session 3F
Page 18
surface, the residual component c
C
, for example, is a
random variable, spatially correlated with the residual
components at adjacent locations. Instead of modelling
the point-to-point variation of the residual component, the
variability of its local averages over local domains within
the random field is considered (Vanmarcke, 1983). The
portion of the slip surface within each soil layer is divided
in the spreadsheet (using principles of analytical
geometry) into segments of lengths l
i
. The statistical
characteristics of local averages of c
C
include the
reduction in variance due to spatial averaging and the
correlation coefficients between averages over adjacent
segments of the slip surface. By taking segment lengths
l
i
equal to or less than the scale of fluctuation, o, the
variances of local averages approach the point variance
and the correlation coefficients between most of the local
averages approach zero (Vanmarcke, 1977 and; 1983).
Assuming the spatial structure of the residual
components follows an exponential autocovariance
function, the scale of fluctuation, o, is equal to twice the
autocorrelation distance; approximately 20 m.
4.3 Critical Slip Surface
Two candidate critical slip surfaces are considered; the
deterministic critical slip surface based on mean values
of input parameters, and the surface corresponding to
the minimum reliability index based on the Hassan and
Wolff (1999) algorithm (Figure 2). Preliminary Monte
Carlo simulations indicated that the probability of
unsatisfactory performance associated with the former is
higher than that of the latter. Hence, the deterministic
critical slip surface is considered the critical surface for
probabilistic slope stability analyses.
4.4 Monte Carlo Simulation
Having completed the spreadsheet, Monte Carlo
simulation is performed with @Risk drawing at random a
value for each input variable from its defined probability
distribution. In sampling the input probability distributions,
the correlations between variables are maintained. The
sampled input values are used to solve the spreadsheet
and calculate the corresponding factor of safety. The
process is repeated sufficient times to estimate the
statistical distribution of the factor of safety. Using 10,000
iterations, the mean and standard deviation of the factor
of safety are estimated to be 1.11 and 0.15, respectively.
The probability of unsatisfactory performance P
u
(probability of factor of safety less than one) is estimated
to be 23.8x10
-2
. Figure 7 shows the histogram of the
factor of safety. Since the simulation process is based on
random sampling of input variables, the calculated
probability of unsatisfactory performance is also a
variable. Based on the results of 25 simulations (using
different seed values), the mean probability of
unsatisfactory performance is 24.1x10
-2
. The reliability
index, another probabilistic safety indicator, is given by
þ={E[FS]-1}/o[FS], where E[FS] and o[FS] are the mean
and standard deviation of the factor of safety. It is
computed to be 0.72.
1.0
P
u
Seed = 31069
E[FS] = 1.11
[FS] = 0.15
P = 23.8x10
o
u
-2
Figure 7 Histogram of factor of safety at failure
(embankment height above ground = 4.7 m)
4.5 Calibration of Probability Figures
Providing a link between current slope design practice
and probabilistic slope stability analyses is essential in
explaining the meanings of the computed probabilities of
unsatisfactory performance and, hence, reduce the
hesitation of practitioners to use such techniques in
practice. The stability of the Muar embankment is
analyzed probabilistically and deterministically at various
stages during construction (i.e. different embankment
heights). Four embankment heights are considered; 2.5,
3.3, 3.5 and 4.0 m. The probabilistic analyses are
conducted following the same approach described
previously. Figure 8 shows the variation of the probability
of unsatisfactory performance and factor of safety with
embankment height. The plot indicates little increase in
the probability of unsatisfactory performance up to a
height of about 3.3 m followed by a sharp increase in
probability value as the embankment height exceeds
3.5m.
Based on the observed field performance of the Muar
embankment and typical factors of safety used in
practice, a 3.3 m high embankment (a deterministic
factor of safety of 1.42) would have been anticipated to
perform adequately in a conventional slope assessment.
Probabilistic slope stability analysis of this embankment
indicates a mean factor of safety of 1.42 with a 0.20
standard deviation. The probability of unsatisfactory
performance is 1.38x10
-2
, and the reliability index is 2.07.
For comparison, the probability of unsatisfactory
performance and the reliability index of the failed
embankment (4.7 m in height) are 24.1x10
-2
and 0.72,
respectively.
Session 3F
Page 19
0
0.1
0.2
0.3
2 3 4 5
Embankment Height Above Ground; H (m)
P
r
o
b
a
b
i
l
i
t
y

o
f

U
n
s
a
t
i
s
f
a
c
t
o
r
y
P
e
r
f
o
r
m
a
n
c
e
;

P
u
0.4
0.8
1.2
1.6
2.0
F
a
c
t
o
r
o
f
S
a
f
e
t
y
;

F
S
Acceptable
Design (H=3.3m)
Embankment
Failed (H=4.7m)
FS
P
u
Figure 8 Variation of the probability of unsatisfactory
performance and the factor of safety with
embankment height.
