Você está na página 1de 19

Color profile: Generic CMYK printer profile

Composite Default screen

665

Probabilistic slope stability analysis for practice


H. El-Ramly, N.R. Morgenstern, and D.M. Cruden

Abstract: The impact of uncertainty on the reliability of slope design and performance assessment is often significant.
Conventional slope practice based on the factor of safety cannot explicitly address uncertainty, thus compromising the
adequacy of projections. Probabilistic techniques are rational means to quantify and incorporate uncertainty into slope
analysis and design. A spreadsheet approach for probabilistic slope stability analysis is developed. The methodology is
based on Monte Carlo simulation using the familiar and readily available software, Microsoft® Excel 97 and @Risk.
The analysis accounts for the spatial variability of the input variables, the statistical uncertainty due to limited data,
and biases in the empirical factors and correlations used. The approach is simple and can be applied in practice with
little effort beyond that needed in a conventional analysis. The methodology is illustrated by a probabilistic slope
analysis of the dykes of the James Bay hydroelectric project. The results are compared with those obtained using the
first-order second-moment method, and the practical insights gained through the analysis are highlighted. The
deficiencies of a simpler probabilistic analysis are illustrated.

Key words: probabilistic analysis, slope stability, Monte Carlo simulation, spatial variability.

Résumé : L'impact de l'incertitude sur la fiabilité de la conception et de l'évaluation de la performance des talus est
souvent significative. La pratique conventionnelle des talus basée sur le coefficient de sécurité ne peut pas traiter de fa-
çon explicite l'incertitude, compromettant ainsi la précision des projections. Les techniques probabilistes constituent des
moyens rationnels pour quantifier et incorporer l'incertitude dans la conception et l'analyse des talus. On développe une
approche de tableur pour l'analyse de la stabilité des talus. La méthodologie est basée sur la simulation de Monte Carlo
au moyen de logiciels familiers et facilement accessibles, Microsoft® Excel 97 et @Risk. L'analyse tient compte de la
variabilité spatiale des variables entrées, de l'incertitude statistique due aux données limitées et aux distorsions des fac-
teurs empiriques et des corrélations utilisés. L'approche est simple et peut être mise ne pratique avec peu d'effort en
plus de celui requis par une analyse conventionnelle. La méthodologie est illustrée par une analyse probabiliste des ta-
lus des digues du projet hydroélectrique de la Baie James. Les résultats sont comparés avec ceux obtenus en utilisant
la méthode du second moment de premier ordre, et les éclaircissements pratiques obtenus par cette analyse sont mis en
lumière. On illustre les déficiences d'une analyse probabiliste plus simple.

Mots clés : analyse probabiliste, stabilité de talus, simulation de Monte Carlo, variabilité spatiale.

[Traduit par la Rédaction] El-Ramly et al. 683

Introduction ods. Probabilistic analyses allow uncertainty to be quantified


and incorporated rationally into the design process. Probabil-
Slope engineering is perhaps the geotechnical subject most istic slope stability analysis (PSSA) was first introduced into
dominated by uncertainty. Geological anomalies, inherent slope engineering in the 1970s (Alonso 1976; Tang et al.
spatial variability of soil properties, scarcity of representative 1976; Harr 1977). Over the last three decades, the concepts
data, changing environmental conditions, unexpected failure and principles of PSSA have developed and are now well es-
mechanisms, simplifications and approximations adopted in tablished in the literature. In addition to accounting for un-
geotechnical models, and human mistakes in design and con- certainty, PSSA is also a useful approach for estimating
struction are all factors contributing to uncertainty. Conven- hazard frequency for site-specific quantitative risk analyses,
tional deterministic slope analysis does not account for particularly in the absence of representative empirical data.
quantified uncertainty in an explicit manner and relies on con- The merits of probabilistic analyses have long been noted
servative parameters and designs to deal with uncertain condi- (Chowdhury 1984; Whitman 1984; Wolff 1996; Christian
tions. The impact of such subjective conservatism cannot be 1996). Despite the uncertainties involved in slope problems
evaluated, and past experience shows that apparently conser- and notwithstanding the benefits gained from a PSSA, the
vative designs are not always safe against failure. profession has been slow in adopting such techniques. The
The evaluation of the role of uncertainty necessarily re- reluctance of practicing engineers to apply probabilistic
quires the implementation of probability concepts and meth- methods is attributed to four factors. First, engineers’ train-

Received 22 May 2001. Accepted 7 December 2001. Published on the NRC Research Press Web site at http://cgj.nrc.ca on 16 May 2002.
H. El-Ramly.1 AMEC Earth and Environmental, 4810-93 Street, Edmonton, AB T6E 5M4, Canada.
N.R. Morgenstern and D.M. Cruden. Geotechnical and Geoenvironmental Group, Department of Civil and Environmental
Engineering, University of Alberta, Edmonton, AB T6G 2G7, Canada.
1
Corresponding author (e-mail: hassan.el-ramly@ualberta.net).
Can. Geotech. J. 39: 665–683 (2002) DOI: 10.1139/T02-034 © 2002 NRC Canada

I:\cgj\Cgj39\Cgj-03\T02-034.vp
Tuesday, May 14, 2002 10:37:12 AM
Color profile: Generic CMYK printer profile
Composite Default screen

666 Can. Geotech. J. Vol. 39, 2002

ing in statistics and probability theory is often limited to ba- σ 2 [ε], independent of location and estimated from the scatter
sic information during their early years of education. Hence, of observations around the trend. The residual component at
they are less comfortable dealing with probabilities than location i is spatially correlated with the residual compo-
they are with deterministic factors of safety. Second, there is nents at surrounding locations.
a common misconception that probabilistic analyses require
significantly more data, time, and effort than deterministic Uncertainty in trend
analyses. Third, few published studies illustrate the imple- Site investigation programs are usually controlled by bud-
mentation and benefits of probabilistic analyses. Lastly, ac- get, which limits the number of tests performed. Since the
ceptable probabilities of unsatisfactory performance (or trend function is estimated from the available sample of ob-
failure probability) are ill-defined, and the link between a servations, it is uncertain. The uncertainty due to the sample
probabilistic assessment and a conventional deterministic as- size (number of observations) is known as statistical uncer-
sessment is absent. This creates difficulties in comprehend- tainty and is typically quantified by considering the regres-
ing the results of a probabilistic analysis. All of these issues sion coefficients of the trend function as variables with
are addressed in detail in El-Ramly (2001). means and non-zero variances, which are inversely propor-
This paper addresses the integration of probabilistic meth- tional to the number of observations, n.
ods into geotechnical practice, focusing on the first three ob- The method of least squares is the most common tech-
stacles of the factors listed in the previous paragraph. First, nique for estimating the means and variances of trend-
an overview of some of the concepts involved in quantifying function coefficients. Neter et al. (1990) discussed estimat-
uncertainty of soil properties is presented. The paper then ing the uncertainties of regression coefficients for linear
describes the development of a practical methodology for trends. Li (1991) pointed out some limitations of the method
probabilistic slope analysis based on Monte Carlo simula- of least squares. Baecher (1987) recommended that trend
tion. The methodology makes use of familiar and readily equations be linear, as the higher the order of the trend func-
available software, namely Microsoft® Excel 97 and @Risk. tion, the more uncertain the regression coefficients. Where
The implementation of the methodology is illustrated the soil properties do not exhibit a clear trend, a constant
through a probabilistic analysis of the stability of the James equal to the mean of the observations is assumed.
Bay dykes. Lastly, the results are compared with those of the Bias is another source of uncertainty that could affect the
first-order second-moment (FOSM) method and the practical estimated trend. The measured soil property can be consis-
insights gained through the analysis are highlighted. tently overestimated or underestimated at all locations. Fac-
tors such as the testing device used, boundary conditions,
Statistical analysis of soil data soil disturbance, or the models and correlations used to in-
terpret the measurements may contribute to biases. As an ex-
Components of statistical analysis ample, Bjerrum (1972) noted that the field vane consistently
Statistical analysis of soil data comprises two main stages: overestimated the undrained shear strength of highly plastic
data review and statistical inference. Data review is largely clays. An empirical correction factor is used to adjust the bi-
judgmental and encompasses a broad range of issues. First, ased measurements. The additional uncertainty introduced
the consistency of the data, known as the condition of by the use of Bjerrum’s vane correction factor is, however,
stationarity, should be ensured. Inconsistency can arise from significant. This could also be the case with other empirical
pooling data belonging to different soil types, stress condi- factors.
tions, testing methods, stress history, or patterns of sample
disturbance (Lacasse and Nadim 1996). The histogram is a Spatial variability
convenient tool in this regard. A multimodal histogram is an Soil composition and properties vary from one location to
indication of inconsistent data, or a nonstationary condition. another, even within homogeneous layers. The variability is
Data review also aims at identifying trends in the data, bi- attributed to factors such as variations in mineralogical com-
ases in measurements, errors and discrepancies in the re- position, conditions during deposition, stress history, and
sults, and outlier data values. Decisions should be made physical and mechanical decomposition processes (Lacasse
whether to reject outliers or to accept them as extreme val- and Nadim 1996). The spatial variability of soil properties is
ues. Baecher (1987) warned that care should be exercised in a major source of uncertainty.
this process to avoid rejecting true, important information. Spatial variability is not a random process, rather it is
Another concern is any clustering of data in a zone within controlled by location in space. Statistical parameters such
the spatial domain of interest which may render the data un- as the mean and variance are one-point statistical parameters
representative. and cannot capture the features of the spatial structure of the
Statistical inferences are commonly based on a simplified soil (Fig. 1). The plot in Fig. 1 compares the spatial variabil-
model (Vanmarcke 1977a; DeGroot and Baecher 1993) which ity of two artificial sets of data generated using the geo-
divides the quantity, xi, at any location, i, into a deterministic statistics software GSLIB (Deutsch and Journel 1998). Both
trend component, ti, and a residual component, ε i , where sets have similar histograms. The upper plot is highly erratic
[1] xi = ti + ε i and the data are almost uncorrelated, whereas the lower plot
is characterized by high spatial continuity. Additional statis-
The trend is evaluated deterministically using regression tical tools, such as the autocovariance and semivariogram,
techniques and is usually a function of location. The residual are essential to describe spatial variability. The pattern of
component is a random variable whose probability density soil variability is characterized by the autocorrelation dis-
function (PDF) has a zero mean and a constant variance, tance (or scale of fluctuation; Vanmarcke 1977a). A large

