You are on page 1of 26

INTERNATIONAL JOURNAL FOR NUMERICAL AND ANALYTICAL METHODS IN GEOMECHANICS

Int. J. Numer. Anal. Meth. Geomech. 2010; 34:1–26


Published online 7 May 2009 in Wiley InterScience (www.interscience.wiley.com). DOI: 10.1002/nag.791

Effect of spatial variability of cross-correlated soil properties


on bearing capacity of strip footing

Sung Eun Cho1, ∗, †, ‡ and Hyung Choon Park2, §


1 Korea
Institute of Water and Environment, Korea Water Resources Corporation, 462-1,
Jeonmin-Dong, Yusung-Gu, Daejon 305-730, South Korea
2 Department of Civil Engineering, Chungnam National University, 220 Gung-Dong, Yusung-Gu,

Daejon 305-764, South Korea

SUMMARY
Geotechnical engineering problems are characterized by many sources of uncertainty. Some of these
sources are connected to the uncertainties of soil properties involved in the analysis. In this paper, a
numerical procedure for a probabilistic analysis that considers the spatial variability of cross-correlated
soil properties is presented and applied to study the bearing capacity of spatially random soil with
different autocorrelation distances in the vertical and horizontal directions. The approach integrates a
commercial finite difference method and random field theory into the framework of a probabilistic analysis.
Two-dimensional cross-correlated non-Gaussian random fields are generated based on a Karhunen–Loève
expansion in a manner consistent with a specified marginal distribution function, an autocorrelation
function, and cross-correlation coefficients. A Monte Carlo simulation is then used to determine the
statistical response based on the random fields. A series of analyses was performed to study the effects
of uncertainty due to the spatial heterogeneity on the bearing capacity of a rough strip footing. The
simulations provide insight into the application of uncertainty treatment to geotechnical problems and show
the importance of the spatial variability of soil properties with regard to the outcome of a probabilistic
assessment. Copyright q 2009 John Wiley & Sons, Ltd.

Received 16 May 2008; Revised 21 December 2008; Accepted 22 February 2009

KEY WORDS: bearing capacity; probabilistic analysis; spatial variability; Monte Carlo simulation

1. INTRODUCTION

Soil properties vary spatially even within homogeneous layers as a result of depositional and
post-depositional processes [1]. Nevertheless, most geotechnical analyses adopt a deterministic

∗ Correspondence to: Sung Eun Cho, Korea Institute of Water and Environment, Korea Water Resources Corporation,
462-1, Jeonmin-Dong, Yusung-Gu, Daejon 305-730, South Korea.

E-mail: drsecho@hanmail.net

Senior Researcher.
§ Assistant Professor.

Copyright q 2009 John Wiley & Sons, Ltd.


2 S. E. CHO AND H. C. PARK

approach based on single soil parameters applied to each distinct layer. In response, numerous
studies have been undertaken in recent years to develop a probabilistic analysis method that deals
with the uncertainties of soil properties in a systematic manner [2–5]. Detailed reviews of these
studies can be found in Mostyn and Li [6], Elkateb et al. [7], and Baecher and Christian [8].
Although probabilistic analysis methods do not consider all of the components of design where
judgment needs to be utilized and do not suggest the level of reliability that should be targeted [9],
working within a probabilistic framework does imply that the reliability of the system can be
considered in a logical manner. Thus, probabilistic models can facilitate the development of
new perspectives concerning risk and reliability, which are outside the scope of conventional
deterministic models.
In particular, the effect of inherent random variations of soil properties on the response of
geotechnical structures has received considerable attention in recent years. Griffiths and Fenton
[10, 11], Fenton and Griffiths [12], and Popescu et al. [13] examined the response of shallow
foundations; Haldar and Babu [14] analyzed the response of a deep foundation under vertical load;
Paice et al. [15] studied settlements of foundations on elastic soil; Griffiths and Fenton [16] studied
slope stability; Popescu et al. [17, 18] and Koutsourelakis et al. [19] studied seismically induced
soil liquefaction; and Kim et al. [20] reported on emergent phenomena related to variability in soil
properties.
In this study, a numerical procedure for a probabilistic analysis that considers the spatial vari-
ability of soil properties is presented. The approach integrates a commercial finite difference
method and random field theory into a probabilistic analysis. Soils with spatially varying shear
strength are modeled as anisotropic random fields with different autocorrelation distances in the
vertical and horizontal directions, and an elasto-plastic finite difference analysis is subsequently
performed to evaluate the effects of spatial variability of cross-correlated strength parameters on
the bearing capacity of a footing.
The framework presented by Vořechovský [21] is adopted to generate non-Gaussian cross-
correlated random fields with a specified marginal distribution function, an autocorrelation function,
and cross-correlation coefficients. The approach is combined into the well-known Karhunen–Loève
(KL) expansion method for simulation of a Gaussian random field. Then the simulated Gaussian
random field is transformed to a non-Gaussian random field.
This study focuses on inherent soil variability, where probabilistic analyses can be employed
to assess the effect of this type of variability on a geotechnical structure. The importance of the
effects of the cross-correlation coefficient and the horizontal and vertical autocorrelation distances
of soil properties on the bearing capacity of a rough strip footing is highlighted.

2. BEARING CAPACITY OF A SHALLOW STRIP FOOTING

Terzaghi [22] suggested the following well-known form of the bearing capacity formula for a
centrally and vertically loaded, shallow strip footing:

B
qu = cNc +D Nq + N (1a)
2
where c is the cohesion of the soil,  is the unit weight of the soil, D and B are the depth and the
width, respectively, of the footing, and Nc , Nq , and N are bearing capacity factors.

