Der - Kiureghian - 2015 - Simulation of Synthetic Ground Motions

EARTHQUAKE ENGINEERING AND STRUCTURAL DYNAMICS
Earthquake Engng Struct. Dyn. 2010; 39:1155–1180

Published online 11 March 2010 in Wiley InterScience (www.interscience.wiley.com). DOI: 10.1002/eqe.997
Simulation of synthetic ground motions for specified earthquake

and site characteristics
Sanaz Rezaeian∗, †, ‡ and Armen Der Kiureghian§

University of California, Berkeley, CA 94720, U.S.A.
SUMMARY
A method for generating a suite of synthetic ground motion time-histories for specified earthquake and
site characteristics defining a design scenario is presented. The method employs a parameterized stochastic
model that is based on a modulated, filtered white-noise process. The model parameters characterize
the evolving intensity, predominant frequency, and bandwidth of the acceleration time-history, and can
be identified by matching the statistics of the model to the statistics of a target-recorded accelerogram.
Sample ‘observations’ of the parameters are obtained by fitting the model to a subset of the NGA database
for far-field strong ground motion records on firm ground. Using this sample, predictive equations are
developed for the model parameters in terms of the faulting mechanism, earthquake magnitude, source-
to-site distance, and the site shear-wave velocity. For any specified set of these earthquake and site
characteristics, sets of the model parameters are generated, which are in turn used in the stochastic model
to generate the ensemble of synthetic ground motions. The resulting synthetic acceleration as well as
corresponding velocity and displacement time-histories capture the main features of real earthquake ground
motions, including the intensity, duration, spectral content, and peak values. Furthermore, the statistics of
their resulting elastic response spectra closely agree with both the median and the variability of response
spectra of recorded ground motions, as reflected in the existing prediction equations based on the NGA
database. The proposed method can be used in seismic design and analysis in conjunction with or instead
of recorded ground motions. Copyright q 2010 John Wiley & Sons, Ltd.
Received 10 September 2009; Revised 10 December 2009; Accepted 17 December 2009
KEY WORDS: earthquake ground motions; model validation; NGA database; simulation; stochastic
models; strong-motion records; synthetic motions
1. INTRODUCTION
Synthetic ground motions have been of interest in the field of earthquake engineering for many
years. These motions are used in the evaluation of seismic demand on structures by response-history
∗ Correspondence to: Sanaz Rezaeian, University of California, Berkeley, CA 94720, U.S.A.

†
E-mail: [email protected]
‡
Doctoral Candidate.
§ Taisei Professor of Civil Engineering.
Contract/grant sponsor: State of California
Copyright q 2010 John Wiley & Sons, Ltd.

1156 S. REZAEIAN AND A. DER KIUREGHIAN
dynamic analysis when appropriate recorded motions are not available. They are also used in the
Monte Carlo simulation studies aimed at seismic reliability assessment. With increasing computing
power and the advent of performance-based earthquake engineering, response-history dynamic
analysis is more often employed by structural engineers, especially when nonlinear behavior is
expected. This has led to the realization of how few recorded ground motions are available for
many possible design scenarios to be considered. To overcome this difficulty, many engineers select
recorded motions from locations other than the project site and modify them by scaling or spectrum
matching [1–3], which are controversial methods [4, 5] and may result in motions with unrealistic
characteristics. Selection of input ground motions, real or synthetic, which incorporate key features
of real earthquakes, is essential for realistic estimation of the seismic demand, particularly when
sophisticated nonlinear structural models are to be used.
In this paper, a new method is developed for generating suites of synthetic ground motions,
which reproduce characteristics of real earthquake ground motions considered to be important
for determining the nonlinear response of structures. These characteristics include the evolving
intensity and predominant frequency of the motion, the duration, and the bandwidth. Each suite of
synthetic motions represents realizations of possible ground motions resulting from a given design
earthquake scenario that is specified in terms of its faulting mechanism, moment magnitude, source-
to-site distance, and site shear-wave velocity. The simulation model is calibrated to a database of
recorded motions so that the variability observed among the synthetic ground motions is consistent
with the variability observed among real recorded motions.
As described in the thorough review by Douglas and Aochi [6], there are three types of synthetic
ground motion models: Those that employ physically-based seismological models of the source
mechanism and wave travel path, those that employ parameterized stochastic models fitted to
recorded ground motions, and hybrid models that combine elements from both seismological and
stochastic models. Seismological models can produce realistic accelerograms at low frequencies
(<1 Hz), but often need to be adjusted for high frequencies by combining with a stochastic
or empirical component (hybrid model). These models require a thorough knowledge of the
source, wave path, and site characteristics, which may not be available to the practicing engineer.
They also typically require extensive computations. Furthermore, as pointed out by Stafford et al.
[7], these models depend on physical parameters that vary significantly from region to region, thus
limiting their use in regions where seismological data are lacking.
Our aim in this paper is to develop a method for generating synthetic ground motions, which
uses information that is readily available to the practicing engineer. For this purpose, we select the
approach based on fitting a parameterized stochastic model to a recorded ground motion. Many
such models have been developed in the past; see our recent paper [8] for a brief review. Most of
these models limit their scope to generating synthetics similar to a target recorded motion. The
novel aspect of our approach is that we relate the parameters of our model to the earthquake and
site characteristics. Furthermore, by accounting for the uncertainty in the model parameters, we
are able to reproduce in the synthetics the variability present in real ground motions, which has
been lacking in the previous models with two exceptions: The paper by Pousse et al. [9], in which
the parameters of an improved version of the model by Sabetta and Pugliese [10] were fitted to
the K-Net Japanese database, and the work by Alamilla et al. [11], in which the parameters of a
model similar to that proposed by Yeh and Wen [12] were fitted to a database of ground motions
corresponding to the subduction zone lying along the southern coast of Mexico. In both cases, the
model parameters were randomized to achieve the variability. Stafford et al. [7] also related the
parameters of their model to the earthquake and site characteristics, but they did not account for
Copyright q 2010 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2010; 39:1155–1180
DOI: 10.1002/eqe
SIMULATION OF SYNTHETIC GROUND MOTIONS 1157
the variability in the model parameters. It is noted that several recent seismological models do
account for the variability in ground motions. Typically, this is done by varying the source param-
eters, as in Liu et al. [13], Hutchings et al. [14], Causse et al. [15], and Ameri et al. [16]. Also
noteworthy is the work by Frankel [17], who compares the elastic response spectra of synthetic
ground motions generated using a seismological model with the Next Generation Attenuation
(NGA) relations—a validation approach that is also used in the present study.
The paper begins with a brief summary of the stochastic ground motion model from our previous
work [8], which forms the core of the model developed here. A set of transformed model parameters
are introduced that directly relate to the physical features of a ground motion. In order to obtain
sample realizations of the model parameters, a subset of the strong ground motion records from the
NGA database is selected. Model parameters for each selected ground motion record are identified
and a probability distribution is assigned to each model parameter based on the identified data.
After transformation to the standard normal space, empirical prediction equations are constructed
for the transformed model parameters in terms of earthquake and site characteristics. The specific
characteristics considered are the type of faulting, earthquake magnitude, source-to-site distance,
and the shear-wave velocity of the site. The correlation coefficients between the model parameters
are computed by correlation analysis of the residuals, i.e. the differences between the observed
and the predicted values of the transformed model parameters. Next, the procedure for generating
a suite of ground motions for specified earthquake and site characteristics is described, including
the case where one or more of the model parameters are prescribed. Comparisons are made
between selected recorded and synthetic ground motions, demonstrating similarity of their features.
Finally, the proposed model is validated by comparing the statistics of elastic response spectra of
simulated motions at given periods with those of NGA ground motion prediction equations [18].
Full details of the methods and results presented in this paper are reported in the first author’s
doctoral dissertation [19].
2. STOCHASTIC GROUND MOTION MODEL
Earthquake ground motions have nonstationary characteristics both in time and frequency domains.
The temporal nonstationarity refers to the variation of the intensity of the ground motion in time
and is usually characterized by an evolving intensity that gradually increases from zero to achieve
a nearly constant value, representing the strong-shaking phase of the motion, and then gradually
decays back to zero. The spectral nonstationarity refers to the variation of the frequency content
of the motion in time, which arises from the evolving nature of the seismic waves arriving at the
site. It is important to properly model both these effects, particularly for inelastic and degrading
structures, which tend to have resonant frequencies that also evolve in time.
2.1. Model formulation