© 2002 NRC Canada

I:\cgj\Cgj39\Cgj-03\T02-034.vp
Tuesday, May 14, 2002 10:37:13 AM
Color profile: Generic CMYK printer profile
Composite Default screen

El-Ramly et al. 667

Fig. 1. A highly erratic spatial structure (upper right) and a highly continuous structure (lower right), both with similar histograms.

autocorrelation distance reflects the spatial continuity shown the extent of the domain over which the soil property is
in the lower plot of Fig. 1, whereas a short distance reflects being averaged increases, the variance decreases. For quanti-
erratic variability (Fig. 1, upper plot). DeGroot (1996) and ties averaging linearly, the amount of variance reduction de-
Lacasse and Nadim (1996) illustrated the estimation of auto- pends on the autocorrelation distance. The mean, however,
correlation distance. remains almost unchanged. The numerical data in Fig. 2 (up-
per right plot) are discretized based on grids of sizes 1 × 1,
Spatial averaging 5 × 5, and 10 × 10. For each scheme, the averages within
The performance of a structure is often controlled by the grid squares are calculated and the histogram of the local av-
average soil properties within a zone of influence, rather than erages is plotted as shown in Fig. 2. As the averaging area is
soil properties at discrete locations. Slope failure is more increased, the coefficient of variation of the local averages
likely to occur when the average shear strength along the fail- drops from 1.55 to 0.28 and the minimum and maximum
ure surface is insufficient rather than due to the presence of values become closer to the mean. If the averaging area is
some local weak pockets. The uncertainty of the average constant, the amount of variance reduction increases as the
shear strength along the slip surface, not the point strength, is autocorrelation distance decreases, i.e., soil variability be-
therefore a more accurate measure of uncertainty. Baecher comes more erratic.
(1987) warned, however, that depending on performance
mode, average properties are not always the controlling factor. Theory of random fields
Internal erosion in dams and progressive failure are examples The theory of random fields (Vanmarcke 1977a, 1977b,
of cases where extreme values control performance. 1983) is a common approach for modeling spatial variability
The variance of the strength spatially averaged over some of soil properties. It is also the basis of our probabilistic
surface is less than the variance of point measurements. As slope analysis methodology; to illustrate its rationale, some

© 2002 NRC Canada

I:\cgj\Cgj39\Cgj-03\T02-034.vp
Tuesday, May 14, 2002 10:37:13 AM
Color profile: Generic CMYK printer profile
Composite Default screen

668 Can. Geotech. J. Vol. 39, 2002

Fig. 2. Variance reduction due to spatial averaging over blocks of sizes 1 × 1, 5 × 5, and 10 × 10.

of the basic concepts of the theory of random fields are pre- pends on their relative locations, regardless of their absolute
sented here. locations within the random field.
At any location within a soil layer, a soil parameter is an The point to point variation of a random field is very diffi-
uncertain quantity, or a random variable, unless it is mea- cult to obtain in practice and is often of no real interest
sured at that particular location. Each random variable is (Vanmarcke 1983). Local averages over a spatial or temporal
characterized by a probability distribution and is correlated local domain are of much greater value. For example, the
with the random variables at adjacent locations. The set of hourly or daily rainfall is of interest to hydrologists rather
random variables at all locations within the layer is referred than the instantaneous rate of fall. Similarly, the average
to as a random field and is characterized by the joint proba- shear strength of a soil over the area of the slip surface is of
bility distribution of all random variables. A random field is more interest to geotechnical engineers than the variation of
called stationary (or homogeneous) if the joint probability strength from point to point within the layer.
distribution that governs the field is invariant when trans- Figure 3 shows a one-dimensional (1D) stationary random
lated over the parameter space. This implies that the cumula- field of a variable, x, with a mean, E[x], a variance, σ 2 , and
tive probability distribution function, the mean, and the a cumulative probability distribution function, F(x); the
variance are constant for any location within the random graph could also portray the variation along a line in the pa-
field and that the covariance of two random variables de- rameter space of a homogeneous k-dimensional random

© 2002 NRC Canada

I:\cgj\Cgj39\Cgj-03\T02-034.vp
Tuesday, May 14, 2002 10:37:13 AM
Color profile: Generic CMYK printer profile
Composite Default screen

El-Ramly et al. 669

Fig. 3. A realization of a 1D random field of a variable x with a mean E[x], variance σ2, and cumulative probability distribution
function F(x), showing local averages over intervals ∆z and ∆z′.

field. For this type of random field, Vanmarcke (1977a) de- where Z1 is the distance from the beginning of the first inter-
fined the dimensionless variance function Γ(∆z) as val to the beginning of the second interval, Z2 is the distance
from the beginning of the first interval to the end of the sec-
σ2
[2] Γ(∆ z) = ond interval, and Z3 is the distance from the end of the first
σ ∆2 z interval to the end of the second interval.
where σ 2∆ z is the variance of the local average, X(∆z), of the Random testing error
variable, x, over an interval of length ∆z. The variance func- Random testing errors arise from factors related to the
tion is a measure of the reduction in the point variance, σ 2 , measuring process (other than bias), such as operator error
under local averaging. For most commonly used correlation or a faulty device. They can be directly measured only
functions, Vanmarcke (1983) showed that Γ(∆z) can be ap- through repeated testing of the same specimen by different
proximated by operators and devices. This is not practical, as most geo-
1 ∆z ≤ δ technical field and laboratory tests disturb the sample, so
 random measurement errors cannot be known. In data analy-
[3] Γ(∆ z) =  δ
∆z ≥ δ sis, measurement error is regarded as a random variable with
 ∆ z
a zero mean and a constant variance, referred to as random
where δ is the scale of fluctuation. In concept, the scale of error variance. The uncertainty due to testing errors is not a
fluctuation has the same meaning as the autocorrelation dis- true variation in the soil property and should be discarded in
tance but differs in numeric value. For the common expo- evaluating parameter variability. Approximate analytical pro-
nential and Gaussian autocorrelation models, δ is 2.0 and cedures (Baecher 1987) are used to estimate the random er-
1.77 times the autocorrelation distance, ro, respectively. ror variance. Jaksa et al. (1997), however, pointed to
The model indicates no variance reduction, Γ(∆z) = 1, due important inaccuracies associated with such procedures.
to local averaging up to an averaging interval ∆z equal to δ. Reliable estimation of random error variance requires data
The rationale of this approximation is that modeling a ran- at very small separation distances (r ≈ 0), not available in
dom phenomenon at a level of aggregation more detailed practice. As a result, the distinction between noise, or ran-
than the way the information about the phenomenon is ac- dom error, and real, short-scale variability (over distances
quired or processed is impractical and unnecessary less than data spacing) is not possible. Hence, the analyti-
(Vanmarcke 1983). In site investigation programs, the spac- cally computed variance is composed of random error vari-
ing or the time interval between observations is often large. ance and short-scale variability. Jaksa et al. (1997) also
Characterizing the correlation structure at small intervals be- showed that the computed value of the random error vari-
comes unreliable and justifies the approximation by a perfect ance is sensitive to data spacing. Kulhawy and Trautmann
autocorrelation. (1996) commented that quantitative assessment of testing er-
Local averages of the variable x over intervals ∆z and ∆z′ rors of in situ measurements is rare because of the difficul-
(Fig. 3), X(∆z) and X′(∆z′), are spatially correlated. The cor- ties in separating natural spatial variability from test
relation coefficient, ρ(X∆z, X ∆′ z′ ), between X(∆z) and X′(∆z′) uncertainty. So, the reliability of the analytical estimation of
is given by eq. [4] (Vanmarcke 1983). It is a function of the random error variance is in question. Random testing errors
lengths of the two intervals, the separation, Zo, between are best addressed through standardization of testing equip-
them (Fig. 3), and the variance function of the variable being ment and procedures, proper personnel training, and contin-
averaged: uous inspection.