Copyright q 2009 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2010; 34:1–26
DOI: 10.1002/nag
EFFECT OF SPATIAL VARIABILITY OF SOIL PROPERTIES 3

While the exact values for N remain unknown, the bearing capacity factors Nc and Nq have
been solved analytically for a weightless soil using the method of characteristics [23] under the
assumption that the soil satisfies an associated flow rule:
 
   tan 
Nq = tan 2
+ e (1b)
4 2

Nc = (Nq −1) cot  (1c)

Under the assumption of a weightless soil, the bearing capacity equation simplifies to the first term
of Equation (1a) (qu = cNc ).
Probabilistic studies on the bearing capacity of a footing have been reported previously for
cohesive soil by considering randomly distributed undrained strength [10, 11]. Fenton and Griffiths
[12] also investigated the influence of cross-correlation between the cohesion and friction angle
on the bearing capacity for c– soil. They assumed the random field to be statistically isotropic
(the same autocorrelation distance in any direction through the soil). Popescu et al. [13] studied the
bearing capacity problem by modeling soil properties as homogeneous non-Gaussian random fields.
They performed parametric studies to assess the influence of various probabilistic characteristics
of soil properties on the bearing capacity of a strip foundation placed at ground level on an
overconsolidated clay layer under undrained conditions. Soubra et al. [24] studied the effect of the
spatial variability of soil properties on the ultimate bearing capacity of a vertically loaded shallow
strip footing. They modeled the cohesion and friction angle as non-normal anisotropic random
fields based on the spectral representation method.
This paper addresses the bearing capacity of a strip footing located on the surface of a c– soil
under the action of a vertical, central load. The analyses were performed by applying a controlled
downward velocity (displacement per calculation step) to the surface nodes on the base of the
footing. Although the spatial variation of soil properties might cause rotation of the footing, which
cannot be predicted in homogeneous soil [13], this aspect is not considered in the present study
for simplicity.
The contact stress is calculated by dividing the sum of the vertical footing nodal forces by the
width of the footing extended to the center of the first element outside the footing, following the
analysis presented in the FLAC manual [25].

3. RANDOM FIELD MODEL

3.1. The spatial variability of soil


One of the main sources of heterogeneity is inherent spatial soil variability, i.e. the variation of soil
properties from one point to another in space due to different depositional conditions and different
loading histories [7].
Spatial variation is not a random process; rather it is controlled by location in space. Statistical
parameters such as the mean and variance are one-point statistical parameters and cannot capture
the features of the spatial structure of the soil [26]. Spatial variations of soil properties can
be effectively described by their correlation structure (i.e. autocorrelation function) within the
framework of random fields [27].

Copyright q 2009 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2010; 34:1–26
DOI: 10.1002/nag
4 S. E. CHO AND H. C. PARK

Two indices of correlation, namely, scale of fluctuation [27] and autocorrelation distance [28],
have been used to describe the spatial extent within which soil properties show a strong correlation.
The autocorrelation distance is defined as the distance to which the autocorrelation function decays
to 1/e, where e is the base of natural logarithms.
The scale of fluctuation  is defined as
 ∞
= () d (2)
−∞

where () is the autocorrelation function and  is the separation of two points. For the exponential
autocorrelation function, the scale of fluctuation function is equal to two times the autocorrelation
distance [29].
A large autocorrelation distance value implies that the soil property is highly correlated over a
large spatial extent, resulting in a smooth variation within the soil profile. On the other hand, a
small value indicates that the fluctuation of the soil property is large.
Although an isotropic correlation structure is often assumed in works reported in the literature,
correlations in the vertical direction tend to have much shorter distances than those in the horizontal
direction due to the geological soil formation process for most natural soil deposits. A ratio of
about 1 to 10 for these autocorrelation distances is common [8].
A Gaussian random field is completely defined by its mean (x), variance 2 (x), and autocor-
relation function (x, x  ). Autocorrelation functions commonly used in geotechnical engineering
have been presented by Li and Lumb [4] and Rackwitz [30]. In this study, an exponential auto-
correlation function is used and different autocorrelation distances in the vertical and horizontal
directions are used as follows:
 
|x − x  | |y − y  |
(x, y) = exp − − (3)
lh lv
where lh and lv are autocorrelation distances in the horizontal and vertical directions, respectively.

3.2. Discretization of random fields


The spatial fluctuations of a parameter cannot be accounted for if the parameter is modeled by
only a single random variable. Therefore, it is reasonable to use random fields for a more accurate
representation of the variations when spatial uncertainty effects are directly included in the analysis.
Because of the discrete nature of numerical methods such as finite element or finite difference
formulation, a continuous-parameter random field must also be discretized into random variables.
This process is commonly known as discretization of a random field.
Several methods have been developed to carry out this task, such as the spatial average method,
the midpoint method, and the shape function method. These early methods are relatively inefficient,
in the sense that a large number of random variables are required to achieve a good approximation
of the field. More efficient approaches for discretization of random fields using series expansion
methods such as the KL expansion, the orthogonal series expansion, and the expansion optimal
linear estimation method have been introduced [31].
A comprehensive review and comparison of these discretization methods have been presented
by Sudret and Der Kiureghian [32] and Matthies et al. [33].
All series expansion methods result in a Gaussian field, which is exactly represented as a
series involving random variables and deterministic spatial functions depending on the correlation

Copyright q 2009 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2010; 34:1–26
DOI: 10.1002/nag
EFFECT OF SPATIAL VARIABILITY OF SOIL PROPERTIES 5

structure of the field. The approximation is then obtained as a truncation of the series. The accuracy
of the representation depends on the number of terms used in the series expansion and the particular
expansion method used.
In this study, the KL expansion is adopted to discretize anisotropic random fields of soil
properties in the two-dimensional space, since the method requires the minimum number of terms
for a specified level of accuracy compared with other mathematical representations [34].

3.3. The KL expansion


A random field H (x,
) is a collection of random variables associated with a continuous index
x ∈  ⊆ R n , where
∈  is the coordinate in the outcome space. The KL expansion of a random
field H (x,
) is based on the spectral decomposition of its autocorrelation function (x, x  ), which
is bounded, symmetric, and positive definite. The set of deterministic functions over which any
realization of the field H (x,
0 ) is expanded is defined by the eigenvalue problem as

(x, x  ) i (x  ) dx  = i i (x) (4)

where i and i , respectively, denote the eigenfunctions and eigenvalues of the autocorrelation
function.
The series of the deterministic set forms the expansion of H (x,
):
∞ 

H (x,
) = + i i (x) i (
), x ∈  (5)
i=1

where i (
) is a set of orthogonal random coefficients (uncorrelated random variables with zero
mean and unit variance).
The KL expansion is mean-square convergent irrespective of the probabilistic structure of the
process being expanded, provided it has a finite variance. The monotony of the decay in the
magnitude is guaranteed by the symmetry of the correlation function, and the rate of the decay is
inversely proportional to the autocorrelation distance of the process being expanded [34].
The approximate random field is defined by truncating the ordered series given in Equation (5):
M 

Ĥ (x,
) = + i i (x) i (
), x ∈ (6)
i=1
The number M to be chosen strongly depends on the desired accuracy and on the autocorrelation
function of the random field.
In the case of an exponential autocorrelation function (Equation (3)) for a one-dimensional case,
the eigenvalue problem (Equation (4)) can be solved analytically. Extension to two-dimensional
fields defined for the correlation function on a rectangular domain can be obtained as well. Detailed
closed-form solutions can be found in Spanos and Ghanem [35] and Ghanem and Spanos [36].
For complicated domains and more general correlation functions, however, the integral eigenvalue
problem has to be solved numerically. A Galerkin-type procedure was suggested by Ghanem and
Spanos [36].

