Content uploaded by Kenneth W. Campbell
Author content
All content in this area was uploaded by Kenneth W. Campbell on Dec 07, 2018
Content may be subject to copyright.
Selection of Ground Motion Prediction
Equations for the Global Earthquake
Model
Jonathan P. Stewart,a) M.EERI, John Douglas,b) Mohammad Javanbarg,c)
Yousef Bozorgnia,d) M.EERI, Norman A. Abrahamson,e) M.EERI,
David M. Boore,f) Kenneth W. Campbell,g) M.EERI, Elise Delavaud,h)
Mustafa Erdik,i) M.EERI, and Peter J. Stafford,j) M.EERI
Ground motion prediction equations (GMPEs) relate ground motion intensity
measures to variables describing earthquake source, path, and site effects. From
many available GMPEs, we select those models recommended for use in seismic
hazard assessments in the Global Earthquake Model. We present a GMPE selec-
tion procedure that evaluates multidimensional ground motion trends (e.g., with
respect to magnitude, distance, and structural period), examines functional forms,
and evaluates published quantitative tests of GMPE performance against inde-
pendent data. Our recommendations include: four models, based principally
on simulations, for stable continental regions; three empirical models for interface
and in-slab subduction zone events; and three empirical models for active shallow
crustal regions. To approximately incorporate epistemic uncertainties, the selec-
tion process accounts for alternate representations of key GMPE attributes, such
as the rate of distance attenuation, which are defensible from available data.
Recommended models for each domain will change over time as additional
GMPEs are developed. [DOI: 10.1193/013013EQS017M]
INTRODUCTION
Ground motion prediction equations (GMPEs) relate a ground motion intensity measure,
for example, peak ground acceleration (PGA), to a set of explanatory variables describing the
earthquake source, wave propagation path, and local site conditions (Douglas 2003). These
independent variables include magnitude, source-to-site distance and some parameterization
of local site conditions, and frequently style of faulting mechanism. Certain recent models
also account for other factors affecting earthquake ground motions; e.g., hanging wall effects.
In the past five decades, many hundreds of GMPEs for the prediction of PGA and linear
Earthquake Spectra, Volume 31, No. 1, pages 19–45, February 2015; © 2015, Earthquake Engineering Research Institute
a)
University of California, Los Angeles, CA; jstewart@seas.ucla.edu
b)
BRGM, Orléans, France
c)
AIG, New York, NY (formerly PEER Center, UC Berkeley)
d)
PEER Center, University of California, Berkeley
e)
PG&E, San Francisco, CA
f)
US Geological Survey, Menlo Park, CA
g)
EQECAT, Inc., Beaverton, OR
h)
ETH, Zurich, Switzerland
i)
KOERI, Istanbul, Turkey
j)
Imperial College London, UK
19
elastic response spectral ordinates have been published, which are summarized in a series of
public reports by the second author (Douglas 2011). Therefore, the seismic hazard analyst is
faced with the difficult task of deciding which GMPEs to use for a given project. This deci-
sion is a critical step in any hazard assessment because the resulting predicted spectra are
strongly dependent on the chosen GMPEs.
We describe the selection process for GMPEs undertaken within the framework of the
Global Earthquake Model (GEM) Global GMPEs project, coordinated by the Pacific Earth-
quake Engineering Research Center (PEER; Di Alessandro et al. 2012). The process began
by preselecting, from available models, the most robust GMPEs as candidates for final selec-
tion. This preselection was based on applying the exclusion criteria of Cotton et al. (2006) to
the complete list of models summarized by Douglas (2011). These quality assurance criteria
exclude models that, for example are superseded by more recent GMPEs; do not allow pre-
dictions for the entire magnitude-distance-structural period range of interest; and employ
independent (e.g., magnitude scale) or dependent (e.g., horizontal component definition)
parameters that would complicate their use in state-of-the-art seismic hazard assessments.