[4] ρ(X ∆ z, X ∆′ z′) =


Probabilistic slope analysis methods
Z 2o Γ(Z o) − Z 12 Γ(Z1) + Z 22 Γ(Z 2 ) − Z 23 Γ(Z 3)
Probabilistic procedures for slope stability analysis vary
2∆ z ∆ z′ [Γ(∆ z)Γ(∆ z′ )]0.5 in assumptions, limitations, capability to handle complex

© 2002 NRC Canada

I:\cgj\Cgj39\Cgj-03\T02-034.vp
Tuesday, May 14, 2002 10:37:14 AM
Color profile: Generic CMYK printer profile
Composite Default screen

670 Can. Geotech. J. Vol. 39, 2002

problems, and mathematical complexity. Most of them, The slope problem (geometry, stratigraphy, soil properties,
however, fall into one of two categories: approximate meth- and slip surface) and the selected method of analysis are first
ods (the FOSM method, the point estimate method), and modeled in a Microsoft® Excel 97 spreadsheet. Available
Monte Carlo simulation. data are examined and uncertainties in input parameters are
Approximate methods make simplifying assumptions that identified and described statistically by representative proba-
often limit their application to specific classes of problems. bility distributions. Only those parameters whose uncertain-
For example, some used very simple slope models such as ties are deemed significant to the analysis need to be treated
the ordinary method of slices (Tang et al. 1976; Vanmarcke as variables. A Monte Carlo simulation is then performed as
1980; Honjo and Kuroda 1991; Bergado et al. 1994). Others illustrated schematically in Fig. 4. @Risk draws at random a
dealt only with frictionless soils (Matsuo and Kuroda 1974; value for each input variable from within the defined proba-
Vanmarcke 1977b; Bergado et al. 1994). A restriction to a bility distributions. These values (input set 1, Fig. 4) are
circular (or cylindrical) slip surface is common in many used to solve the spreadsheet and calculate the correspond-
studies (Alonso 1976; Vanmarcke 1977b; Yucemen and Al- ing factor of safety. The process is repeated sufficient times,
Homoud 1990; Bergado et al. 1994). The spatial variability m, to estimate the statistical distribution of the factor of
of soil properties and pore-water pressure is often ignored, safety. Statistical analysis of this distribution allows estimat-
assuming perfect autocorrelation (Nguyen and Chowdhury ing the mean and variance of the factor of safety and the
1984; Wolff and Harr 1987; Duncan 2000). probability of the factor of safety being less than one. The
Approximate methods allow the estimation of the mean procedure is applicable to any method of limit equilibrium
and variance of the factor of safety but do not provide any analysis that can be represented in a spreadsheet.
information about the shape of the probability density func-
tion, so the failure probability can only be obtained by as- Statistical characterization of input variables
suming a parametric probability distribution of the factor of
safety (typically normal or log-normal). Estimates of low Probability distributions
probabilities, required for safe structures, are sensitive to the Different approaches can be adopted to estimate the prob-
assumed distribution. ability distribution of each input variable. Where there are
Few studies used Monte Carlo simulation in slope analysis adequate amounts of data, the cumulative distribution func-
when compared to approximate methods. The extensive com- tion (CDF) of the measurements can be used directly in the
putational efforts involved in the simulations required research- simulation process. In geotechnical practice, the observed
ers to develop their own software to solve slope stability CDF (or histogram) may show spikes that would not appear
problems. Computer times needed for the simulations were were more observations available. The CDF is better ob-
also significant. As a result, a Monte Carlo simulation was con- tained by resetting the probability associated with each ob-
sidered, until recently, to be uneconomic (Tobutt 1982; Priest servation to the average of its cumulative probability and
and Brown 1983). Studies applying Monte Carlo simulation that associated with the next lower observation (Deutsch and
also rarely addressed the spatial variability of soil properties Journel 1998). This procedure smooths unwanted spikes and
(Major et al. 1978; Tobutt 1982; Nguyen and Chowdhury allows assigning finite probabilities for values of the input
1985) because of difficulties in generating random values in parameter less than the minimum value measured and more
ways that preserved their correlations. Today, rapid develop- than the maximum value.
ments in software are significantly changing this situation. Where observations are scarce or absent, parametric dis-
The limitations, and sometimes the complexities, of tributions can be assumed from the literature. Studies have
probabilistic methods combined with the poor training of estimated coefficients of variation and probability density
most engineers in statistics and probability theory have sub- functions of soil properties (Lumb 1966; Chowdhury 1984;
stantially inhibited the adoption of probabilistic slope stabil- Harr 1987; Kulhawy et al. 1991; Lacasse and Nadim 1996).
ity analysis in practice. To facilitate the integration of Care should be exercised, however, to ensure that the mini-
probabilistic methods into practice, the methodologies and mum and maximum values of the selected distribution are
techniques, while being consistent with principles of logic consistent with the physical limits of the parameter being
and mechanics, should be simple, capable of solving real modeled. For example, shear strength parameters should not
slope problems, and in formats familiar to engineers. take negative values. If the selected distribution implies nega-
tive values, then the distribution is truncated at a practical
minimum threshold. Alternatively, variability can be assumed
An integrated probabilistic slope analysis to follow a triangular distribution with estimates of minimum,
methodology maximum, and most likely values based on expert opinion.
@Risk built-in functions allow great flexibility in model-
Outline of the methodology ing input variables. Parametric and nonparametric distribu-
The probabilistic slope analysis methodology based on tions (e.g., experimental CDF) can be modeled using @Risk
Monte Carlo simulation developed here has the advantages functions. Desired truncations can be easily imposed on any
of being simple and not requiring a comprehensive statistical distribution either through @Risk or using Microsoft® Ex-
and mathematical background. The methodology is spread- cel 97 functions.
sheet based and makes use of the familiar and readily avail-
able Microsoft® Excel 97 (Microsoft Corporation 1997) and Modeling parameter uncertainty
@Risk (Palisade Corporation 1996) software. Other similar The probabilistic slope analysis methodology accounts for
software, e.g., Lotus 1-2-3 and Crystal Ball (Decisioneering the uncertainty of the input parameters based on the concepts
Inc. 1996), is likely capable of undertaking the task. discussed in previous sections and the following equation:
© 2002 NRC Canada

I:\cgj\Cgj39\Cgj-03\T02-034.vp
Tuesday, May 14, 2002 10:37:14 AM
Color profile: Generic CMYK printer profile
Composite Default screen

El-Ramly et al. 671

Fig. 4. Monte Carlo simulation procedure using Microsoft® Excel 97 and @Risk software.

[5] V = B(t + ε) This is because the variance function Γ(l), eq. [3], equals
one for l ≤ δ. The correlation coefficients between the vari-
where V is the input variable adjusted for statistical errors ables representing the local averages over different segments
and bias, B is a bias correction factor, t is the trend compo- of the slip surface are estimated using eq. [4]. Choosing the
nent, and ε is the residual component. The statistical uncer- length of segments equal to δ eliminates the correlation coef-
tainty in the trend is estimated by standard statistical ficients between most of the variables and greatly simplifies
methods (Ang and Tang 1975; Neter et al. 1990). The vari- the simulation. Figure 5 is a schematic illustration of the
ance of the residual component is equal to the variance of model, where ρ is the correlation coefficient between vari-
the measured data, with no explicit consideration of the un- ables.
certainty due to random testing errors for two reasons. Esti-
mates of random error variance are unreliable, as discussed
in an earlier section, so any reduction of the observed vari- Critical slip surface
ance could be unsafe. In addition, random errors fluctuate in For any slope, there is an unlimited number of potential
magnitude above and below zero and are uncorrelated spa- slip surfaces. The slope may fail, or perform unsatisfactorily,
tially. By taking the spatial average of the input variable over along any of these surfaces. The total probability of unsatis-
the whole area of interest, such as the slip surface, positive factory performance of a slope is, thus, the joint probability
and negative random errors at the different locations within of failure occurring along the admissible slip surfaces. These
the averaging domain tend to cancel out. As a result, the ran- surfaces, however, are highly correlated, since they are all
dom error variance associated with the averaged quantity is analyzed using the same input variables and the same analyt-
largely reduced. It is emphasized, however, that a critical ical model. Furthermore, the close proximity of many of
review of the observations should ensure that appropriate them add an element of spatial correlation. Evaluating the
testing standards and procedures are followed. The bias cor- total probability of unsatisfactory performance of the slope
rection factor is also considered a variable with a mean and is a mathematically formidable task.
variance evaluated by experience and comparison with field The strong correlation between slip surfaces, however,
performance or by observations from a more accurate testing significantly reduces the difference between the total proba-
procedure. It does not have to be a multiplier as in eq. [5]; it bility of unsatisfactory performance and that of the most
could be an addition [V = B + (t + ε)]. critical surface (Mostyn and Li 1993). Hence, the probability
The spatial variability of an input variable is represented estimate from the most critical slip surface is considered a
by the correlation structure of the residual component, ε, and reasonable estimate of slope reliability (Vanmarcke 1977b;
modeled using the theory of random fields. The spatial vari- Alonso 1976; Yucemen and Al-Homoud 1990). In this study,
ability of ε along the slip surface is approximated by a 1D the probability of unsatisfactory performance is obtained by
stationary random field. Vanmarcke’s (1983) approximate independently analyzing a number of predetermined slip sur-
model (eq. [3]) is adopted for the analysis of the random faces. The highest estimated probability value is considered
field. The point to point variability of the residual compo- representative of the total probability of unsatisfactory per-
nent along the slip surface is resembled by the variability of formance of the slope. The question, however, is which sur-
its local averages over segments of the surface. The portion faces to consider?
of the slip surface within the subject layer is divided into Chowdhury and Tang (1987) and Hassan and Wolff (1999)
segments of length l not exceeding the scale of fluctuation δ. indicated that the deterministic critical slip surface is not al-
The local average of the residual component over the length ways the most critical surface in a probabilistic analysis.
of any of these segments, ε(l), is considered a variable. The Hassan and Wolff (1999) proposed a search algorithm for lo-
CDF of the variable ε(l) is the same distribution function of cating the slip circle with the minimum reliability index. It
the residual component, F(ε), with no variance reduction. searches for the slip surface with the longest length within
© 2002 NRC Canada

