Soil Dynamics and Earthquake Engineering 90 (2016) 158–168

Contents lists available at ScienceDirect

Soil Dynamics and Earthquake Engineering

journal homepage: www.elsevier.com/locate/soildyn

Probabilistic assessment of ground motions intensity considering soil

properties uncertainty
Sara Hamidpour, Masoud Soltani n
Faculty of Civil and Environmental Engineering, Tarbiat Modares University, Jalaal Ale Ahmad Highway, Tehran, Iran

art ic l e i nf o a b s t r a c t

Article history: This paper aims to evaluate the effects of soil uncertainty and soil depth to bedrock on the ground
Received 15 September 2014 motions intensity. For this, three different depths are considered and the variability of the ground mo-
Received in revised form tions characteristics traveling from depth to surface is investigated. Soil maximum shear modulus, G0,
24 May 2016
which mainly controls soil stiffness and strength characteristics is considered as the uncertain soil
Accepted 20 June 2016
Available online 1 September 2016
material property. By employing the Monte Carlo (MC) simulation technique and for a defined soil depth,
the effect of G0 variability on the ground motions intensity is investigated. Furthermore, the accuracy of
Keywords: approximate method of First Order Second Moment (FOSM) for response variability estimation is eval-
Uncertainty uated. Using an approximate method is important because probabilistic analysis methods are commonly
Soil properties
very time consuming. By conducting investigations, it's observed that the seismic responses of soil do-
Soil depth
main including PGA, Amplification Factor (AF) and spectral responses of Single Degree Of Freedom
Ground motions
Probabilistic analysis (SODF) system are strongly dependent on the soil depth. Moreover, by comparing the results of FOSM
Seismic response with MC, it is observed that FOSM is able to estimate the responses' variability with acceptable accuracy.
Thus, FOSM method could be reasonably used instead of MC simulation technique for predicting the
seismic response of the soil domain considering soil G0 uncertainty.
& 2016 Elsevier Ltd. All rights reserved.

1. Introduction
In the recent years, consideration of uncertainties has played an
increasing role in the geotechnical engineering. Uncertainties are
due to the lack of knowledge of physical phenomena and should
be properly considered to achieve acceptable accuracy in seismic
assessments. The evaluation of the structural response when the
dispersion of the modeling parameters is included is of a great
importance since it is directly correlated with the structural de-
struction and damage. The geotechnical characteristics of a site,
which control the dynamic response of soil domain at the time of
the earthquake, are not certainly known in most cases. The im-
portance of the issue has motivated researchers to examine the
effect of the geotechnical parameters on the seismic motions in-
tensity [1–7], as well as the structural responses [8–12] in the
framework of the probabilistic seismic assessment.
Local ground characteristics and underlying soil layers' depth at
a site can alter the intensity and frequency content of ground
motions. Soil is a natural material that experiences uncertainty in
its physical and mechanical properties. The soil properties
n
Corresponding author.
E-mail address: msoltani@modares.ac.ir (M. Soltani).
uncertainty arises from the estimation of the basic soil stiffness
and strength parameters. The proper determination of them de-
pends on the inherent uncertainty of soil as a natural material-
mostly known as a heterogeneous material-as well as soil speci-
mens sampling, testing methods and measurements in field and
laboratory. However, soil depth to bedrock is mainly known as a
varying parameter from one place to another rather than an un-
certain parameter. Earthquake uncertainties are attributed to two
main items: (1) choosing of a proper Intensity Measure (IM) as an
indicator of earthquake intensity and (2) selecting a set of earth-
quake records as seismic excitations that could properly represent
the seismicity specifications of a particular site. The involved un-
certainties in the process of analysis and design may lessen the
trust in the accuracy of the computed results. In order to achieve a
proper seismic assessment, the impact of the involved un-
certainties should be properly considered.
In this paper, the seismic response of soil domain considering
three different depths is investigated. For this purpose, the dy-
namic analysis is carried out for soil medium with depths of 10, 30
and 100 m. Amplification factor, PGA and spectral responses of
SDOF system, as representative characteristic parameters of
earthquake, at the soil surface are the interested quantities to
study. The term AF is defined as the ratio of spectral acceleration of

