Sei sulla pagina 1di 13

Soil Dynamics and Earthquake Engineering 69 (2015) 207–219

Contents lists available at ScienceDirect

Soil Dynamics and Earthquake Engineering


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

Comparison of 1D linear, equivalent-linear, and nonlinear site response


models at six KiK-net validation sites
James Kaklamanos n, Laurie G. Baise, Eric M. Thompson 1, Luis Dorfmann
Department of Civil and Environmental Engineering, Tufts University, 113 Anderson Hall, Medford, MA 02155, USA

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

Article history: Vertical seismometer arrays represent a unique interaction between observed and predicted ground
Received 30 September 2013 motions, and they are especially helpful for validating and comparing site response models. In this study,
Received in revised form we perform comprehensive linear, equivalent-linear, and nonlinear site response analyses of 191 ground
22 September 2014
motions recorded at six validation sites in the Kiban–Kyoshin network (KiK-net) of vertical seismometer
Accepted 19 October 2014
arrays in Japan. These sites, which span a range of geologic conditions, are selected because they meet
the basic assumptions of one-dimensional (1D) wave propagation, and are therefore ideal for validating
Keywords: and calibrating 1D nonlinear soil models. We employ the equivalent-linear site response program
Earthquake ground motion SHAKE, the nonlinear site response program DEEPSOIL, and a nonlinear site response overlay model
Seismic analysis
within the general finite element program Abaqus/Explicit. Using the results from this broad range of
Seismic effects
ground motions, we quantify the uncertainties of the alternative site response models, measure the
Nonlinear soil behavior
Numerical modeling strain levels at which the models break down, and provide general recommendations for performing site
response analyses. Specifically, we find that at peak shear strains from 0.01% to 0.1%, linear site response
models fail to accurately predict short-period ground motions; equivalent-linear and nonlinear models
offer a significant improvement at strains beyond this level, with nonlinear models exhibiting a slight
improvement over equivalent-linear models at strains greater than approximately 0.05%.
& 2014 Elsevier Ltd. All rights reserved.

1. Introduction linear approximation becomes inadequate, and fully nonlinear site


response analyses are needed to accurately predict surface ground
A fundamental step in any seismic hazard analysis is the motions [5–13]. Fully nonlinear site response analyses are per-
quantification of the expected levels of ground motions for formed in the time-domain by integrating the equation of motion
potential earthquakes. Analytical site response analyses are per- in small time steps. Numerous nonlinear site response programs
formed by propagating an input motion through the soil profile in have been developed, although compared to equivalent-linear
order to predict the ground motion at the surface of a site. Site models, their usage in standard engineering practice is relatively
response analyses can also be performed empirically by using limited. Examples include D-MOD2000 [14], DEEPSOIL [15], TESS
seismic site coefficients such as those in the National Earthquake [16], SUMDES [17], OpenSees [18], and NOAHW [7].
Hazard Reduction Program (NEHRP) seismic provisions [1]. For Given a range of model complexities (linear, equivalent-linear,
analytical site response analyses (the focus of this paper), the state and nonlinear) and specific site response codes, engineers would
of the practice in earthquake engineering is to approximate non- benefit from increased insight to (a) the ranges of ground motions
linear soil behavior using equivalent-linear (EQL) site response over which each type of model is accurate, and (b) which site
models, such as SHAKE [2–4] or STRATA [5]. However, once the response codes offer the strongest goodness-of-fit between pre-
shear strains in the soil exceed some critical level, the equivalent- dicted and observed ground motions. Vertical seismometer arrays,
which have both surface and downhole recordings are an excellent
opportunity for validating site response models. This study uses
n the Kiban–Kyoshin (KiK-net) strong-motion network of vertical
Correspondence to: Department of Civil and Mechanical Engineering, Merri-
mack College, 315 Turnpike Street, North Andover, MA 01845, USA. seismic arrays in Japan [19,20]. This data-rich network provides
Tel.: þ 1 978 837 3401; fax: þ 1 978 837 5029. numerous surface–downhole station pairs that have recorded
E-mail addresses: KaklamanosJ@merrrimack.edu (J. Kaklamanos), many earthquakes with varying levels of ground motion. A
Laurie.Baise@tufts.edu (L.G. Baise), ethompson@mail.sdsu.edu (E.M. Thompson), number of studies have used downhole arrays such as KiK-net to
Luis.Dorfmann@tufts.edu (L. Dorfmann).
1
Present affiliation: Department of Geological Sciences, San Diego State
quantify nonlinear soil behavior and site response model uncer-
University, 5500 Campanile Dr, 237 Geology Mathematics and Computer Science tainty. Assimaki et al. [9] developed a framework for quantifying
Building, San Diego, CA 92182. the susceptibility of nonlinear behavior using data from three

http://dx.doi.org/10.1016/j.soildyn.2014.10.016
0267-7261/& 2014 Elsevier Ltd. All rights reserved.
208 J. Kaklamanos et al. / Soil Dynamics and Earthquake Engineering 69 (2015) 207–219

Table 1
Information on KiK-net stations used in this study.

Station name FKSH11 FKSH14 IWTH08 IWTH27 NMRH04 TKCH08

Latitude (deg) 37.1976 37.0233 40.2658 39.0278 43.3953 42.4847


Longitude (deg) 140.3420 140.9736 141.7867 141.5356 145.1264 143.1564
Avg. shear-wave velocity, VS30 (m/s) 240 237 305 670 168 353
NEHRP site class D D D C E D
Depth to bedrock, Zrock (m) 35 61 20 4 185 78
Install depth of downhole seismometer, Zmax (m) 115 147 100 100 216 100
Frequency of fundamental mode, fo (Hz) 1.20 1.20 2.80 6.23 0.40 1.80
Damping ratio for linear site response, ξ (%) 2.86 2.50 2.00 5.00 1.67 2.50
Number of recorded events, nev 39 52 18 25 23 34