4.6 Impact of Autocorrelation Distance
The sensitivity of the estimated probability of
unsatisfactory performance of the 3.3 m high
embankment to the value of the autocorrelation distance,
r
o
, is investigated by repeating the probabilistic analysis
using different autocorrelation distances. Five
autocorrelation distances within the postulated range of
5 to 15 m (Section 4.2) are considered. The probabilities
of unsatisfactory performance range between 1.1x10
-2
for
r
o
=5 m, and 1.7x10
-2
for r
o
=15 m. From a practical
perspective, the probability of unsatisfactory performance
of the Muar embankment does not seem to be sensitive
to the autocorrelation distance of the marine clay
foundation. Addressing the stability of a tailings dyke on
a marine clay shale foundation, El-Ramly et al. (2003)
noted a similar conclusion.
4.7 Sensitivity Analysis
Using @Risk software, a sensitivity analysis is
undertaken to assess the relative contributions of input
variables to the uncertainty in the factor of safety.
Spearman rank correlation coefficient is used as the
analysis tool. It measures the strength of linear
correlations between input variables and the factor of
safety. Its numeric value ranges between 1 and –1. A
value of 1 indicates a perfect positive correlation, a value
of –1 indicates a perfect negative correlation, and zero
indicates no correlation. Figure 9 shows Spearman rank
correlation coefficients for all input variables.
The plot indicates that the impact of the uncertainty in
Bjerrum’s vane correction factor on the reliability of the
computed factor of safety is significant, and comparable
to that of the spatial variability of the undrained shear
strength. In other words, the use of Bjerrum’s factor has
introduced an additional element of uncertainty that is as
significant as the inherent spatial variability of the
undrained shear strength. This highlights the drawbacks
of using empirical factors and correlations without proper
understanding of their limitations and, more importantly,
their reliability. The plot also shows that the uncertainty in
the contribution of fill strength to stability has a large
impact on the reliability of the factor of safety; a
longstanding issue still facing practitioners in the design
of embankments on soft soils.
In conventional slope design practice, deterministic
estimates of the above two quantities, namely; the
reduction (if any) in the measured vane strength and the
contribution of fill strength to stability, ought to be made.
These estimates are largely subjective and depend on
the designer’s judgement and experience, which
explains, at least in part, the significant scatter in the
predictions of embankment height at failure in Figure 1.
a
o-C
c
C
- Segment 1
S
u-Fill
u
C
0.0 0.2 0.4 0.6 0.8
I
n
p
u
t

V
a
r
i
a
b
l
e
Spearman Rank Correlation Coefficient
u
Cr
c
C
- Segment 2
a
1-C
c
Cr
- Segment 2
a
1-Cr
a
o-Cr
c
Cr
- Segment 1
NOTE: “ - Segment 1” is the local
average of over the length of a
segment of the critical slip surface
c
c
C
C
Figure 9 Spearman rank correlation coefficients
between input variables and factor of safety
As expected, the uncertainty in the strength of the
weathered clay crust has a small impact on the reliability
of the computed factor of safety. Because of the
relatively large number of field vane measurements, the
statistical uncertainty in the mean trend of the undrained
shear strength of the soft clays also has a minimal
impact on the analysis.
Session 3F
Page 20
5. CONCLUDING REMARKS
In a similar study, El-Ramly et al. (2002b) conducted a
probabilistic slope analysis of the Lodalen slide. Based
on a back analysis of the failure, they redesigned the cut
slope (hypothetically) to a flatter inclination deemed
stable and undertook a probabilistic assessment of its
stability. In another study, El-Ramly et al. (2003)
presented a probabilistic analysis of a tailings dyke on a
pre-sheared clay-shale foundation. The dyke
performance was monitored during and after construction
and is deemed adequate. Table 3 summarizes the
results of the two case studies together with the results
of the Muar embankment at an intermediate height,
3.3 m, deemed stable.
Table 3. Safety indices of three case histories
Soil Layer E[FS] o[FS] P
u þ
Lodalen Slope 1.33 0.07 ~ 0.00 4.85
Tailings Dyke 1.31 0.14 0.16x10
-2
2.31
Muar Embankment 1.42 0.20 1.38x10
-2
2.07
The results in Table 3 indicate two important
observations. First, the probabilities of unsatisfactory
performance of adequate slopes appear to be much
higher than some of the recommendations reported in
the literature. For example, the US Corps of Engineers
(1995) recommended a probability of unsatisfactory
performance of 3x10
-5
or a minimum reliability index of
4.0 as design targets for a “Good” performance level.
Second, the factor of safety of the Muar embankment is
higher than those of the two other cases, yet its
probability of unsatisfactory performance is substantially
higher. This inconsistency is attributed to the significant
uncertainties in the analysis of the Muar embankment,
including Bjerrum’s vane correction factor and
embankment strength. Neither of these issues would be
addressed in conventional slope stability analyses based
on the deterministic factor of safety.
REFERENCES
Aas, G., Lacasse, S., Lunne, T., and Hoeg, K. 1986. Use
of in situ tests for foundation design on clay. In Use of in
Situ Tests in Geotechnical Engineering, Proceedings of
In Situ’86. Geotechnical Special Publication No. 6,
ASCE, Blacksburg, USA, June 23-25, pp 1-30.