http://dx.doi.org/10.1016/j.soildyn.2016.06.010
0267-7261/& 2016 Elsevier Ltd. All rights reserved.
 S. Hamidpour, M. Soltani / Soil Dynamics and Earthquake Engineering 90 (2016) 158–168
Fig. 1. (a) Spectral acceleration response and (b) scatter gram of the magnitude and rupture distance of input records.
a specific record at the surface to the spectral acceleration of the
same record at the bedrock at corresponding periods which shows
how a record is amplified as it reaches the surface from depth. PGA
is a common informative parameter of earthquake intensity and
SDOF spectral responses are presenters of frequency content of
ground motion records. This paper also studies the influence of
soil maximum shear modulus, G0, uncertainty on the ground
motions characteristics. Utilizing MC simulation technique, sto-
chastic soil samples with varying G0 values and defined 30 m
depth are simulated and analyzed. In order to perform such ana-
lyses, a sufficient number of random possible soil samples are
generated and the variability of the soil medium response under
seismic excitation is studied. Furthermore, considering the time-
consuming feature of probabilistic analyses, the efficiency of FOSM
as an approximate analytical tool, which needs fewer simulations
and less analysis time in comparison to MC, in predicting the
probabilistic seismic response of soil, is investigated.
The soil medium is modeled using COM3 [13], a finite element
code developed at the University of Tokyo, Japan, which is able to
simulate the high nonlinear soil model. The soil profile is modeled
using 8-node cubic elements defined as Solid elements in the
software. Solid element is capable of assuming plane strain con-
dition in order to consider the infinity effects of the soil medium. It
can also take the homogeneity of soil into account and simulate
the nonlinear mechanics of soil by taking its mechanical and
physical properties. This software provides a model that can study
the nonlinear response of the soil domain under the earthquake
excitation [13,14].
The scientific interest of this kind of studies is to acquire the
variability of ground motions that hit the structure at the time of
earthquake as a design input variable, which will definitely affect
the structural response.
2. Input ground motions' characteristics
In order to account for the variability of input seismic motions,
a set of 20 real rock records from 8 different earthquakes was used
as seismic excitation for probabilistic analysis. This number of
adopted records was found to be sufficient to meet the goal of the
nonlinear soil dynamic analysis in this research work. The records
are acceleration time histories and selected based on moment
magnitude of the event (M) and closest distance to the rupture
zone (R) according to Shome and Cornell observations [15]. The
selected records are rather strong ground motions which belong to
159
the category of large magnitude (46.5) with small rupture dis-
tance (o30 km) records according to Shome and Cornell. The
range of records' magnitude was between M6.5 to M7.4 and the
range of distance to rupture zone was between R8.0 to R 29.1 km.
The major horizontal component of all records was scaled to
single defined value (0.35g in this case) and used as input seismic
excitation. The purpose of scaling was to eliminate the effects of
the varying PGA of the records on the PGA values obtained on the
soil surface since we intend to study the changes in the PGA value
as an informative intensity measure of earthquake due to the soil
depth variability and soil uncertainty in the forthcoming sections.
It is to be noted that the soil medium is affected by the all char-
acteristics of seismic excitation like PGA, energy and frequency
content as the applied records reach the surface from depth.
Choosing 0.35g as the scaling acceleration, regardless of its effects
on the energy and frequency content of records, was according to
the seismic zoning of Iran (the country in which this study was
implemented). On the basis of the seismic zoning maps of Iran, the
major part of the country area is of high seismic hazard regions for
which the local seismic references propose the PGA of 0.35g as the
earthquake acceleration.
The accelerograms were directly applied to the bottom
boundary at the soil medium base where connected to bedrock.
The surface motion was then recorded on the soil surface while
the shear wave was propagating upward through the soil profile.
Fig. 1 displays the 5% damped acceleration response spectra of the
selected input records along with mean, median and mean 7 s
curves and the scatter gram of the M and R of records. The stan-
dard deviation, s, measures the amount of variation from the
mean value. The s quantity was calculated by taking the square
root of the mean of the squared differences of the spectral accel-
eration response values at each period from their mean value at
the same period. The information belonging to earthquake records
could be found in Table A1 in the Appendix A.
3. Soil medium description
3.1. Soil medium finite element modeling
The studied soil site consists of single layer of non-cohesive and
relatively dense sand. The soil deposit is dry and non-liquefiable
and is underlain by the bedrock. The soil medium is assumed
homogenous in the whole depth and length and the physical and
mechanical properties of soil are constant throughout the entire
160 S. Hamidpour, M. Soltani / Soil Dynamics and Earthquake Engineering 90 (2016) 158–168
domain. As previously stated, soil medium models were simulated
by the aid of software COM3. The multi-yield surface plasticity
model is used in this software to formulate stress-strain re-
lationship of the soil. In this model soil is idealized as an assembly
of finite numbers of elasto-perfectly plastic components, which
are conceptually connected in parallel. As each component is given
different yield strengths, all components subsequently begin to
yield at different total shear strains, which result in a gradual in-