downhole array sites in southern California, and concluded that response program SHAKE, the nonlinear site response program
fully nonlinear analyses are necessary when the rock-outcrop DEEPSOIL, and a nonlinear site response overlay model within the
(input) peak ground acceleration (PGA) exceeds 0.2 g for soft general finite element program Abaqus/Explicit, introduced by Kak-
(NEHRP class E) sites. Kwok et al. [10] performed a blind site lamanos et al. [22]. Using the results from this broad range of ground
response prediction test at the Turkey Flat downhole array using motions, we quantify the prediction accuracies of the alternative site
data from the 2004 Mw 6.0 Parkfield, California, earthquake. They response models, and provide recommendations for modeling site
tested five nonlinear site response models, and found that the response in engineering practice.
models underestimated the site response amplifications at high
frequencies, and overestimated the site response amplifications at
frequencies near the fundamental mode of the site [10]. Kim and 2. Data
Hashash [12] assessed KiK-net downhole array data from the 2011
Mw 9.0 Tohoku earthquake, and also found that equivalent-linear This study focuses on the six KiK-net sites detailed in Table 1. The
and nonlinear site response models underpredicted the site response raw site data available from the KiK-net website [23] are the seismic
amplifications at high frequencies, especially at soft sites. They found velocity profiles and geologic profiles with soil/rock descriptions.
that the accuracies of equivalent-linear and nonlinear models were Table 1 includes basic information on the stations: the time-averaged
generally similar, but the predictions deviate when maximum shear shear-wave velocity over the top 30 m of the subsurface VS30, NEHRP
strains exceed  0.3%. Yee et al. [13] studied surface–downhole site class [1], depth to first rock layer Zrock, installed depth of the
ground motions from the 2007 Mw 6.6 Niigata-ken Chuetsu-oki downhole seismometer Zmax, frequency of the fundamental peak of
earthquake, and found that equivalent-linear and nonlinear models the 1D theoretical surface–downhole transfer function fo, damping
offered similar predictions for maximum shear strains up to  0.2%, ratio used in the linear site response analysis ξ [21], and number of
but that predictions deviated at greater strains. recorded events used in the analyses nev. Because we do not have
In prior work [11], we used the KiK-net database to analyze the site-specific estimates of ξ, we choose a value of ξ that provides the
accuracy (bias) and variability (precision) resulting from common site best fit for the weak motions recorded at each site, as explained by
response modeling assumptions, and we identified critical parameters Thompson et al. [21]; at a given site, we assume a constant value of ξ
that significantly contribute to the uncertainty in site response for all layers in the profile. The shear-wave velocity (VS) profiles for
analyses. We performed linear and equivalent-linear site response the six sites are provided in Fig. 1; the profiles were obtained from
analyses at 100 KiK-net sites using 3720 ground motions ranging from surface–downhole logging and are posted on the KiK-net website
weak to strong in amplitude, in particular, with 204 records having [23]. Fig. 2 displays the 1D theoretical amplification spectra H ðf Þ
PGA40.3 g at the ground surface. The present work builds upon the computed using the Thomson–Haskell matrix method [24,25], which
study of Kaklamanos et al. [11] by adding nonlinear analyses and is coded in the program Nrattle included in the Boore [26] suite of
additional equivalent-linear analyses at a subset of 6 of the 100 KiK- ground motion simulation programs. Also shown are the 1D empiri-
net sites originally studied. The six sites were selected using the cal amplification spectra computed from the surface-to-downhole
methodology of Thompson et al. [21], which developed a classification ratios of the recorded weak motions at each site (having PGAo0.1 g).
scheme for downhole arrays that identifies stations where the one- The six selected KiK-net sites span a range of geologic conditions
dimensional (1D) wave propagation assumption is valid. A station's and site classes. Stations FKSH11 and FKSH14 both have similar
classification is a function of (a) its inter-event variability and (b) the values of VS30 (240 m/s and 237 m/s, respectively), and therefore both
similarity between the empirical and 1D theoretical transfer functions fall into the NEHRP-D site category. Both stations are located in
(amplification spectra). Of the 100 KiK-net sites used by Thompson Fukushima prefecture (on east-central Honshu island) and recorded
et al. [21] and Kaklamanos et al. [11], 16 sites fall into the LG category, strong ground motions from the 2011/3/11 Mw 9.0 Tohoku earth-
having Low inter-event variability and Good fit between the empirical quake. Station FKSH11 consists of 35 m of gravel over tuffaceous rock,
and theoretical transfer functions, and therefore are ideal for calibra- and includes a velocity inversion due to a layer of welded tuff at 35–
tion and validation of 1D site response models. By selecting appro- 57 m depth. The shear-wave velocity profile at FKSH14 is slightly less
priate validation sites, confounding errors can be avoided, such as complex, consisting of 61 m of sand (mixed with gravel below 21 m),
calibrating a 1D site response model at locations significantly affected overlying siltstone and sandstone. Stations IWTH08 and IWTH27 are
by three-dimensional (3D) effects [21]. both located in Iwate prefecture (on northeastern Honshu island).
In the present study, we perform linear, equivalent-linear, and Station IWTH08, a NEHRP class D site, consists of 83 m of weathered
nonlinear site response analyses of 191 ground motions recorded at granite over competent granite, and was also studied by Kaklamanos
six selected LG sites. This study focuses on 1D total stress site et al. [27] and Kaklamanos [28]. Due to the low VS values in the upper
response models, which are the most commonly used site response 20 m (VS r280 m/s), we assume that the upper 20 m is composed of
models in engineering practice. We employ the equivalent-linear site residual soil. Station IWTH27 (site class C) is the stiffest site explored
J. Kaklamanos et al. / Soil Dynamics and Earthquake Engineering 69 (2015) 207–219 209

in this study, and consists of 4 m of granular fill overlying tuffaceous PGA–VS30 distribution of the ground-motion records; individual
rock. IWTH27 is an interesting site because of its close proximity to histograms of these variables are shown in the margins of the
several major earthquakes, including the 2011 Tohoku event. Of the plots. More details on the set of 191 ground motions are available
191 ground motions used in this study (described in the following in Table S1 in the electronic supplement (Appendix) to this article.
paragraph), the three strongest ground motions in the dataset
occurred at IWTH27. Each of these three records have PGA40.75 g;
therefore, the shallow fill layer at IWTH27 may exhibit significant 3. Methods
nonlinear response. Station NMRH04 (site class E), located in Nemuro
subprefecture on eastern Hokkaido island, is the softest and deepest 3.1. Linear and equivalent-linear site response models
site considered in this study. The site consists of a deep 185 m layer
of Quaternary soil (sand interbedded with sandy gravel) over The simplest linear and equivalent-linear site response models
interbedded layers of sandstone and siltstone. Station TKCH08 (site involve calculations in the frequency-domain; in this study, the
class D; near the C–D boundary) is located in Tokachi subprefecture program SHAKE was used for all frequency-domain calculations.
on southern Hokkaido island, and consists of 78 m of Quaternary These models use the properties of the soil profile (density ρ,
sandy gravel over Cretaceous sandstone. shear-wave velocity VS, and damping ratio ξ) and the input motion
The ground motions used in this study represent the catalog of
recordings at the six stations from 2000 (when the first of these
stations went online) through mid-2011. Fig. 3 illustrates a map of
the six stations and 154 earthquake epicenters used in this study,
which represent a wide range of sources and paths. Fig. 4 gives the

Fig. 1. Shear-wave velocity profiles at the six KiK-net stations, arranged in order of Fig. 3. Map of the six KiK-net stations and 154 earthquake epicenters used in
increasing VS30. this study.

Fig. 2. Theoretical and empirical 1D linear amplification spectra (surface–downhole transfer functions) for the six KiK-net stations, arranged in order of increasing VS30. The
median and 95% confidence intervals of the empirical transfer functions are shown, as determined from the recorded weak (PGA o 0.1 g) motions at each site.
210 J. Kaklamanos et al. / Soil Dynamics and Earthquake Engineering 69 (2015) 207–219

Fig. 4. Distributions of the geometric mean PGA versus VS30 for the 191 ground-
motion records used in this study. Histograms of the individual variables are shown
in the margins of the plots.