In this paper, we employ the stochastic ground motion model proposed in our earlier work [8].
This model accounts for both the temporal and spectral nonstationarities of the motion. Furthermore,
it has the important advantage of completely separating the temporal and the spectral nonstationary
characteristics of the process, thereby greatly facilitating estimation of the model parameters.
The simulated acceleration process is passed through a high-pass filter to assure zero residual
velocity and displacement, as well as to produce reliable response spectral values at long periods.
DOI: 10.1002/eqe
White-noise Unit-variance process with

w(τ) spectral non-stationarity
Linear filter Normalization by Time
with standard-deviation modulating
time-varying of the filtered filter
parameters white-noise process
Fully non-stationary process Simulated ground acceleration

x(t ) z (t )
High-pass
filter
Figure 1. Procedure for simulating a single ground motion.
Figure 1 illustrates the steps involved in simulating a single ground acceleration time-history for
a given set of model parameters.
As can be seen in Figure 1, before high-pass filtering, the ground acceleration process x(t)
is obtained by time-modulating a normalized filtered white-noise process with the filter having
time-varying parameters. In a continuous form, the model is defined as
t
1
x(t) = q(t, a) h[t −, k()]w() d (1)
h (t) −∞
In this expression, q(t, a) is a deterministic, nonnegative, time-modulating function with parameters

a controlling its shape and intensity; w() is a white-noise process; the integral inside the curved
brackets is a filtered white-noise process with h[t −, k()] denoting the
timpulse-response function
(IRF) of the filter with time-varying parameters k(); and 2h (t) = −∞ h 2 [t −, k()] d is the
variance of the integral process. Because of the normalization by h (t), the process inside the
curved brackets has unit variance. As a result, q(t, a) equals the standard deviation of the resulting
process x(t). It should be clear that the modulating function q(t, a) completely defines the temporal
characteristics of the process, whereas the form of the filter IRF and its time-varying parameters
define the spectral characteristics of the process.
For the current study, we select a ‘gamma’ modulating function:
q(t, a) = 1 t 2 −1 exp(−3 t) (2)
where a = (1 , 2 , 3 ), 0<1 , 3 , and 1<2 . Of the three parameters, 1 controls the intensity of
the process, 2 controls the shape of the modulating function, and 3 controls the duration of the
motion. For the filter IRF, we select a form that corresponds to the pseudo-acceleration response
of a single-degree-of-freedom linear oscillator:

f ()
h[t −, k()] = exp[−f ()f ()(t −)]×sin[f () 1−2f ()(t −)], t
1−2f (t) (3)
=0 otherwise
DOI: 10.1002/eqe
where k() = (f (), f ()) is the set of time-varying parameters of the IRF with f () denoting
the frequency of the filter and f () denoting its damping ratio. Note that these parameters depend
on the time of application of the pulse. The parameters f () and f (), respectively, control the
evolutionary predominant frequency and bandwidth of the process.
In order to facilitate digital simulation, the stochastic model in (1) is discretized in time domain
into the form

n
x̂(t) = q(t, a) {si (t, k(ti ))u i } (4)
i=1
where ti , i = 0, 1, . . . , n, are a set of equally spaced time points with t0 = 0 and tn representing the
total duration of the motion; u i are a set of standard normal random variables representing random
pulses at the discrete time points ti , and for 0<ik where 1kn −1,
h[t −ti , k(ti )]
si (t, k(ti )) = , tk t<tk+1
k 2 [t −t , k(t )]
j=1 h j j (5)
= 0, t<ti
For a given modulating function and filter IRF, a realization of the process in (4) is obtained by
simulating the random variables u i and performing the indicated superposition.
As mentioned earlier and illustrated in Figure 1, the simulated process is eventually high-pass
filtered. This is necessary to assure zero residual velocity and displacement of the motion, as well
as realistic response spectral values at long periods. As has been shown in our previous paper
[8], without such filtering, stochastically generated ground motions tend to overestimate response
spectral values in the long period range (usually greater than 2 s). As in [8], a critically damped
oscillator is selected for the high-pass filter. The corrected acceleration record, denoted by z̈(t), is
obtained as the solution of the differential equation
z̈(t)+2c ż(t)+2c z(t) = x̂(t) (6)
where c is the frequency of the filter and x̂(t) is the discretized acceleration process as defined in
(4). This filter is motivated by Brune’s [20, 21] source model, in which c is the ‘corner frequency’
and is related to the geometry of the seismic source and the shear-wave velocity. It is noted that
this filtering has little influence on the frequency content of the acceleration process beyond the
corner frequency c .
Further details of the above stochastic model are given in [8, 19].
2.2. Model parameterization

Since we wish to relate the parameters of the modulating function to the earthquake and site
characteristics of recorded motions, it is desirable that these parameters be defined in terms of
ground motion properties that have physical meanings. For this reason, (1 , 2 , 3 ) are related to
three physically based variables ( I¯a , D5−95 , tmid ). The first variable, I¯a , represents the expected
Arias intensity [22] of the acceleration process x̂(t)—a measure of the total energy contained in
the motion—and is defined as
tn
¯Ia = E x̂ (t) dt =
2 tn 2
q (t, a) dt (7)
2g 0 2g 0
DOI: 10.1002/eqe
where g is the gravitational acceleration. The second equality above is obtained by changing the
orders of the expectation and the integration operations and noting that q 2 (t, a) is the variance of
the process. D5−95 represents the effective duration of the motion. Here, motivated by the work
of Trifunac and Brady [23], we define D5−95 as the time interval between the instants at which
the 5 and 95% of the expected Arias intensities are reached. This definition is selected since it
relates to the strong-shaking phase of the time-history, which is critical to nonlinear response of
structures. tmid is the time at the middle of the strong-shaking phase. Based on the investigation
of many ground motions in our database, we have selected tmid as the time at which 45% level of
the expected Arias intensity is reached.
For the selected modulating function in (2), one can show that the variance function q 2 (t, a) is
proportional to a gamma probability density function (PDF) having parameter values 22 −1 and
23 [24]. Let t p represent the p-percentile variate of the gamma cumulative distribution function
Then t p is given in terms of the inverse of the gamma cumulative distribution function at probability
value p%. It follows that t p is uniquely given in terms of the parameters 2 and 3 and probability
p%. We can write
D5−95 = t95 −t5 (8)

tmid = t45 (9)
For given values of D5−95 and tmid , parameters 2 and 3 can be numerically computed from the
above two equations. Furthermore, one can easily show that 1 is directly related to the expected
Arias intensity through

22 −1
(23 )
1 = I¯a (10)
(22 −1)
where (.) is the gamma function. In the remainder of this paper, we only work with
( I¯a , D5−95 , tmid ) as the modulating function parameters. Any simulated values of these parameters
are used in (8)–(10) to back-calculate the corresponding values of (1 , 2 , 3 ), which are then
used to compute the modulating function (2).
As a simple approximation and based on the analysis of a large number of accelerograms, we
have adopted a linear function for the filter frequency and a constant value for the filter damping
ratio:
f () = mid + (−tmid ) (11)

f () = f (12)
Here, mid represents the filter frequency at tmid , and represents the rate of change of the filter
frequency with the time of application of the pulse.
In summary, the physically based parameters ( I¯a , D5−95 , tmid ) and (mid , , f ) completely
define the time modulation and the evolutionary frequency content of the nonstationary ground
motion model. Our simulation procedure is based on generating samples of these parameters for
given earthquake and site characteristics.
DOI: 10.1002/eqe
3. IDENTIFICATION OF MODEL PARAMETERS FOR A TARGET ACCELEROGRAM
Given a target accelerogram, the model parameters are identified by matching the properties of
the recorded motion with the corresponding statistical measures of the process. The modulating
function parameters ( I¯a , D5−95 , tmid ) are naturally matched with the Arias intensity, the effective
duration (time between 5 and 95% levels of Arias intensity), and the time to the middle of
the strong-shaking phase (time to the 45% level of Arias intensity) of the record, respectively.
In determining tmid for a recorded accelerogram, sometimes it is necessary to make a time shift.
This is because the zero point along the time axis of a record is rather arbitrary (There is no
standard as to where to set the initial point of an acceleration signal.) In fact, some records in the
NGA database have long stretches of zero motion in their beginning. In such cases, a better fit is
achieved by identifying an additional parameter, t0 0. This is done by replacing t in (2) by t −t0
and replacing (9) by D5−45 = t45 −t5 , where D5−45 is the time interval between 5 and 45% levels
of Arias intensity of the record. After identification of a, t0 is computed as the difference between
the time to the 45% level of Arias intensity of the record and t45 . Finally, tmid is determined as
tmid = t45 −t0 .
As mentioned earlier, the model parameters (mid , , f ) control the evolving predominant
frequency and bandwidth of the process. As a measure of the evolving predominant frequency of
the recorded motion, as in [8], we consider the rate of zero-level up-crossings, and as a measure
of its bandwidth, we consider the rate of negative maxima (peaks) and positive minima (valleys).
In [8], the evolution of the predominant frequency was determined by minimizing the difference
between the cumulative mean number of zero-level up-crossings of the process in time with the
cumulative count of the zero-level up-crossings of the recorded accelerogram. The bandwidth
parameter, f , was determined by minimizing the difference between the mean rate of negative
maxima and positive minima with the observed rate of the same in the recorded accelerogram.
The process required an iterative scheme, since the predominant frequency and bandwidth of the
process are interrelated. That method is ideal if the purpose is to closely match the statistical
characteristics of a target accelerogram. For the purpose of identifying the model parameters
for a large number of earthquakes, such high level of accuracy is not necessary. Instead, the
following simpler method is adopted to reduce computational effort, while providing sufficient
accuracy.
One can show [25] that the mean zero-level up-crossing rate of the stationary response of a
second-order filter (i.e. the filter defined by (3) with time-invariant parameters) to a white-noise
excitation is equal to the filter frequency. This motivates the idea of approximating the filter
frequency f () by the rate of change of the cumulative count of zero-level up-crossings of the
target accelerogram (see Figure 2(a)). In order to identify the two parameters mid and for a
given record, a second-order polynomial is fitted to the cumulative count of zero-level up-crossings
of the accelerogram. This is done in a least-squares sense at equally spaced time points starting
from the time at 1% level of Arias intensity to the time at 99% level of Arias intensity (a total
of nine points were selected in this study). The fitted polynomial is then differentiated to obtain a
linear estimate of the filter frequency as a function of time. The value of this line at tmid represents
the estimate of mid , and its slope represents the estimate of . Figure 2(a) demonstrates this
fitting process for the component 090 of the accelerogram recorded at the Silent Valley-Poppet
Flat station during the 1992 Landers earthquake. Comparisons of the estimated filter frequency
with those computed by the more exact method in [8] for several accelerograms revealed that the
method is sufficiently accurate for the intended purpose.
DOI: 10.1002/eqe
200
Cumulative count of zero-level up-crossings
Target
Target
positive-minima and negative-maxima