crease of entire nonlinearity. The nonlinear behavior appears
naturally as a combined response of all components. The general
Drucker-Prager yield surface is implemented for each of the yield
surfaces to represent the relation of the deviatoric shear stress and
the mean stress [14].
The required soil medium length was determined with atten-
tion to the infinity effects of domain in longitudinal direction so
that the computed responses converged regardless of the domain
length. Hence, the soil medium length as large as 10 times the
depth was considered. The thickness of the soil profile was as-
sumed unity as for the plane strain condition of the Solid elements
which are capable of simulating the infinity effects of the soil
profile in the transverse direction.
The free-field vibration situation was considered in the lateral
boundaries of the soil medium by using Line elements in COM3.
Line elements have the length of the soil medium and they bond
the corresponding nodes of the same elevation at two extreme
ends of the soil medium. Line elements are axially rigid. Because of
the existing free-field situation, the soil strata can slide over each
other and deform while the input excitation is not trapped and not
reflected into the medium and the boundaries are open for it to
exit. Boundary condition at the bottom of the soil medium where
connected to bedrock was considered as fixed nodes to accom-
modate the uniform and stiff nature of the bedrock. The responses
at the soil surface were obtained in the middle of the soil domain
so the probable impact of distance asymmetry between two ends
of domain on the responses was eliminated. The finite element
equations governed the numerical solution of the problem.
3.2. Soil material property uncertainty
A useful approach to define an uncertain parameter is the
statistical definition of it. In the probabilistic response assessment
of this study, the uncertain parameter was considered as random
variable. The considered uncertain soil parameter was soil max-
imum shear modulus at small strains, G0, which is an important
property of soil engaged with the soil stiffness and strength
characteristics. Four other characteristic parameters of soil in-
cluding viscous damping ratio, relative density, material unit
weight, and Poisson's ratio, were also considered which were as-
sumed as non-variable input parameters for soil medium simula-
tions. Mean value of the soil characteristic parameters and the
Coefficient of Variation (CV) of G0 were determined using past
research works accomplished based on both extensive laboratory
and field investigations [16–23] regarding the considered soil type
of this study. Darendeli [20], for instance, has studied numerous
soil samples from different geologic locations gathering informa-
tion around the stiffness and strength parameters of different soil
types; and Jones et al. [21] have combined results of a variety of
studies concerning soil characteristic parameters and have pre-
sented possible sources of soil uncertainties as well as values of
physical and mechanical properties for different soil types.
The soil property G0, was considered as a log-Gaussian random
variable parameter. In the probability theories, Log-Gaussian is in
fact a Non-Gaussian distribution which is attributed to random
values that do not obey the normal distribution. However, the G0
was not considered as a spatially random variable since the spatial
variability study of uncertain soil properties was not the objective
Table 1
Mean and coefficient of variation of soil properties.
Soil properties Symbol Mean CV (%)
Maximum shear modulus (kg/cm2) G0 1700 50
Material unit weight (kg/cm3) γ 0.002 –
Relative density (%) Dr 70 –
Poisson's ratio ν 0.35 –
Viscous damping ratio (%) μ 2 –
of this research. It was previously explained that the soil consists
of single homogeneous layer and properties are constant
throughout the entire domain as it is the case in many probabil-
istic research works [6,8,10,23]. The mean and coefficient of var-
iation of soil G0 along with the mean value of the other soil
properties are presented in Table 1. Coefficient of variation is de-
fined as the ratio of the standard deviation to the mean value of a
specific parameter. It is to be noted that, a relation was assumed
between the soil shear modulus and the friction angle, ϕ, through
Mohr-Coulomb theory and soil backbone curve. Hence the varia-
bility of ϕ has been taken into account within the soil G0 un-
certainty [6]. Relative density, Poisson's ratio, viscous damping
ratio, material unit weight and maximum shear modulus of soil
were considered independent from each other. The only two
parameters that were correlated in this study were the maximum
shear modulus and the friction angle of soil.
Soil backbone curve and shear modulus degradation ( G/G0 )
curve were considered according to Darendeli's observations for
the same soil type at the confining pressure of σ = 1 atm [20]. Soil
backbone curve consists of shear stress-strain pairs and defines
the soil behavior on the basis of the applied shear strain to the soil
medium due to seismic loading. Soil shear modulus is defined as a
ratio of shear stress to shear strain, G=τ /γ (the slope of the back-
bone curve). Similarly the soil maximum shear modulus, G0, is the
ratio at very small strains. The failure stress can also be de-
termined from the backbone (τ). On the other hand, Mohr-Cou-
lomb theory relates the soil failure shear stress (τ) to soil cohesion
(c), confining stress (s) and friction angle (ϕ) according the fol-
lowing equation.
τ = c + σ tan φ (1)
Based on the assumption of this study, the soil was a non-co-
hesive sand ( c = 0) and soil backbone curve was chosen with the
confining stress of σ = 1 atm according to Darendeli's study [20].
By simultaneous usage of the soil backbone curve and G0 for small
strain, ϕ and Mohr-Coulomb for behavior at failure strain and
empirical degradation curve for interpolation, a triangulated re-