applied at the base of the profile (obtained from the downhole


recording at the KiK-net site) to predict the ground motion at the
surface. Linear site response analyses assume a strain-constant
shear modulus Gmax ¼ ρV 2S based on assumed small-strain soil
properties, and constant damping ratio ξ. Linear site response
analyses can also be performed in the time domain; we performed
Fig. 5. Examples of equivalent-linear (a) modulus-reduction and (b) damping
linear time-domain site response analyses using DEEPSOIL and curves from Zhang et al. [29] and Darendeli [30] for the bottom of the 4-meter-
Abaqus to serve as the baseline for the fully nonlinear site thick surficial layer at site NMRH04, having VS ¼ 100 m/s and mean effective
response analyses performed using these programs (described in confining pressure (σ m ) ¼ 76 kPa, The curves representing a linear site response
the next section). In the equivalent-linear formulation, the values analysis are also shown (ξ ¼1.67% for NMRH04).
of G and ξ are iteratively adjusted to be consistent with the
effective level of shear strain in each soil layer; the converged [33]. In DEEPSOIL, the MRDF pressure-dependent hyperbolic
values are then used in the site response calculation. A number of model procedure [15] is used to obtain the fitted nonlinear curves
modulus-reduction and damping relationships are available for from the Zhang et al. [29] modulus-reduction and damping curves.
use in engineering practice; in this study, we apply the Zhang et al. This procedure includes a reduction factor developed by Phillips
[29] and Darendeli [30] models to compare the effect of the and Hashash [34] that modifies the extended Masing hysteretic
modulus-reduction and damping relationship on the performance behavior to match modulus-reduction and damping curves simul-
of an equivalent-linear method. In [11], we also used the Zhang taneously over a wide range of shear strains.
et al. [29] modulus-reduction and damping relationships, which The overlay model of Kaklamanos et al. [22] is a simple
require the mean effective confining pressure σ m , plasticity index methodology for modeling earthquake site response within a
PI, and geologic age (Quaternary, Tertiary, etc.) as input para- general finite element model (here, Abaqus/Explicit), and allows
meters. The Darendeli [30] relationships require σ m , PI, over- for a representation of any backbone stress–strain curve, along
consolidation ratio, loading frequency, and number of loading with hysteretic unloading–reloading (extended Masing) behavior
cycles. As an example, comparisons of the two models are [35]. To represent overlay elements in a finite element model, the
illustrated in Fig. 5 for the bottom of the 4-meter-thick surficial user defines a number of elements and assigns each of them
layer at NMRH04. Details on the selection of the input parameters identical node numbers and different shear moduli and yield
for the Zhang et al. [29] and Darendeli [30] models are available in stresses [22,36]. In this study, the backbone curves corresponding
the Appendix to this article. to the Zhang et al. [29] modulus reduction curves are used to
derive the shear moduli and yield stresses, using the equations of
3.2. Nonlinear site response models Kaklamanos et al. [22]. For each node, we use N ¼20 overlay
elements. The associated constitutive model is that of Iwan [37]
When performing fully nonlinear analyses in the time domain, and Mroz [38], which represents stress–strain behavior with a set
G and ξ vary throughout the duration of loading. In this study, we of elastoplastic springs connected in parallel. The overlay model
compare the linear and equivalent-linear model predictions with for site response requires basic geotechnical data, can be imple-
two nonlinear 1D site response models: (1) the program DEEPSOIL mented in nearly any existing finite element code, and can be
[15]; and (2) a nonlinear overlay model introduced by Kaklamanos easily adapted to model more complex behavior (such as cyclic
et al. [22] within the finite element program Abaqus/Explicit [31]. hardening and softening, and 3D site response [39]), without the
In this study, the Zhang et al. [29] modulus-reduction and damp- specification of any complicated constitutive models.
ing curves are used as the target curves for the two nonlinear
programs. In both programs, a rigid boundary condition is 3.3. Model validation methods
employed at the base of the model because the input motion is
obtained from a downhole recording [8,28]. To quantify the goodness-of-fit of the site response models, we
The constitutive model in DEEPSOIL is an extended version of compare the observed surface response spectra, PSAobs ðTÞ, to the
the MKZ model developed by Matasović and Vucetic [32], which predicted surface response spectra from the site response mod-
modified the original KZ model developed by Kondner and Zelasko el, PSApred ðTÞ, where PSA is the 5%-damped pseudo-acceleration
J. Kaklamanos et al. / Soil Dynamics and Earthquake Engineering 69 (2015) 207–219 211

response spectra as a function of spectral period T. We compute bias at the shortest periods. At periods greater than 0.1 s, the fixed
the residual between the observed and predicted PSA values in effect becomes more erratic, especially at the periods correspond-
natural logarithmic space as ing to the fundamental peaks of the individual sites' 1D theoretical
  transfer functions. These periods are marked by vertical lines in
PSAresid ðT Þ ¼ ln½PSAobs ðT Þ  ln PSApred ðT Þ ; ð1Þ
Fig. 6, and they are usually characterized by troughs in the plot of
where the geometric mean is used to combine the two orthogonal model bias, indicating that the models are overpredicting the
horizontal components of recorded ground motion. Positive resi- response at the fundamental modes (also observed by Kwok et al.
duals indicate underpredictions, and negative residuals indicate [10]). In general, the 6-site results for the fixed effect are similar to
overpredictions. those of the Zhang et al. [29] equivalent-linear model tested in
In order to obtain statistically significant inferences about site Kaklamanos et al. [11] at 100 sites. The largest difference is that
response, we must account for the dependence between multiple the Kaklamanos et al. [11] results have smoother trends; the
recordings at a single site (i.e. repeatable site effects). Mixed- individual site influences (at specific periods) tend to cancel each
effects regression [40] is a statistical procedure that allows for the other out when such a large number of sites are considered. In the
estimation of the repeatable biases and variances when the data aggregate, as observed in Fig. 6(a), most site response models tend
are grouped according to one or more classification factors; here, to underpredict short-period ground motions, perhaps due to
the data are grouped by site. The mixed-effects regression model breakdowns of the model assumptions, such as the assumed soil
incorporates both fixed effects (parameters associated with an parameters, viscous damping, boundary conditions, and limita-
entire population or with certain repeatable levels of classification tions of the 1D model [11]. Similar results were observed by Kwok
factors, e.g., sites), and random effects (parameters associated with et al. [10] and Kim and Hashash [12], particularly for soft sites.
individual units drawn at random from the population). At a given Fig. 6(b–d) characterize the variability of the model residuals.
spectral period T, the residuals are modeled by the mixed-effects The differences in the standard deviations between the linear,
regression equation equivalent-linear, and nonlinear site response models are not
significant. All models exhibit increased variability at the spectral
PSAresid ðT Þi;j ¼ a þ ηSi þ ϵi;j ; ð2Þ
periods corresponding to the fundamental peaks of the 1D theo-
where a is the population mean of PSAresid ðT Þ (i.e., the fixed effect), retical transfer functions at the six sites. Again, the Kaklamanos
which represents the average bias in the site response model et al. [11] 100-site trends are smoother than the 6-site trends of
across all sites and ground motions; ηSi is the inter-site residual this study. The intra-site standard deviations σ 0 are generally
(i.e. the between-site residual), which gives the deviation from the within the range of 0.2–0.3 natural log units, and are less erratic
population mean of the mean residual for the ith site; and ϵi;j is the than the inter-site standard deviations τS . The intra-site standard
intra-site residual (i.e. the within-site residual), which represents deviations for the 6-site results are not significantly different than
the deviation for ground-motion observation j at site i from the the intra-site standard deviations for the 100-site results of
mean residual at site i. In other words, the intra-site residual ϵi;j is Kaklamanos et al. [11]. Given that σ 0 represents the inherent
the residual after accounting for all repeatable effects: The inter- ground-motion variability once all repeatable effects are removed,
and intra-site residuals are normally distributed zero-mean ran- we would not expect a significant difference between the two
dom variables with standard deviations τS and σ 0 , respectively, to datasets. However, the inter-site standard deviations and the total
be determined by the regression. The three unknown parameters standard deviations are noticeably smaller for the 6-site dataset.
of the linear mixed-effects regression model in Eq. (2) are the fixed Recall that the six selected sites in this study are classified as LG by
effect a, inter-site standard deviation τS , and intra-site standard Thompson et al. [21], and therefore it is not surprising that values
deviation σ 0 (which represents the inherent ground-motion varia- of τS (and thus σ Y ) in this study are smaller than those in the
bility; i.e. the variability after all repeatable effects have been broader dataset. Although only six sites have been considered in
removed). The site response model residuals Y ¼ PSAresid ðT Þ have a this analysis, there are many recordings per site (Table 1), and so
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi the usage of the inter-site residuals is reasonable. The limited
total standard deviation given by σ Y ¼ σ 20 þ τ2S . The mixed-effects
number of sites might, however, be an issue if we were making
regression parameters may be used to examine the bias and conclusions about trends with VS30, because there are a small
precision of site response models [11,41]. number of sites for a given range of VS30 values. Therefore, we
avoid drawing conclusions about trends in model residuals versus
VS30.
4. Results
4.2. Analysis of intra-site model residuals: all sites
4.1. Bias and variability as a function of spectral period
Kaklamanos et al. [11] found that the maximum shear strain in
The period dependence of the bias and variability of the site the soil profile (γmax) is the best indicator of model accuracy, and
response models is analyzed in this section. In Fig. 6, we plot the that the observed peak ground acceleration at the ground surface
parameters of the linear mixed-effects regression model versus could be used as a proxy for strain. Therefore, we plot the model
spectral period (T), following Eq. (2) and the notation of Kaklamanos residuals against γmax, which was found to have the strongest
et al. [11]: (a) fixed effect a, (b) total standard deviation σ Y , (c) intra- residual trends. In Fig. 7, we display plots of the 191 intra-site
site standard deviation σ 0 , and (d) inter-site standard deviation τS . In residuals (ϵi;j ) for PSA at a spectral period of T ¼0.1 s, versus γmax
addition to the results from the site response models tested at the six calculated from the site response analyses. The results are shown
sites in this study, the equivalent-linear results from the 100-site for each of the 191 ground motions using seven different site
Kaklamanos et al. [11] study are also plotted for comparison. response models: linear analyses in (a) SHAKE, (b) DEEPSOIL, and
The fixed effects plotted in Fig. 6(a) allow us to characterize the (c) Abaqus; equivalent-linear analyses in SHAKE using the
model biases at different spectral periods. Fig. 6(a) indicates that modulus-reduction and damping curves of (d) Zhang et al. [29]
all site response models generally have positive bias (underpredic- and (e) Darendeli [30]; and nonlinear analyses in (f) DEEPSOIL and
tion of ground motions) at spectral periods less than 0.2 s. (g) Abaqus/Explicit.
However, the nonlinear site response model in Abaqus shows less As in Kaklamanos et al. [11], the linear site response models
bias than the other models, and this model notably has little to no (Fig. 7(a–c)) are strongly biased at large strains: the residuals
212 J. Kaklamanos et al. / Soil Dynamics and Earthquake Engineering 69 (2015) 207–219