As described by Douglas et al. (2012), this screening process within Task 2 of the GEM-
PEER project led to the identification of roughly ten GMPEs for each of three major tectonic
regimes/domains: active crustal regions (ACRs), subduction zones (SZs), and stable conti-
nental regions (SCRs). GMPEs from the recently completed NGA-West2 project (Bozorgnia
et al. 2014) were not available for consideration during the selection process.
Global applications within GEM require approximately three to four recommended
GMPEs for each major tectonic regime for practical reasons; e.g., calculation times. Ideally,
the selection of those GMPEs should account for regional differences within the ACR, SZ,
and SCR regimes, which takes the form of variable GMPE attributes, such as rate of distance
attenuation. In this article, we describe the work undertaken in Task 3 of the GEM-PEER
project to balance these competing objectives in the selection process of having few relatively
robust models that approximately represent epistemic uncertainty in ground-motion predic-
tion. The process, supporting plots, and results are described in more detail in a PEER report
(Stewart et al. 2013a).
Previous GMPE selection tasks have been undertaken for global applications, including
GEM1 pilot project (Douglas et al. 2009), the Seismic Hazard Harmonization in Europe
(SHARE) project (Delavaud et al. 2012), and the Earthquake Model of the Middle East
(EMME) project (S. Akkar, personal communication, 2012). This project is differentiated
from prior work in its global reach, which only GEM1 had previously attempted, and
the approach that was developed to make the selections. In this article, we emphasize
the selection process, which can have long-term applicability, even after the GMPEs that
we have selected are superseded.
Subsequent sections present the procedure followed in GEM-PEER Task 3, including the
composition of the expert panel and the information considered during the selection process.
The next two sections describe the primary tools used in the selection process, which are
trellis plots that compare GMPE predictions for various earthquake scenarios, and a review
of published studies quantitatively comparing predicted and observed response spectral
accelerations in recent earthquakes. We then provide our recommended GMPEs for
GEM global applications, along with the rationales for their selection. For brevity, only
20 STEWART ET AL.
a small subset of the material used by the experts to make the final selection is presented
herein. A more complete, but still abridged, set of plots is provided in an electronic supple-
ment, whereas complete plots are given in Stewart et al. (2013a).
SELECTION PROCEDURE AND FACTORS CONSIDERED
In this section, we present the overall procedure developed to select GMPEs for the three
principal tectonic regimes: SZs, ACRs, and SCRs. The project was overseen by a core group
of experts and a wider expert panel that comprised all members of the project team (Table A1
in the online Appendix). The core group was responsible for preparing initial GMPE recom-
mendations for the three regimes, which were then presented to the wider expert panel for
discussion and potential revision.
We identified criteria for GMPE selection in SZ and ACR regimes as follows:
1. More emphasis given to GMPEs derived from international than local data sets.
Exceptions can be made when a GMPE derived from a local data set has been ver-
ified internationally and found to perform well.
2. More emphasis given to GMPEs that have attributes of their functional form that we
consider desirable, including saturation with magnitude, magnitude-dependent dis-
tance scaling, and terms that mimic the effects of anelastic attenuation.
3. If there are multiple GMPEs that are accurately constrained by data but exhibit dif-
ferent trends, it is desirable to capture those trends in the selected GMPEs to prop-
erly represent epistemic uncertainty.
For SCRs, where strong-motion data are scarce, these criteria were modified as follows:
1. SCR GMPEs are derived principally from the results of numerical simulations.
However, the manner in which the limited available data is used to constrain the
input parameters for the simulations is critical. The empirical calibration may
influence, for example, stress drop parameters and site attenuation (κ0). We prefer
GMPEs judged to effectively use the available data to constrain model parameters.
2. Same as the second criterion for SZ and ACR regimes (desirable attributes of func-
tional form). Because data are limited for SCRs, it is particularly important that the
selected models extrapolate in a reasonable manner beyond the data range.