AIT, 1989. Laboratory test data on soil samples from the
Muar flats test embankment. Asian Institute of
Technology. Research Report – Phase II
Bishop, A.W. 1955. The use of the slip circle in the
stability analysis of slopes. Geotechnique, 5(1): 7-17.
Bjerrum, L. 1972. Embankments on soft ground.
Proceedings of ASCE Specialty Conference on
Performance of Earth and Earth-Supported Structures.
Purdue University, Lafayette, USA, June 11-14. Volume
2. pp 1-54.
Bjerrum, L. 1973. Problems of soil mechanics and
construction on soft clays. Proceedings of 8th
International Conference on Soil Mechanics and
Foundation Engineering. Moscow. Volume 3. pp 111-
159.
Brand, E.W., and Premchitt, J. 1989. Comparison of the
predicted and observed performance of the Muar test
embankment. Proceedings of the Symposium on Trial
Embankments on Malaysian Marine Clays. Kuala
Lumpur, Malaysia, November. Volume 3.
El-Ramly, H. 2001. Probabilistic analyses of landslide
hazards and risks: Bridging theory and practice. Ph.D.
thesis, University of Alberta, Canada.
El-Ramly, H., Morgenstern, N.R., and Cruden, D. 2002a.
Probabilistic slope stability analysis for practice.
Canadian Geotechnical Journal, 39: 665-683.
El-Ramly, H., Morgenstern, N.R., and Cruden, D. 2002b.
Probabilistic Stability Analysis of Lodalen Slide.
Proceedings, 55th Canadian Geotechnical Conference,
Niagara Falls, Canada, October 20-23. Canadian
Geotechnical Society, pp 1053-1060.
El-Ramly, H., Morgenstern, N.R., and Cruden, D. 2003.
Probabilistic stability analysis of a tailings dyke on
presheared clay-shale. Canadian Geotechnical Journal,
40: 192-208.
Ferkh, Z., and Fell, R. 1994. Design of embankments on
soft clay. Proceedings of 13th International Conference
on Soil Mechanics and Foundation Engineering, New
Delhi, India, January 5-10. Volume 2. pp 733-738.
Hassan, A., and Wolff, T. 1999. Search algorithm for
minimum reliability index of earth slopes. Journal of
Geotechnical and Geoenvironmental Engineering,
ASCE, 125(4): 301-308.
Kay, J.N. 1993. Probabilistic design of foundations and
earth structures. In Probabilistic Methods in Geotechnical
Engineering, Li & Lo (eds.). Balkema, Rotterdam. pp
49-62.
Lefebvre, G., Pare, J.J., and Dascal, O. 1987. Undrained
shear strength in the surficial weathered crust. Canadian
Geotechnical Journal, 24: 23-24.
Microsoft, 1997. Microsoft Excel 97. Microsoft
Corporation, Redmond, Washington, USA.
Palisade Corporation, 1996. @Risk: risk analysis and
simulation add-in for Microsoft Excel or Lotus 1-2-3.
Palisade Corporation, Newfield, NY, USA.
Poulos, H.G., Lee, Y.C., and Small, J.C. 1990. Predicted
and observed behaviour of a test embankment on
Malaysian soft clays. Research Report No. R620,
University of Sydney.
Vanmarcke, E.H. 1977. Probabilistic modeling of soil
profiles. Journal of the Geotechnical Engineering
Division, ASCE, 103(11): 1227-1246.
US Army Corps of Engineers 1995. Introduction to
Probability and Reliability Methods for Use in
Geotechnical Engineering. Engineering Technical Letter
No. 1110-2-547, Washington DC.
Vanmarcke, E.H. 1983. Random fields: analysis and
synthesis. MIT Press, Cambridge, Massachusetts, USA.