3.4. Cross-correlated Gaussian random fields


Typically, more than one random property is involved in geotechnical problems. For example,
Young’s modulus, Poisson’s ratio, the cohesion, and the friction angle can be considered as random

Copyright q 2009 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2010; 34:1–26
DOI: 10.1002/nag
6 S. E. CHO AND H. C. PARK

properties in the problem of bearing capacity. In a probabilistic concept, in principle all these
quantities can be modeled by random fields.
The present study deals with cases where all fields simulated over a region  share an identical
autocorrelation function over , and the cross-correlation structure between each pair of simulated
fields is simply defined by a cross-correlation coefficient. This is reasonable since the spatial
correlation structure is caused by changes in the constitutive nature of the soil over space [12].
Therefore, the modal decomposition of the given autocorrelation function is done only once.
The cross-correlated fields are then expanded using the same spectrum of eigenfunctions and
eigenvalues, but the sets of random variables used for the expansion of each field are cross-
correlated [21].
In this study, each field of cohesion and friction angle is expanded using a set of independent
random variables, and these sets are then correlated with respect to the cross-correlation matrix
between two expanded random fields according to the framework presented by Vořechovský [21].
Let block sample matrix vD , which consists of two blocks, be a jointly normally distributed
random vector. Each block viD (i = c, ) represents a Gaussian random vector with M standard
Gaussian independent random variables, while the vectors vD c , v are cross-correlated with the
D

cross-correlation coefficient between the cohesion and the friction angle.


Each approximate Gaussian random field Ĥi is then expanded using each block viD of the
random vector vD as follows:


M 
Ĥi (x,
) = i + i j j (x)i,Dj (
) (for i = c, ) (7)
j=1

To generate cross-correlated non-Gaussian random fields by the KL expansion, input random


variables must be transformed into standardized Gaussian random variables to assess the cross-
correlation and autocorrelation characteristics for the standardized variables. If the autocorrelation
function in the original (non-Gaussian) space is given, corrections must be made for each field
over the whole range of autocorrelation coefficients ˜ of each pair of non-Gaussian variables to
transform the original correlations into the Gaussian space, since the spectral decomposition of
the autocorrelation structure is carried out in the Gaussian space. If the target cross-correlation
coefficient in the original (non-Gaussian) space is given, a corrected cross-correlation coefficient
must also be found prior to simulation of the cross-correlated random vector vD . In this study, the
Nataf model [37] was used to transform a non-Gaussian multivariate distribution into a standard-
ized Gaussian distribution.

3.5. Sampling strategies of random variables


To generate a random field, it is necessary to simulate the random vector. In this study, the
Latin hypercube sampling technique is used to generate the block sample matrix vD . This
technique can be viewed as a stratified sampling scheme designed to ensure that the upper
or lower ends of the distributions are well represented. Latin hypercube sampling is generally
recommended over simple random sampling when the model is complex or when time is an
issue.
In this study, the method proposed by Stein [38] for inducing correlation among the variables
based on the rank of a target multivariate distribution is implemented in Matlab.

Copyright q 2009 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2010; 34:1–26
DOI: 10.1002/nag
EFFECT OF SPATIAL VARIABILITY OF SOIL PROPERTIES 7

3.6. Transformation to non-Gaussian random fields


Although a Gaussian random field is often used to model uncertainties with spatial variability for
reasons of convenience and lack of available data [13], the Gaussian model is not applicable in
many situations where the random variable is always positive.
For convenience, we find an underlying Gaussian random field H that can be easily transformed
into the target field H̃. If the random variables are considered to be lognormally distributed, then
appropriate lognormal random fields can be obtained by exponentiating the approximate Gaussian
field from Equation (7) as follows:


M 
H̃i (x,
) = exp i + i j j (x)i,Dj (
) (for i = c, ) (8)
j=1
In this study, random variables are assumed to be characterized statistically by a lognormal distri-
bution.
The procedure for the stochastic analysis of bearing capacity with random fields is presented in
Figure 1.

Statistical input
marginal distribution
autocorrelation
cross-correlation

Solve eigenvalue Latin hypercube


problem Eq. (4) Sampling, χ D

Cross-correlated random field


Simulate
Nsim realizations

FDM analysis

Curve fitting

Statistical response
probability density
cumulative probability
probability of failure

Figure 1. The procedure for the stochastic analysis of bearing capacity with random fields.

Copyright q 2009 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2010; 34:1–26
DOI: 10.1002/nag
8 S. E. CHO AND H. C. PARK

4. PROBABILISTIC ANALYSIS

4.1. Limit state function and computational method


The problem of the probabilistic analysis is formulated by a vector, X = [X 1 , X 2 , X 3 , . . . , X k ],
representing a set of random variables. From the uncertain variables, a limit state function g(X)
is formulated to describe the limit state in the space of X. In the n-dimensional hyperspace of
the basic variables, g(X) = 0 is the boundary between the region in which the allowable bearing
capacity is not exceeded and the region in which it is exceeded. The probability of failure of the
footing is then given by the following integral:

P f = P[g(X0)] = f X (X) dX (9)
g(X)0

where f X (X) represents the joint probability density function and the integral is carried out over
the failure domain.
The limit state function concerned with the maximum load that can be placed on the footing
just prior to a bearing capacity failure is typically defined by the difference between the capacity
C and demand D [39]:

g(X) = C − D (10)

where D is the allowable bearing capacity evaluated deterministically with the mean values of
c and  and using the factor of safety. An Finite Difference Method (FDM) analysis is used to
describe the above limit state function by calculating the bearing capacity, C.
The probability of failure is then calculated from Equation (9) as

P f = P[CD] = P[Cqu /FS] (11)

For practical problems, direct evaluation of the k-fold integral in Equation (9) is virtually impossible.
The difficulty lies in the fact that complete probabilistic information on the soil properties is
not available and the domain of integration is a complicated function. Therefore, approximate
techniques have been developed to evaluate this integral.
Although various stochastic methods have been proposed in the literature, the only currently
available universal method for accurate solution of geotechnical problems is the Monte Carlo
technique, mainly due to the large variability and strong non-linearity of soil properties [13]. In a
Monte Carlo simulation, a series of random fields are generated in a manner consistent with their
probability distribution and correlation structure, and the response is calculated for each generated
set. The process is repeated many times to evaluate the probability of failure by determining
whether the limit state functions are exceeded. However, the Monte Carlo Simulation method is
not limited to the calculation of the probability of failure. Various statistical properties evaluated
after the process of simulation, such as mean, variance, coefficient of skewness, probability density
functions, and cumulative probability distribution functions, can provide a broader perspective and
a more comprehensive description of a given system.