300 Simulation when zeta=0.9
Fitted points
160 =0.8
Second degree polynomial
=0.7
1 =0.6
Cumulative count of
ω f (t ) =0.5
120 =0.4
200
=0.3
=0.2
=0.1
80
100
40
0 0
t
0 10 20 30 40 50 60 5 10 15 20 25 30 35 40
(a) Time, sec (b) Time, sec (from 5% to 95% level of Arias intensity)
Figure 2. Identification of filter parameters. (a) Matching the cumulative number of zero-level up-crossings
results in mid /2 = 6.92 Hz and /2 = −0.14 Hz/s and (b) matching the cumulative count of negative
maxima and positive minima gives f = 0.40.
To estimate the filter damping ratio, the cumulative number of negative maxima plus positive
minima for the target accelerogram is determined. This value is compared with the estimated
averages of the same quantity for sets of 20 simulations of the theoretical model with the already
approximated filter frequency and the damping values f = 0.1, 0.2, . . . , 0.9 (see Figure 2(b)). Inter-
polation between the curves is used to determine the optimal value of f that best matches the curve
for the target accelerogram. For this analysis, only the time interval between 5 and 95% levels of
Arias intensity is considered, where it is more likely for f to remain constant. This procedure is
a simplification of the more refined fitting method used in [8], as it neglects the influence of the
filter damping on the predominant frequency. Figure 2(b) shows application of this method to the
Landers earthquake record mentioned above.
It is important to note that the modulating function has no influence on the zero-level up-
crossings, or the number of negative maxima and positive minima of the process. This facilitates
estimation of the filter parameters after determining the modulating function parameters.
4. METHODOLOGY FOR DEVELOPING PREDICTIVE EQUATIONS
Our interest is in simulating ground motions for a given set of earthquake and site characteristics
(e.g. fault mechanism, earthquake magnitude, source-to-site distance, local soil type). In this
context, parameters identified for a specific recorded ground motion are regarded as a single
realization of the parameter values that could arise from earthquakes of similar characteristics
on similar sites. To develop a predictive model of the ground motion, it is necessary to relate
the model parameters to the earthquake and site characteristics. For this purpose, we identify
the model parameters for a data set of recorded ground motions with known earthquake and site
characteristics. Using these data, regression models are then developed to relate the stochastic
model parameters to the earthquake and site characteristics.
It is a common practice in developing predictive equations of ground motion intensities to work
with the logarithm of the data to satisfy the normality requirement of the regression error. This
DOI: 10.1002/eqe
transformation implies the lognormal distribution for the predicted intensity. In our case, the data
for several of the model parameters show distinctly nonlognormal behavior, including negative
values and bounds, which cannot be resolved by a logarithmic transformation. To account for
this behavior, each model parameter is assigned a marginal probability distribution based on its
observed histogram. This distribution is then used to transform the data to the normal space, where
empirical prediction models are constructed. In effect, this is a generalization of the logarithmic
transformation.
For convenience, let i denote the ith parameter in the set ( I¯a , D5−95 , tmid , mid /2, /2, f ),
i = 1, . . . , 6, and Fi (i ) denote the marginal cumulative distribution fitted to the data for i .
The marginal transformations
i = −1 [Fi (i )], i = 1, . . . , 6 (13)
where −1 [.] is the inverse of the standard normal cumulative probability function, then define the
set of standard normal random variables i . In our approach, the data on i are transformed to data
on i , which are then regressed against variables defining the earthquake and site characteristics.
This leads to predictive equations of the form
i =
i (Earthquake, Site, bi )+ei , i = 1, . . . , 6 (14)
where
i is a selected predictive formula for the conditional mean of i given the earthquake and
site characteristics, bi is the vector of regression coefficients, and ei represents the regression error
that is zero-mean normally distributed. The following sections elaborate on the selected earthquake
data set, the fitted distributions Fi (i ), the functional forms of the predictive equations (14), the
method of analysis used to estimate the regression coefficients and the error variance, and the
analysis of cross-correlations between the model errors.
5. STRONG MOTION DATABASE
The strong motion database used in this study is a small subset of the PEER NGA (Pacific Earth-
quake Engineering Research Center: Next Generation Attentuation of Ground Motions Project;
see http://peer.berkeley.edu/smcat/.) database, and a subset of the data used in the development of
the Campbell–Bozorgnia (CB) NGA ground motion model [26]. These data were collected for the
Western United States (WUS), but some well-recorded, large-magnitude earthquakes from other
regions, which were deemed to be applicable to the WUS, are also included [18]. As in Camp-
bell and Bozorgnia [26], the database employed in this study excludes aftershocks. Furthermore,
the accelerograms in the database are representatives of ‘free-field’ ground motions recorded in
shallow crustal events in tectonically active regions.
The NGA database lists many characteristics of each earthquake and recording site. Considering
the type of information that is commonly available to a design engineer, four parameters were
selected for the present study: (F, M, Rrup , Vs30 ). F corresponds to the type of faulting with F = 0
denoting a strike-slip fault and F = 1 denoting a reverse fault (normal faults were not considered
since few recordings are available); M represents the moment magnitude of the earthquake;
Rrup represents the closest distance from the recording site to the ruptured area, and Vs30 represents
the shear-wave velocity of the top 30 m of the site soil. Among these parameters, F and M
characterize the earthquake source, Rrup characterizes the location of the site relative to the
DOI: 10.1002/eqe
earthquake source, and Vs30 characterizes the local soil conditions. These parameters are believed
to have the most significant influences on the ground motion at a site. Additional parameters can
be included to refine the predictive equations in future studies.
During the design process, two levels of ground motion are commonly considered: the service-
level ground motion and the maximum considered earthquake (MCE) ground motion. While
response spectrum analysis is sufficient to analyze a structure for the service-level motion during
which the structure is expected to remain elastic, response-history dynamic analysis is usually
recommended or required to capture the likely nonlinear behavior of a structure subjected to
the MCE motion. Many predictive models are available that provide the spectral ordinates of
ground motion required for the response spectrum analysis, including the recently developed and
commonly used NGA ground motion prediction equations by Abrahamson and Silva [27], Boore
and Atkinson [28], Campbell and Bozorgnia [26], Chiou and Youngs [29], and Idriss [30]. Aiming
at a predictive model of ground motion time-histories for the MCE event, we decided to only
consider earthquakes having 6.0M. By limiting the database to large earthquakes, the predictive
equations presented in this study are customized for earthquakes that are capable of damage and
can cause nonlinear behavior in structures.
In the interest of separating the effects of near-fault ground motions, such as the directivity
and fling effects, which could dominate the spectral content of the ground motion [31], only
earthquakes with 10 kmRrup are considered. A separate study for simulation of near-field ground
motions is currently underway. Furthermore, an upper limit Rrup 100 km is selected to exclude
ground motions of small intensity.
In the interest of separating the effect of soil nonlinearity, which can also strongly influence
the spectral content of the ground motion, the lower limit 600 m/sVs30 is selected. For smaller
Vs30 values, one can generate appropriate motions at the firm soil layer and propagate through the
softer soil deposits using standard methods of soil dynamics that account for the nonlinearity in
the shear modulus and damping of the soil.
Figure 3 shows a summary of the selected earthquakes from the CB-NGA database within
the above-stated limits. These constraints reduced the data set used in the analysis to 31 pairs
of horizontal recordings from 12 earthquakes for strike-slip type of faulting, and 72 pairs of
horizontal recordings from 7 earthquakes for reverse type of faulting. The selected earthquakes
and the number of recordings for each earthquake are listed in Table I.
Although the imposed constraints on the earthquake and site characteristics have reduced the
number of recordings in our database, the resulting predictive equations are simpler (additional
terms that reflect influences of low magnitude earthquakes, near-fault ground motions, distant
earthquakes, and nonlinearity of soft soil are not required) and more reliable for the intended
application of nonlinear response-history analysis for the MCE event. Note that the selection of
the database in no way limits the methodology presented in this paper.
6. CONSTRUCTION OF PREDICTIVE EQUATIONS
In this section, we construct empirical prediction equations for each of the model parameters in
terms of the earthquake and site characteristic variables F, M, Rrup , and Vs30 through regression
analysis of the computed dataset of fitted parameter values. The correlations between the predicted
model parameters are also estimated. For simplicity of the notation, hereafter Rrup and Vs30 are
denoted as R and V .
DOI: 10.1002/eqe
8
Strike-slip
Reverse
7.5
Moment Magnitude
6.5
6
10 20 30 40 50 60 70 80 90 100
Closest Distance to Rupture, km
Figure 3. Distribution of moment magnitude and source-to-site distance in the considered database.
Table I. Selected earthquakes from the CB-NGA database, type of faulting,