Fig. 6. Period dependence of the parameters of the linear mixed effects regression model of Eq. (2): (a) fixed effect a, (b) total standard deviation σ Y , (c) intra-site standard
deviation σ 0 , and (d) inter-site standard deviation τS . In addition to the results from the different site response models tested at the six sites in this study, the equivalent-
linear results from the 100-site Kaklamanos et al. [11] study is also plotted for comparison. A vertical line is drawn at the period corresponding to the fundamental peak of
the 1D theoretical transfer function for each site (from left to right: IWTH27, IWTH08, TKCH08, FKSH11/FKSH14 [fundamental periods are equal], and NMRH04).

display a strong downward slope. Because linear models do not increase in the variability of the residuals for values of γmax greater
account for the deamplifications associated with nonlinear beha- than approximately 0.05%. The increased variability in the resi-
vior, they characteristically overpredict large-strain motions. The duals represents a greater uncertainty associated with predicting
linear site response residuals begin to deviate from zero at large-strain ground motions.
approximately γmax ¼0.01%, consistent with the prior results for In addition to T ¼0.1 s, which is represented in Fig. 7, analogous
short periods. plots for PSA at additional spectral periods are displayed as Figs.
Fig. 7(d–g) illustrate that the equivalent-linear and nonlinear S1–S8 in the electronic supplement to this article. These plots
site response models generally offer more accurate predictions of show that the Darendeli [30] equivalent-linear model is extremely
large-strain ground motions. The equivalent-linear and nonlinear biased at large strains for spectral periods between 0.1 and 0.3 s.
site response residuals do not have a systematic bias, with the However, at larger spectral periods (beginning at 0.5–1.0 s), the
exception of the Darendeli [30] modulus-reduction and damping linear, equivalent-linear, and nonlinear site response models do
curves. The Darendeli [30] model residuals display an upward not display significant biases at large strains. This finding is
slope for strains greater than 0.05%, meaning that the model is consistent with Kaklamanos et al. [11], in which we found that
underpredicting the large-strain ground motions. The plots of the site response residuals at spectral periods greater than 0.5 s do not
damping curves in Fig. 5 provide some insight into the Darendeli systematically display noticeable effects of nonlinear soil behavior.
[30] model: for a given level of shear strain40.01%, the Darendeli
damping ratio is generally larger than the Zhang et al. [29] model 4.3. Analysis of intra-site model residuals: separation by site
(except at very large strains not encountered in this study). When
compared to the observed ground motions in this study, the We now develop more site-specific conclusions by studying the
Darendeli [30] model seems to overdamp—and hence underpre- residual trends at each of the six sites individually. Fig. 8 displays
dict—ground motions for this level of shear strain. On the other the intra-site residuals (ϵi;j ) versus γmax for PSA at T¼ 0.1 s for
hand, at least for the levels of shear strain experienced in this the six sites, ordered by increasing VS30. Each panel displays the
study (up to 0.3%), the Zhang et al. [29] modulus-reduction and residuals for the linear site response model (from SHAKE),
damping curves do not seem to be associated with any systematic the equivalent-linear site response model (from SHAKE, using
bias. Therefore, the Zhang et al. [29] modulus reduction curves the Zhang et al. [29] model), and the nonlinear site response
were used to derive the backbone curves and material parameters model (from Abaqus).
for the nonlinear site response analyses in this study. In panels (d), In general, the differences between the equivalent-linear and
(f), and (g) of Fig. 7, no significant differences between the nonlinear site response models are small with respect to the
equivalent-linear [29] residuals and the nonlinear residuals are differences between these models and the linear model. However,
noticed. Additional site- and record-specific plots (to be presented for the softest site (NMRH04) and stiffest site (IWTH27) explored
shortly) are necessary in order to decipher the differences in this study, the equivalent-linear and nonlinear residuals slope
between equivalent-linear and nonlinear site response models. upward for large-strain events, indicating an underprediction of
Although these residuals generally trend near zero, there is a slight ground motion. As summarized in the VS profile in Fig. 1 and
J. Kaklamanos et al. / Soil Dynamics and Earthquake Engineering 69 (2015) 207–219 213

Fig. 7. Plots of the intra-site residuals (ϵi;j ) for PSA at a spectral period of T ¼ 0.1 s, versus the maximum calculated shear strain in the soil profile (γmax). The results are shown
for each of the 191 ground motions using seven different site response models: linear analyses in (a) SHAKE, (b) DEEPSOIL, and (c) Abaqus; equivalent-linear analyses in
SHAKE using the modulus-reduction and damping curves of (d) Zhang et al. [29] and (e) Darendeli [30]; and nonlinear analyses in (f) DEEPSOIL and (g) Abaqus using
backbone curves derived from the Zhang et al. [29] model. Comparable plots for additional spectral periods (besides T ¼ 0.1 s) are presented as Figs. S1–S8 in the electronic
supplement to this article.