Session 3F
Page 21

hence. It was constructed on a crust of weathered clay. Figure 3 shows the variation of the undrained shear strength with depth.0 tC -12.73*Elev.0 1 2. Two clay layers of different properties were identified. including biases in empirical factors used in deducing strength values and statistical uncertainties due to limited amounts of data available. a cross-section of the embankment is shown in Figure 2. SOURCES OF UNCERTAINTY -15.0m thick.5 Clay Crust 0. undrained conditions prevailed at failure. + 2.0 Dense Clayey Sand -17.0 m of highly plastic. and the shear strength of embankment fill. uncertainties in input parameters whose impacts on the stability of the embankment could be significant are quantified.21 -7. 1989) and the summary paper by Brand and Premchitt (1989). In the following sections. Su (kPa) Figure 3 Profile of field vane shear strength Construction of the embankment took 100 days and. and decreases from about 50 kPa at surface to 8 kPa at 2 m depth.5 0 20 40 60 Undrained Shear Strength. Elevation (m) The Muar Trial embankment had a base area of about 55 m x 90 m and side slopes of 2h:1v.0 tC = -1.77*Elev. Figure 3 shows soil stratigraphy and Table 1 summarizes the physical and mechanical characteristics of clay layers. and systematic sources of uncertainty Session 3F Page 15 . 6.5 kPa at 2 m below surface to 34 kPa at 17 m depth. The strengths of the underlying clay layers increase almost linearly with depth from about 6. Detailed descriptions of the geology and soil conditions are given in the report prepared by the Asian Institute of Technology (AIT. underlain by very soft to soft clays.79 -2.5 Soft Silty Clay C 3.0 tCr = 15. + 7.7 m of peat on top of dense clayey sand.15 10 Elevation (m) : 2. STRATIGRAPHY AND SOIL CHARACTERISTICS The undrained shear strengths of subsurface soils were assessed through 9 field vane soundings. The stability of the embankment is governed by the undrained shear strength of foundation soils. very soft clay were encountered followed by 9. 2. The strength of the desiccated crust is much higher than that of the underlying very soft clay.5 Very Soft Silty Clay -5.5 m of soft clay of lower plasticity and slightly higher strength.0 Deterministic critical slip surface Hassan and Wolff (1999) surface 5 Embankment :1 Weathered Clay Crust 0 Very Soft Silty Clay -5 Soft Silty Clay -10 0 10 20 30 Horizontal Distance (m) Figure 2 Geometry and stratigraphy of the Muar embankment at failure 40 50 60 2. Immediately below the weathered crust. The clay sequence was underlain by 0. about 2. Uncertainties in shear strength parameters are attributed to the inherent spatial variability of the soil layers.5 -10.

Using Equations 2 through 4.45 kPa n = 61 where Su-C is the undrained shear strength at location i. 3. bulk unit weight. of the trend line are -1. S u -C C S u-M C (t C C [1] E[ C] = 0. It is only a modelling approach to simplify the statistical treatment of the data.00 kPa [ C] = 2. statistical theory 2 (Neter et al.63 kPa. a1-C ) E[ z ] 2 [ a1 ] [ a o ] [ a1 ] [4] where [-] is the standard deviation. respectively. plastic limit. and Su-M is the measured vane strength at the same location. The trend component is removed from all vane measurements to estimate the statistical population of the residual component C.4 e 1. Bjerrum (1972. The correlation coefficient (a1-C. ( a o-C .45 kPa. The mean values of the slope. Figure 3 shows hardly any distinction between the undrained shear strengths of the very soft and the soft clay layers in terms of trend and scatter of measurements. b. a linear trend is fitted to the vane data (Figure 3). The residual component has a constant mean of zero and a standard deviation. Since E[a1-C] and E[ao-C] are estimated based on a limited amount of data. as shown in Figure 3. PL. all parameters pertaining to the subject clay layers are identified with a “C“ subscript. C. ao-C) is 0.M 1 n E[ z ] 2 2 b (kN/m ) 15. 2 2 Figure 4 Histogram of residual component of undrained shear strength of clay layers. 1973) recommended applying an empirical correction factor. C Based on a review of several embankment and cut slope failures. ao-C) is given by. to vane measurements [ a1-C ] Su . C is Bjerrum’s vane correction factor (Bjerrum. 1972). of 2. It should be noted that this approximation does not reflect the geology on site as indicated by the different physical properties of the two layers (Table 1).60 1. e. the data of the two layers are combined and the undrained shear strengths are treated as one population. liquidity. tC.6 1. The measured vane strength is divided into a trend component.0 3 ( z i E[ z ]) [3] LL.5 14. and the correlation coefficient (a1-C. index. and intercept. void ratio. LI. The estimates a1-C and ao-C. E[ao-C].07 kPa/m and 0. Summary of soil properties of clay layers 2 Soil Layer LL (%) Clay Crust 90 Very Soft Silty 80 Clay Soft Silty Clay 55 PL (%) 30 27 23 LI (%) 0.M E[ z ]) 2 (zi [2] Session 3F Page 16 .21 kPa.0 16. The validity of this simplification is examined in the following paragraphs. 1990) allows estimating the variances [a1-C] 2 and [ao-C] as follows. the standard deviations [a1-C] and [ao-C] are 0. respectively. respectively. are usually correlated.77 kPa/m and 7. The undrained shear strength of the clays is considered a random variable and modelled using Equation 1. . [ C].85 2.5 1.Table 1.1 Very Soft and Soft Clay Layers where E[-] and 2[-] are the mean and variance. E[a1-C]. that characterizes the spatial variation of strength over and above the trend value. and n is the number of vane measurements. Using the method of least squares. dependent on location. Assuming a1-C and ao-C are normally distributed random variables. For illustration purposes. liquid limit.65 [ a o-C ] 2 Su . they are also uncertain. and a residual component. zi is the depth. Hence.86. It exhibits a single peak indicating that vane shear strengths of the two clay layers are consistent in a statistical sense. Figure 4 shows the histogram of C.