magnitude, and number of records.
Earthquake Earthquake ID Moment Number
name in NGA database Faulting mechanism magnitude of records
1 Imperial Valley-06 0050 Strike-slip 6.53 2
2 Victoria, Mexico 0064 Strike-slip 6.33 2
3 Morgan Hill 0090 Strike-slip 6.19 10
4 Landers 0125 Strike-slip 7.28 4
5 Big Bear-01 0126 Strike-slip 6.46 10
6 Kobe, Japan 0129 Strike-slip 6.9 4
7 Kocaeli, Turkey 0136 Strike-slip 7.51 4
8 Duzce, Turkey 0138 Strike-slip 7.14 2
9 Sitka, Alaska 0140 Strike-slip 7.68 2
10 Manjil, Iran 0144 Strike-slip 7.37 2
11 Hector Mine 0158 Strike-slip 7.13 16
12 Denali, Alaska 0169 Strike-slip 7.9 4
13 San Fernando 0030 Reverse 6.61 14
14 Tabas, Iran 0046 Reverse 7.35 2
15 Coalinga-01 0076 Reverse 6.36 2
16 N. Palm Springs 0101 Reverse 6.06 12
17 Loma Prieta 0118 Reverse 6.93 28
18 Northridge-01 0127 Reverse 6.69 38
19 Chi-Chi, Taiwan 0137 Reverse 7.62 48
DOI: 10.1002/eqe
0.5 0.08 0.08

Observed Data
Fitted PDF
0.4
Normal 0.06 Beta 0.06 Beta
0.3
0.04 0.04
0.2
0.02 0.02
Normalized Frequency (Total:206)
0.1
0 0 0
-7.5 -5.5 -3.5 -1.5 0 0 10 20 30 40 50 0 10 20 30 40
ln(Ia, sec.g) D5-95, sec tmid, sec
0.2 5 4
4
0.15 Two -Sided 3 Beta
Gamma
3 Exponential
0.1 2
2
0.05 1
1
0 0 0
0 5 10 15 20 25 -2 -1.5 -1 -0.5 0 0.5 0 0.5 1
ωmid/(2π), Hz ω'/(2π), Hz/sec ζf
Figure 4. Probability density functions superimposed on observed normalized frequency

diagrams of model parameters.
6.1. Fitted distributions

After identifying the model parameter values by fitting to each recorded ground motion in the
database, a probability distribution is assigned to the sample of values of each parameter. The form
of this distribution is inferred by visually inspecting the corresponding histogram and examining
the fit to the corresponding empirical cumulative distribution function. The parameters of the
chosen probability distribution are then estimated by the method of maximum likelihood.
Figure 4 shows the normalized frequency diagrams of the fitted model parameters for the entire
data set (combined strike-slip and reverse faulting mechanisms). The fitted PDFs are superimposed
and their distribution types and parameter values are listed in Table II. For /2, the fitted
distribution is a two-sided exponential with the PDF
⎧
⎪
⎪ 4.85 exp(6.77 /2), −2< /2<0
⎨
f /2 ( /2) = 4.85 exp(−17.10 /2), 0< /2<0.5 (15)
⎪
⎪
⎩0 otherwise
As commonly assumed in the current practice, the data for I¯a is found to be well represented
by the lognormal distribution (ln( I¯a ) is normally distributed). It is observed that the duration
parameter D5−95 varies between 5.4 and 41.3 s, with a mean of about 17 s. The parameter tmid
assumes values between a fraction of a second to about 35.2 s with a mean of about 12 s. Rounded
bounds for the corresponding distributions are assumed to reflect these limits (see Table II). For
some records, tmid is found to be greater than D5−95 due to a long stretch of low-intensity motion
in the beginning of the record.
It is interesting to note that the observed predominant frequency at the middle of strong shaking,
mid /2, ranges from 1.31 to 21.6 Hz for the records in the data set, with a mean value of 5.87 Hz.
The fact that only rock and stiff soils are considered is the reason for this high mean value. It is also
DOI: 10.1002/eqe
Table II. Distribution models assigned to the model parameters.

Parameter Fitted distribution Distribution bounds Mean Std. dev.
I¯a (s g) Lognormal (0, ∞) 0.0468 0.164

D5−95 (s) Beta [5, 45] 17.3 9.31
tmid (s) Beta [0.5, 40] 12.4 7.44
mid /2 (Hz) Gamma (0, ∞) 5.87 3.11
/2 (Hz/s) See (15) [−2, 0.5] −0.0892 0.185
f Beta [0.02, 1] 0.213 0.143
ν1 ν2 ν3
2
0
Quantiles of Observed Data
-2
-4
ν4 ν5 ν6
2
-2
-4
-4 -2 0 2 4 -4 -2 0 2 4 -4 -2 0 2 4
Standard Normal Quantiles
Figure 5. Quantile plots of transformed data for each model parameter. Hollow
circles indicate the first and the third quartiles.
interesting to note that /2 is more likely to be negative than positive (see the middle bottom
graph in Figure 4), i.e. the predominant frequency of the ground motion during the strong motion
phase is more likely to decrease than increase with time. This is consistent with our expectation.
However, a small fraction of the recorded motions in the database shows positive but small /2
(i.e. the target plot similar to the one in Figure 2(a) shows a slightly positive or, in rare cases, an
irregular nonconstant curvature). Finally, the observed filter damping ratio f , which is a measure
of the bandwidth of the ground motion, is found to range from 0.03 to 0.77 with a mean of 0.21.
Figure 5 shows quantile plots of the data for each parameter, after transformation according
to (13), versus the corresponding normal quantiles. It is observed that in most cases the data
within the first and third quartiles (marked by hollow circles) closely follow a straight line, thus
DOI: 10.1002/eqe
confirming that the transformed data follow the normal distribution. The worst fit belongs to the
Arias intensity, for which the commonly assumed lognormal distribution was adopted. We conclude
that the selected distributions provide an effective means for transforming the data to the normal
space. This process helps us to satisfy the normality assumption underlying the regression analysis
that is used to develop empirical prediction equations for the model parameters, as described in
the next section.
6.2. Regression analysis

As seen in Table I, the database contains different numbers of records for different earthquakes.
The records associated with each earthquake correspond to different source-to-site distances, soil
types, or orientations (two orthogonal horizontal components are available for each recording
station). While there are 48 records from the Chi-Chi earthquake, several earthquakes contribute
only two records. This uneven clustering of data must be accounted for in the regression analysis,
so that the results are not overly influenced by an individual earthquake with many records. Further-
more, each earthquake is expected to have its own particular effect on its resulting ground motions.
This effect is random and varies from earthquake to earthquake. Therefore, the data corresponding
to ground motions from the same earthquake have a common factor and are correlated, whereas the
data corresponding to different earthquakes are statistically independent observations. To address
these issues, a random-effects regression analysis method is employed. This method effectively
handles the problem of weighing observations and, unlike ordinary regression analysis, assumes
that data within earthquake clusters are statistically dependent. We employ the random-effects
regression model in the form
i, jk =
i (F j , M j , R jk , Vk , bi )+ ˆ i, j + î, jk (16)
where i = 1, . . . , 6 indexes the model parameters, j = 1, . . . , 19 indexes the earthquakes, and k =