Copyright q 2009 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2010; 34:1–26
DOI: 10.1002/nag
EFFECT OF SPATIAL VARIABILITY OF SOIL PROPERTIES 9

5. EXAMPLE ANALYSIS

This section describes plane strain finite difference analyses carried out to calculate the bearing
capacity of a rigid strip footing founded on a weightless soil with shear strength parameters
c and  represented by spatially varying and cross-correlated random fields. The analysis is two-
dimensional, corresponding to a strip footing with infinite autocorrelation distance in the out of
plane direction and assuming elastic–perfectly plastic behavior of the soil material with Mohr–
Coulomb yield criterion.

5.1. Deterministic analysis


To assess the ability of the numerical analysis to predict the bearing capacity, a deterministic
analysis was performed using the mean values of shear strength for homogeneous soil. A strip
footing with a width of B = 2 m is located on a c– soil having properties as given in Table I. The
finite difference grid consists of 1050 zones and is 14 m wide by 6 m deep, as shown in Figure 2.
Horizontal movements on the vertical boundaries of the grid were restrained, while the base of
the grid was not allowed to move in either horizontal or vertical direction. A rough strip footing
was simulated by setting the horizontal velocity of the nodes representing the footing to zero.
The results of the deterministic analysis are shown in Figure 3. The bearing capacity was
estimated to be 1.01 MPa from the bearing pressure–settlement curve presented in Figure 6(a). This
shows relatively good agreement with the value (1.04 MPa) obtained from Equation (1), despite
that the equation is only approximate due to the disagreement between the angle of dilation and
the friction angle.

Table I. Statistical properties of soil parameters.


Parameter X COV Correlation coefficient
Cohesion c (kPa) 50 0.3 −0.7r 0.5
Friction angle  (deg.) 25 0.2
Shear modulus G (MPa) 100 — —
Bulk modulus K (MPa) 200 — —
Dilation angle  (deg.) 0 — —
(5lh 30 and 1lv 10).

2m

6m

14 m

Figure 2. Grid used in the analysis of bearing capacity.

Copyright q 2009 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2010; 34:1–26
DOI: 10.1002/nag
10 S. E. CHO AND H. C. PARK

Figure 3. Results of deterministic analysis for mean values: (a) velocity vector and
(b) maximum shear strain rate.

Vectors of velocity and contours of maximum shear strain rate at steady plastic flow for the
analysis are shown in Figure 3. These show a well-defined wedge-shaped zone remaining elastic
immediately below the center of the footing.
More accurate solutions can be obtained by using a more refined mesh around the edge of the
footing. However, these solutions involve larger computational times owing to the singularity at
the edge of the footing. Hence, considering accuracy and efficiency, the mesh shown in Figure 2
has been retained for the work described in this study.

5.2. Stochastic simulations


In this section, application of the presented procedure is illustrated through a series of simulations.
To obtain accurate statistical responses such as the mean, standard deviation, and probability
density function, 5000 sets of random fields were generated for each case based on the statistical
information. A series of analyses was then performed based on the generated random fields.
A FISH (the built-in programming language of FLAC) function was written to generate random
fields from Equation (8) based on the solutions of the eigenvalue problem and the sampled
vector vD .
As the strength of the soil is spatially distributed, random variable soil strength parameters related
to the bearing capacity of the footing, including the friction angle and cohesion, are considered as
random fields. The bulk modulus K and shear modulus G were assumed to be deterministic since
the bearing capacity is not sensitive to these variables. Table I summarizes the statistical properties
of the soil parameters.

Copyright q 2009 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2010; 34:1–26
DOI: 10.1002/nag
EFFECT OF SPATIAL VARIABILITY OF SOIL PROPERTIES 11

Random variables, the cohesion and the friction angle, are assumed to be characterized statisti-
cally by a lognormal distribution defined by a mean  X and a standard deviation X . The lognormal
distribution ranges between zero and infinity, skewed to the low range, and is therefore particularly
suited for parameters that cannot take on negative values. Once the mean and standard deviation
are expressed in terms of the dimensionless coefficient of variation (COV), defined as V X = X / X ,
the mean and standard deviation of the underlying normal distribution of ln X are then given by

ln X = ln{1+ V X2 } (12)

ln X = ln  X −0.5 2ln X (13)


In order to incorporate the dependence between the strength parameters, the cross-correlation
coefficient r (c, ) is needed. Wolff [40] reported the correlation between c and  for CU tests
as r = 0.25 and for CD tests as r = −0.47. Yucemen et al. [41] reported values in a range of
−0.49r −0.24, while Lumb [42] noted values of −0.7r −0.37. A negative correlation
implies that low values of cohesion are associated with high values of friction angle and vice versa.
In other words, a negative correlation between the cohesion and the friction angle means that the
uncertainty in the calculated shear strength is smaller than the combined uncertainty in the two
parameter values used to model the shear strength. This observation arises from the fact that the
variance of the shear strength is reduced if there is a negative correlation between the cohesion
and the friction angle [12]. In this study, a value of −0.5 is considered as a base set, and the range
of −0.7r 0.5 around the base value is considered.
According to the results of a literature review by El-Ramly et al. [29], the autocorrelation
distance is within a range of 10–40 m in the horizontal direction, while in the vertical direction it
ranges from 1 to 3 m. These values are consistent with those noted by Phoon and Kulhawy [43].
Based on the knowledge of the above statistical input, the effects of varying autocorrelation
distance and cross-correlation were investigated.
In this study the autocorrelation distance l in the Gaussian field is considered to maximally
exploit the analytical solution of Equation (4). If the autocorrelation distance in the original non-
Gaussian field is used, then the original autocorrelation structure should be corrected according
to the Nataf model. Figure 4 shows the relationship between the corrected correlation  and
the original correlation ˜ with a common lognormal distribution for various COVs, where the
relationship is represented by the Nataf model as follows [44]:
ln(1+ ˜ i,i V X2i )
i,i =  (14)
ln(1+ V X2i ) ln(1+ V X2i )