I:\cgj\Cgj39\Cgj-03\T02-034.vp
Tuesday, May 14, 2002 10:37:15 AM
Color profile: Generic CMYK printer profile
Composite Default screen

672 Can. Geotech. J. Vol. 39, 2002

Fig. 5. Modeling the spatial variability of an input parameter variance reduction, since the variance function Γ(∆z ≤ δ) is
along the failure surface using Vanmarcke’s (1983) approxima- equal to one (eq. [3]). @Risk functions are used to assign
tion of a 1D random field. appropriate probability distributions for the spreadsheet cells
containing the input variables. The correlations between
variables are modeled using @Risk functions “IndepC” and
“DepC” or by creating a correlation coefficients matrix us-
ing the “Correlate” command. In the simulation process,
correlated variables are sampled so as to preserve input cor-
relation coefficients.
Slope analysis methods, such as the Bishop, Janbu, and
Spencer methods, can be easily modeled in a spreadsheet.
However, more advanced and complex methods such as the
Morgenstern–Price method and three-dimensional (3D)
methods, are, for now, cumbersome to model in a spread-
sheet. In this paper, the Bishop method of slices is used. The
spreadsheet model is built by creating a calculation table
similar to that proposed by Bishop (1955) for mechanical
the soil layer whose uncertainty contributes the most to the computations. The appropriate soil parameters for each slice
uncertainty of the factor of safety. Their algorithm, however, are automatically chosen by a series of nested “IF” state-
does not directly account for the spatial variability of soil ments that compares the X coordinate of the mid-base point
properties and pore-water pressure. In locating the probabil- with the coordinates of the intersections of the slip surface
istic critical slip surface, variance reduction due to spatial with soil layers. The Microsoft® Excel 97 “circular refer-
averaging of soil parameters over the length of the slip sur- ence” feature is used for the iteration process until the differ-
face has to be considered. The reduction depends on the ence between the assumed and the calculated factors of safety
autocorrelation distance and the length of slip surface, which is within the user-specified range. Further discussions and an
is not known beforehand. As a result, an increase in the vari- illustration of the structure of the spreadsheet are presented in
ance of the factor of safety, due to a longer portion within a the context of the case study analyzed in a later section.
highly uncertain layer, could be offset by a variance reduc-
tion due to spatial averaging. Hence, in slopes dominated by Issues in simulation
uncertainty due to spatial variability, the slip surface based Monte Carlo simulation requires the generation of random
on the Hassan and Wolff algorithm may not be the most crit- numbers, which are used in sampling the CDFs of the input
ical surface in a probabilistic analysis. variables. @Risk software (Palisade Corporation 1996) al-
An essential part of the probabilistic analysis is, thus, to lows two sampling techniques, Monte Carlo sampling (or
consider all slip surfaces that may be hazardous. These may random sampling) and Latin Hypercube sampling. The latter
include the deterministic critical slip surface, the minimum allows sampling the entire CDF curve, including the tails of
reliability index surface according to Hassan and Wolff the distribution, with fewer iterations, and consequently less
(1999), noncircular, structurally controlled surfaces, and sur- computer time. It is adopted in this study.
faces through weak or presheared layers. The output of a Monte Carlo simulation is sensitive to the
number of iterations, m. When m is large, the random samples
Spreadsheet modeling drawn for each input variable are also large and the match be-
The spreadsheet software Microsoft® Excel 97 is used in tween the CDF recreated by sampling and the original input
modeling the slope geometry, stratigraphy, and slope analysis CDF is more accurate. Hence, the level of noise in the simu-
method. Using analytical geometry, the equations describing lation diminishes and the output becomes more stable at the
the ground profile, the boundaries between soil layers, and expense of increasing computer time. The optimum number
the slip surface are first established with reference to a coor- of iterations depends on the sizes of the uncertainties in the
dinate system. The equations are then modeled in a input parameters, the correlations between input variables,
Microsoft® Excel 97 spreadsheet. The coordinates of the and the output parameter being estimated. A practical way to
points of intersection of the different boundaries, for exam- optimize the simulation process is to repeat the simulation us-
ple, where the slip surface meets the borders between layers, ing the same seed value and an increasing number of itera-
can be obtained within the spreadsheet and used in calculat- tions. A plot of the number of iterations, m, against the
ing the slice information (width, coordinates of mid-base probability of unsatisfactory performance indicates the mini-
point, total height, and thickness in each soil type) for factor mum number of iterations at which the probability value sta-
of safety computations. bilizes.
Next, the slip surface within each soil layer is divided into
segments of length δ in addition to a residual segment of Interpretation of the output
smaller length, similar to the illustration in Fig. 5. The local
averages of a soil parameter over the length of each segment Probability of unsatisfactory performance
are considered variables. The spatial variability of the soil The output of the simulation is the probability density func-
parameter along the slip surface is modeled by the correla- tion of the factor of safety from which the mean and standard
tions between these variables. The variables are character- deviation can be estimated. The probabilistic safety measure
ized by the CDF of the residual component, F(ε), with no used in this study is the failure probability. It is the probabil-

© 2002 NRC Canada

I:\cgj\Cgj39\Cgj-03\T02-034.vp
Tuesday, May 14, 2002 10:37:15 AM
Color profile: Generic CMYK printer profile
Composite Default screen

El-Ramly et al. 673

ity of the factor of safety, FS, being less than one, or simply probability could be a lower bound to the actual probability
the number of iterations with FS ≤ 1.0 relative to the total of unsatisfactory performance. That is why comparison with
number of iterations, m. The term failure, however, implies computed probabilities of different designs is believed to be
that the collapse of the slope is the event of concern to the de- of greater value.
signer, which is not necessarily the case. The serviceability of Having estimated the probability of unsatisfactory perfor-
the slope is as important as slope collapse and requires thor- mance, the next step is to assess whether it is acceptable.
ough evaluation and assessment. Serviceability issues include, Typically, this is achieved through comparing the computed
among others, slope movement and cracking (without the values with a probabilistic slope design criterion. Commonly,
slope collapsing), high water seepage, and surface erosion. a criterion based on the observed frequency of slope failures
The term failure may also imply to clients, particularly non- and judgement is used. A major drawback to this approach
professional clients, that the slope would fail. We recommend is that the site- and case-specific features such as slope ge-
using a different terminology. The U.S. Army Corps of Engi- ometry, site conditions, and sources and levels of uncertainty
neers (1992, 1995) used probability of unsatisfactory perfor- of the case histories constituting the database are not ad-
mance, Pu, instead of failure probability. The same term, dressed. Applying such a global criterion to any slope is a
unsatisfactory performance, is adopted here to address failure significant generalization.
mechanisms. Another safety measure, perhaps probability of The authors are not aware of any well-founded probabilis-
unsatisfactory serviceability, should be considered to address tic criteria for the acceptability of a design. A reliable ap-
serviceability criteria. The evaluation of slope serviceability proach to estimate such criteria is to calibrate the computed
is, however, beyond the scope of this study. probabilities of unsatisfactory performance of slopes with
Because the simulation process uses random sampling of their observed field performance. This requires probabilistic
the input variables, the calculated probability of unsatisfac- slope stability analyses of cases of failed slopes and of
tory performance is also a variable. The probability value slopes performing satisfactorily. Through the comparison of
from a single simulation could differ from the true value. the probabilities associated with the two classes, a probabil-
Law and McComas (1986) described relying on the results istic slope design criterion could be established. This ap-
of a single simulation run as one of the most common and proach has been developed by El-Ramly (2001) and will be
potentially dangerous simulation practices. It is, therefore, published separately.
essential to repeat the simulation using different seed values
to assess the consistency of the estimates. By running the Reliability index
simulation many times, the histogram of the probability of The reliability index, β, another common probabilistic
unsatisfactory performance, the mean probability, and the safety measure, is given by
95% confidence interval around the mean could also be esti- E[FS] − 1
mated. Using @Risk “macro” functions, the process of run- [6] β=
σ[FS]
ning a number of simulations can be fully automated. A
simple macro file can be designed to run a number of simula- where E[FS] and σ[FS] are the mean and standard deviation
tions and save the output of each simulation in a separate file. of the factor of safety, respectively. The above definition is
In reflecting on the computed probability of unsatisfactory accurate if the performance function (factor of safety equa-
performance, three points should be noted. First, the com- tion) is linear, which is not the case for slope analysis mod-
puted probability of unsatisfactory performance does not els. Mostyn and Li (1993) suggested, however, that the
address temporal changes in input variables. If pore-water performance functions of slopes are reasonably linear and rec-
pressures, as an example, vary with time, the impact of such ommended ignoring the nonlinearity when calculating β. Ang
variations on stability can only be assessed by undertaking and Tang (1984) provided a convenient theoretical background
further analyses using the new pore-pressure distributions. of the meaning and estimation of the reliability index.
Second, the computed probability of unsatisfactory perfor-
mance represents the probability of occurrence of a hazardous Sensitivity analysis
event over the lifetime of the conditions considered in the Through the software @Risk, Spearman rank correlation
analysis, such as loading conditions and environmental condi- coefficients between the factor of safety and the input vari-
tions. The calculated probability is, thus, associated with a ables can be calculated for a sensitivity analysis. The
time frame that varies from one problem to another. If the an- Spearman coefficient is a correlation coefficient based on the
nual probability of unsatisfactory performance is required, as rank of the data values within the minimum–maximum
is the case in quantitative risk analyses, the reference time range, not the actual values themselves. It varies between 1
needs to be estimated. For example, if the pore-water pressure and –1. A value of zero indicates no correlation between the
used in the analysis is the result of a rainstorm, the annual input variable and the output, a value of 1 indicates a com-
probability of unsatisfactory performance is estimated based plete positive correlation, and a value of –1 indicates a com-
on the return period of the storm. In the case that none of the plete negative correlation. Spearman correlation coefficients
input variables is time dependent, the annual probability could could be used as measures of the relative contribution of
be referenced to the lifetime of the slope. each input variable to the uncertainty in the factor of safety.
Third, despite efforts to include all sources of uncertainty, This contribution comprises two elements, the degree of un-
there is the possibility of undetected uncertainties such as certainty of the input parameter and the sensitivity of the
human mistakes affecting slope performance. The contribu- factor of safety to changes in that parameter.
tion of these unknown uncertainties to the probability of un- The results of sensitivity analyses are of significant practi-
satisfactory performance is not considered, so the computed cal value, as they quantify the contributions of the various