lation was developed to estimate behavior in failure strains based
on the small strain measurements in soil simulations.
By knowing the value of soil G0, and utilizing G/G0 curve, the
value of G could be determined at any strain. Therefore, soil
backbone curve could be generated for different values of G0. In
this way, for every specific value of soil G0 there will be a specific
soil backbone curve and friction angle value. The mean value of
maximum shear modulus was also obtained from Darendeli's
study.
The soil backbone curve considering mean value for soil G0 and
soil modulus reduction curve are displayed in Fig. 2. Only one G/G0
curve was considered in this study for random simulations with
certain value of G/G0 at shear strains. It is to be noted that, the soil
models in software COM3 are pressure dependent such that the
software takes the backbone stress and strain pairs at the con-
fining pressure of σ = 1 atm and develops new soil backbone
curves for the other values of s by its change with the change of
 S. Hamidpour, M. Soltani / Soil Dynamics and Earthquake Engineering 90 (2016) 158–168
Fig. 2. Considered (a) soil backbone and (b) shear modulus reduction curve.
the soil depth. The software also considers the hysteresis damping
for soil in the numerical nonlinear solution of the problem which
is modified by increasing of the soil shear strain due to applied
seismic excitation within the software.
The mean value of the soil properties is known as the Best
engineering Estimate (BE) in this study. Knowing the BE value
along with the CV, the values of limit states including Upper
Bound (UB) and Lower Bound (LB) were developed as below:
UB = ( 1+CV ) × BE
LB = ( 1−CV ) × BE
4. Soil depth variability
In seismology it is well-known that the depth in which the
bedrock lies under the soil layers is an important factor and can
alter the ground motions intensity and frequency content. The soil
depth can affect the seismic response of both soil and on the
ground structure by changing the characteristics of seismic mo-
tions. In the literature, considering of soil depth of 100 m is more
common. Meanwhile, a depth of 30 m is the most commonly used
depth for determining the geotechnical classifications of a soil site
in many regulations. In this study, in addition to depth of 30 and
100 m, in order to account for the close to the ground bedrock
situation, a depth of 10 m was also considered. Therefore, the ef-
fect of soil depth was studied assuming three deterministic depths.
In order to evaluate the effect of soil depth on the variability of
its seismic response, three soil samples with similar physical and
mechanical characteristics but different depths were investigated.
In this section, the soil samples were considered characteristically
deterministic giving the soil G0 and other soil properties their
constant mean value. This assumption was made to eliminate the
effect of soil uncertainty on the response, so we could exclusively
investigate the influence of the soil depth on the soil domain
seismic responses. To ensure achieving reliable estimation and
limiting the earthquake record to record variability, the average
result of 20 records was investigated. In this way, the seismic input
variability was also taken into account. The accelerograms scaled
based on 0.35g were applied at the bedrock, driven through the
soil medium and recorded at the surface. As previously discussed,
scaling of all records to a similar PGA helped to better presentation
of the PGA variability at the soil surface due to soil depth and
161
records energy and frequency content excluding the variability of
PGA value at the bedrock. The shear wave velocity was considered
constant at depth.
There are many factors that may accompany depth randomness
in changing the seismic waves' behavior as they travel from bed-
rock to surface, such as soil layers status and position, surface
topography, soil type, site category and the characteristic of
earthquake excitation itself; however, the present work in this
section examines the characteristics of ground motions at the
(2) surface only as the consequence of the soil depth changes.
(3)
5. Results of depth effects study
5.1. Results of three soil depths analyses
By obtaining the surface motions, the effect of soil depth on the
PGA, AF, and the SDOF spectral responses was studied and the
mean and variability of each parameter were determined. The AF
at the surface shows how the seismic motion at the bedrock ex-
periences amplification or deamplification as the seismic waves
pass the different depths of the soil profile upward to the surface.
For three soil depths, 20 amplification factor and spectral response
curves were obtained by applying 20 records to the soil medium.
The AF and 5% damped spectral acceleration curves, along with
mean, median and mean 7s are displayed in Figs. 3–5 for three
soil depths. The maximum AF of records on average was equal to
2.1, 2.4 and 3.2 and occurred at the periods of 0.25, 0.6 and 1.6 s for
10, 30 and 100 m depths, respectively. In all three cases, the
maximum dispersion occurred near the curve's peak, where the
response was maximum. In the case of the AF, in overall, the
variability of AF increased with increasing of the soil depth.
5.2. Influence of depth on soil responses
Fig. 6 compares the mean AF curve as well as mean spectral
acceleration, velocity and displacement response curves for 10, 30
and 100 m soil depths. The mean spectral response curves of the
input earthquake records at the bedrock are also included in the
graphs. The mean curves were obtained averaging 20 records'
results. According to Fig. 6(a), by increasing the soil depth the
value of the maximum AF increased, which indicates that the
greater depth amplified the ground motion more than the shallow
one. The peak of the AF curves moved toward the longer periods as
the depth increased. It can be reasonably explained that the
162 S. Hamidpour, M. Soltani / Soil Dynamics and Earthquake Engineering 90 (2016) 158–168