The figure indicates that the correction factor is only slightly greater than 1.0 over the range of
possible autocorrelations, and thus the difference between the correlation structures of Gaussian
and lognormal random fields is very small.
Analyses were carried out with the same grid used for the deterministic analysis. As explained in
Section 3.3, a continuous random field can be obtained for an exponential autocorrelation function
on a rectangular domain based on the analytical solution of the eigenvalue problem by the KL
expansion method. Therefore, the random field discretization is independent of the shape of the
mesh. The accuracy of the random field generated by the KL expansion method depends on the
number of terms used in the series expansion, not on the mesh size. The mesh size only controls

Copyright q 2009 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2010; 34:1–26
DOI: 10.1002/nag
12 S. E. CHO AND H. C. PARK

1.0

0.8

0.6

ρ
0.4

COV=0.45
0.2 COV=0.30
COV=0.10

ρ =ρ
0.0
0 0.2 0.4 0.6 0.8 1

ρ

Figure 4. Relationships between original autocorrelation ˜ and corrected autocorrelation


 for a common lognormal distribution.

40
lv=1m
lv=2m
30
Eigenvalue

20

10

0
0 20 40 60 80 100
Number of eigenmode

Figure 5. Eigenvalues of the autocorrelation function.

the accuracy of numerical analysis; therefore, if an acceptable accuracy of numerical analysis can
be obtained for a mesh, then the mesh can be used for the expansion of the random field by
the KL expansion method. As the spatial mesh is regulated by the stress gradients of the response,
the mesh presented in Figure 2 was introduced. Then the spatial discretization is able to model the
variability of the random field effectively through the expansion of the random field by the KL
expansion method.
The mean values of cohesion and friction angle were fixed, while the COV, autocorrelation
distance, and cross-correlation coefficient were varied.
The number of terms in the truncated series should be carefully chosen so as to accurately
reflect the spatial variability of the random field being expanded. The number of eigenmodes to

Copyright q 2009 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2010; 34:1–26
DOI: 10.1002/nag
EFFECT OF SPATIAL VARIABILITY OF SOIL PROPERTIES 13

be retained while discretizing a random field depends on the magnitudes of the corresponding
eigenvalues. Figure 5 represents the decaying trends of the eigenvalues obtained by solving the KL
integral eigenvalue problem. The figure shows that a larger number of terms in the KL expansion
are necessary to accurately represent the random field for smaller autocorrelation distances. In this
study 100 terms of eigenmode were used to represent the random fields of cohesion and friction
angle.
Figure 6(a) shows typical bearing pressure–settlement curves for the first 100 realizations as
part of the result for the case where r (c, ) = −0.5, lh = 10 m, and lv = 1 m. Figure 7(a) and (b)
shows the convergence of the estimated mean and standard deviation of the bearing capacity.
Random properties can be calculated at any point in the domain of analysis such as centroid or
integration point in the elements by the KL expansion method since the method offers a random
field to be represented in terms of a continuous function. In this study, random properties are
calculated at the centroid of grid zone (element) for the finite difference analysis. Hence, the
random field used in the finite difference analyses is not the originally generated continuous
random field, but a discontinuous random field with element-wise constant properties. It may
seem that there is no difference to the simple and straightforward midpoint method approach;
however, the use of the KL expansion method has comparative advantages over the midpoint
method.
The midpoint method is very straightforward to implement; however, it results in a very large
number of random variables equal to the number of elements used for spatial discretization. The
method puts limitation on the mesh size compared with the autocorrelation distance and requires
almost regular discretization. For very short autocorrelation distance, the random field discretization
has to be very fine, which increases the numerical effort dramatically. The KL expansion method
automatically captures the variability using a small number of random variables. In this study,
the KL expansion method requires 100 random variables, but the midpoint method requires 1050
random variables corresponding to the number of elements. Although the more dense elements
are adopted, the KL expansion method requires the same 100 random variables without loss of
accuracy of random field discretization since the random field discretization procedure does not
change with the choice of element formulations and boundary conditions. This makes it possible to
use the same mesh for different autocorrelation distances, which is advantageous for the parametric
studies.
Figure 8 shows two typical realizations of random fields. In the figure, the lighter regions
denote a larger strength parameter value (stronger soil) and darker regions indicate a smaller
strength parameter value (weaker soil). It can be observed that the cohesion and the friction angle
show a negative correlation. Figure 8(a) shows that the failure region developed to the left of the
footing through the weak strength path. A non-symmetric failure mechanism, caused by the spatial
heterogeneity, is not manifested in the deterministic analysis or the probabilistic analysis with
a single random variable due to the representation of a homogeneous soil medium. Figure 8(b)
shows another realization of random fields of shear strength parameters and the corresponding
failure mechanism, which indicates a shallower slip path due to the weaker region of soil around
the surface.
The histograms of the bearing capacity were determined and fitted to a lognormal distribution
for the simulation results obtained from 5000 realizations. The histogram shown in Figure 6(b)
captures the major trends and shows that the fit appears reasonable.
The sensitivity of the statistical response to the autocorrelation distance was examined. The
effects of the autocorrelation distance are summarized in Figures 9 and 10 for anisotropic random

Copyright q 2009 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2010; 34:1–26
DOI: 10.1002/nag
14 S. E. CHO AND H. C. PARK

1600

1200

Bearing pressure (kPa)


800

400

Deterministic curve

0
0 0.01 0.02 0.03 0.04 0.05
(a) Settlement (m)
2.5

1.5
Pdf (× 10-3)

0.5

0
0 500 1000 1500 2000 2500
(b) Bearing capacity (kPa)

Figure 6. Typical results of the Monte Carlo simulation (r (c, ) = −0.5, lh = 10 m, and lv = 1 m):
(a) bearing pressure–settlement curves for the first 100 realizations and (b) probability density function.

fields. As indicated in the figures, the mean values show slight increases but the values of standard
deviation show greater increases with an increase in the autocorrelation distance (see Figure
9(a) and (b) for the influence of horizontal correlation distance and Figure 10(a) and (b) for
the influence of vertical correlation distance). An infinite value of the autocorrelation distance

Copyright q 2009 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2010; 34:1–26
DOI: 10.1002/nag
EFFECT OF SPATIAL VARIABILITY OF SOIL PROPERTIES 15

1200

Mean bearing capacity (kPa)


1100

1000

900

800
0 1000 2000 3000 4000 5000
(a) Number of simulation
Standard deviation bearing capacity

300

260

220
(kPa)