62 kPa. the mean and standard deviation of the residual component. The tests yielded undrained shear strength envelopes with cohesion intercepts ranging between 64 and 19 kPa.15. The scatter of the data around Bjerrum’s recommended relationship is substantial. E[ Cr] = 0. Preliminary slope stability analyses of the Muar embankment using an upper strength threshold of 14 kPa for the weathered crust showed minimal impact. An approximate estimate of fill strength is obtained by averaging shear strengths corresponding to a confining stress equivalent to mid height of the embankment from failure envelopes of three triaxial tests. the strains experienced by the compacted fill. It is estimated to be 60 kPa. The Session 3F Page 17 . are 15. are zero and 4. 1987) suggested that the field vane overestimates the operational shear strength of overconsolidated clay crusts due to the presence of pre-existing fissures.55 kPa. are 2. irrespective of the plasticity index.2 Bjerrum. on the factor of safety. Assuming to be normally distributed. the procedure used above to estimate the average shear strength is approximate.1 is used to quantify the uncertainties in the undrained shear strength of the clay crust. As part of the site investigation program.73 kPa/m and 2. Based on the plasticity indices of the subject clay layers. A similar approach to that in Section 3. E[ C].6 Bjerrum. all parameters pertaining to the clay crust are denoted with a “Cr” subscript.75 (Figure 5) and a standard deviation of 0.79 kPa. Figure 5 indicates a relatively uniform scatter of data points around Bjerrum’s mean curve. the mean value of Bjerrum’s vane correction factor. Cr. Figure 5 shows Bjerrum’s database. less than 4 percent. Cr 3. [a1-Cr] and [ao-Cr]. 1. no threshold is applied to the measured vane strength of the clay crust.. The uncertainty in the operational shear strength of embankment fill is attributed to two factors.2 Clay Crust Published studies (e. Lefebvre et al.49 kPa/m and 3.15. respectively.92. the vane correction factor can be considered a random variable with a mean equal to Bjerrum’s recommended curve and a constant standard deviation deduced from the scatter of data around the mean curve. and corresponding friction angles ranging between 12 and 26 degrees. the standard deviations. due to the 3. the uncertainty surrounding the vane correction factor is significant and should be addressed in any probabilistic assessment. Figure 6 shows the histogram of the residual component. Assuming normal distributions. E[a1-Cr] and E[ao-Cr].8 0.(Equation 1) depending on the plasticity index.g.55 kPa n = 11 1. three unconsolidated undrained triaxial tests were performed on samples of fill material.80. So. A linear trend is fitted to the vane data as shown in Figure 3. the standard deviation [ ] is 0.4 0 50 100 150 Plasticity Index (%) Figure 6 Histogram of residual component of undrained shear strength of clay crust. The correlation coefficient (a1-Cr. Ferkh and Fell (1994) recommended the use of a reduced strength for the crust. 1972 0. ao-Cr) is -0. Hence. For clarity. mean values of the slope and intercept.3 Embankment Fill Material Figure 5 Bjerrum’s vane correction factor The embankment was constructed using compacted clayey sand to sandy clay fill. 1986 Vane Correction Factor. particularly since the initial stresses within the embankment are not geostatic and depend largely on the compaction energy. First. is estimated to be 0. (1986). Removing trend values from vane measurements.00 kPa [ Cr] = 4. Hence. The uncertainty in Bjerrum’s vane correction factor is represented by a normal probability distribution with a mean of 0.. as well as additional cases compiled by Aas et al. Second. 1972 & 1973 Compiled by Aas et al.

each parameter is represented by a number of variables. the operational shear strength of the fill at failure is an unknown value between the peak and residual strengths. mean. exceeds the strain levels at peak strength. -N.49 3. Hence. experimental probability distribution. . standard deviation.55 ---2. stationary random fields (Vanmarcke. The standard deviation is assigned. and selected method of slope analysis are modelled in an Excel spreadsheet. 4.2..distortion of the underlying soft clay. Very soft/soft clay layers Trend a1-C (kPa/m) ao-C (kPa) C Residuals. --Exper. rather than different horizontal and vertical distances. each of these input parameters is assigned appropriate probability distributions. ro. A range of 30-40 m is postulated as a possible range for the horizontal autocorrelation distance and 1-3 m as a possible range for the vertical autocorrelation distance. Exper. The sensitivity of the outputs to the autocorrelation distance is investigated in a latter section. residual components. whose histograms are shown in Figures 4 and 6. 2003). the equivalent isotropic autocorrelation distance. The biases in Bjerrum’s vane correction factors and the statistical uncertainties in regression coefficients of the linear trends of the undrained shear strengths are represented by normal probability distributions with the means and standard deviations estimated in Section 3. Session 3F Page 18 . the shear strength of the embankment fill.07 0. 1996) software. probability density function. a1 and ao. The Bishop method of slices (Bishop. of the marine clays in the foundation is in the range of 5-15 m. PDF.62 -0.80 -2. an estimate of the autocorrelation of in-situ soils is required. To account for the various sources of uncertainty addressed in Section 3. N. Using @Risk functions. (Figure 6) ---Exper. Table 2 summarizes the statistical parameters of all input variables. The spatial variability of the residual components of undrained shear strengths along the slip surface is modelled using one-dimensional. In the absence of adequate data to conduct a site specific assessment.0 12. Su-Fill (kPa) Weathered clay Crust Trend Soil spatial variability E[--] PDF [--] 60.15 -N. slope and intercept of linear trends of undrained shear strengths. N.21 -0. empirical estimates of horizontal and vertical autocorrelations distances are made based on typical values in the literature (El-Ramly et al. The correlations between slopes and intercepts of trend lines are established in the spreadsheet using @Risk tools. a coefficient of variation of 0. The uncertainties due to the spatial variability of clay layers are represented by the observed probability distribution functions of the Table 2.0 --4. Only brief descriptions of the methodology are provided here and the reader is referred to the original paper for more details. 