3. We seek GMPEs that meet the preceding criteria and that collectively (1) represent
diverse geographic regions and (2) use alternative simulation methodologies. This is
intended to represent epistemic uncertainty in the selected GMPEs.
In the selection process, we decided not to down-weight GMPEs with difficult-to-
implement parameters, such as basin depth terms or depth to top of rupture, because
those concerns can be overcome with appropriate parameter selection protocols (Kaklamanos
et al. 2011). We also did not down-weight GMPEs that either lack site terms or whose mod-
eling of site response is nonoptimal, such as lacking nonlinearity, because GMPEs can be
evaluated for a reference rock site condition in hazard analysis and site effects subsequently
added in a hybrid process (Cramer 2003,Goulet and Stewart 2009).
The principal resources developed for GMPE selection were a synthesis of functional
forms, plots showing comparative ground motion scaling with predictive parameters
SELECTION OF GMPEs FOR THE GLOBAL EARTHQUAKE MODEL 21
(distance, magnitude, period, and site condition), and model-data comparisons from the
literature. The latter two are described in the following subsections. In the synthesis of
functional forms, we repeat the equations using consistent terminology across GMPEs
(Stewart et al. 2013a). Certain models assume simple linear scaling with magnitude and
1∕Rdistance decay (where Ris site-to-source distance), whereas others account for more
complex effects, such as magnitude-saturation and magnitude-dependent distance scaling.
These effects are discussed from comparative GMPE scaling plots in the next section.
COMPARATIVE GMPE SCALING
ACTIVE CRUSTAL REGIONS
Trellis charts were drawn to display the multidimensional (magnitude, source-to-site dis-
tance, and structural period) predicted ground motion space in various ways to provide insight
into the preselected GMPEs. The aim is to help identify outliers with clearly nonphysical
behavior, but also to guide the selection of models to capture epistemic uncertainty; that
is, distance attenuation rates are regionally dependent, so it is important to capture this varia-
tion. The charts prepared for rock site conditions and site effects are considered separately.
Trellis plots for ACRs are given for pseudo-spectral acceleration (PSA) versus period in
Figure 1(where acronyms used to refer to specific GMPEs are defined), PSA versus mag-
nitude (M) in Figure 2, and PSA versus distance in Figure 3. Plots of the standard deviation
terms from these models are given in Figure A1 in the online Appendix. Figure 1shows that
the predicted spectra of the nine ACR GMPEs show relatively less model-to-model varia-
bility than those from the other two tectonic regimes, shown subsequently in Figures 6
and 10. The predicted spectra from MEA06 for M5 earthquakes are considerably higher
than the others, perhaps because this magnitude is below the minimum magnitude recom-
mended for application. This characteristic makes this GMPE less appealing because possible
overprediction of ground motions from moderate earthquakes may have a large impact on the
results of hazard analyses and risk assessments, particularly when short return periods are
important or seismicity rates are relatively low, but still qualifying as active. Predictions from
the FEA10 model often fall below the majority of models and display a different spectral
shape, with two shallow peaks at longer source-to-site distances. This may be because it is
based on a limited number of records having rock-like site conditions.
Figure 2shows magnitude scaling of the ACR GMPEs. The models of KEA06 and
FEA10 lack magnitude saturation (i.e., PSA scales linearly with M), which argues against
their selection because they can lead to the prediction of unphysically large or small ground
motions at the edges of the magnitude-distance range of interest. Figure 3shows the distance
attenuation of the ACR GMPEs. All of the models have magnitude-dependent attenuation
terms, but a point of differentiation is that some include effective anelastic attenuation leading
to steeper attenuation for distances beyond 70–100 km (BA08, CY08, MEA06, and ZEA06)
and some do not (AS08, AB10, CB08, FEA10, and KEA06).