Fig. 3. (a) Amplification factor and (b) spectral acceleration response in 10 m soil depth.

Fig. 4. (a) Amplification factor and (b) spectral acceleration response in 30 m soil depth.

Fig. 5. (a) Amplification factor and (b) spectral acceleration response in 100 m soil depth.

greater soil depth had longer period of oscillation; therefore, the curves increased by increasing of the soil depth and this indicates
maximum amplification which could be attributed to the soil- that the greater depth of soil experienced more resonances.
earthquake system resonance, occurred at the longer periods as According to Fig. 6(b)–(d), the maximum spectral acceleration
the soil depth increased. Moreover, the number of peaks in AF belonged to the soil depths of 10, 30 and 100 respectively while
 S. Hamidpour, M. Soltani / Soil Dynamics and Earthquake Engineering 90 (2016) 158–168
Fig. 6. Comparison of (a) amplification factor, (b) spectral acceleration, (c) spectral velocity and (d) spectral displacement in 10, 30 and 100 m soil depth.
Fig. 7. Cumulative probability of PGA occurrence in 10, 30 and 100 m soil depth.
the maximum spectral velocity and displacement showed an in-
verse result. By comparing the spectral responses, the maximum
value of spectral acceleration, velocity and displacement occurred
in shorter periods respectively as expected according to Mohraz's
studies [24]. Mohraz introduces three regions of amplifications
that are recognized in a response spectrum; the short-period or
163
acceleration region, medium-period or velocity region and long-
period or displacement region. In each region, the corresponding
response is amplified the most [24]. Similar results were observed
with respect to the obtained spectral responses in this study.
The PGA, as an important earthquake IM, was the other para-
meter that severely depended on the soil depth to bedrock. The
spots and the fitted curve in Fig. 7, respectively display the cu-
mulative probability of occurrence acquired directly from the
probabilistic analysis results and based on log-Gaussian distribu-
tion formulas regarding PGA values. Smaller values for PGA were
found as the soil depth increased. The PGA was on average equal to
0.42g, 0.36g and 0.24g at the surface and varied with CV values
equal to 0.12, 0.13 and 0.22 in 10, 30 and 100 m depth soil medium,
respectively. Comparing the average PGA values for different
depths, it is interesting that in the case of shallow bedrock of 10 m
depth, an amplification for peak ground acceleration was observed
while a deamplification was observed for the deep bedrock of
100 m depth. Furthermore, the variability of PGA values increased
by increasing the soil depth. According to Fig. 7, the good agree-
ment between the spots and the fitted curve of log-Gaussian dis-
tribution implied that PGA values also obeyed log-Gaussian
distribution.
164 S. Hamidpour, M. Soltani / Soil Dynamics and Earthquake Engineering 90 (2016) 158–168
6. Methodology of probabilistic analysis approaches
6.1. Monte Carlo simulation technique
Monte Carlo simulation technique is a set of computational
algorithms which relies on repetitive random sampling for result
calculation. This technique converts an uncertain problem into
problems with definite responses and attains the overall result
gathering the definite problems' responses. MC technique is ac-
complished in several steps including defining a domain of pos-
sible inputs, generating random inputs from the domain, im-
plementing a deterministic computation on the inputs and cal-
culating the result. In the present study, the random inputs from
the domain were generated by a random number generating code
with the aid of the computer programing language MATLAB. The
code was simple and generated a set of random values based on
the mean, s and the probability distribution (log-Gaussian in this
study) of desired parameter.
6.2. FOSM method
First Order Second Moment is a method in statistical en-
gineering analysis for calculating the probabilistic response of an
uncertain problem. The most popular form of FOSM is on the basis
of partial derivative or Taylor series expansion method [25]. In the
FOSM method, the problem is defined as a mono- or multi-vari-
able functional relationship as Eq. (4), in which the problem, Y, is a
function of the varying input parameter, x. Providing that deriva-
tives of the function are available at any point of x, solution to this
problem is obtained by finding a relationship for relating the
moments of the problem, to the moments of the input variable
[25]. Utilizing the method, it's simpler to propagate two first
moments than the exact solution. The mean and variance of the
problem are obtained based on the mean and variance of the input
varying parameter. The parameter x in the function, is a random
variable with the mean value of μx and variance of s2x .
The function is rewritten on the basis of Taylor series expansion
method about the point x0 as Eq. (5). The first moment of the
function, μY, is defined as the mean value for Y in Eq. (6), and the
second moment, s2Y, is defined as the variance of Y in Eq. (7). In the
following equations, term (dg /dx )0 indicates the response sensi-
tivity to changes of variable x.
Y = g ( x), x={x 0 , x1,…,x n}
Y ≈ g0 + ( dg /dx) (x − x 0)
0 rence and the cumulative probability of its occurrence are also
μY = E[g (x)] ≈ E⎡⎣g0 + (dg /dx)0(x − x 0)⎤⎦
≈ E⎡⎣g0⎤⎦+(dg /dx)0 E[x − x 0] ≈ g0 + (dg /dx)0 μx − x 0 ( )
σ 2Y = E⎡⎣ g 2( x)⎤⎦ − μ2Y
⎡ 2
≈ E⎢ g02 +
⎣ { ( dg /dx) ( x − x )}
0 0 + 2g0( dg /dx) ( x − x 0)⎥ − μ2Y
0 ⎦
2 2
≈ g02 + ( dg /dx) E( x − x 0) + 2g0( dg /dx) E( x − x 0) − μ2Y
0 0
2 2
≈ g02 + ( dg /dx) σ x + 2g0( dg /dx) μx − x 0 − μ2Y
( )
0 0
For specific case of x0 = μx , Eqs. (6) and (7) are expressed as Eqs.
(8) and (9), respectively, as below:
μY = g (μx )
2 this, the soil samples were increasingly simulated and analyzed to
σY 2 ≈ ( dg /dx) σx2
0 obtain the results. The cumulative mean and s of results at each
when there is a given problem with a given uncertainty, there
is a response with non-zero probability of occurrence for the
problem. According to the Eqs. (8) and (9), the mean and the
variance of the problem are determined by knowing the same
values for input variable.
7. Results of MC simulation analyses
Utilizing MC simulation method, soil responses were obtained
by performing dynamic analysis for a number of randomly varying
soil samples. The random samples of MC simulation were gener-
ated based on the random values of the soil G0. In this way, in the
simulations of the soil medium samples, all of the soil character-
istic properties were considered constant excluding the soil shear
modulus which varied from one sample to another. In other
words, the stiffness and strength characteristics of soil medium
which mainly control the nonlinear behavior of it, varied de-
pending on the soil G0 value by deriving new soil backbone curves
for each soil G0 as explained in Section 3.2. In the simulations, the
soil medium samples were modeled with the depth of 30 m since
the depth is widely used for classifying soil sites in many regula-
tions and also mainly used in research works. Single earthquake
record was used as the input seismic excitation for the random soil
samples analyses. The record was the one that showed the most
similar results to the mean results obtained out of 20 records in
Section 5.1.
PGA as an informative parameter of ground motions' intensity
along with spectral responses of SDOF oscillator at the soil surface
were studied in MC simulation analyses. The scatter gram of PGA
on the basis of soil G0, its probability distribution and the mean
and s values were investigated. By gathering the results of MC
simulations, we obtained a set of deterministic seismic responses
of soil media with different G0 quantities. It is believed that the
results of MC could be an explainer of the real condition of a
problem since it takes the possible conditions of the problem into
account.
Fig. 8 displays the MC simulation results for 30 m depth soil
media with varying G0 values. Fig. 8(a) depicts the scatter gram of
PGA based on the soil G0 quantities obtained from 150 MC simu-
lations. It displays the possible observed quantities for PGA at the
soil surface considering the changes in the value of maximum
shear modulus as an uncertain soil parameter which leads to dif-
(4) ferent soil stiffness and strength characteristics. It was observed
that the stiffer soil (with larger value of soil G0) experienced larger
quantities of PGA at the surface. The frequency of the PGA occur-
(5)
displayed in Fig. 8(b) and (c). Studies showed that by taking the
soil G0 uncertainty in to account, the PGA values varied between
0.19g and 0.48g with the highest frequency of occurrence between
(6) 0.35g and 0.45g as displayed in the bar chart in Fig. 8(b). PGA
values had log-Gaussian distribution with the mean and CV values
equal to 0.384g and 0.128, respectively. The maximum and mini-
⎤ mum PGA values at the surface were amplified by 1.36 and 0.56
and the PGA on average was amplified by 1.1 relative to the PGA of
0.35g at the bedrock. By comparing the coefficient of variation of
PGA regarding the soil uncertainty in this section and the earth-
quake record to record variability for 30 m depth soil in Section
(7) 5.1, it could be concluded that both equally were important
sources of uncertainty.
The required number of simulations to reach acceptable con-
sequence in MC analysis was determined by performing a con-
(8) vergence test for the results. The convergence test can be im-
plemented for both values of mean and standard deviation. For
(9)
 S. Hamidpour, M. Soltani / Soil Dynamics and Earthquake Engineering 90 (2016) 158–168 165