© 2002 NRC Canada

I:\cgj\Cgj39\Cgj-03\T02-034.vp
Tuesday, May 14, 2002 10:37:15 AM
Color profile: Generic CMYK printer profile
Composite Default screen

674 Can. Geotech. J. Vol. 39, 2002

sources of uncertainty to the overall design uncertainty. Re- The variances of the unit weight and friction angle of the
sources, whether intellectual or physical, can thus be ratio- embankment material are evaluated judgementally (Christian
nally allocated towards reducing the uncertainty of the et al. 1994) to reflect potential variability in fill properties.
variables with large impacts on design. Also, the relative im- The bias in vane measurements is adjusted by Bjerrum’s
pacts of systematic uncertainty (statistical uncertainty and vane correction factor, which is also considered uncertain
bias) and uncertainty due to inherent spatial variability can due to the scatter in Bjerrum’s database. The uncertainty in
be estimated. This information is of interest because system- the depth of the till layer, Dtill, is entirely statistical uncer-
atic uncertainty has a consistent effect at all locations within tainty due to the limited number of borings. In the absence
the domain of the problem and can have a major impact on of the actual probability distributions of the data, all vari-
design. Unlike the uncertainty due to inherent spatial vari- ables are assumed normal. The spatial variability of all un-
ability, systematic uncertainty can be reduced by increasing certain soil parameters is characterized by an isotropic
the size of data sample and by avoiding highly uncertain em- autocorrelation distance of about 15 m, which was inferred
pirical correlations and factors. from Christian et al. (1994).
As the strength of the lacustrine clay is relatively low, the
depth to the till layer controls the location of the critical slip
Probabilistic analyses of James Bay dykes circle. The uncertainty in Dtill, thus, introduces uncertainty
in the location of the slip circle. To examine the impact of
Spreadsheet-based probabilistic slope analysis this uncertainty, deterministic stability analyses were per-
The application of the probabilistic slope analysis methodol- formed varying Dtill incrementally between 15.5 and 21.5 m
ogy described earlier in the paper is illustrated through the (±3 standard deviations). The minimum factor of safety var-
analysis of the well-documented dykes of the James Bay hy- ied between 1.69 and 1.30, implying that the uncertainty in
droelectric project in Quebec, Canada. Although the dykes Dtill could have an important impact on the reliability of the
were not built, the design was the subject of extensive studies design. All the critical slip circles were tangent to the top of
quantifying the sources of uncertainty, analyzing the spatial the till, daylighted within a short distance at the top of the
variability of soil properties (Ladd et al. 1983; Soulié et al. embankment, and shared almost the same x coordinate for
1990), and evaluating the stability of the dyke probabilistically the centres.
using the FOSM method (Christian et al. 1994). The equations describing the dyke profile and soil layers
The project called for the construction of 50 km of earth with reference to a coordinate system are estimated and
dykes in the James Bay area. Various design options were modeled in a Microsoft® Excel 97 spreadsheet. The general
investigated. Only one design is considered in this study, the layout and structure of the spreadsheet are illustrated in Ap-
single stage construction of a 12 m high embankment with a pendix A. Since the depth of the till is considered a variable,
slope angle of 18.4° (3h:1v) and a 56 m wide berm at mid- a different value is sampled for each simulation iteration and
height. Figure 6 shows the geometry of the embankment and consequently the critical slip circle varies from one iteration
the underlying stratigraphy. to another. To minimize the computer time, some restrictions
The embankment is on a clay crust, about 4.0 m thick, were imposed on the geometry of the slip circles based on
overlying a sensitive marine clay which in turn is underlain the results of the previous parametric study. The slip circles
by a lacustrine clay. The marine clay is about 8.0 m thick are assumed tangent to the till layer, daylight at a fixed point
and the lacustrine clay is about 6.5 m thick. The undrained at the top of the embankment (X1 = 4.9, Y1 = 36.0), and have
shear strength of both clays, measured by field vane tests at a common X coordinate for the centres (X0 = 85.9). The
1.0 m depth intervals, exhibited a large scatter. The mean equation of the slip circle is then added to the spreadsheet as
undrained shear strength of the marine clay is about 35 kPa a function of the Y coordinate of the centre, Y0, and the ra-
and that of the lacustrine clay is about 31 kPa. The lacus- dius, R; both depend on the sampled value of Dtill. In each
trine clay is underlain by a stiff till. iteration the sampled value of Dtill thus defines only one crit-
The uncertainty in soil parameters was quantified by Ladd ical slip circle. Using the principles of analytical geometry,
et al. (1983) and Christian et al. (1994). They identified the the intersections of the slip circle and the boundaries be-
input parameters whose uncertainties are deemed important tween layers and the breakage points in ground profile are
to the stability of the dykes and estimated their statistical pa- computed.
rameters. The following probabilistic stability analyses are The spatial variabilities of soil parameters are modeled as
based on the conclusions of these studies. Other case studies 1D random fields, assuming exponential autocovariance
demonstrating the complete analysis starting with field and functions. The spatial variability of the unit weight of the
laboratory data, through the quantification of parameter un- embankment is, in fact, a two-dimensional (2D) random
certainty, the probabilistic assessment, and the estimation of field. Vanmarcke (1983) showed that the process of averag-
the probability of unsatisfactory performance will be pub- ing a 2D random field over a rectangular area with one side
lished separately by the present authors. of the area smaller than the scale of fluctuation in the same
Eight input parameters are considered variables: the unit direction could be approximated by averaging a 1D random
weight and friction angle of embankment material, the thick- field in the perpendicular direction. If the cross section of
ness of the clay crust, the undrained shear strength and the embankment is regarded as a rectangle, the variability of
Bjerrum vane correction factors for the marine and lacus- unit weight could be approximated by a 1D random field in
trine clays, and the depth of till layer. Table 1 summarizes the horizontal direction. This approximation is not likely to
the means and variances of all input variables. have any effect on the analysis, as the impact of the spatial

© 2002 NRC Canada

I:\cgj\Cgj39\Cgj-03\T02-034.vp
Tuesday, May 14, 2002 10:37:16 AM
Color profile: Generic CMYK printer profile
Composite Default screen

El-Ramly et al. 675

Fig. 6. Cross section and stratigraphy of the James Bay dykes, showing the approach adopted to account for spatial variability of soil
properties.

Table 1. Input variables and their statistical parameters (based on Christian et al. 1994)

Soil parameters
Variance Bias factors
Input variable Mean Inherent variability Statistical uncertainty Mean Variance
Fill friction angle, φfill (°) 30.0 1.00 3.00 — —
Fill unit weight, γfill (kN/m3) 20.0 1.00 1.00 — —
Clay crust thickness, tcr (m) 4.0 0.19a 0.04 — —
Shear strength of marine clay, SuM (kPa) 34.5 66.26 0.90 — —
Bjerrum vane correction factor for SuM, µM — — — 1.0 0.006
Shear strength of lacustrine clay, SuL (kPa) 31.2 74.82 3.00 — —
Bjerrum vane correction factor for SuL, µL — — — 1.0 0.023
Depth of till, Dtill (m) 18.5 — 1.00 — —
a
Variance is reduced by 80% to account for spatial averaging based on the assessment of Christian et al. (1994).

variability of unit weight on design reliability is minimal Fig. 7. Estimating the optimum number of iterations for Monte
(Alonso 1976; Nguyen and Chowdhury 1984). This is attrib- Carlo simulation.
uted to the small spatial variability of the unit weight (a co-
efficient of variation of only 5% in the James Bay case) and
the insensitivity of stability calculations to variations in unit
weight.
The slip surface within each soil layer is divided into seg-
ments of lengths not exceeding δ = 30 m (δ = 2ro for expo-
nential autocovariance functions), as illustrated in Fig. 6.
Hence, the average undrained shear strength over the length
of each segment has the same variance as the input data (Ta-
ble 1) with no reduction. The undrained shear strength of the
lacustrine clay, for example, is modeled by three variables
representing the average strength over three segments of the
slip surface, as illustrated in Fig. A1 in Appendix A. Simi-
larly, the embankment fill is divided into five zones and the
average unit weight within each zone is considered a random
variable. The correlation coefficients between the variables
are estimated using eq. [4], and the correlation coefficient
matrices are shown in Appendix A. Additional variables are
added to the spreadsheet to model statistical uncertainties
and the uncertainties in Bjerrum vane correction factors. In

© 2002 NRC Canada

I:\cgj\Cgj39\Cgj-03\T02-034.vp
Tuesday, May 14, 2002 10:37:16 AM
Color profile: Generic CMYK printer profile
Composite Default screen

676 Can. Geotech. J. Vol. 39, 2002

Fig. 8. Histogram and probability distribution function of the factor of safety.