1, . . . , n j indexes the records associated with the jth earthquake with n j denoting the number of
records from that earthquake. The transformed model parameter, i, jk , is chosen as the response
parameter of the regression.
i and bi are defined in (14). The former is more precisely denoted
as
i |F,M,R,V , the predictive (conditional mean) value of i for given F, M, R, and V . Having
random effects necessitates a more careful definition of the residuals. Therefore, the total residual,
defined as the difference between the observed and predicted values of the response variable, is
represented as the sum of ˆ i, j and î, jk , respectively referred to as the inter-event (random effect
for the jth earthquake) and intra-event residuals. The superposed hats indicate that these residuals
are observed values of independent, zero-mean, normally distributed error terms i and i with
variances i2 and i2 , respectively. With this arrangement, the total error for the ith model parameter
is a zero-mean normally distributed random variable with variance i2 +i2 .
One may argue that since the two horizontal components of each record are correlated, an
additional random effect term needs to be included in (16). This is not necessary because the two
components are included for all the records of the database. In effect, the dependence between
the pairs of components at each site is accounted for through the random-effect term for all the
recordings of the same earthquake. Therefore, the resulting parameter estimates are unbiased.
Eventually, the resulting sample correlations between the data corresponding to the two horizontal
components of ground motion at each site provide a means for simulating pairs of ground motion
components at a site of interest.
DOI: 10.1002/eqe
Random-effects modeling is sometimes referred to as variance-components modeling because for

a given database, in addition to estimating the regression coefficients bi , one needs to individually
estimate the error variances i2 and i2 . In this study, we employ the maximum likelihood technique
to obtain estimates of all the regression coefficients and variances at one step. Although this
method requires the use of a numerical optimization technique, it is not computationally intensive
and, unlike other proposed methods (e.g. [32, 33]), does not require a complicated algorithm that
calculates the regression coefficients and the variance components in separate, iterative steps.
The likelihood function is formulated by noting that the observed values of the total residuals are
jointly normal with a zero-mean vector and a block-diagonal covariance matrix. Therefore, for the
ith model parameter, the likelihood function of the regression coefficients and variance components
is equal to the joint normal PDF evaluated for the observed values of the total residuals. Writing
the total residuals as i, jk −
i (F j , M j , R jk , Vk , bi ) and collecting the values for all j and k into
vectors vi and li (bi ), the likelihood function assumes the form

1 1 T −1
L(bi , i2 , i2 ) = exp − [vi −li (bi )] i [vi −li (bi )] (17)
|i |1/2 2
where i is the covariance matrix, which is expressed as a function of the variance components
i2 and i2 in the form
⎡ ⎤
i2 In 1 +i2 1n 1 0 ... 0
⎢ ⎥
⎢ 0 i2 In 2 +i2 1n 2 ... 0 ⎥
⎢ ⎥
i (i , i ) = ⎢
2 2
⎢ .. .. ..
⎥
⎥ (18)
⎢ .. ⎥
⎣ . . . . ⎦
0 0 ... i2 In 19 +i2 1n 19 N ×N
In the above, In is the identity matrix of size n, 1n is an n ×n matrix of 1’s, and N is the total
number of observations (206 in the present case). The above formulation takes into consideration
the fact that data corresponding to records from different earthquakes are uncorrelated (off-diagonal
blocks are zero), data corresponding to the records from the same earthquake have correlation
i2 /(i2 +i2 ) (off-diagonal elements of the diagonal blocks), and each data point is fully correlated
with itself (diagonal elements equal i2 +i2 ). Maximum likelihood estimates of the parameters bi ,
i2 , and i2 for each transformed stochastic model parameter i are obtained by maximizing the
function in (17) relative to these parameters. The MATLAB optimization toolbox is used for this
purpose.
To assess the sufficiency of the selected functional forms for each regression law, plots of the
residuals versus predictor variables are inspected. This process, which is commonly known as the
analysis of residuals, requires partitioning of the total residuals into inter-event and intra-event
residuals. Estimating the inter-event residuals for each group (data corresponding to the records
of a single earthquake) involves calculating the raw residuals (average of the total residuals in that
group) and multiplying it with a shrinkage factor that reflects the relative size of the variation of
that group to the total variation in the database [34]. The intra-event residuals are then estimated
as the difference between the total and the inter-event residuals.
DOI: 10.1002/eqe
6.3. Regression results

For the sake of simplicity and considering the relatively narrow range of earthquake magnitudes, a
linear form of the regression equation for each transformed model parameter in terms of explana-
tory functions representing the type of faulting, earthquake magnitude, source-to-site distance,
and soil effect was employed. Various linear and nonlinear forms of the explanatory functions
were examined. In view of the availability of previous predictive formulas for Arias intensity
and duration [35, 36], more possible forms of the explanatory functions for these two parame-
ters were investigated. For the other model parameters, alternative forms were considered only
if the linear form revealed inadequate behavior of the residuals. For each model parameter, the
relative performances of the resulting functional forms were assessed by inspecting the resid-
uals and the estimates of the variance components. The functional forms with smaller variances
that demonstrated adequate behavior of the residuals (i.e. lack of systematic patterns in the plots
of residuals versus the predictor variables) were selected. The resulting predictive equations are
given in (19) and (20) with the estimated regression parameters and standard deviations listed in
Table III. Standard significance tests verified the adequacy of the regression for each model param-
eter at the 90% and higher confidence levels. Furthermore, the regression coefficients i,1 , i,2 , i,3 ,
and i,4 (i = 1, . . . , 6) were individually tested; those with statistical significance at the 95% confi-
dence level are shown in bold in Table III (confidence intervals and p values are reported in [19]).
However, in the subsequent analysis, all the coefficients in Table III are used to randomly generate
the model parameters and simulate ground motions.

M R Vs
1 = 1,0 + 1,1 (F)+ 1,2 + 1,3 ln + 1,4 ln +1 +1 (19)
7.0 25 km 750 m/s

M R Vs
i = i,0 + i,1 (F)+ i,2 + i,3 + i,4 +i +i , i = 2, . . . 6 (20)
7.0 25 km 750 m/s
The moment magnitude, source-to-site distance, and shear-wave velocity terms in the predictive
equations have each been normalized by a typical value for engineering purposes. This normal-
ization renders the regression coefficients dimensionless.
The estimated parameters in Table III provide some interesting insight. For example, we observe,
as expected, that Arias intensity tends to increase with magnitude and decrease with distance and
site stiffness. The effective duration as well as tmid tend to increase with magnitude and distance
(more distant sites tend to experience longer motions) and tend to decrease with site stiffness.
These findings are consistent with prior observations [23, 35, 36]. The results also suggest that
Table III. Maximum likelihood estimates of regression coefficients and standard error components.
i i,0 i,1 i,2 i,3 i,4 i i
1 −1.844 −0.071 2.944 −1.356 −0.265 0.274 0.594
2 −6.195 −0.703 6.792 0.219 −0.523 0.457 0.569
3 −5.011 −0.345 4.638 0.348 −0.185 0.511 0.414
4 2.253 −0.081 −1.810 −0.211 0.012 0.692 0.723
5 −2.489 0.044 2.408 0.065 −0.081 0.129 0.953
6 −0.258 −0.477 0.905 −0.289 0.316 0.682 0.760
DOI: 10.1002/eqe
the effective duration and tmid tend to be shorter for reverse faulting compared with strike-slip
faulting. Furthermore, the results indicate that the predominant frequency at the middle of strong
shaking tends to decrease with increasing magnitude and source-to-site distance, whereas the rate
of change of the predominant frequency (which has a negative mean) tends to increase, i.e. a slower
change with increasing magnitude and distance. Finally, the filter damping, which is a measure
of the bandwidth of the ground motion, tends to increase with the moment magnitude and site
stiffness and decrease with source-to-site distance. These trends are in general consistent with our
expectations.
Figure 6 shows the diagnostic scatter plots of the residuals versus the predictor variables. These
plots show that the residuals are evenly scattered above and below the zero level with no obvious
systematic trends. This implies lack of bias and a good fit of the regression models to the data.
6.4. Correlation analysis