180

140

100
0 1000 2000 3000 4000 5000
(b) Number of simulation

Figure 7. Convergence of the estimated mean and standard deviation (r (c, ) = −0.5, lh = 10 m, and
lv = 1 m): (a) mean vs number of trials and (b) standard deviation vs number of trials.

implies a perfectly correlated random field or a single random variable and provides the maximum
value of the mean and standard deviation. A smaller autocorrelation distance results in weak
correlation between soil parameters over the possible failure surface, which induces significant
fluctuation of the soil properties. Therefore, the variability of bearing capacity decreases, since the
fluctuations are averaged to a mean value along the possible failure surface. On the contrary, the
variability of bearing capacity increases as the autocorrelation distance is increased (see Figures 9(c)
and 10(c)). This could be anticipated since a higher autocorrelation distance value indicates that
the random variables are more strongly correlated, thereby reducing the averaging effects. For the
range considered in this study, the horizontal autocorrelation distance does not significantly affect
the statistical response, as can be seen in Figure 9(d) and (e). On the contrary, Figure 10 shows
that the vertical autocorrelation distance has greater influence on the statistical response. As shown
in Figure 10(d) and (e), the estimated responses converge to those of a single random variable as
the vertical autocorrelation distance approaches infinity.
Figure 11 shows the effect of cross-correlation in the original (lognormal) space on the estimated
statistical response. In this case the mean value of the bearing capacity shows a slight increase with
an increase of the negative value of the cross-correlation coefficient, but the standard deviation
decreases considerably (see (a) and (b)). Since the increase of one parameter value decreases the

Copyright q 2009 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2010; 34:1–26
DOI: 10.1002/nag
16 S. E. CHO AND H. C. PARK

(a) (b)

Figure 8. Typical realization of random field and corresponding analysis results


(r (c, ) = −0.5, lh = 10 m, and lv = 1 m).

Copyright q 2009 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2010; 34:1–26
DOI: 10.1002/nag
EFFECT OF SPATIAL VARIABILITY OF SOIL PROPERTIES 17

994

Mean bearing capacity (kPa)


992

990

988

986

984

982
0 10 20 30
(a) lh(m)
220
Standard deviation bearing capacity

210

200
(kPa)

190

180

170
0 10 20 30
(b) lh(m)

0.25
Coefficient of variation

0.20

0.15
0 10 20 30
(c) lh(m)

Figure 9. Influence of horizontal autocorrelation distance on the estimated statistical response obtained
from simulation (r (c, ) = −0.5, COVc = 0.3, COV = 0.2, and fixed lv = 1 m): (a) mean bearing capacity;
(b) standard deviation of the bearing capacity; (c) COV of the bearing capacity; (d) probability density
function; and (e) probability distribution of the bearing capacity.

Copyright q 2009 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2010; 34:1–26
DOI: 10.1002/nag
18 S. E. CHO AND H. C. PARK

2.5
l h= 5 m

l h= 10m

l h= 20m
2
l h= 30m

l h=

1.5
Pdf (× 10-3)

0.5

0
0 1 000 2000 3000
(d) Bearing capacity (kPa)
1

0.8
Cumulative probability

0.6
FS=1.5

0.4
l h= 5 m

l h=10m

l h=20m
0.2
l h=30m
FS=3.0 FS=1.0
l h=

0
0 1000 2000 3000
(e) Bearing capacity (kPa)

Figure 9. Continued.

other value, the variation of the total shear strength is reduced, and consequently the variation of
the bearing capacity also decreases (see (c)). An opposite effect for the case of a positive value
of the correlation coefficient is observed, since the increase of one parameter induces an increase
of the other, which results in an increase of variation of the total shear strength.

Copyright q 2009 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2010; 34:1–26
DOI: 10.1002/nag
EFFECT OF SPATIAL VARIABILITY OF SOIL PROPERTIES 19

1060

Mean bearing capacity (kPa)


1040

1020

1000

980
0 2 4 6 8 10
(a) lv (m)
320
Standard deviation bearing capacity

280
(kPa)

240

200

160
0 2 4 6 8 10
(b) lv (m)
0.35
Coefficient of variation

0.30

0.25

0.20

0.15
0 2 4 6 8 10
(c) lv (m)

Figure 10. Influence of vertical autocorrelation distance on the estimated statistical response obtained from
simulation (r (c, ) = −0.5, COVc = 0.3, COV = 0.2, and fixed lh = 10 m): (a) mean bearing capacity;
(b) standard deviation of the bearing capacity; (c) COV of the bearing capacity; (d) probability density
function; and (e) probability distribution of the bearing capacity.

Copyright q 2009 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2010; 34:1–26
DOI: 10.1002/nag
20 S. E. CHO AND H. C. PARK

2.5
l v= 1m

l v= 2m

2 l v= 5m

l v=10m

l v= ∞

1.5
Pdf (× 10-3)

0.5

0
0 1000 2000 3000
(d) Bearing capacity (kPa)
1

0.8
Cumulative probability

0.6

FS=1.5

0.4
l v= 1m

l v= 2m

l v= 5m
0.2
l v=10m
FS=3.0 FS=1.0 l v= ∞

0
0 1000 2000 3000
(e) Bearing capacity (kPa)

Figure 10. Continued.

Figure 11(d) shows that the probability density functions of the bearing capacity have a positive
coefficient of skewness. This implies that the probability density functions have a longer tail to
the right than to the left. The coefficient of skewness decreases with an increase of the negative

Copyright q 2009 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2010; 34:1–26
DOI: 10.1002/nag
EFFECT OF SPATIAL VARIABILITY OF SOIL PROPERTIES 21

1000

Mean bearing capacity (kPa)


990

980

970

960

950
-0.8 -0.4 0 0.4 0.8
(a) Cross-correlation coefficient r (c,φ)
320
Standard deviation bearing capacity (kPa)

280

240

200

160

120
-0.8 -0.4 0 0.4 0.8
(b) Cross-correlation coefficient r (c,φ)
0.36

0.32
Coefficient of variation

0.28

0.24

0.20

0.16

0.12
-0.8 -0.4 0 0.4 0.8
(c) Cross-correlation coefficient r (c,φ)

Figure 11. Influence of the cross-correlation coefficient on the estimated statistical response obtained from
simulation (COVc = 0.3, COV = 0.2, lh = 10 m, and lv = 1 m): (a) mean bearing capacity; (b) standard
deviation of the bearing capacity; (c) COV of the bearing capacity; (d) probability density function; and
(e) probability distribution of the bearing capacity.