Fig. 9. Histogram of the probability of unsatisfactory perfor- shown in Fig. A2 in Appendix A. For each set of sampled
mance based on the results of 25 simulations. input parameters, the iterative process for factor of safety cal-
culations is repeated until the difference between the factors
of safety in two consecutive calculations is less than 0.01.
Trial simulations indicated that the optimum number of it-
erations is 32 000 (Fig. 7). Using a seed number of 31 069
(an arbitrary value), the mean factor of safety is estimated to
be 1.46, with a standard deviation of 0.20. The probability of
unsatisfactory performance is 4.70 × 10–3. Figure 8 shows
the histogram and the probability distribution function of the
factor of safety. The histogram is slightly right skewed, with
a coefficient of skewness of 0.29. After 25 simulations, the
mean factor of safety is 1.46, with a standard deviation of
0.20 and a coefficient of skewness of 0.30. The mean probabil-
ity of unsatisfactory performance is 4.70 × 10–3, with the 95%
confidence interval between 4.50 × 10–3 and 4.90 × 10–3. Fig-
ure 9 shows the histogram of the probability of unsatisfactory
performance. The reliability index is calculated to be 2.32.
A sensitivity analysis (Fig. 10) shows Spearman rank cor-
relation coefficients for all 19 input variables. It is interest-
ing that many of the factors with major contributions to the
uncertainty of the factor of safety are not related to soil
property measurements. For example, Bjerrum’s correction
factor for the undrained shear strength of the lacustrine clay,
the statistical uncertainty in the depth of the till, and the sta-
tistical uncertainty in the unit weight of the fill (which was
total, 19 variables are defined in the spreadsheet (shaded evaluated judgementally) are among the main factors affect-
cells in Fig. A1 in Appendix A) to account for the various ing the reliability of the design. The results highlight the
components of parameter uncertainty. significance of the additional uncertainty that could be intro-
Using @Risk options, truncation limits are imposed on duced by the designer through the use of empirical factors
the probability distributions of the undrained shear strength and subjective estimates of uncertainty. Although subjective
to prevent sampling negative values. Truncation limits at the estimates based on experience are acceptable in practice,
mean ±3σ are also imposed on the probability distributions quantitative estimates of uncertainty, such as variance, are
of the thickness of the clay crust and the depth of the till not yet reliable. Therefore care should be exercised in mak-
layer. The purpose of these limits is to avoid sampling ex- ing such judgements.
treme low or high values that may cause discrepancies in the
sequence of layers. First-order second-moment (FOSM) analysis
The stability calculations are based on the Bishop method The FOSM method is an approximate approach based on
of slices. The tables used for factor of safety computations are Taylor’s series expansion of the performance function g(x1,
© 2002 NRC Canada

I:\cgj\Cgj39\Cgj-03\T02-034.vp
Tuesday, May 14, 2002 10:37:17 AM
Color profile: Generic CMYK printer profile
Composite Default screen

El-Ramly et al. 677

Fig. 10. Sensitivity analysis results. Spearman rank correlation coefficients for all input variables.

x2,…,xn) around its mean value. The performance function g 2ro


[10] f ≅
could be Bishop’s method of slices, which is a function of a L
number of input variables xn that represents soil properties,
pore pressure, and slope geometry. For uncorrelated input where L is the length over which the parameter of interest is
variables, the mean and variance of the factor of safety are being averaged.
given by eqs. [7] and [8], respectively: When the stability of James Bay dykes is analyzed
probabilistically using the FOSM method, the mean factor of
[7] E[FS] ≅ g(E[x1], E[x2],…,E[xn]) safety is estimated to be 1.46. Table 2 summarizes the calcu-
j 2 lations of the variance of the factor of safety. For computing
 ∂g  2
σ [FS] ≅ ∑ 
the variance reduction factor f, the length L is taken equal to
[8] 2
 σ [x i ]
i =1  ∂ x i 
the length of the slip surface within each layer, except for
the fill unit weight, where L is taken equal to the length of
where E[–] and σ 2 [–] denote the mean and variance, respec- the embankment in cross section. The standard deviation of
tively. For most geotechnical models, the analytical evalua- the factor of safety is computed to be 0.19, thus the reliabil-
tion of the derivatives (∂g/∂x i ) is cumbersome. A finite ity index is 2.42. To estimate the probability of unsatisfac-
difference approach is commonly used to approximate the tory performance, a form of the probability density function
partial derivatives as follows: of the factor of safety must be assumed. For normal and log-
normal probability distributions, the probability of unsatis-
∂g FS1 − FS 2 ∆ FS factory performance is estimated to be 8.4 × 10–3 and 2.5 ×
[9] ≅ =
∂ xi xi − xi
1 2
∆ xi 10–3, respectively. Table 3 compares the results of the FOSM
method and the spreadsheet-based Monte Carlo simulation.
where FS1 and FS2 are the factor of safety calculated for val- The FOSM method appears to be a reasonable approach for
ues of the input variable xi equal to x1i and x i2 , respectively, estimating the mean and variance of the factor of safety.
with all other input variables assigned their mean values. However, the uncertainty about the shape of the probability
Usually, x1i and x i2 are taken equal to one standard deviation density function of the factor of safety introduces uncertain-
above and below the mean E[xi]. To account for spatial vari- ties in estimating the probability of unsatisfactory perfor-
ability and the reduction in the variance of the average pa- mance, as illustrated in Table 3.
rameters, the variance of measured data, σ 2 [xi], is reduced
by a reduction factor f. Vanmarcke (1977a) suggested that Simplified analysis
the variance reduction factor can be approximated by The spatial variability of soil properties and pore-water
© 2002 NRC Canada

I:\cgj\Cgj39\Cgj-03\T02-034.vp
Tuesday, May 14, 2002 10:37:17 AM
Color profile: Generic CMYK printer profile
Composite Default screen

678 Can. Geotech. J. Vol. 39, 2002

Table 2. FOSM calculations of the variance of the factor of safety.


Variance, σ 2 [x]
Inherent Systematic Inherent variability Systematic uncertainty
Input variable variability uncertainty ∆FS/∆x f ≈ 2ro /L (∆FS/∆x)2fσ 2 [x] (∆FS/∆x)2σ 2 [x]
φ fill (°) 1.00 3.00 0.009 1.00 0.0001 0.0003
γ fill (kN/m3) 1.00 1.00 0.061 0.24 0.0009 0.0037
tcr (m) 0.19a 0.04 0.007 — 0.0000 0.0000
SuM-1 (kPa) 66.26 7.60 0.005 1.00 0.0019 0.0002
SuM-2 (kPa) 66.26 7.60 0.005 1.00 0.0019 0.0002
SuL (kPa) 74.82 24.90 0.022 0.37 0.0129 0.0115
Dtill (m) 0.00 1.00 0.055 0.00 0.0000 0.0030
Output
Σ 0.0180 0.0190
σ 2 [FS] 0.0370
σ[FS] 0.1920
a
Variance is reduced by f = 0.2 based on the assessment of Christian et al. (1994).

Table 3. Comparing the outputs of different analysis approaches.


Method of analysis E[FS] σ[FS] Skewness Pu β
Spreadsheet-based probabilistic slope analysis 1.46 0.20 0.30 4.70×10–3 2.32
FOSMa 1.46 0.19 Not available 8.40×10–3; 2.50×10–3 2.42
Simplified analysis 1.46 0.25 0.32 2.37×10–2 1.84
a
The value 8.40 × 10–3 is based on the assumption that the probability density function of the factor of safety is normal, and the value 2.50 × 10–3
assumes the probability density function of the factor of safety is log-normal.