explained in the Section 2, site IWTH27 consists of 4 m of granular display positive trends for the largest-strain events at intermediate
fill (VS ¼ 150 m/s) overlying tuffaceous rock (VS Z1100 m/s). Given sites FKSH11 and TKCH08, in addition to NMRH04 (as noted
the sharp impedance contrast, therefore virtually all of the non- previously). At T¼ 0.2 s, the various model residuals at IWTH08
linear soil behavior must occur in the top 4 m. The effects of the and IWTH27 have leveled off, although the linear, equivalent-
resonances of trapped seismic waves within this shallow layer linear, and nonlinear model residuals at FKSH14 still display a
may also be difficult to predict; similar results were observed for strong downward trend. For longer spectral periods (up to 0.5 s),
the KiK-net shallow soil site MYGH11 studied in Baise et al. [42]. the model residuals for all sites are more or less constant with
Large amounts of nonlinearity in the soft surficial layers overlying γmax, although there is a slight downward trend remaining at
KiK-net rock sites have been also been documented by Assimaki FKSH14. The residual patterns at FKSH14 suggest that a greater-
et al. [43], Ghofrani et al. [44], and Régnier et al. [45]. than-expected amount of nonlinearity is occurring at this site,
To extend the site-specific results beyond T ¼0.1 s (as shown in perhaps due to cyclic mobility in the loose saturated sand layer
Fig. 8), analogous plots for additional spectral periods are dis- near the ground surface.
played as Figs. S9–S16 in the electronic supplement to this article.
The trends for PGA (Fig. S9) are similar to those for PSA at T¼ 0.1 s 4.4. Detailed study of nonlinear ground motions
(Figs. 8 and S10). For PSA at T ¼0.2 s (Fig. S12), the linear residuals
do not display as strong of a negative trend as they do for shorter Further site-specific results are available in Figs. S17–S31 in the
spectral periods, but the equivalent-linear and nonlinear residuals electronic supplement to this article, where detailed ground-motion
214 J. Kaklamanos et al. / Soil Dynamics and Earthquake Engineering 69 (2015) 207–219

Fig. 8. Plots of the intra-site residuals (ϵi;j ) for PSA at a spectral period of T ¼0.1 s, versus the maximum calculated shear strain in the soil profile (γmax). The residuals for the
six sites are displayed in panels (a)–(f), respectively (ordered by increasing VS30). Each panel displays the residuals for the linear site response model (SHAKE), the equivalent-
linear site response model (SHAKE, using the Zhang et al. [29] modulus-reduction and damping curves), and the nonlinear site response model (Abaqus/Explicit).
Comparable plots for additional spectral periods (besides T ¼ 0.1 s) are presented as Figs. S9–S16 in the electronic supplement to this article.

plots are presented for the 15 strongest ground motions consid- 1.2 Hz (or 0.83 s). The equivalent-linear and nonlinear site
ered in this study, as ranked by γmax in Table S1. The figures response analyses move the predicted peak downward closer to
include acceleration time series, stress–strain curves, response the observation, although these models are still overpredicting the
spectra, and amplification spectra from the various models. All six amplification. This overprediction could suggest interference with
sites are represented in this set of 15 figures, and each record has the downgoing wave (such as scattering due to subsurface
γ max Z 0:05% for both nonlinear models. Due to space limitations, heterogeneities), or perhaps an error in the KiK-net VS profile
we discuss only one strong ground motion in detail here (Fig. 9): (such as the inversion of the velocity profile for depths 35–57 m).
the strongest event at station FKSH11, which is the 2011/3/11 Mw Nevertheless, the plots indicate that the equivalent-linear and
9.0 Tohoku earthquake (event no. 15 in Table S1; PGA ¼0.502 g, nonlinear models generally offer similar predictions. At high
γmax ¼0.294%). In Fig. 9, we present the (a) surface acceleration frequencies, however, as seen in panel (c), the equivalent-linear
time series, (b) stress–strain curves at the depth of maximum model notably underpredicts the ground motions for frequencies
shear strain, (c) surface/downhole amplification ratios, (d) surface greater than 8 Hz; the overdamping (underprediction) of high-
response spectra, and (e) surface response spectra residuals. frequency components of ground motion is a well-known problem
The different panels of Fig. 9 allow us to compare the model with equivalent-linear models that were also apparent in the
predictions for the Tohoku recording at station FKSH11. First, as results of Kaklamanos et al. [11]. This set of figures illustrates the
seen in panels (c)–(e), the linear site response model most greatly importance of considering multiple factors when analyzing site
overpredicts the amplification at the fundamental mode near response models (e.g., response spectra, amplification spectra),
J. Kaklamanos et al. / Soil Dynamics and Earthquake Engineering 69 (2015) 207–219 215

Fig. 9. Detailed results figure for the 2011/3/11 Mw 9.0 Tohoku event recorded at station FKSH11. Results are displayed for the linear site response model (SHAKE), the
equivalent-linear site response model (SHAKE, using the Zhang et al. [29] modulus-reduction and damping curves), and the nonlinear site response model (Abaqus/Explicit):
(a) observed and predicted surface acceleration time series centered in a 5 s window around the time of the absolute maximum acceleration (north–south component);
(b) predicted stress–strain loops at 31.5 m, the depth corresponding to the maximum shear strain in the profile (north–south component); (c) observed and predicted
surface/downhole amplification ratios; (d) observed and predicted surface response spectra (PSA); and (e) the corresponding PSA residuals (PSAresid). Panels (c)–(e) use the
geometric mean to combine the two orthogonal horizontal components of recorded ground motion. Comparable plots for additional strong-motion events are presented as
Figs. S17–S31 in the electronic supplement to this article.

and also the importance of considering the period dependence of models, computed by pooling the amplification spectra across all
the models' accuracies. 6 sites and 191 ground motions to arrive at a single value of r for
each model. The correlation coefficients are also computed using
4.5. Prediction accuracies of site response models subsets of the pooled amplification spectra corresponding to
records with γmax exceeding 0.01%, 0.02%, 0.05%, and 0.1%. The
In this section, we quantify the prediction accuracies of the trends in Table 3 are graphically displayed in Fig. 10. When all 191
linear, equivalent-linear, and nonlinear site response models ground motions are considered, all seven models have values of r
across all ground motions and sites. First, similar to Thompson in the 0.5–0.6 range. At first, it may seem surprising that the
et al. [21], Pearson's correlation coefficient r is used to compare the benefits of the equivalent-linear and nonlinear models are not
observed and predicted surface/downhole amplification spectra. apparent, but this is because goodness-of-fit calculations in this
For each site, r is calculated using the pooled amplification spectra table are performed for all ground motions at each site, including
from all events, for n ¼200 logarithmically spaced frequencies small-strain motions. In time-domain analyses, one issue is that
between the first and fourth peak of the site's theoretical linear 1D small-strain damping is often poorly constrained (for example,
transfer function (Fig. 2), which is the range of frequencies that when Rayleigh damping is used), although the novel frequency-
will likely dominate the seismic response at each site. Frequencies independent small-strain damping formulation in DEEPSOIL elim-
above 20 Hz are not included in the calculation of r, even if the inates the need to select Rayleigh damping coefficients [34].
fourth peak is at a greater frequency (this occurs at station Nonlinear analyses are also affected by the fitting procedure of
IWTH27). The frequencies used in the calculations at each site the nonlinear curves from the target modulus-reduction and
are shown in Table 2, ordered by VS30. Also shown are the total damping curves.
number of records ntotal, and the number of records at each site The true benefit of equivalent-linear and nonlinear site
surpassing various thresholds of γmax ( Z0.01%, Z0.02%, Z0.05%, response models is observed when the correlation coefficients
and Z0.1%). are computed using only the large-strain ground motions; clear
In Table 3, we present the correlation coefficients between the increases in r are observed when the model type is advanced from
observed and predicted amplification spectra for the site response linear, to equivalent-linear, and to nonlinear. Table 3 and Fig. 10
216 J. Kaklamanos et al. / Soil Dynamics and Earthquake Engineering 69 (2015) 207–219

Table 2
Details of goodness-of-fit calculations at each station.