Fig. 8. MC analysis results (a) PGA dispersion based on soil G0 (b) frequency of PGA occurrence, (c) cumulative probability of PGA occurrence, (d) convergence test of the
mean value of PGA and (e) convergence test of the standard deviation of PGA.

step were divided by the total mean and s of all the analyzed
samples and a ratio was obtained. The convergence test outcome is
plotted on a two-axial graph with one axis for the obtained ratio
and the other one for the number of the analyzed samples as
displayed in Fig. 8(d) and (e). The convergence test outcome can be
made to be as close to 1 as desired by analyzing increasing number
of samples to produce a sufficient number of simulations. The
convergence of the results is obtained as soon as the test ratio is
close enough to 1. In this way the number of simulations is as-
sumed sufficient and the results are considered reliable. According
to Fig. 8(d) and (e), the results of MC simulations in this study
were achieved with good accuracy by analyzing 150 soil samples.
8. Comparison between the results of MC and FOSM
MC simulation technique as an exact and full probabilistic
analytical approach needs a long period of time to be completed as
well as a large amount of output storage capacity to be recorded.
On the other hand, FOSM method as an approximate analytical
166 S. Hamidpour, M. Soltani / Soil Dynamics and Earthquake Engineering 90 (2016) 158–168

Fig. 9. Comparison between MC and FOSM for calculating the mean (a) amplification factor, (b) spectral acceleration, (c) spectral velocity and (d) spectral displacement.

tool in comparison to MC is very time and cost saving in use. To Table 2

ensure the accuracy of FOSM method in predicting the probabil- Mean and s of soil medium responses using MC and FOSM.
istic seismic responses of soil, its results were compared to the
Response Monte Carlo FOSM
results of MC analyses. FOSM method estimates the mean and
variance of responses and if its results are accurate and reliable, it Mean s Mean s
will be very useful for the probabilistic analysis.
PGA (g) 0.384 0.049 0.394 0.059
In the FOSM method analysis in this study, similar to MC si-
Maximum amplification factor 2.19 0.26 2.26 0.25
mulations, the soil G0 was considered as the single input variable, Maximum spectral acceleration (g) 1.61 0.31 1.57 0.4
x. To accomplish the FOSM method for the objective of this study, Maximum spectral velocity (cm/s) 126.34 22.83 118.97 21.66
three soil samples with different G0 values including BE, LB and UB Maximum spectral displacement (cm) 24.99 7.16 24.23 6.71
and of 30 m depth were analyzed. In the simulations, the other soil
characteristic parameters were given their constant mean value
according to Table 1. The soil seismic responses were obtained
applying the same earthquake record to the soil medium as the representing the deviation from the mean.
one used in MC simulations. According to Eq. (8), the mean seismic A typical calculation procedure for the mean and s values of
response for the soil belongs to the sample simulated considering PGA is presented below; the procedure was the same for the other
the mean value, BE, of soil G0. The response sensitivity to G0 responses.
variability, (dg /dx ), is then calculated based on the samples si- The value of PGA obtained from soil sample analysis consider-
mulated considering the limits, LB and UB, of soil G0 as defined in ing BE, LB and UB for soil G0 were equal to 0.394g, 0.32g and
Eqs. (2) and (3). The mean and CV of soil G0 were equal to 1700 0.438g, respectively. Using Eqs. (8) and (9), the mean and variance
and 0.5, so the limits of G0 were equal to 850 and 2550 kg/cm2, of PGA were calculated as below:
respectively. According to Eq. (9), the variance of the soil seismic
response, s2Y, is calculated based on the term (dg /dx ) and the var- μPGA =0.394g
iance of soil G0, s2x. Statistically defining, the standard deviation is
the square root of the variance. Since standard deviation has the
dg /dx = ( 0.438 − 0.32)/( 2*850) = 6.94e−5
same dimension as the data, it is simpler to use for comparing and
 S. Hamidpour, M. Soltani / Soil Dynamics and Earthquake Engineering 90 (2016) 158–168 167