Table 4. Input variables and statistical parameters for the simplified analysis (based on Christian
et al. 1994).
Soil parameters Bias factors
Input variable Mean Variance Mean Variance
Fill friction angle, φ fill (°) 30 4 — —
Fill unit weight, γ fill (kN/m3) 20 2 — —
Clay crust thickness, tcr (m) 4 1 — —
Shear strength of marine clay, SuM (kPa) 34.5 66.26 — —
Bjerrum vane correction factor for SuM, µ M — — 1 0.006
Shear strength of lacustrine clay, SuL (kPa) 31.2 74.82 — —
Bjerrum vane correction factor for SuL, µ L — — 1 0.023

pressure is a major source of parameter uncertainty. How- the algorithm of Hassan and Wolff (1999). Based on trial
ever, probabilistic analyses ignoring the issues of spatial simulations, the latter slip surface yielded a slightly higher
variability and statistical uncertainty and relying only on the probability of unsatisfactory performance and was subse-
means, variances, and probability distributions of measured quently used in the probabilistic assessment. A Microsoft®
data are not uncommon (e.g., Nguyen and Chowdhury 1984; Excel 97 model and a Monte Carlo simulation using 32 000
Wolff and Harr 1987; Duncan 2000). Such an approach, re- iterations and a seed number of 31 069 gave a mean factor
ferred to hereafter as a simplified analysis, can be erroneous of safety of 1.46, with a standard deviation of 0.25 and a co-
and misleading. efficient of skewness of 0.32. The probability of unsatisfac-
To illustrate the errors incurred by the simplified approach, tory performance is calculated to be 2.39 × 10–2. Based on
the James Bay dykes are reanalyzed probabilistically, ignor- the results of 25 simulations, the mean probability of unsat-
ing the spatial variability of soil properties. The analysis is isfactory performance is 2.37 × 10–2, with the 95% confi-
based directly on the probability distributions of the data dence interval between 2.34 × 10–2 and 2.41 × 10–2. The
with no variance reduction and no considerations of statisti- reliability index is 1.84. Table 3 compares the results of the
cal uncertainties. Table 4 summarizes the input variables and simplified analysis with those of the previous analyses. The
the statistical parameters used in the analysis, based on simplified analysis overestimates the probability of unsatis-
Christian et al. (1994). All variables are assumed to be nor- factory performance by a factor of 5. This is attributed to the
mally distributed. significant reduction in the uncertainty due to soil variability
The deterministic critical slip surface almost coincided as a result of spatial averaging, which is taken into account
with the surface of minimum reliability index according to in all the analyses but the simplified analysis. In slopes dom-

© 2002 NRC Canada

I:\cgj\Cgj39\Cgj-03\T02-034.vp
Tuesday, May 14, 2002 10:37:18 AM
Color profile: Generic CMYK printer profile
Composite Default screen

El-Ramly et al. 679

inated by uncertainties due to the spatial variability of soil of a design could be significantly reduced by the use of
properties, the simplified analysis could significantly overes- empirical factors and correlations without the designer even
timate the probability of unsatisfactory performance. realizing this. Understanding the limitations and, more im-
portantly, the reliability of such factors and correlations
prior to using them is essential.
Conclusions It is our view that combining conventional deterministic
Conventional slope design practice addresses uncertainty slope analysis and probabilistic analysis will be beneficial to
only implicitly and in a subjective manner, thus compromis- slope engineering practice and will enhance the decision-
ing the adequacy of projections. Without proper consider- making process. It is important to note, however, that simpli-
ation of uncertainty, the factor of safety alone can give a fied probabilistic analyses can be erroneous and misleading.
misleading sense of safety and is not a sufficient safety indi- For example, ignoring the spatial variability of soil properties
cator. Probabilistic slope stability analysis is a rational and assuming perfect autocorrelations as in our simplified
means to incorporate quantified uncertainties into the design analysis can significantly overestimate the probability of un-
process. An important conclusion of this study is that proba- satisfactory performance.
bilistic analyses can be applied in practice without an exten-
sive effort beyond that needed in a conventional analysis.
The stated obstacles impeding the adoption of such tech- Acknowledgements
niques into geotechnical practice are more apparent than real. The authors would like to thank the Natural Sciences and
The developed probabilistic spreadsheet approach, based Engineering Research Council of Canada for providing the
on Monte Carlo simulation, makes use of Microsoft® Excel financial support for this research. We are also grateful for
97 and @Risk software, which are familiar and readily avail- useful discussions with colleagues at the University of Al-
able to most engineers. The underlying procedures and con- berta and elsewhere.
cepts are simple and transparent, requiring only fundamental
knowledge of statistics and probability theory. At the same
time, the analysis accounts for the spatial variability of the References
input variables, the statistical uncertainty due to limited data,
Alonso, E.E. 1976. Risk analysis of slopes and its application to
and the bias in the empirical factors and correlations used. slopes in Canadian sensitive clays. Géotechnique, 26: 453–472.
The approach is flexible in handling real slope problems, in- Ang, A.H-S., and Tang, W.H. 1975. Probability concepts in engi-
cluding various loading conditions, complex stratigraphy, c– neering planning and design. Vol. I. Basic principles. John Wiley,
φ soils, and circular and noncircular slip surfaces. New York.
Probabilistic slope stability analysis provides practicing Ang, A.H-S., and Tang, W.H. 1984. Probability concepts in engi-
engineers with valuable insights that cannot be reached oth- neering planning and design. Vol. II. Decision, risk and reliabil-
erwise. For example, the level of uncertainty in the factor of ity. John Wiley, New York.
safety is quantified through the variance of the factor of Baecher, G.B. 1987. Statistical analysis of geotechnical data. Final
safety and the probability of unsatisfactory performance. Report GL-87-1, U.S. Army Corps of Engineers, Waterways Ex-
This could have an important impact on decisions about a periment Station, Vicksburg, Miss.
design factor of safety. If the reliability of the computed fac- Bergado, D.T., Patron, B.C., Youyongwatana, W., Chai, J.C., and
tor of safety is deemed high, the profession may be willing Yudhbir. 1994. Reliability-based analysis of embankment on
to adopt lower design factors of safety than usual, provided soft Bangkok clay. Structural Safety, 13: 247–266.
that the serviceability of the slope is not compromised. With Bishop, A.W. 1955. The use of the slip circle in the stability analy-
increasingly sparse funds, many agencies and organizations sis of slopes. Géotechnique, 5: 7–17.
prioritize slope repair and maintenance expenditures, as is Bjerrum, L. 1972. Embankments on soft ground. In Proceedings of
the case of hydraulic structures, based on safety levels. the American Society of Civil Engineers Specialty Conference on
Comparing the safety of different structures based on the Performance of Earth and Earth-Supported Structures, Purdue
factor of safety alone is inadequate because the underlying University, Lafayette, Ind., June 11–14, Vol. 2, pp. 1–54.
sources and levels of uncertainty are not addressed. By com- Chowdhury, R.N. 1984. Recent developments in landslide studies:
bining the most likely value of the factor of safety and the probabilistic methods. In Proceedings of the 4th International
Symposium on Landslides, Toronto, Ont., September 16–21. Ca-
uncertainty in that value, the probability of unsatisfactory
nadian Geotechnical Society, Vol. 1, pp. 209–228.
performance and the reliability index provide a sounder ba-
Chowdhury, R.N., and Tang, W.H. 1987. Comparison of risk mod-
sis for the comparison.
els for slopes. In Proceedings of the 5th International Confer-
The practical value of quantifying the relative contribu- ence on Applications of Statistics and Probability in Soil and
tions of the various sources of uncertainty to the overall un- Structural Engineering, Vancouver, B.C., May 25–29, Vol. 2,
certainty of the factor of safety through sensitivity analyses, pp. 863–869.
using Spearman rank correlation coefficient, cannot be un- Christian, J.T. 1996. Reliability methods for stability of existing
derestimated. Such information allows available resources, slopes. In Proceedings of Uncertainty ‘96. Geotechnical Special
whether intellectual or physical, to be rationally allocated to- Publication 58, Vol. 2, pp. 409–419.
wards reducing the uncertainties of the variables with the Christian, J.T., Ladd, C.C., and Baecher, G.B. 1994. Reliability and
largest impact on design. The sensitivity analyses undertaken probability in stability analysis. Journal of Geotechnical Engi-
in this study showed that the uncertainty of Bjerrum’s vane neering, ASCE, 120: 1071–1111.
correction factor is substantial and could have a large impact Decisioneering Inc. 1996. Crystal ball: Monte Carlo simulation software
on the reliability of the design. This warns that the reliability for spreadsheet risk analysis. Decisioneering Inc., Denver, Colo.