The site response functions in the ACR GMPEs are shown in Figure 4, which shows VS30
scaling, where VS30 is the time-averaged shear wave velocity in the upper 30 m of a site; and
Figure 5, which shows soil nonlinearity. Starting with VS30 scaling, three of the models
(FEA10, KEA06, and ZEA06) are predominantly derived from Japanese data, yet have
22 STEWART ET AL.
significantly different scaling at mid-to-short periods, with FEA10 and KEA06 being very
strong relative to worldwide models and ZEA06 slightly weaker. Based on results from the
NGA-West2 project (e.g., Seyhan and Stewart 2014), the ZEA06 trend is considered more
representative for Japan. The VS30 scaling from international models (e.g., AS08, BA08,
CB08, and CY08) at short periods is stronger, indicating a potential regional dependency
in site amplification, which should be considered when selecting GMPEs for ACRs. Regard-
ing nonlinearity (Figure 5), the models of AB10, FEA10, KEA06, MEA06, and ZEA06 are
Figure 1. Trellis chart showing predicted PSAs for preselected ACR GMPEs for various earth-
quake scenarios under rock site conditions. Dashed lines indicate where the scenario falls outside
the magnitude-distance range of validity of the model. The model abbreviations given in the
legend are used henceforth.
SELECTION OF GMPEs FOR THE GLOBAL EARTHQUAKE MODEL 23
linear, whereas the others are nonlinear at short periods. A lack of nonlinearity leads to sig-
nificant overestimation of ground motions for strong levels of input motions for soil site
conditions and mid-to-short-periods. For soft soil conditions, there are large differences,
up to a factor of ten, in the predicted amplifications for high shaking levels.
SUBDUCTION ZONES
Preselection criteria for SZs required that the models distinguish between interface events
at the plate boundary and in-slab events. We prepared separate sets of trellis plots for both
event types, but this study emphasizes interface events for brevity (similar in-slab plots are
discussed in Stewart et al. 2013a). Interface SZ trellis plots are given for PSA against period
in Figure 6(GMPE acronyms defined in legend), PSA versus Min Figure 7, and PSA with
respect to distance in Figure 8. Plots of the standard deviation terms from these models are
given in Figure A2 in the online Appendix. The trellis charts for the interface SZ GMPEs
show that the KEA06 model is an outlier, particularly at long periods, when evaluated for
large magnitude earthquakes (Figure 6) because linear magnitude scaling is assumed
(Figure 7). This suggests that this model is not a good candidate because this behavior
may lead to erroneous hazard analyses for locations where very large events are possible.
Figure 2. Trellis chart showing magnitude scaling of predicted PSAs for preselected ACR
GMPEs for various structural periods and source-to-site distances under rock site conditions.
Dashed lines indicate where the scenario falls outside the magnitude-distance range of validity
of the model.
24 STEWART ET AL.
Linear magnitude scaling is also used by LL08, AEA10, and AB03, but these models also
have a magnitude-dependent distance decay that effectively produces nonlinear magnitude
scaling, as shown in Figure 7.
As shown in Figure 8, distance attenuation rates are quite variable among the GMPEs,
particularly at magnitudes of 8 and 9. At these large magnitudes, the AB03 model for inter-
face events shows relatively flat attenuation rates, whereas AEA15, KEA06, and ZEA06
have relatively steep attenuation rates. These differences may reflect regional variations,
that is genuine epistemic uncertainty, because the AB03 model is drawn heavily from Central
and South American data, whereas AEA15, KEA06, and ZEA06 are based largely or entirely
upon data from Japan. This concern is explored further in the model-data comparisons
presented in the next section. All of the models have magnitude-dependent distance attenua-
tion rates.
Predictions from the AB03 model for interface events typically represent a lower bound
on estimates from the other considered GMPEs (Figure 6), except at long distances from very
large earthquakes, where the flat decay curve leads to high predicted PSAs (Figure 8). The
models of AEA15 and ZEA06 often predict spectral ordinates at the upper end of the spread
of spectra. Predictions from the other GMPEs are more grouped, particularly within the rough
Figure 3. Trellis chart showing distance decay of predicted PSAs for preselected ACR GMPEs
for various structural periods and magnitudes under rock site conditions. Dashed lines indicate
where the scenario falls outside the magnitude-distance range of validity of the model.