2 Furthermore, by increasing the soil depth more resonant peaks in

( )
σPGA2 = 6.94e−5 *8502=0.003481
1/2
σPGA = ( 0.003481) =0.059g
In this way, the mean and s of the interested soil seismic re-
sponses were determined using the FOSM method and the results
were compared to those obtained from MC simulation analyses.
Fig. 9 and Table 2 display the results of FOSM method analyses
in comparison to MC simulation technique. The compared re-
sponses included the mean and s of PGA, AF and spectral re-
sponses of SDOF at the soil surface. The mean value of MC simu-
lation responses was obtained averaging 150 simulations results
and was compared to the obtained mean response of the FOSM, as
already described. According to the results, the mean and s values
of the all responses obtained from two approaches were almost
the same. It could be concluded that in the case of the studied soil
site and the considered uncertainty, FOSM could be reasonably
used instead of MC simulation technique for the assessment of
seismic soil responses. According to the results of the present
study, despite the approximation of FOSM method in predicting
the responses, since the full probabilistic analyses methods in
most cases require a great deal of work and time, FOSM method
could be properly used for probabilistic seismic assessment of soil
sites.
9. Conclusions
In this study, the effect of soil uncertainty and depth to bedrock
randomness on the ground motions intensity was investigated.
Three different depths of soil medium with similar mechanical and
physical characteristics were modeled and analyzed subjected to
20 earthquake records. It was observed that by increasing of the
soil depth the value of PGA at the soil surface decreased while its
dispersion increased. Maximum AF increased by increasing the soil
depth and occurred in longer periods since the greater depth had
longer period of oscillation in which the resonance occurred.
AF curve were observed.
The maximum spectral acceleration was observed in the soil
depths of 10, 30 and 100 respectively, while the maximum spectral
velocity and displacement belonged to 100, 30 and 10 m of depth,
respectively. Also, the maximum value of spectral acceleration,
velocity and displacement occurred in shorter periods, respec-
tively. Therefore, it could be concluded that the soil depth severely
affects the frequency content and characteristic parameters of
ground motions.
Probabilistic analysis was implemented for studying the soil
uncertainty effect on its response using MC simulation technique.
150 soil samples of 30 m depth with different G0 values were
analyzed and results were obtained. The results showed that the
PGA values had log-Gaussian distribution considering the soil G0
uncertainty. It was also observed that PGA was equally affected by
the soil uncertainty and record-to-record variability of input
earthquakes in the studied 30 m depth soil medium. Moreover, the
PGA value increased by increasing the soil stiffness and strength as
a result of the increase of soil G0.
Because of the time consuming feature of full probabilistic
analysis approaches like MC, the efficiency and the accuracy of
FOSM method in predicting the soil responses due to soil un-
certainty was examined. The comparisons showed that FOSM
approximate method was reasonably accepted for evaluating the
seismic response of nonlinear soil site with uncertain soil G0
property. The mean and the dispersion of the soil domain re-
sponses could be estimated with good accuracy requiring much
less simulations by using FOSM method instead of numerous MC
simulations.
Appendix A
See Appendix Table A1.

Table A.1
Characteristics of input ground motions database.

No. Event Station Year M R (km) Geothecnic PGA (g)