For a given set of earthquake and site characteristics (F, M, R, V ), the parameters i and, therefore,
i are correlated. These are estimated as the correlations between the total residuals ˆ i + î . Table IV
lists the correlation coefficients between the jointly normal variables i . Several of these estimated
correlations provide interesting insight. Observe that there is negative correlation between 1 and
2 (corresponding to I¯a and D5−95 ). This is somewhat surprising, since one would expect a higher
Arias intensity for a longer duration. However, since Arias intensity is more strongly related to the
amplitude of the motion than to the duration (it is related to the square of the amplitude but linear
in duration), this result may be due to the tendency of motions with high amplitude to have shorter
durations. This negative correlation has also been observed by Trifunac and Brady [23]. Second, a
strong positive correlation is observed between 2 and 3 (corresponding to D5−5 and tmid ), which
is as expected. Interestingly, 4 (corresponding to mid ) has negative correlations (though small)
with all three previous parameters. Thus, higher intensity and longer duration motions tend to have
lower predominant frequency. The correlation between 4 and 5 (corresponding to mid and )
is negative, indicating that motions with higher predominant frequency tend to have a faster decay
of the frequency with time. Finally, the positive correlation between 4 and 6 (corresponding to
mid and f ) suggests that high-frequency motions tend to have broader bandwidth.
7. SIMULATION OF GROUND MOTIONS FOR A GIVEN DESIGN SCENARIO
Given a design scenario expressed in terms of F, M, R, and V , any number of synthetic ground
motions can be generated based on the information provided in the preceding sections and without
the need for any previously recorded motion. We observe that, given F, M, R, and V , the random
variables i , i = 1, . . . , 6, are jointly normal with means
i (F, M, R, V, bi ), as expressed in (19)–
(20) with bi values listed in Table III, variances i2 +i2 , which can be computed from values
listed in Table III, and the correlation matrix in Table IV. This information is sufficient to generate
samples of these random variables. Each set of simulated values of these six variables is used in
(13) together with the assigned marginal distributions of Table II to compute the corresponding set
of model parameters (1 , 2 , 3 , 4 , 5 , 6 ) = ( I¯a , D5−95 , tmid , mid /2, /2, f ). The first three
parameters are then converted to the modulating function parameters a = (1 , 2 , 3 ) by use of
(8)–(10), yielding the set (1 , 2 , 3 , mid /2, /2, f ). These parameter values together with a
set of n statistically independent standard normal random variables u i , i = 1, . . . , n, are then used in
DOI: 10.1002/eqe
Figure 6. Scatter plots of residuals against earthquake magnitude, source-to-site distance, and shear-wave
velocity for each model parameter.
the stochastic model in (4) and the high-pass filter in (6) to generate a synthetic accelerogram, z̈(t).
Any number of accelerograms for the given earthquake and site characteristics can be synthesized
by generating new realizations of i and u i .
DOI: 10.1002/eqe
Table IV. Sample correlations between the error terms of model parameters.
1 2 3 4 5 6
1 1 −0.36 0.01 −0.15 0.13 −0.01
2 1 0.67 −0.13 −0.16 −0.20
3 1 −0.28 −0.20 −0.22
4 1 −0.20 0.28
5 Sym. 1 −0.01
6 1
The simulation method described above maintains the natural variability of ground motions
for a given set of earthquake and site characteristics. To demonstrate this, in Figures 7(a),(b),
and (c) we show three sets of ground motions for given values of F, M, R, and V . Each set
includes one recorded motion and two simulated motions and for each motion the acceleration,
velocity, and displacement time-histories are shown. Also listed in the figure is the set of parameters
for each motion (numerically evaluated for the recorded motion and randomly generated for the
synthetic motions). For the synthetic motions, a discretization step of 0.02 s and the high-pass
filter frequency c /2 = 0.1 Hz are used. Observe that although the three events have almost
identical earthquake and site characteristics (all are reverse faulting; M = 6.61, 6.93, and 6.69;
R = 19.3, 18.3, and 19.1 km; and V = 602, 663, and 706 m/s), the three recorded motions are
vastly different in their characteristics. Specifically, their Arias intensities range from 0.0401 to
0.109 s g, effective durations range from 5.95 to 12.62 s, predominant frequencies range from 3.97
to 14.58 Hz, and bandwidth parameters range from 0.03 to 0.24. Furthermore, the acceleration,
velocity, and displacement traces and their peak values are vastly different. Similar variability
can be observed among the simulated motions (compare the parameter values and the traces).
Furthermore, observe that the general features of the simulated motions are similar in character to
those of the recorded motions.
In practice, it might be of interest to simulate ground motions with given values for a subset of
the model parameters, e.g. the Arias intensity, the effective duration, or the predominant frequency.
In such cases, the corresponding i variables are fixed while the remaining i variables are generated
using the conditional mean vector and covariance matrix for the given values of the fixed variables.
These conditional values can be computed from the well-known formulas in probability theory
(see, e.g. [37]). As an example, Figure 8 shows the recorded motion in Figure 7(c) together with
four synthetic accelerograms, which are conditioned to have the Arias intensity of the recorded
motion. These are obtained by generating sets of the five variables 2 to 6 for the given value
1 = ln(0.1087) of the first variable.
8. MODEL VALIDATION THROUGH EXAMINATION OF ELASTIC RESPONSE SPECTRA
8.1. Against real recorded motions

Figure 9 shows 5% damped elastic response spectra of the two components of accelerogram
recorded at the LA-Wonderland Ave station during the 1994 Northridge earthquake against 5%
damped elastic response spectra of 50 synthetic ground motions having the same fault type,
moment magnitude, source-to-site distance, and V value as the recorded motions. It is seen that
DOI: 10.1002/eqe
0.2 Recorded
Realizations of model parameter:
0
-0.2
I D t /2π ' /2 π
a 5-95 mid mid f
0.2
Acceleration, g
Simulated
sec.g sec sec Hz Hz/sec
0
-0.2
0.0401 5. 95 0.93 14.58 0.53 0.18
0.2 Simulated 0.0330 8. 83 3.44 3.74 0.05 0.80

0
-0.2 0.0801 11.24 5.86 5.42 0.16 0.12
0 5 10 15 20 25
0.05 Recorded 0.01 Recorded

0
0
-0.01
-0.05
Displacement, m
Velocity, m/sec
0.1 Simulated 0.05 Simulated
0 0
-0.1 -0.05

0 0
-0.1
-0.1
-0.2
0 5 10 15 20 25 0 5 10 15 20 25
(a) Time, sec Time, sec
0.2
Recorded
Realizations of model parameter:
0
-0.2 I D t /2π ' /2 π

a 5-95 mid mid f
0.5
Acceleration, g
Simulated
0
-0.5 0.0451 12.62 4.73 3.97 0.08 0.03

0.1
Simulated 0.2144 8. 03 4.15 5.44 0.02 0.08
0
0.0083 27. 09 9.72 3.78 0.04 0.12
-0.1
0 5 10 15 20 25 30 35 40
0.2 0.05
Recorded Recorded
0 0
-0.2 -0.05
Displacement, m
0.2
Velocity, m/sec
Simulated 0.05 Simulated

Simulated
0
0 -0.05
-0.1
-0.2
0.1 0.05
Simulated Simulated
0 0
-0.1 -0.05
0 5 10 15 20 25 30 35 40 0 5 10 15 20 25 30 35 40
(b) Time, sec Time, sec
Figure 7. (a) Recorded and synthetic motions corresponding to F = 1 (Reverse faulting), M = 6.61,
R = 19.3 km, V = 602 m/s. The recorded motion is component 291 of the 1971 San Fernando earthquake
at the Lake Hughes #12 station; (b) recorded and synthetic motions corresponding to F = 1 (Reverse
faulting), M = 6.93, R = 18.3 km, V = 663 m/s. The recorded motion is component 090 of the 1989 Loma
Prieta earthquake at the Gilroy Array #6 station; and (c) recorded and synthetic motions corresponding
to F = 1 (Reverse faulting), M = 6.69, R = 19.1 km, V = 706 m/s. The recorded motion is component 090
of the 1994 Northridge earthquake at the LA 00 station.
DOI: 10.1002/eqe
0.2 Recorded
0 Realizations of model parameter:
-0.2
Ia D5-95 tmid /2π ' /2 π

Acceleration, g
0.2 mid f
Simulated
0 sec.g sec sec Hz Hz/sec
-0.2
0.1087 7. 96 7.78 4.66 0.09 0.24
0.1
Simulated
0
0.0471 5. 86 2.74 3.99 0.09 0.60
-0.1 0.0179 20.76 13.62 6.18 0.06 0.12

0 10 20 30 40 50 60
0.05
0.4 Recorded Recorded
0.2 0
0
-0.2 -0.05
Displacement, m
Velocity, m/sec
0.4
0 0.1
0
-0.2
-0.1
0.05 0.05
Simulated Simulated
0 0
-0.05 -0.05
0 10 20 30 40 50 60 0 10 20 30 40 50 60
(c) Time, sec Time, sec
Figure 7. Continued.
0.2 Recorded
0
-0.2
Realizations of model parameters:
0.2 Simulated
Ia D5-95 tmid ωmid/2π ω’/2π
0
-0.2
Acceleration, g
0.2 Simulated
0.1087 7. 996 7.78 4.66
4 0.09 0.24
0
-0.2
0.1087 6. 668 6.60 7.68
7 0.13 0.43
0.2 Simulated
0 0.1087 17. 448 9.90 88.36 0.09 0.08
-0.2
0.1087 8. 669 8.69 6 .71 0.04 0.25