Copyright q 2009 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2010; 34:1–26
DOI: 10.1002/nag
22 S. E. CHO AND H. C. PARK

3
r (c, φ)= −0.7
r (c, φ)= −0.5
2.5
r (c, φ)= 0
r (c, φ)= 0.25
r (c, φ)= 0.5
2

Pdf (× 10-3)
1.5

0.5

0
0 1000 2000 3000
(d) Bearing capacity (kPa)
1

0.8
Cumulative probability

0.6 FS=1.5

0.4

r (c, φ)= −0.7


r (c, φ)= −0.5
0.2 r (c, φ)= 0
FS=3.0
FS=1.0 r (c, φ)= 0.25
r (c, φ)= 0.5

0
0 1000 2000 3000
(e) Bearing capacity (kPa)

Figure 11. Continued.

value of the cross-correlation coefficient. Namely, the shape of the probability density function
becomes narrower and the uncertainty in the bearing capacity decreases.
Although the effect of the variation of soil properties is not presented in this paper, increasing
the COV of cohesion and friction angle resulted in a reduction of the mean value and an increase
of the standard deviation of the bearing capacity [10, 12].

Copyright q 2009 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2010; 34:1–26
DOI: 10.1002/nag
EFFECT OF SPATIAL VARIABILITY OF SOIL PROPERTIES 23

1.0x100
1.0x10-1
1.0x10-2
1.0x10-3

Pf
1.0x10-4
1.0x10-5
1.0x10-6
FS=1.0
FS=1.5
1.0x10-7
FS=2.0
1.0x10-8
-0.8 -0.4 0 0.4 0.8
(a) Cross-correlation coefficient r (c,φ)
1.0x100

1.0x10-1
Pf

1.0x10-2

1.0x10-3
FS=1.0
FS=1.5
FS=2.0
1.0x10-4
0 2 4 6 8 10
(b) lv (m)

Figure 12. Estimated probability of failure for bearing capacity: (a) influence of cross-correlation
and (b) influence of vertical correlation distance.

Figure 12 presents a summary of the probability of failure for the bearing capacity. This indicates
that the simulation based on random fields gives a probability of exceeding the deterministic bearing
capacity (FS = 1) greater than 50%. This in turn means that the median bearing capacity given
by the simulation is always smaller than the deterministic bearing capacity, which has also been
observed by other investigators [10, 12, 13]. Additionally, the probability of failure was negligible
in the case of FS = 3 for all cases studied in this paper.
Figure 12(a) shows the estimated probability of failure against the cross-correlation. The prob-
ability of failure decreases with an increase of the negative correlation coefficient. Therefore,
the assumption of independence between cohesion and friction angle gives conservative results
if the actual correlation is negative, but slightly unconservative results are obtained if the actual
correlation is positive.
The effects of the autocorrelation distance on the probability of failure are summarized in
Figure 12(b). As indicated in the figure, with an increase in the vertical autocorrelation distance,
the probability of failure decreases slightly if FS = 1.0, but increases if a factor of safety
of 1.5 or 2.0 is used.

Copyright q 2009 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2010; 34:1–26
DOI: 10.1002/nag
24 S. E. CHO AND H. C. PARK

6. CONCLUSIONS

The effect of spatial variability of cross-correlated shear strength parameters on the bearing capacity
of a strip footing was studied using random field theory integrated into a commercial finite difference
package. An exponential autocorrelation function that considers different autocorrelation distances
in the vertical and horizontal directions is used to describe the spatial variability of the soil.
Two-dimensional cross-correlated non-Gaussian random fields are generated by a KL expansion,
which is the most efficient method requiring the smallest number of random variables to represent
the field within a given level of accuracy. A series of Monte Carlo simulations is then conducted
to determine the statistical response of the bearing capacity of a shallow footing. The simulation
observed various non-symmetric failure mechanisms caused by the spatial heterogeneity.
In the problem studied in this paper the cross-correlation between cohesion and friction angle
and the autocorrelation distance in the vertical direction were found to be significant factors in
the stochastic behavior of bearing capacity. In particular, when a negative cross-correlation is
considered, the effect on the probability of failure is important. The probability of failure decreases
with an increase of the negative correlation coefficient. Therefore, the assumption of independence
between cohesion and friction angle gives conservative results if the actual correlation is negative,
but slightly unconservative results are obtained if the actual correlation is positive.
With a decrease in the vertical autocorrelation distance, the probability density function of the
bearing capacity becomes narrower and the uncertainty in the bearing capacity decreases. On the
contrary, for the range considered in this study, the horizontal autocorrelation distance does not
significantly affect the statistical response.
The obtained results provide insight regarding the stochastic analysis in the field of geotechnical
engineering and show the importance of the spatial variability of soil properties in the outcomes
of a probabilistic assessment.

REFERENCES
1. Lacasse S, Nadim F. Uncertainties in characterizing soil properties. In Uncertainty in the Geologic Environment:
From Theory to Practice, Shackleford CD, Nelson PP, Roth MJS (eds). ASCE Geotechnical Special Publication
No. 58. ASCE: New York, 1996; 49–75.
2. Alonso EE. Risk analysis of slopes and its application to slopes in Canadian sensitive clays. Géotechnique 1976;
26:453–472.
3. Vanmarcke EH. Probabilistic modeling of soil profiles. Journal of Geotechnical Engineering (ASCE) 1977;
103:1227–1246.
4. Li KS, Lumb P. Probabilistic design of slopes. Canadian Geotechnical Journal 1987; 24:520–535.
5. Duncan JM. Factors of safety and reliability in geotechnical engineering. Journal of Geotechnical and
Geoenvironmental Engineering 2000; 126:307–316.
6. Mostyn GR, Li KS. Probabilistic slope stability—state of play. In Conference on Probabilistic Methods in
Geotechnical Engineering, Li KS, Lo S-CR (eds). Balkema: Rotterdam, The Netherlands, 1993; 89–110.
7. Elkateb T, Chalaturnyk R, Robertson PK. An overview of soil heterogeneity: quantification and implications on
geotechnical field problems. Canadian Geotechnical Journal 2002; 40:1–15.
8. Baecher GB, Christian JT. Reliability and Statistics in Geotechnical Engineering. Wiley: New York, 2003.
9. D’Andrea R. Discussion of ‘Search algorithm for minimum reliability index of earth slopes.’ Journal of
Geotechnical and Geoenvironmental Engineering 2001; 127:195–197.
10. Griffiths DV, Fenton GA, Manoharan N. Bearing capacity of rough rigid strip footing on cohesive soil: probabilistic
study. Journal of Geotechnical and Geoenvironmental Engineering 2002; 128:743–755.
11. Griffiths DV, Fenton GA. Bearing capacity of spatially random soil: the undrained clay Prandtl problem revisited.
Géotechnique 2001; 51:351–359.