4. is adopted.79 -0. Using the approximate procedure proposed by El-Ramly (2001) and the typical ranges above. Note: E[-]. At any given location along the slip The stability of the Muar embankment at failure is analysed using the probabilistic methodology developed by El-Ramly et al. In the stability analyses in the following sections. Three input parameters are considered random variables. (kPa) C Bjerrum factor. -N. an equivalent isotropic autocorrelation distance. The probabilistic analyses in the following sections are based on an intermediate value of 10 m. a1-Cr (kPa/m) --ao-Cr (kPa) Cr Residuals.. soil properties. Statistical parameters of input variables Input variable Embankment fill.75 -1.63 -0. (Figure 4) -Systematic uncertainty E[--] PDF [--] -15.15 0.2 Autocorrelation Distance and Spatial Variability 4. 1955) is used in the spreadsheet.00 ---0.73 2. the average shear strength of embankment fill along the slip surface is considered a random variable and modelled using a normal probability distribution with a mean equal to 60 kPa. 1997) and @Risk (Palisade. stratigraphy. The embankment geometry.1 PROBABILISTIC SLOPE ANALYSIS Probabilistic Methodology and Input Variables To account for the spatial variability of undrained shear strengths along the slip surface. N. residual component.00 -- Bjerrum factor. judgementally. normal probability distribution. critical slip surface(s). a value of 12 kPa. The probabilistic methodology is spreadsheet-based and makes use of the Microsoft Excel (Microsoft.45 -N. (2002a).77 7. N. [-]. (kPa) Cr 0. 1983). and the undrained shear strengths of the clay crust and the underlying soft clays. To simplify the analysis of spatial variability.

the variances of local averages approach the point variance and the correlation coefficients between most of the local averages approach zero (Vanmarcke. a 3. . the scale of fluctuation.11 [FS] = 0. Based on the observed field performance of the Muar embankment and typical factors of safety used in practice. the deterministic critical slip surface is considered the critical surface for probabilistic slope stability analyses.5m. the mean probability of unsatisfactory performance is 24.72.e.1x10-2. the deterministic critical slip surface based on mean values of input parameters. The statistical characteristics of local averages of C include the reduction in variance due to spatial averaging and the correlation coefficients between averages over adjacent segments of the slip surface.72. The plot indicates little increase in the probability of unsatisfactory performance up to a height of about 3.4 Monte Carlo Simulation 4.38x10-2.15 -2 Pu = 23. 1983). and the surface corresponding to the minimum reliability index based on the Hassan and Wolff (1999) algorithm (Figure 2). The probabilistic analyses are conducted following the same approach described previously.5 and 4.20 standard deviation. the mean and standard deviation of the factor of safety are estimated to be 1.0 Figure 7 Histogram of factor of safety at failure (embankment height above ground = 4. In sampling the input probability distributions.5. 4. the calculated probability of unsatisfactory performance is also a variable. respectively. for example. The stability of the Muar embankment is analyzed probabilistically and deterministically at various stages during construction (i. By taking segment lengths li equal to or less than the scale of fluctuation. respectively. The portion of the slip surface within each soil layer is divided in the spreadsheet (using principles of analytical geometry) into segments of lengths li. .3 Critical Slip Surface Seed = 31069 E[FS] = 1. It is computed to be 0. 2. the probability of unsatisfactory performance and the reliability index of the failed embankment (4. 1983). the correlations between variables are maintained.3 m followed by a sharp increase in probability value as the embankment height exceeds 3. Monte Carlo simulation is performed with @Risk drawing at random a value for each input variable from its defined probability distribution. and the reliability index is 2. Using 10. Probabilistic slope stability analysis of this embankment indicates a mean factor of safety of 1.7 m in height) are 24.3 m high embankment (a deterministic factor of safety of 1. Providing a link between current slope design practice and probabilistic slope stability analyses is essential in explaining the meanings of the computed probabilities of unsatisfactory performance and.8x10 Pu 1. reduce the hesitation of practitioners to use such techniques in practice. spatially correlated with the residual components at adjacent locations. where E[FS] and [FS] are the mean and standard deviation of the factor of safety. 4. For comparison. the variability of its local averages over local domains within the random field is considered (Vanmarcke.8x10-2.42) would have been anticipated to perform adequately in a conventional slope assessment. different embankment heights). Preliminary Monte Carlo simulations indicated that the probability of unsatisfactory performance associated with the former is higher than that of the latter.15. 3.000 iterations. Figure 7 shows the histogram of the factor of safety.3.11 and 0.7 m) Two candidate critical slip surfaces are considered. The probability of unsatisfactory performance is 1. 1977 and. hence. Assuming the spatial structure of the residual components follows an exponential autocovariance function. Four embankment heights are considered.5 Calibration of Probability Figures Having completed the spreadsheet. the residual component C.1x10-2 and 0. Hence. 3. The sampled input values are used to solve the spreadsheet and calculate the corresponding factor of safety. The probability of unsatisfactory performance Pu (probability of factor of safety less than one) is estimated to be 23. is equal to twice the autocorrelation distance.0 m. Instead of modelling the point-to-point variation of the residual component. is a random variable. The process is repeated sufficient times to estimate the statistical distribution of the factor of safety.42 with a 0. Session 3F Page 19 . is given by ={E[FS]-1}/ [FS]. The reliability index. approximately 20 m. Based on the results of 25 simulations (using different seed values). another probabilistic safety indicator.surface.07. Since the simulation process is based on random sampling of input variables. Figure 8 shows the variation of the probability of unsatisfactory performance and factor of safety with embankment height.