1 Sanfernando Santa Felita Dam(outlet) 1971 6.6 27.5 Rock 0.151

2 Sanfernando Lake Hughes#4 1971 6.6 24.2 Rock 0.192
3 Sanfernando Fairmont Dam 1971 6.6 29.1 Rock 0.108
4 Sanfernando Lake Hughes#9 1971 6.6 23.5 Rock 0.157
5 Sanfernando Santa Anita Dam 1971 6.6 27.0 Rock 0.212
6 Imperial valley Cerro Prieto 1979 6.5 26.5 Rock 0.169
7 Imperial valley Superstition Mtn Camera 1979 6.5 26.0 Rock 0.195
8 Kocaeli, Turkey Gebze 1999 7.4 17.0 Rock 0.244
9 Duzce, Turkey Lamont531 1999 7.1 11.4 Rock 0.159
10 Lomaprieta BRAN 1989 6.9 10.3 Rock 0.501
11 Lomaprieta Gilroy array#1 1989 6.9 11.2 Rock 0.473
12 Lomaprieta Coyote Lake Dam 1989 6.9 21.8 Rock 0.483
13 Lomaprieta UCSC Lick Observatory 1989 6.9 17.9 Rock 0.450
14 Lomaprieta Anderson Dam(L Abut) 1989 6.9 21.4 Rock 0.077
15 Nahanni Site3 1985 6.8 16.0 Rock 0.147
16 Cape Mendocino Cape Mendocino 1992 7.1 8.5 Rock 1.497
17 Northridge LA Wonderland 1994 6.7 22.7 Rock 0.172
18 Northridge Vasquez Rocks Park 1994 6.7 24.2 Rock 0.151
19 Northridge Pacoima Dam(Downstr) 1994 6.7 8.0 Rock 0.434
20 Northridge Lake Hughes#9 1994 6.7 26.8 Rock 0.216
168 S. Hamidpour, M. Soltani / Soil Dynamics and Earthquake Engineering 90 (2016) 158–168
References
[1] Wang S, Hao H. Effects of random variations of soil properties on site ampli-
fication of seismic ground motions. J Soil Dyn Earthq Eng 2002;22:551–64.
[2] Bazzurro P, Cornell CA. Nonlinear soil-site effects in probabilistic seismic-ha-
zard analysis. Bull Seismol Soc Am 2004;94:2110–23.
[3] Lopez F, Modaressi A. Assessment of variability and uncertainties effect on the
seismic response of a liquefiable soil profile. J Soil Dyn Earthq Eng
2010;30:600–13.
[4] Marano GC, Trentadue F, Morronoe E, Amara L. Sensitivity analysis of optimum
stochastic nonstationary response spectra under uncertain soil parameters. J
Soil Dyn Earthq Eng 2008;28:1078–93.
[5] Tsai CC, Probabilistic P. Seismic hazard analysis considering nonlinear site
effects. Bull Seismol Soc Am 2000;90:1–7.
[6] Bazzurro P, Cornell CA. Ground-motion amplification in nonlinear soil sites
with uncertain properties. Bull Seismol Soc Am 2004;94:2090–109.
[7] Badaoui M, Berrah MK, Mebaraki M. Depth to bedrock randomness effect on
design spectra in the city of Algiers (Algeria). J Eng Struct 2010;32:590–9.
[8] Rachowdhury P. Effect of soil parameter uncertainty on seismic demand of
low-rise steel building on dense silty sand. J Soil Dyn Earthq Eng
2009;29:1367–78.
[9] Na UJ, Ray Chaudhuri S, Shinozuka M. Effect of spatial variation of soil prop-
erties on seismic performance of port structures. J Soil Dyn Earthq Eng
2009;29:537–45.
[10] Tang Y, Zhang J. Probabilistic seismic demand analysis of a slender RC shear
wall considering soil-structure interaction effects. J Eng Struct 2011;33:218–
29.
[11] Ray Chaudhuri S, Gupta VK. Variability in seismic response of secondary sys-
tems due to uncertain soil properties. J Eng Struct 2002;24:1601–13.
[12] Jin S, Lutes LD, Sarkani S. Response variability for a structure with soil-struc-
ture interactions and uncertain soil properties. J Probab Eng Mech
2000;15:175–83.
[13] Maekawa K, Ishida T. DUCOM-COM3 user manual and guide.Japan: Concrete
Laboratory Department of Civil Engineering, The Univ of Tokyo; 2010.
[14] Maekawa K, Pimanmas A, Okamuea H. Nonlinear mechanics of reinforced
concrete.London: Spon Press; 2003.
[15] Shome N, Cornell CA. Probabilistic seismic demand analysis of nonlinear
structures. Reliability of marine structures report. Stanford: Department of
Civil and Environmental Engineering, Univ of California; 1999 No. RMS-35.
[16] Das BM. Principles of geotechnical engineering.EL Paso: The University of
Texas; 1941.
[17] Seed H, Wong R, Idriss I, Tokimatsu K. Moduli and damping factors for dy-
namic analysis of cohessionless soils. Berkeley: College of Engineering, Univ of
California; 1984 Report No UCB/EERC-84/14.
[18] Kramer SL. Geotechnical earthquake engineering. Int series in civil engineer-
ing and engineering mechanics.New York: Prentice Hall; 1996.
[19] Bazzurro P, Cornell CA. Nonlinear soil-site effects in probabilistic seismic-ha-
zard analysis. Bull Seismol Soc Am 2004;96:2110–23.
[20] Darendeli MB. Development of new family of normalized modulus reduction
and material damping [PhD dissertation]. Austin: Faculty of Graduate School
of the Univ of Texas; 2001.
[21] Jones AL, Kramer SL, Arduino P. Estimation of uncertainty in geotechnical
properties for performance-based earthquake engineering. Pacific Earthquake
Engineering Research center, Rept 2002/16, Univ of California, Berkeley, 2002.
[22] Kumar K. Basic geotechnical earthquake engineering.New Delhi: New Age Int
Publishers; 2008.
[23] Rieck PJ, Houston TW. Seismic response of statistically varied soil column
profiles for SSI Analysis. Transactions, SMiRT 16, Washington DC, 2001.
[24] Mohraz B. A study of earthquake response spectra for different geological
conditions. Bull Seismol Soc Am 1976;66:915–35.
[25] Wong SF. First order second moment methods. J Comput Struct 1985;20:
779–791.