Copyright q 2009 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2010; 34:1–26
DOI: 10.1002/nag
EFFECT OF SPATIAL VARIABILITY OF SOIL PROPERTIES 25

12. Fenton GA, Griffiths DV. Bearing capacity prediction of spatially random c– soils. Canadian Geotechnical
Journal 2003; 40:54–65.
13. Popescu R, Deodatis G, Nobahar A. Effects of random heterogeneity of soil properties on bearing capacity.
Probabilistic Engineering Mechanics 2005; 20:324–341.
14. Haldar S, Babu GLS. Effect of soil variability on the response of laterally loaded pile in undrained clay.
Computers and Geotechnics 2007; 35:537–547.
15. Paice GM, Griffiths GV, Fenton GA. Finite element modeling of settlement on spatially random soil. Journal of
Geotechnical Engineering (ASCE) 1996; 122:777–779.
16. Griffiths DV, Fenton GA. Influence of soil strength spatial variability on the stability of an undrained clay slope
by finite elements. Slope Stability 2000. Geotechnical Special Publications No. 101. ASCE: New York, 2000;
184–193.
17. Popescu R, Prevost JH, Deodatis G. Effects of spatial variability on soil liquefaction: some design
recommendations. Géotechnique 1997; 47:1019–1036.
18. Popescu R, Prevost JH, Deodatis G. 3D effects in seismic liquefaction of stochastically variable soil deposits.
Géotechnique 2005; 55:21–31.
19. Koutsourelakis S, Prevost JH, Deodatis G. Risk assessment of an interacting structure–soil system due to
liquefaction. Earthquake Engineering and Structural Dynamics 2002; 31:851–879.
20. Kim HK, Narsilio GA, Santamarina JC. Emergent phenomena in spatially varying soils. Probabilistic Applications
in Geotechnical Engineering. ASCE Geotechnical Special Publication No. 170. ASCE: New York, 2007; 1–10.
DOI: 10.1061/40914(233)10.
21. Vořechovský M. Simulation of simply cross correlated random fields by series expansion methods. Structural
Safety 2008; 30:337–363.
22. Terzaghi K. Theoretical Soil Mechanics. Wiley: New York, 1943.
23. Prandtl L. Über die Härte plastischer Körper. Nachrichten von der Königlichen Gesellschaft der Wissenschaften
zu Göttingen, Mathematisch-Physikalische Klasse 1920; 74–85.
24. Soubra AH, Youssef Abdel Massih DS, Kalfa M. Bearing capacity of foundations resting on a spatially random
soil. GeoCongress 2008: Geosustainability and Geohazard Mitigation. ASCE Geotechnical Special Publication
No. 178. ASCE: New York, 2008; 66–73.
25. ITASCA Consulting Group Inc. FLAC Fast Lagrangian Analysis of Continua. Minneapolis, MN, U.S.A., 2002.
26. El-Ramly H, Morgenstern NR, Cruden DM. Probabilistic slope stability analysis for practice. Canadian
Geotechnical Journal 2002; 39:665–683.
27. Vanmarcke EH. Random Fields: Analysis and Synthesis. MIT Press: Cambridge, MA, 1983.
28. DeGroot DJ, Baecher GB. Estimating autocovariance of in-situ soil properties. Journal of the Geotechnical
Engineering (ASCE) 1993; 119:147–166.
29. El-Ramly H, Morgenstern NR, Cruden DM. Probabilistic stability analysis of a tailings dyke on presheared
clay-shale. Canadian Geotechnical Journal 2003; 40:192–208.
30. Rackwitz R. Reviewing probabilistic soils modeling. Computers and Geotechnics 2000; 26:199–223.
31. Sudret B, Der Kiureghian A. Comparison of finite element reliability methods. Probabilistic Engineering Mechanics
2002; 17:337–348.
32. Sudret B, Der Kiureghian A. Stochastic finite element methods and reliability: a state-of-the-art report. Technical
Report No. UCB/SEMM-2000/08, Department of Civil and Environmental Engineering, Institute of Structural
Engineering, Mechanics and Materials, University of California, Berkeley, 2000.
33. Matthies HG, Brenner CE, Bucher CG, Soares CG. Uncertainties in probabilistic numerical analysis of structures
and solids—stochastic finite elements. Structural Safety 1997; 19:283–336.
34. Ghiocel DM, Ghanem RG. Stochastic finite-element analysis of seismic soil–structure interaction. Journal of
Engineering Mechanics 2002; 128:66–77.
35. Spanos PD, Ghanem RG. Stochastic finite element expansion for random media. Journal of Engineering Mechanics
1989; 115:1035–1053.
36. Ghanem RG, Spanos PD. Stochastic Finite Element—A Spectral Approach. Springer: New York, 1991.
37. Liu PL, Der Kiureghian A. Multivariate distribution models with prescribed marginals and covariances.
Probabilistic Engineering Mechanics 1986; 1:105–112.
38. Stein ML. Large sample properties of simulations using Latin hypercube sampling. Technometrics 1987; 29:
143–151.
39. Cherubini C. Reliability evaluation of shallow foundation bearing capacity on c ,  soils. Canadian Geotechnical
Journal 2000; 37:264–269.

Copyright q 2009 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2010; 34:1–26
DOI: 10.1002/nag
26 S. E. CHO AND H. C. PARK

40. Wolff TH. Analysis and design of embankment dam slopes: a probabilistic approach. Ph.D. Thesis, Purdue
University, Lafayette, IN, 1985.
41. Yucemen MS, Tang WH, Ang AHS. A Probabilistic Study of Safety and Design of Earth Slopes. Civil Engineering
Studies, Structural Research Series, vol. 402. University of Illinois: Urbana, IL, 1973.
42. Lumb P. Safety factors and the probability distribution of soil strength. Canadian Geotechnical Journal 1970;
7:225–242.
43. Phoon KK, Kulhawy FH. Characterization of geotechnical variability. Canadian Geotechnical Journal 1999;
36:612–624.
44. Most T. Stochastic crack growth simulation in reinforced concrete structures by means of coupled finite element
and meshless methods. Ph.D. Thesis, Bauhaus-University Weimar, Germany, 2005.

Copyright q 2009 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2010; 34:1–26
DOI: 10.1002/nag

You might also like