Station name

NMRH04 FKSH14 FKSH11 IWTH08 TKCH08 IWTH27

Frequencies used in the goodness-of-fit calculations:


Minimum frequency, fmin (Hz) 0.40 1.20 1.20 2.80 1.80 6.23
Maximum frequency, fmax (Hz) 2.34 6.19 6.07 11.59 9.77 20.00
Numbers of ground motions: Total
Total number of records, ntotal 23 52 39 18 34 25 191
Records with γ max Z 0:01% 21 20 15 5 10 9 80
Records with γ max Z 0:02% 10 11 7 2 5 7 42
Records with γ max Z 0:05% 3 4 2 1 1 4 15
Records with γ max Z 0:1% 1 1 2 1 1 3 9

Table 3
Correlation coefficients between observed and predicted amplification spectra: All sites combined.

Model Correlation coefficient, r

All records Records with γ max Z 0:01% Records with γ max Z 0:02% Records with γ max Z 0:05% Records with γ max Z 0:1%

Linear: Abaqus 0.544 0.533 0.534 0.524 0.519


Linear: DEEPSOIL 0.584 0.587 0.576 0.558 0.558
Linear: SHAKE 0.587 0.589 0.575 0.558 0.555
Equivalent-Linear (SHAKE):Darendeli [30] 0.571 0.553 0.521 0.575 0.599
Equivalent-Linear (SHAKE):Zhang et al. [29] 0.583 0.633 0.665 0.771 0.767
Nonlinear: DEEPSOIL 0.585 0.650 0.702 0.817 0.821
Nonlinear: Abaqus 0.545 0.631 0.699 0.831 0.818

For large-strain ground motions, the values of r for the linear


analyses remain in the 0.5–0.6 range, along with the value of r for
the equivalent-linear analyses using the Darendeli [30] curves. For
the two nonlinear models, as well as the equivalent-linear ana-
lyses performed using the Zhang et al. [29] curves, noticeable
improvements in the goodness-of-fit are clear for the γmax Z0.01%
and γmax 40.02% thresholds. However, the most significant bene-
fits of the more advanced models are manifested when γmax
exceeds 0.05%. For records with γmax Z 0.05%, the correlation
coefficient advances to 0.771 for the Zhang et al. [29] equivalent-
linear analyses, and further advances to 0.817 and 0.831 for the
nonlinear site response analyses in DEEPSOIL and Abaqus, respec-
tively. At these large-strain levels, the nonlinear models have a
slightly stronger predictive edge over the equivalent-linear
models.
Fig. 10. Goodness-of-fit between observed and predicted amplification spectra for
the site response models, computed by pooling the amplification spectra across all
sites and 191 ground motions, as well as by pooling the records meeting specific 5. Discussion
thresholds of γmax. The numbers in the legend correspond to the numbers of ground
motions in each category.
5.1. Accuracy as a function of maximum shear strain

In Kaklamanos et al. [11], we concluded that in terms of γmax,


the linear site response model begins to become inaccurate at
display a large difference between the Darendeli [30] and Zhang strains in the range of 0.01–0.1%. At shear strains greater than
et al. [29] models in terms of the equivalent-linear goodness-of-fit. these values, and less than γmax E0.1–0.4%, the equivalent-linear
These results underscore the extreme influence of the modulus- site response formulation improves the accuracy of site response
reduction and damping relationship on the accuracy of equivalent- predictions. In the present study, these conclusions are supported
linear analyses. One possible reason for the underperformance of using the results from additional linear and equivalent-linear site
the Darendeli [30] model is the larger number of input parameters response models, although the Darendeli [30] modulus-reduction
that are not well-constrained (such as overconsolidation ratio, and damping curves have noticeably lower goodness-of-fit than
loading frequency, and equivalent number of cycles). Alternatively, the Zhang et al. [29] curves. Equivalent-linear and nonlinear site
the Zhang et al. [29] model may have a greater accuracy because response models perform similarly across most levels of ground
the model requires one input parameter that the Darendeli [30] motion (as noted, for example, by Kim and Hashash [12] and Yee
model does not contain: geologic age. The Zhang et al. [29] model et al. [13]), but Table 3 and Fig. 10 illustrate that the nonlinear site
specifically accounts for Quaternary soil, which often comprises response models offer slight improvements over equivalent-linear
loose geologic material in which strong nonlinear behavior occurs. analyses for shear strains beyond 0.05%. The maximum shear
J. Kaklamanos et al. / Soil Dynamics and Earthquake Engineering 69 (2015) 207–219 217

strain encountered in the present study is 0.3%; we therefore are findings is that the differences in accuracy are largest between the
precluded from drawing conclusions about larger strain levels. linear model and the other models, and that there are relatively
small differences in accuracy between equivalent-linear and non-
5.2. Damping in nonlinear site response analyses linear site response models. The critical level of maximum shear
strain (γmax) at which the linear site response model breaks down
Nonlinear site response models have two types of damping: is 0.01–0.1% (generally corresponding to PGA E0.1–0.3 g), consis-
(a) hysteretic damping, which is associated with energy dissipa- tent with the results of Kaklamanos et al. [11]; similar trends were
tion within hysteresis loops, and (b) viscous (velocity-propor- observed in linear frequency-domain and linear time-domain site
tional) damping, which is associated with dashpots embedded response models. Beyond shear strains of approximately 0.01% (to
within the material elements. In a nonlinear analysis, there will be maximum strains of 0.3% considered in this study), equivalent-
nearly zero damping at small strains, because the backbone curve linear and nonlinear site response models both offer significant
is nearly linear and there is very little width to the hysteresis loop. improvements over linear site response models, and nonlinear site
Many nonlinear site response codes (e.g., D-MOD2000 [14], response models are shown to exhibit a slight improvement over
DEEPSOIL [15], SUMDES [17], and OpenSees [19]) include viscous equivalent-linear site response models for shear strains greater
damping in order to maintain some level of damping at small than approximately 0.05%. It is important to note, however, that
strains. Although the true nature of soil damping at small strains is the performance of the equivalent-linear model is heavily depen-
not truly viscous, the assumption of viscous damping is a con- dent on the assumed modulus-reduction and damping relation-
venient representation for simulation purposes [8]. The programs ship, as are nonlinear models, which also depend on the assumed
NONLI3 [46,47], NERA [48], NOAHW [7], and the overlay finite target curves. Although fully nonlinear site response models
element model of Kaklamanos et al. [22], which are all based the represent an improvement over linear and equivalent-linear
Iwan [37] and Mroz [38] material model, include only hysteretic models, capturing nonlinear soil behavior is just one step towards
damping. The results of this paper show that the exclusion of improving predictions of complex site response behavior.
viscous damping may mitigate the issue of overdamping at large
strains. Fig. 6 indicates that the overlay model (which includes
only hysteretic damping) is less biased than DEEPSOIL (which Acknowledgments
includes both hysteretic and viscous damping) for short-period
ground motions; Table 3 and Fig. 10 show that this improvement is This research was supported under National Science Foundation
largely manifested at strains greater than 0.05%. Relative to (NSF) Grant no. 1000210; we gratefully acknowledge this support.
DEEPSOIL, the increased accuracy of the overlay model at large We are appreciative of two anonymous reviewers for their insightful
strains comes at a slight cost: decreased accuracy at small strains feedback that improved the quality of this manuscript. We also thank
(Table 3, Fig. 10), where viscous damping has the strongest benefit. the following individuals for providing helpful feedback on an earlier
version of this manuscript: Dr. David M. Boore (U.S. Geological
5.3. Limitations of 1D total stress site response models Survey); and Drs. David M. Garman and Richard M. Vogel of Tufts
University. In addition, we would like to acknowledge the National
The site response models considered in this study all share a Research Institute for Earth Science and Disaster Prevention (NIED)
common trait: they are 1D total stress analyses. Compounding for making the KiK-net strong-motion data and velocity profiles
factors such as basin waves, path effects, soil heterogeneity, freely available.
nonvertical incidence, pore-pressure generation, and poorly con-
strained soil properties can greatly reduce the accuracies of 1D
total-stress site response models [41,49]. Accounting for these Appendix A. Determination of site response model parameters
compounding factors can help improve predictions of site
response, although the complexity of the site response model is In this appendix, we describe how the input parameters for the
increased in doing so. First, there is significant uncertainty Zhang et al. [29] and Darendeli [30] models were selected. The
associated with the assumed 1D velocity profiles at the KiK-net Zhang et al. [29] modulus-reduction and damping relationships
sites. Previous researchers, e.g., Assimaki et al. [50], have chosen to require the mean effective confining pressure σ m , plasticity index
vary the soil profiles from the measured values to improve the fit. PI, and geologic age (Quaternary, Tertiary and older, or residual/
Second, more complicated nonlinear constitutive models could be saprolite soil) as input parameters. The Darendeli [30] relation-
employed, but given the lack of nonlinear material data and pore- ships require σ m , PI, overconsolidation ratio, loading frequency,
pressure data at the KiK-net sites (and many sites in engineering and number of loading cycles. In the equivalent-linear analyses,
practice), we would induce additional uncertainty by using a more the models of Zhang et al. [29] or Darendeli [30] were used to
complicated constitutive model. represent the stress–strain response of the soil at depths above the
soil-bedrock interface (Zrock, given in Table 1). For the bedrock
layers below Zrock, and down to the maximum depth Zmax where
6. Conclusions the downhole ground motion is applied, we assumed linear
stress–strain behavior for all analyses (G/Gmax ¼1, with strain-
In this study, we performed linear, equivalent-linear, and non- independent damping).
linear site response analyses of 191 ground motions recorded at six The layer density ρ required for all site response analyses was
selected sites. Because we focused on sites that are well-modeled determined from the P-wave velocity (VP) using the procedure of
by 1D wave propagation, the observed misfit for strong motions Boore [51]. In order to compute the mean effective confining
can mostly be attributed to uncertainties in soil properties and the pressure σ m , we assumed the groundwater table is located where
constitutive model. Across all sites, ground motions, and intensity VP first surpasses 1500 m/s; numerous studies [52-54] have found
levels, all models generally displayed positive bias (underpredic- that P-wave velocities of 1000–2000 m/s are characteristic of
tion of ground motions) at short spectral periods (o 0.2 s), saturated soil. The coefficient of at-rest lateral earth pressure
although the nonlinear site response analysis in Abaqus displayed (Ko) is required for the computation of the horizontal effective
lesser bias than the other models over this range of periods. From stresses from the vertical effective stresses. We computed Ko from
the analysis of the model residuals, one of the most consistent Poisson's ratio (ν) using the theoretical relationship K o ¼ ν=ð1  νÞ,
218 J. Kaklamanos et al. / Soil Dynamics and Earthquake Engineering 69 (2015) 207–219