0.2 Simulated
0
-0.2 0.1087 8. 440 8.40 3.34 0.15 0.12
0 5 10 15 20 25 30 35 40
Time, sec
Figure 8. Recorded and synthetic motions with specified Arias intensity. The recorded motion and
earthquake and site characteristics are the same as in Figure 7(c).
DOI: 10.1002/eqe
101 100
Pseudo-Acceleration Response Spectrum, g
Recorded
Deformation Response Spectrum, m

Simulated
100 10-1
10-1 10-2
10-2 10-3
10
-3
10-4
10-1 100 5×100 10-1 100 5×100
Period, sec Period, sec
Figure 9. Elastic response spectra (5% damped) of two horizontal components of the 1994 Northridge
earthquake recorded at the LA-Wonderland Ave and of 50 synthetic motions. The motions correspond to
F = 1 (Reverse faulting), M = 6.69, R = 20.3 km, V = 1223 m/s.
the spectra of the recorded motions, which can be regarded as just one pair of realization of
possible ground motions resulting from an earthquake of similar characteristics, are well within
the range of variability of the spectra of the simulated synthetic motions throughout the period
range considered. The variability observed in the spectra of the synthetic motions is representative
of the variability inherent in ground motions for the given earthquake and site characteristics.
8.2. Against NGA attenuation models

The synthetic ground motions are intended for use in engineering practice as predictions of future
earthquake ground motions at a given site. Therefore, a reasonable validation approach is to
investigate how these motions compare with existing ground motion prediction equations used
in practice. For this purpose, we compare the statistics of 5% damped elastic response spectra
of a set of 500 synthetic accelerograms with the corresponding statistics of response spectrum
prediction equations developed by Abrahamson and Silva [27], Boore and Atkinson [28], Campbell
and Bozorgnia [26], and Chiou and Youngs [29], which are all based on various subsets of the
NGA database. Figure 10 compares the median and median ± one logarithmic standard deviation
values of the two sets of response spectra for periods up to 5 s for selected moment magnitude
and source-to-site distance values. Strike-slip faulting and V = 760 m/s are selected in all cases.
Also shown as dashed lines are the averages of the four selected NGA prediction equations.
It can be seen that, except for the case of M = 6.0, both the median curves and the dispersions of
the synthetic response spectra are in close agreement with the corresponding statistics of the four
NGA-based prediction equations. The case of M = 6.0 coincides with the lower boundary of our
database, where few records are available (see Figure 3) and the model fit is not as good. In any
case, this magnitude level is not of interest for design against the MCE. For all other magnitudes
and for all distances, the observed deviations from the prediction equations are much smaller
than the variability present in the prediction equations themselves. Thus, we conclude that the
method presented in this study for generating synthetic ground motions for given earthquake and
site characteristics is viable and consistent with the existing prediction equations for source-to-site
distances R10 km and moment magnitudes greater than about M = 6.5.
DOI: 10.1002/eqe
100 Campbell-Bozorgnia 100 Simulation

Abrahamson-Silva Average NGA
Chiou-Youngs
Boore-Atkinson
10-1 10-1
10-2 10-2
M=6.0, R=20km M=7.0, R=10km
10-3 10-3
10-1 100 5×100 10-1 100 5×100
Pseudo-Acceleration Response Spectrum, g
100 100
10-1 10-1
10-2 10-2
M=7.0, R=20km M=7.0, R=20km
10-3 10-3
10-1 100 5×100 10-1 100 5×100
100 100
10-1 10-1
10-2 10-2
M=8.0, R=20km M=7.0, R=40km
10-3 10-3
10-1 100 5×100 10-1 100 5×100
(a) Period, sec (b) Period, sec
Figure 10. Median and median ± one logarithmic standard deviation of 5% damped
pseudo-acceleration response spectra of 500 synthetic motions and corresponding values predicted
by the average of four NGA-based prediction models for selected moment magnitudes, source-to-site
distances, F = 0 (Strike-slip faulting), and V = 760 m/s. Estimated NGA values are based on a
rupture width of 20 km, depth to top of rupture of 1 km, Z 2.5 = 1 km for Campbell and Bozorgnia
[26], Z 1.0 = 0.034 km for Abrahamson and Silva [27], and Z 1.0 = 0.024 km for Chiou and Youngs
[29]: (a) constant R, varying M and (b) constant M, varying R.
DOI: 10.1002/eqe
9. CONCLUSION
A new method for generating a suite of synthetic ground motions for a given set of earthquake and
site characteristics is presented. Prediction equations are developed that relate the parameters of a
stochastic model to the earthquake fault mechanism, moment magnitude, source-to-site distance,
and the site shear-wave velocity. By taking into account the uncertainty of model parameters, the
simulated suite of ground motions captures the natural variability of real recorded ground motions.
The method employs a previously developed stochastic model that is based on a modu-
lated, filtered white-noise process and incorporates both temporal and spectral nonstationarities.
The model is defined in terms of a set of parameters that characterize the evolving intensity,
predominant frequency, and bandwidth of the ground acceleration process. The model parameters
are assigned probability distributions based on empirical data obtained from fitting the stochastic
model to a subset of the NGA strong motion database. Using the assigned probability distribu-
tions, model parameters are transformed to the standard normal space to satisfy the normality
requirement of regression analysis. Because the database contains different numbers of recordings
from different earthquakes, a random-effects regression analysis method is employed to separately
account for the intra- and inter-event uncertainties. The maximum likelihood method is used to
estimate the regression coefficients and the error variances, resulting in an empirical prediction
equation for each transformed model parameter. Correlation analysis is performed to determine the
correlations among the transformed model parameters. Finally, the empirical prediction equations
and correlations are used to randomly generate sets of realizations of the model parameters, each
set providing one realization of a possible ground motion for the specified earthquake and site
characteristics.
Examples of synthetic acceleration, velocity, and displacement time-histories are compared with
real recorded motions, indicating similar characteristics and variability. Comparison of synthetic
with real elastic response spectra for a specific earthquake shows that the spectra of recorded
motions are well within the range of variability of the spectra of synthetic motions. Finally,
comparisons are made between median and median ± one logarithmic standard deviation elastic
response spectra of synthetic motions with similar statistics of NGA-based prediction models for
selected magnitude and distance values, thus indicating the range of validity of the proposed
prediction models.
ACKNOWLEDGEMENTS
This work is supported by the State of California through the Transportation Systems Research Program
of the Pacific Earthquake Engineering Research Center (PEER). Any opinions, findings, and conclusions
or recommendations expressed in this material are those of the authors and do not necessarily reflect those
of the funding agency. Special thanks to Yousef Bozorgnia for valuable assistance and advice during the
course of this study.
REFERENCES
1. Watson-Lamprey JA. Selection and scaling of ground motion time series. Ph.D. Thesis (in Engineering—Civil
and Environmental Engineering), University of California, Berkeley, CA, 2007.
2. Bommer JJ, Acevedo AB. The use of real earthquake accelerograms as input to dynamic analysis. Journal of
Earthquake Engineering 2004; 8:43–91.
DOI: 10.1002/eqe
3. Hancock J, Watson-Lamprey JA, Abrahamson NA, Bommer JJ, Markatis A, McCoy E, Mendis R. An improved
method of matching response spectra of recorded earthquake ground motion using wavelets. Journal of Earthquake
Engineering 2006; 10:67–89.
4. Naeim F, Lew M. On the use of design spectrum compatible time histories. Earthquake Spectra 1995; 11:111–127.
5. Luco N, Bazzurro P. Does amplitude scaling of ground motion records result in biased nonlinear structural drift
responses? Earthquake Engineering and Structural Dynamics 2007; 36:1813–1835.
6. Douglas J, Aochi H. A survey of techniques for predicting earthquake ground motions for engineering purposes.
Surveys in Geophysics 2008; 29:187–220.
7. Stafford PJ, Sgobba S, Marano GC. An energy-based envelope function for the stochastic simulation of earthquake
accelerograms. Soil Dynamics and Earthquake Engineering 2009; 29:1123–1133.
8. Rezaeian S, Der Kiureghian A. A stochastic ground motion model with separable temporal and spectral
nonstationarity. Earthquake Engineering and Structural Dynamics 2008; 37:1565–1584.
9. Pousse G, Bonilla LF, Cotton F, Margerin L. Nonstationary stochastic simulation of strong ground motion time
histories including natural variability: application to the K-net Japanese database. Bulletin of the Seismological
Society of America 2006; 96:2103–2117.
10. Sabetta F, Pugliese A. Estimation of response spectra and simulation of nonstationary earthquake ground motions.
Bulletin of the Seismological Society of America 1996; 86:337–352.
11. Alamilla L, Esteva L, Garcia-Perez J, Diaz-Lopez O. Evolutionary properties of stochastic models of earthquake
accelerograms: their dependence on magnitude and distance. Journal of Seismology 2001; 5:1–21.
12. Yeh CH, Wen YK. Modeling of nonstationary earthquake ground motion and biaxial and torsional response of
inelastic structures. Civil Engineering Studies, Structural Research Series Report No. 546, University of Illinois,
1989.
13. Liu P, Archuleta RJ, Hartzell SH. Prediction of broadband ground-motion time histories: hybrid low/high-
frequency method with correlated random source parameters. Bulletin of the Seismological Society of America
2006; 96:2118–2130.
14. Hutchings L, Ioannidou E, Foxall W, Voulgaris N, Savy J, Kalogeras I, Scognamiglio L, Stavrakakis G.
A physically based strong ground-motion prediction methodology; application to PSHA and the 1999 Mw = 6.0
Athens earthquake. Geophysical Journal International 2007; 168:659–680.
15. Causse M, Cotton F, Cornou C, Bard PY. Calibrating median and uncertainty estimates for a practical use of
empirical Green’s funtions technique. Bulletin of the Seismological Society of America 2008; 98:344–353.
16. Ameri G, Gallovic F, Pacor F, Emolo A. Uncertainties in strong ground-motion prediction with finite-fault
synthetic seismograms: an application to the 1984 M 5.7 Gubbio, central Italy, earthquake. Bulletin of the
Seismological Society of America 2009; 99:647–663.
17. Frankel A. A constant stress-drop model for producing broadband synthetic seismograms: comparison with the
Next Generation Attenuation relations. Bulletin of the Seismological Society of America 2009; 99:664–680.
18. Abrahamson N, Atkinson G, Boore D, Bozorgnia Y, Campbell K, Chiou B, Idriss IM, Silva W, Youngs R.
Comparisons of the NGA ground-motion relations. Earthquake Spectra 2008; 24:45–66.
19. Rezaeian S. Stochastic modeling and simulation of ground motions for performance-based earthquake engineering.
Ph.D. Thesis (in Engineering—Civil and Environmental Engineering), University of California, Berkeley, CA,
2010.
20. Brune JN. Tectonic stress and the spectra of seismic shear waves from earthquakes. Journal of Geophysical
Research 1970; 75:4997–5009.
21. Brune JN. Correction. Journal of Geophysical Research 1971; 76:5002.
22. Arias A. A measure of earthquake intensity. In Seismic Design for Nuclear Power Plants, Hansen RJ (ed.). MIT
Press: Cambridge, MA, 1970; 438–483.
23. Trifunac MD, Brady AG. A study on the duration of strong earthquake ground motion. Bulletin of the Seismological
24. Ang A, Tang W. Probability Concepts in Engineering: Emphasis on Applications to Civil and Environmental
Engineering (2nd edn). Wiley: Hoboken, NJ, 2006.
25. Lutes LD, Sarkani S. Random Vibrations: Analysis of Structural and Mechanical Systems. Elsevier, Butterworth-
Heinemann: Burlington, MA, 2004.
26. Campbell KW, Bozorgnia Y. NGA ground motion model for the geometric mean horizontal component of PGA,
PGV, PGD and 5% damped linear elastic response spectra for periods ranging from 0.01 to 10 s. Earthquake
Spectra 2008; 24:139–171.
27. Abrahamson N, Silva W. Summary of the Abrahamson & Silva NGA ground-motion relations. Earthquake
Spectra 2008; 24:67–97.
DOI: 10.1002/eqe
28. Boore D, Atkinson G. Ground-motion prediction equations for the average horizontal component of PGA, PGV,
and 5%-damped PSA at spectral periods between 0.01 s and 10.0 s. Earthquake Spectra 2008; 24:99–138.
29. Chiou B, Youngs R. An NGA model for the average horizontal component of peak ground motion and response
spectra. Earthquake Spectra 2008; 24:173–215.
30. Idriss IM. An NGA empirical model for estimating the horizontal spectral values generated by shallow crustal
earthquakes. Earthquake Spectra 2008; 24:217–242.
31. Tothong P, Cornell CA, Baker JW. Explicit directivity-pulse inclusion in probabilistic seismic hazard analysis.
Earthquake Spectra 2007; 23:867–891.
32. Abrahamson N, Youngs R. A stable algorithm for regression analyses using the random effects model. Bulletin
of the Seismological Society of America 1992; 82:505–510.
33. Brillinger DR, Preisler HK. Further analysis of the Joyner–Boore attenuation data. Bulletin of the Seismological
34. University of Bristol. Residuals, Center for Multilevel Modeling. Available from: http://www.cmm.bristol.ac.
uk/learning-training/videos/RP/Residuals.shtml [10 April 2009].
35. Travasarou T, Bray JD, Abrahamson NA. Empirical attenuation relationship for Arias Intensity. Earthquake
Engineering and Structural Dynamics 2003; 32:1133–1155.
36. Abrahamson NA, Silva WJ. Empirical ground motion models, Section 5: Empirical model for duration of strong
ground motion. Report, Brookhaven National Laboratory, 1996.
37. Anderson TW. An Introduction to Multivariate Statistical Analysis. Wiley: New York, 1958; Theorem 2.5.1.
DOI: 10.1002/eqe