In conventional slope design practice.Segment 1 The sensitivity of the estimated probability of unsatisfactory performance of the 3.1x10-2 for ro=5 m. the probability of unsatisfactory performance of the Muar embankment does not seem to be sensitive to the autocorrelation distance of the marine clay foundation.2 FS 2. It measures the strength of linear correlations between input variables and the factor of safety. is investigated by repeating the probabilistic analysis using different autocorrelation distances. Session 3F Page 20 . Pu Acceptable Design (H=3. In other words. and comparable to that of the spatial variability of the undrained shear strength. The plot indicates that the impact of the uncertainty in Bjerrum’s vane correction factor on the reliability of the computed factor of safety is significant. and zero indicates no correlation.Segment 2 Input Variable Cr ao-C Cr .Segment 2 Cr a1-C a1-Cr ao-Cr 0.0.2 NOTE: “ C . the use of Bjerrum’s factor has introduced an additional element of uncertainty that is as significant as the inherent spatial variability of the 0.3 m high embankment to the value of the autocorrelation distance. which explains. Figure 8 Variation of the probability of unsatisfactory performance and the factor of safety with embankment height. at least in part. Five autocorrelation distances within the postulated range of 5 to 15 m (Section 4. the reduction (if any) in the measured vane strength and the contribution of fill strength to stability. (2003) noted a similar conclusion. The probabilities of unsatisfactory performance range between 1. Its numeric value ranges between 1 and –1.Segment 1” is the local average of C over the length of a segment of the critical slip surface Using @Risk software. Spearman rank correlation coefficient is used as the analysis tool. These estimates are largely subjective and depend on the designer’s judgement and experience. and 1.8 Embankment Failed (H=4. El-Ramly et al. FS C Su-Fill 4.3 Probability of Unsatisfactory Performance.7 Sensitivity Analysis . Factor of Safety. more importantly. 4. ought to be made.2 0. ro.0 0. their reliability.8 Spearman Rank Correlation Coefficient Figure 9 Spearman rank correlation coefficients between input variables and factor of safety As expected.4 0. Because of the relatively large number of field vane measurements. This highlights the drawbacks of using empirical factors and correlations without proper understanding of their limitations and.7x10-2 for ro=15 m. a sensitivity analysis is undertaken to assess the relative contributions of input variables to the uncertainty in the factor of safety.3m) 0.1 Pu 0 2 3 4 5 0. Figure 9 shows Spearman rank correlation coefficients for all input variables. the statistical uncertainty in the mean trend of the undrained shear strength of the soft clays also has a minimal impact on the analysis.2) are considered. Addressing the stability of a tailings dyke on a marine clay shale foundation. a longstanding issue still facing practitioners in the design of embankments on soft soils. the significant scatter in the predictions of embankment height at failure in Figure 1.4 Embankment Height Above Ground.Segment 1 .0 1.6 1. deterministic estimates of the above two quantities.7m) 0. a value of –1 indicates a perfect negative correlation.6 Impact of Autocorrelation Distance C C . From a practical perspective. A value of 1 indicates a perfect positive correlation.6 0. H (m) undrained shear strength. the uncertainty in the strength of the weathered clay crust has a small impact on the reliability of the computed factor of safety. namely. The plot also shows that the uncertainty in the contribution of fill strength to stability has a large impact on the reliability of the factor of safety.