where ν ¼ ðV 2P  2V 2S Þ=ð2V 2P  2V 2S Þ. The plasticity index (PI) [19] Aoi S, Obara K, Hori S, Kasahara K, Okada Y. New Japanese uphole-downhole
required for the modulus-reduction and damping relationships strong-motion observation network: KiK-net. Seismological Research Letters
2000;72:239.
was assumed to be 0 for sand or gravel, 15 for silt, and 30 for clay [20] Okada Y, Kasahara K, Hori S, Obara K, Sekiguchi S, Fujiwara H, Yamamoto A.
(generic values based on soil type). The selection of geologic age Recent progress of seismic observation networks in Japan – Hi-net, F-net, K-
for the Zhang et al. [29] model was inferred from the geologic net and KiK-net. Earth Planets Space 2004;56:xv–xviii.
[21] Thompson EM, Baise LG, Tanaka Y, Kayen RE. A taxonomy of site response
profile provided on the KiK-net website [23]. For the Darendeli complexity. Soil Dyn Earthq Eng 2012;41:32–43.
[30] model, the overconsolidation ratio was reasonably assumed to [22] Kaklamanos J, Dorfmann L, Baise L.G. Modeling dynamic site response
be unity, and the loading frequency and number of loading cycles using the overlay concept. In: Abu-Farsakh M, Yu X, Hoyos LR. (editors).
Proceedings of Geo-Congress 2014: geo-characterization and modeling
were assumed to be 10 and 1 Hz, respectively, using the default for sustainability. Atlanta, Georgia: American Society of Civil Engineers
values recommended by Darendeli [30] for site response analyses. (ASCE); 23–26 February 2014. p. 1167–1176, Geotechnical special publication
no. 234, available at /http://cedb.asce.org/cgi/WWWdisplay.cgi?315410S.
[23] National Research Institute for Earth Science and Disaster Prevention [NIED].
In: Strong-motion seismograph networks (KiK-net), available at 〈http://www.
Appendix B. Supplementary material kyoshin.bosai.go.jp/〉; 2012 [accessed May 2014].
[24] Haskell NA. The dispersion of surface waves on multilayered media. Bull
Seismol Soc Am 1953;72:17–34.
The electronic supplement to this manuscript contains one [25] Thomson WT. Transmission of elastic waves through a stratified solid. J Appl
table and 31 figures that contain additional metadata and results Phys 1950;21:89–93.
[26] Boore DM. SMSIM – Fortran programs for simulating ground motions from
at the six KiK-net stations used in this study. Supplementary data
earthquakes, version 2.3 – a revision of OFR 96-80-A. (Open-file report 2008–
associated with this article can be found in the online version at 1128). Reston, Virginia: US Geological Survey; 2008 (55 pp).
http://dx.doi.org/10.1016/j.soildyn.2014.10.016. [27] Kaklamanos J, Baise LG. Dorfmann L. Quantification of uncertainty in non-
linear soil models at a representative seismic array. In: Deodatis G, Ellingwood
BR, Frangopol DM (editors). Proceedings of the 11th international conference
on structural safety and reliability (ICOSSAR 2013). New York City, New York;
References 16-20 June 2013. p. 4189-4196.
[28] Kaklamanos J. Quantifying uncertainty in earthquake site response models
[1] Building Seismic Safety Council [BSSC]. NEHRP recommended provisions for using the KiK-net database. ([Ph.D. dissertation]). Medford, Massachusetts:
seismic regulations for new buildings and other structures. (FEMA P-750 Tufts University; 2012 (301 pp).
report). 2009 edition. Washington, D.C.: Federal Emergency Management [29] Zhang J, Andrus RD, Juang CH. Normalized shear modulus and material
Agency; 2009. damping ratio relationships. J Geotech Geoenviron Eng 2005;131:453–64.
[2] Schnabel PB, Lysmer J, Seed HB. SHAKE: a computer program for earthquake [30] Darendeli MB. Development of a new family of normalized modulus reduction
response analysis of horizontally layered sites. (Report No. UCB/EERC-72/12). and material damping curves. ([Ph.D. thesis]). Austin, Texas: University of
Berkeley, California: Earthquake Engineering Research Center, University of Texas at Austin; 2001 (396 pp).
California Berkeley; 1972 (102 pp). [31] Dassault Systèmes. Abaqus analysis user's manual. (Version 6.9-2). Dassault
[3] Idriss IM, Sun JI. SHAKE91: a computer program for conducting equivalent Systèmes, 2009, Providence, Rhode Island.
linear seismic response analyses of horizontally layered soil deposits (User's [32] Matasović N, Vucetic M. Cyclic characterization of liquefiable sands. J Geotech
Manual). Davis, California: University of California Davis; 1992 (37 pp). Eng 1993;119:1805–22.
[4] Ordóñez GA. SHAKE2000: a computer program for the 1-D analysis of [33] Kondner RL, Zelasko JS. A hyperbolic stress-strain formulation of sands. In:
geotechnical earthquake engineering problems (User's Manual). Lacey, Proceedings of the 2nd pan American conference on soil mechanics and
Washington: GeoMotions, LLC; 2010 (262 pp). foundation engineering. São Paulo, Brazil: Brazilian Association of Soil
[5] Kottke AR, Rathje EM. Technical manual for Strata. (PEER report 2008/10). Mechanics; 1963. p. 289–324.
Berkeley, California: Pacific Earthquake Engineering Research Center, Univer- [34] Phillips C, Hashash YMA. Damping formulation for nonlinear 1D site response
sity of California, Berkeley; 2008 (100 pp). analyses. Soil Dyn Earthq Eng 2009;29:1143–58.
[6] Hashash YMA, Park D. Non-linear one-dimensional seismic ground motion [35] Masing G. Eigenspannungen and verfertigung beim messing. In: Proceedings
propagation in the Mississippi embayment. Eng Geol 2001;62:185–206. of the 2nd international congress on applied mechanics. Zurich, Switzerland;
[7] Hartzell S, Bonilla LF, Williams RA. Prediction of nonlinear soil effects. Bull 1926.