Der - Kiureghian - 2015 - Simulation of Synthetic Ground Motions

Uploaded by

Copyright:

Available Formats

Der - Kiureghian - 2015 - Simulation of Synthetic Ground Motions

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Der - Kiureghian - 2015 - Simulation of Synthetic Ground Motions

Uploaded by

Copyright:

Available Formats

EARTHQUAKE ENGINEERING AND STRUCTURAL DYNAMICS

Earthquake Engng Struct. Dyn. 2010; 39:1155–1180

Simulation of synthetic ground motions for specified earthquake

Sanaz Rezaeian∗, †, ‡ and Armen Der Kiureghian§

Received 10 September 2009; Revised 10 December 2009; Accepted 17 December 2009

∗ Correspondence to: Sanaz Rezaeian, University of California, Berkeley, CA 94720, U.S.A.

Contract/grant sponsor: State of California

Copyright q 2010 John Wiley & Sons, Ltd.

2. STOCHASTIC GROUND MOTION MODEL

2.1. Model formulation

White-noise Unit-variance process with

Fully non-stationary process Simulated ground acceleration

Figure 1. Procedure for simulating a single ground motion.

In this expression, q(t, a) is a deterministic, nonnegative, time-modulating function with parameters

q(t, a) = 1 t 2 −1 exp(−3 t) (2)

2.2. Model parameterization

D5−95 = t95 −t5 (8)

f () = mid + (−tmid ) (11)

3. IDENTIFICATION OF MODEL PARAMETERS FOR A TARGET ACCELEROGRAM

positive-minima and negative-maxima

4. METHODOLOGY FOR DEVELOPING PREDICTIVE EQUATIONS

5. STRONG MOTION DATABASE

6. CONSTRUCTION OF PREDICTIVE EQUATIONS

Table I. Selected earthquakes from the CB-NGA database, type of faulting,

0.5 0.08 0.08

Figure 4. Probability density functions superimposed on observed normalized frequency

6.1. Fitted distributions

Table II. Distribution models assigned to the model parameters.

I¯a (s g) Lognormal (0, ∞) 0.0468 0.164

6.2. Regression analysis

where i = 1, . . . , 6 indexes the model parameters, j = 1, . . . , 19 indexes the earthquakes, and k =

Random-effects modeling is sometimes referred to as variance-components modeling because for

6.3. Regression results

6.4. Correlation analysis

7. SIMULATION OF GROUND MOTIONS FOR A GIVEN DESIGN SCENARIO

8. MODEL VALIDATION THROUGH EXAMINATION OF ELASTIC RESPONSE SPECTRA

8.1. Against real recorded motions

0.2 Simulated 0.0330 8. 83 3.44 3.74 0.05 0.80

0.05 Recorded 0.01 Recorded

0.1 Simulated 0.05 Simulated

0.1 Simulated 0.1 Simulated

-0.2 I D t /2π ' /2 π

-0.5 0.0451 12.62 4.73 3.97 0.08 0.03

Simulated 0.05 Simulated

Ia D5-95 tmid /2π ' /2 π

-0.1 0.0179 20.76 13.62 6.18 0.06 0.12

0.1087 8. 669 8.69 6 .71 0.04 0.25

Deformation Response Spectrum, m

8.2. Against NGA attenuation models

100 Campbell-Bozorgnia 100 Simulation

You might also like

f () = mid + (−tmid ) (11)