Cambridge. (2003) presented a probabilistic analysis of a tailings dyke on a pre-sheared clay-shale foundation.C. 1990.H. 1996. June 23-25. NY.G. J. Volume 2.38x10 4. N. Based on a back analysis of the failure. N. Laboratory test data on soil samples from the Muar flats test embankment. Malaysia. Session 3F Page 21 . 5(1): 7-17. E. Probabilistic design of foundations and earth structures. In Probabilistic Methods in Geotechnical Engineering. D. REFERENCES Aas. and Cruden. Geotechnical Special Publication No. H. El-Ramly et al. Embankments on soft ground. Lacasse. Y... Palisade Corporation. Moscow. University of Sydney. H. Undrained shear strength in the surficial weathered crust.14 0. Proceedings of ASCE Specialty Conference on Performance of Earth and Earth-Supported Structures. Proceedings of 8th International Conference on Soil Mechanics and Foundation Engineering. El-Ramly. First. Bjerrum. El-Ramly et al. 1972. J. Morgenstern. and Fell.. USA. Canadian Geotechnical Society.31 2. Vanmarcke. 24: 23-24. Morgenstern. Newfield. G.. Probabilistic Stability Analysis of Lodalen Slide. 40: 192-208. Lee. Proceedings of the Symposium on Trial Embankments on Malaysian Marine Clays.. USA. and Cruden. (2002b) conducted a probabilistic slope analysis of the Lodalen slide. Corporation. 2003. 1987. D.R. Redmond.. In Use of in Situ Tests in Geotechnical Engineering. USA. Massachusetts. Problems of soil mechanics and construction on soft clays. Palisade Corporation. Research Report No. USA. and Dascal. 2001.. June 11-14. Morgenstern. the probabilities of unsatisfactory performance of adequate slopes appear to be much higher than some of the recommendations reported in the literature. Brand. 1993. 1977. Table 3 summarizes the results of the two case studies together with the results of the Muar embankment at an intermediate height. 1983. ASCE. Ferkh. 1989. and Cruden. Proceedings. Research Report – Phase II Bishop.). pp 49-62. University of Alberta. 1989. The dyke performance was monitored during and after construction and is deemed adequate. E. Proceedings of In Situ’86. Journal of the Geotechnical Engineering Division. Comparison of the predicted and observed performance of the Muar test embankment. Volume 3. ASCE. Proceedings of 13th International Conference on Soil Mechanics and Foundation Engineering. D. T. Canada. thesis. the US Corps of Engineers (1995) recommended a probability of unsatisfactory performance of 3x10-5 or a minimum reliability index of 4. A.R.. USA.D.0 as design targets for a “Good” performance level. Rotterdam. Probabilistic modeling of soil profiles. Niagara Falls. Poulos. Journal of Geotechnical and Geoenvironmental Engineering. Search algorithm for minimum reliability index of earth slopes. H. R620. 1110-2-547. Volume 3. El-Ramly. Asian Institute of Technology. MIT Press. Z. S. Blacksburg. 6. the factor of safety of the Muar embankment is higher than those of the two other cases. H. AIT. G.00 -2 0. Canadian Geotechnical Journal. O. Probabilistic slope stability analysis for practice. Random fields: analysis and synthesis.07 0.85 2..16x10 -2 1. Purdue University. 1986. In another study. January 5-10. Li & Lo (eds.31 1. Probabilistic stability analysis of a tailings dyke on presheared clay-shale. deemed stable. This inconsistency is attributed to the significant uncertainties in the analysis of the Muar embankment. Canadian Geotechnical Journal. CONCLUDING REMARKS In a similar study. yet its probability of unsatisfactory performance is substantially higher. Lafayette.20 Pu 0. Canadian Geotechnical Journal. Bjerrum. 103(11): 1227-1246.. The use of the slip circle in the stability analysis of slopes. Kuala Lumpur. A. US Army Corps of Engineers 1995. India. H. Kay. pp 1-54. 1997.33 1. @Risk: risk analysis and simulation add-in for Microsoft Excel or Lotus 1-2-3. Canada. Engineering Technical Letter No. and Hoeg. they redesigned the cut slope (hypothetically) to a flatter inclination deemed stable and undertook a probabilistic assessment of its stability.N.. including Bjerrum’s vane correction factor and embankment strength.42 [FS] 0. Balkema.H. Washington. and Premchitt. Design of embankments on soft clay. Predicted and observed behaviour of a test embankment on Malaysian soft clays. K. El-Ramly. Second. Safety indices of three case histories Soil Layer Lodalen Slope Tailings Dyke Muar Embankment E[FS] 1.R. Microsoft. Microsoft Excel 97.. ASCE. 3. El-Ramly. E. Use of in situ tests for foundation design on clay. L. Washington DC. 1973. pp 733-738. and Wolff. November.J. Geotechnique.. and Small. J.3 m. Neither of these issues would be addressed in conventional slope stability analyses based on the deterministic factor of safety.... 2002a. pp 1-30. New Delhi. 1955. 125(4): 301-308. Microsoft Table 3. R. J. 55th Canadian Geotechnical Conference. Pare. pp 1053-1060. N. 39: 665-683.W. Lefebvre. Lunne.5. October 20-23. L. Vanmarcke. 1994. 2002b. Probabilistic analyses of landslide hazards and risks: Bridging theory and practice. T. Hassan. Introduction to Probability and Reliability Methods for Use in Geotechnical Engineering. 1999. Volume 2.W. pp 111159.07 The results in Table 3 indicate two important observations. For example. Ph.C.