Seismol Soc Am 2004;94:1609–29. [36] Nelson RB, Dorfmann A. Parallel elasto-plastic models of inelastic material
[8] Kwok AOL, Stewart JP, Hashash YMA, Matasovic N, Pyke R, Wang Z, Yang Z. Use behavior. J Eng Mech 1995;121:1089–97.
of exact solutions of wave propagation problems to guide implementation of [37] Iwan WD. On a class of models for the yielding behavior of continuous and
nonlinear, time-domain ground response analysis routines. J Geotech Geoen- composite systems. J Appl Mech 1967;34:612–7.
viron Eng 2007;133:1385–98. [38] Mroz Z. On the description of anisotropic workhardening. J Mech Phys Solid
[9] Assimaki D, Li W, Steidl J, Schmedes J. Quantifying nonlinearity susceptibility 1967;15:163–75.
via site response modeling uncertainty at three sites in the Los Angeles basin. [39] Kaklamanos J, Dorfmann L, Baise LG. A simple approach to site response
Bull Seismol Soc Am 2008;98:2364–90. modeling: the overlay concept. Seismol Res Lett 2015 (in press).
[10] Kwok AOL, Stewart JP, Hashash YMA. Nonlinear ground-response analysis of [40] Pinheiro JC, Bates DM. Mixed-Effects Models in S and S-PLUS. New York:
Turkey Flat shallow stiff soil site to strong ground motion. Bull Seismol Soc Am Springer-Verlag; 2000 (528 pp).
2008;98:331–43. [41] Bradley BA. A framework for validation of seismic response analyses using
[11] Kaklamanos J, Bradley BA, Thompson EM, Baise LG. Critical parameters seismometer array recordings. Soil Dyn Earthq Eng 2011;31:512–20.
affecting bias and variability in site response analyses using KiK-net downhole [42] Baise LG, Thompson EM, Kaklamanos J, Dorfmann L. Complex site response:
array data. Bull Seismol Soc Am 2013;103:1733–49. does one-dimensional site response work? In: Proceedings of the interna-
[12] Kim B, Hashash YMA. Site response analysis using downhole array recordings tional symposium on the effects of surface geology on seismic motion (ESG4)
during the March 2011 Tohoku-Oki earthquake and the effect of long-duration on 4th international association of seismology and physics of the Earth's
ground motions. Earthq Spectra 2013;29:S37–54. interior (IASPEI)/ international association of earthquake engineering (IAEE).
[13] Yee E, Stewart JP, Tokimatsu K. Elastic and large-strain nonlinear seismic site Santa Barbara, California, 23-26 August 2011 (12 pp).
response from analysis of vertical array recordings. J Geotech Geoenviron Eng [43] Assimaki D, Li W, Steidl J, Tsuda K. Site amplification and attenuation via
2013;139:1789–801. downhole array seismogram inversion: a comparative study of the 2003
[14] Matasović N, Ordóñez GA. D-MOD2000: a computer program for seismic Miyagi-Oki aftershock sequence. Bull Seismol Soc Am 2008;98:301–30.
response analysis of horizontally layered soil deposits, earthfill dams, and [44] Ghofrani H, Atkinson GM, Goda K. Implications of the 2011 M9.0 Tohoku Japan
solid waste landfills (User's Manual). Lacey, Washington: GeoMotions, LLC; earthquake for the treatment of site effects in large earthquakes. Bull Earthq
2010 (182 pp). Eng 2013;11:171–203.
[15] Hashash YMA, Groholski DR, Phillips CA, Park D, Musgrove M. DEEPSOIL 5.0, [45] Régnier J, Cadet H, Bonilla LF, Bertrand E, Semblat J-F. Assessing nonlinear
user manual and tutorial. Champaign, Illinois: University of Illinois at Urbana- behavior of soils in seismic site response: statistical analysis on KiK-net
Champaign; 2011 (107 pp). strong-motion data. Bull Seismol Soc Am 2013;103:1750–70.
[16] Pyke RM. TESS: a computer program for nonlinear ground response analyses. [46] Joyner WB, Chen ATF. Calculation of nonlinear ground response in earth-
Lafayette, California: TAGA Engineering Systems and Software; 2000. quakes. Bull Seismol Soc Am 1975;65:1315–36.
[17] Li XS, Wang ZL, Shen CK. SUMDES: a nonlinear procedure for response analysis [47] Joyner WB. A FORTRAN program for calculating nonlinear seismic ground
of horizontally-layered sites subjected to multidirectional earthquake loading. response. (Open-file report No. 77-671). Reston, Virginia: US Geological
Davis, California: Department of Civil Engineering, University of California Survey; 1977 (50 pp).
Davis; 1992. [48] Bardet JP, Tobita T. NERA: a computer program for nonlinear earthquake
[18] McKenna F, Fenves GL. The OpenSees command language manual. (Version site response analysis of layered soil deposits. Los Angeles, California:
1.2). Berkeley, California: Pacific Earthquake Engineering Research Center, Deptartment of Civil Engineering, University of Southern California; 2001
University of California Berkeley; 2001 ([accessed May 2014]). (44 pp).
J. Kaklamanos et al. / Soil Dynamics and Earthquake Engineering 69 (2015) 207–219 219

[49] Thompson EM, Baise LG, Kayen RE, Guzina BB. Impediments to predicting site [52] Hasselstroem B. Water prospecting and rock-investigation by the seismic
response: seismic property estimation and modeling simplifications. Bull refraction method. Geoexploration 1969;7:213.
Seismol Soc Am 2009;99:2927–49. [53] Haeni FP. Application of seismic refraction methods in groundwater modelling
[50] Assimaki D, Steidl J, Liu PC. Attenuation and velocity structure for site studies in New England. Geophysics 1986;51:236–49.
response analyses via downhole seismogram inversion. Pure Appl Geophys [54] Grelle G, Guadagno FM. Seismic refraction methodology for groundwater level
2006;163:81–118. determination: water seismic index. J Appl Geophys 2009;68:301–20.
[51] Boore DM. Some thoughts on relating density to velocity, available at 〈http://
daveboore.com/daves_notes.php〉; 2007 [accessed May 2014], 12 pp.

Potrebbero piacerti anche