© 2002 NRC Canada

I:\cgj\Cgj39\Cgj-03\T02-034.vp
Tuesday, May 14, 2002 10:37:18 AM
Color profile: Generic CMYK printer profile
Composite Default screen

680 Can. Geotech. J. Vol. 39, 2002

DeGroot, D.J. 1996. Analyzing spatial variability of in situ soil Neter, J., Wasserman, W., and Kutner, M.H. 1990. Applied linear
properties. In Proceedings, Uncertainty ‘96. Geotechnical Spe- statistical models. Richard D. Irwin Inc., Boston.
cial Publication 58, Vol. 1, pp. 210–238. Nguyen, V.U., and Chowdhury, R.N. 1984. Probabilistic study of
DeGroot, D.J., and Baecher, G.B. 1993. Estimating autocovariance spoil pile stability in strip coal mines — two techniques com-
of in-situ soil properties. Journal of Geotechnical Engineering, pared. International Journal of Rock Mechanics, Mining Science
ASCE, 119(1): 147–166. and Geomechanics, 21: 303–312.
Deutsch, C.V., and Journel, A.G. 1998. GSLIB: geostatistical software Nguyen, V.U., and Chowdhury, R.N. 1985. Simulation for risk
library and user’s guide. Oxford University Press, New York, analysis with correlated variables. Géotechnique, 35: 47–58.
543 pp. Palisade Corporation. 1996. @Risk: risk analysis and simulation
Duncan, J.M. 2000. Factors of safety and reliability in geotechnical add-in for Microsoft Excel or Lotus 1-2-3. Palisade Corporation,
engineering. Journal of Geotechnical and Geoenvironmental En- Newfield, N.Y.
gineering, ASCE, 126: 307–316. Priest, D.S., and Brown, E.T. 1983. Probabilistic stability analysis
El-Ramly, H. 2001. Probabilistic analyses of landslide hazards and of variable rock slopes. Transactions of the Institute of Mining
risks: bridging theory and practice. Ph.D. thesis, University of and Metallurgy, 92: A1–A12.
Alberta, Edmonton, Alta. Soulié, M., Montes, P., and Silvestri, V. 1990. Modelling spatial
Harr, M.E. 1977. Mechanics of particulate media — a probabilistic variability of soil parameters. Canadian Geotechnical Journal,
approach. McGraw-Hill, New York, 543 pp. 27: 617–630.
Harr, M.E. 1987. Reliability-based design in civil engineering. Tang, W.H., Yucemen, M.S., and Ang, A.H.S. 1976. Probability-
McGraw-Hill Inc., New York, 290 pp. based short-term design of slopes. Canadian Geotechnical Jour-
Hassan, A., and Wolff, T. 1999. Search algorithm for minimum re- nal, 13: 201–215.
liability index of earth slopes. Journal of Geotechnical and Geo- Tobutt, D.C. 1982. Monte Carlo simulation methods for slope sta-
environmental Engineering, ASCE, 125: 301–308. bility. Computers and Geosciences, 8: 199–208.
Honjo, Y., and Kuroda, K. 1991. A new look at fluctuating geo- U.S. Army Corps of Engineers. 1992. Reliability assessment of
technical data for reliability design. Soils and Foundations, 31: navigation structures. Engineering Technical Letter 1110-2-532,
110–120. U.S. Army Corps of Engineers, Washington, D.C.
Jaksa, M.B., Brooker, P.I., and Kaggwa, W.S. 1997. Inaccuracies U.S. Army Corps of Engineers. 1995. Introduction to probability
associated with estimating random measurement errors. Journal and reliability methods for use in geotechnical engineering. En-
of Geotechnical and Geoenvironmental Engineering, ASCE, gineering Technical Letter 1110-2-547, U.S. Army Corps of En-
123: 393–401. gineers, Washington, D.C.
Kulhawy, F.H., and Trautmann, C.H. 1996. Estimation of in-situ test Vanmarcke, E.H. 1977a. Probabilistic modeling of soil profiles.
uncertainty. In Proceedings of Uncertainty ‘96. Geotechnical Spe- Journal of the Geotechnical Engineering Division, ASCE, 103:
cial Publication 58, Vol. 1, pp. 269–286. 1227–1246.
Kulhawy, F.H., Roth, M.S., and Grigoriu, M.D. 1991. Some statistical Vanmarcke, E.H. 1977b. Reliability of earth slopes. Journal of the
evaluation of geotechnical properties. In Proceedings of the 6th In- Geotechnical Engineering Division, ASCE, 103: 1247–1265.
ternational Conference on Applications of Statistics and Probability Vanmarcke, E.H. 1980. Probabilistic stability analysis of earth slopes.
in Civil Engineering, Mexico City, Vol. 2, pp. 705–712. Engineering Geology, 16: 29–50.
Lacasse, S., and Nadim, F. 1996. Uncertainties in characterizing Vanmarcke, E.H. 1983. Random fields: analysis and synthesis. MIT
soil properties. In Proceedings of Uncertainty ‘96. Geotechnical Press, Cambridge, Mass.
Special Publication 58, Vol. 1, pp. 49–75. Whitman, V.W. 1984. Evaluating calculated risk in geotechnical
Ladd, C.C., Dascal, O., Law, K.T., Lefebrve, G., Lessard, G., Mesri, engineering. Journal of Geotechnical Engineering, ASCE, 110:
G., and Tavenas, F. 1983. Report of the subcommittee on em- 145–188.
bankment stability — annex II. Committee of Specialists on Sen- Wolff, T.F. 1996. Probabilistic slope stability in theory and prac-
sitive Clays on the NBR Complex, Société d’Energie de la Baie tice. In Proceedings of Uncertainty ‘96. Geotechnical Special
James, Montréal, Que. Publication 58, Vol. 2, pp. 419–433.
Law, A.M., and McComas, M.G. 1986. Pitfalls in the simulation of Wolff, T.F., and Harr, M.E. 1987. Slope design for earth dams. In
manufacturing systems. In Proceedings of the Winter Simulation Proceedings of the 5th International Conference on Applications
Conference, Washington, D.C., pp. 539–542. of Statistics and Probability in Soil and Structural Engineering,
Li, K.S. 1991. Discussion: probabilistic potentiometric surface map- Vancouver, B.C., May 25–29, Vol. 2, pp. 725–732.
ping. Journal of Geotechnical Engineering, ASCE, 117: 1457–1458. Yucemen, M.S., and Al-Homoud, A.S. 1990. Probabilistic three-
Lumb, P. 1966. The variability of natural soils. Canadian Geotechnical dimensional stability analysis of slopes. Structural Safety, 9: 1–20.
Journal, 3: 74–97.
Major, G., Ross-Brown, D.M., and Kim, H-S. 1978. A general
probabilistic analysis for three-dimensional wedge failures. In
Proceedings of the 19th U.S. Symposium on Rock Mechanics, List of symbols
Stateline, Nev., 1–3 May, pp. 45–56.
Matsuo, M., and Kuroda, K. 1974. Probabilistic approach to design b slice width
of embankments. Soils and Foundations, 14: 1–17. B bias correction factor
Microsoft Corporation. 1997. Microsoft® Excel 97. Microsoft Cor- C cohesion
poration, Redmond, Wash. Dtill depth of till
Mostyn, G.R., and Li, K.S. 1993. Probabilistic slope stability anal- E[–] mean
ysis — state-of-play. In Probabilistic methods in geotechnical f variance reduction factor
engineering. Edited by K.S. Li and S.-C.R. Lo. A.A. Balkema, F(x) cumulative probability distribution function of variable x
Rotterdam. pp. 89–109. FS factor of safety

© 2002 NRC Canada

I:\cgj\Cgj39\Cgj-03\T02-034.vp
Tuesday, May 14, 2002 10:37:18 AM
Color profile: Generic CMYK printer profile
Composite Default screen

El-Ramly et al. 681

g(x1, x2, performance function Z2 distance from the beginning of the first interval to the
…, xn) end of the second interval
h total height of slice Z3 distance from the end of the first interval to the end of
hcr height in clay crust the second interval
hfill height in embankment fill α angle to the horizontal of slice base
hL height in lacustrine clay β reliability index
hM height in marine clay δ scale of fluctuation
l1, l2, l3 segments of slip surface of lengths l1, l2, and l3 ε residual component, equal to difference between x and t
L length over which soil parameters are averaged ε i residual component at location i
m number of iterations in Monte Carlo simulation φ friction angle
n number of observations φ′ effective friction angle
Pu probability of unsatisfactory performance φ fill fill friction angle
r separation distance between two variables in the space Φ variable representing soil stratigraphy
of a random field γ bulk unit weight
ro autocorrelation distance γ fill fill unit weight
ru pore-pressure ratio ξ soil parameter
Su undrained shear strength µ Bjerrum vane correction factor
SuL undrained shear strength of lacustrine clay µ L , µ M Bjerrum vane correction factor for SuL and SuM, respec-
SuM undrained shear strength of marine clay tively
t trend component for variable x ρ correlation coefficient between two variables
tcr clay crust thickness σ[–] standard deviation
ti trend component at location i σ 2 [–] variance
u pore-water pressure σ 2∆ z variance of X(∆z)
V input variable adjusted for statistical uncertainty and bias ∆z interval of length ∆z
W total height of slice ∆z′ interval of length ∆z′
x random variable Γ(∆z) variance function
xi value of random variable x at location i
X(∆z) local average of variable x over a length ∆z Appendix A
z distance
Z0 separation distance between two intervals ∆z, ∆z′ This appendix illustrates the general structure of the
Z1 distance from the beginning of the first interval to the spreadsheet model used in the probabilistic analysis of the
beginning of the second interval James Bay dyke.

© 2002 NRC Canada

I:\cgj\Cgj39\Cgj-03\T02-034.vp
Tuesday, May 14, 2002 10:37:19 AM
Color profile: Generic CMYK printer profile
Composite Default screen

682 Can. Geotech. J. Vol. 39, 2002

Fig. A1. Spreadsheet model, section 1: dyke geometry, stratigraphy, and input variables.

© 2002 NRC Canada

I:\cgj\Cgj39\Cgj-03\T02-034.vp
Tuesday, May 14, 2002 10:37:19 AM
Color profile: Generic CMYK printer profile
Composite Default screen

El-Ramly et al. 683

Fig. A2. Spreadsheet model, section 2: tables of factor of safety computations. G.S., ground surface; S.S., slip surface.

© 2002 NRC Canada

I:\cgj\Cgj39\Cgj-03\T02-034.vp
Tuesday, May 14, 2002 10:37:20 AM
View publication stats

Você também pode gostar