SELECTION OF GMPEs FOR THE GLOBAL EARTHQUAKE MODEL 25
center of the distribution of available data from interface SZ events, with M6 to 7 and Rfrom
50 km to 150 km (Figure 6).
Figure 9shows attributes of VS30 scaling in site response functions for SZ GMPEs. The
SZ GMPEs predict similar VS30 scaling, except for the KEA06 model, which predicts higher
amplification for slow VS30 than the other GMPEs. Only three of the GMPEs under consid-
eration account for nonlinear site response: AEA15, AB03, and MEA06. In GEM, ground
motions need to be predicted on soil sites close to the largest subduction events; therefore,
models that include a nonlinear site term are favored.
STABLE CONTINENTAL REGIONS
Figure 10 shows the PSAs from the ten preselected GMPEs (GMPE acronyms defined in
legend). The variations among predictions is large in comparison to other regimes, sometimes
up to a factor of ten, particularly at higher magnitudes and closer distances. This is rather
expected because there are practically no strong-motion records from earthquakes in SCRs
for these magnitude-distance ranges and the manner in which the models extrapolate will
vary substantially between investigators. This comparison also shows that certain models
predict greatly different PSAs than the majority of GMPEs at given distances and magni-
tudes. For example, DEA06 predicts much lower spectra at close distances, whereas the pre-
dicted spectra from SEA09 (Craton model) show a bump at approximately 1 s. Both these
features are the results of choices in modeling to capture local characteristics in the areas for
which these GMPEs were derived. DEA06 assumed particularly deep focal depths when
deriving their model, using Joyner-Boore distance as a predictor variable in the GMPEs
for southern Norway, which leads to low near-source motions. SEA09 developed their
model for the Yilgarn Craton in Western Australia, which has a specific combination of
Figure 4. Trellis chart showing VS30 scaling of the ACR GMPEs for a reference rock peak accel-
eration of PGAr¼0.1g. Amplification has been computed relative to a consistent reference
velocity of Vref ¼1;000 m∕s, regardless of the reference condition used in the GMPE. Stepped
relationships (e.g., AB10) describe site response relative to discrete categories, whereas contin-
uous relations use VS30 directly as the site parameter.
26 STEWART ET AL.
shallow earthquakes and a crustal structure that leads to large surface waves. The local pecu-
liarities of these models mean that they may not be applicable for other SCRs that do not have
these characteristics.
Figure 11 shows that the magnitude scaling of the SCR GMPEs is quite variable with
respect to magnitude saturation. Weak magnitude saturation occurs in DEA06, FEA96,
SEA09, and RKI07, which in certain cases leads to the prediction of potentially unrealis-
tically large PSAs from large earthquakes, particularly at long periods. Other
models include stronger magnitude saturation terms, which may be preferable for GEM
application.
Figure 12 shows the predicted distance attenuation of the ten models, which again are
quite variable. Many of these models were developed for central and eastern North America
Figure 5. Trellis chart showing variation of site amplification with reference rock peak accel-
eration (for Vref ¼1;000 m∕s) for various site classes and period. Representative velocities for
each site class are based on category medians in the NGA-West2 database (Seyhan and Stewart
2012).
SELECTION OF GMPEs FOR THE GLOBAL EARTHQUAKE MODEL 27
(ENA), and reflect a change toward flatter attenuation associated with Moho bounce effects
between 70 and 140 km (AB06′, C03, FEA96, and PEA11). Other models for this same
region, SEA02 and TEA97, do not model such effects. To account for epistemic uncertainty
in modeling the effect of crustal structure and the requirement of global applicability of the
selected GMPEs, it was considered desirable to select models that apply to both these cate-
gories. Another observation that can be made from Figure 12 is that for very large earth-
quakes, AB06 is often a lower bound on the predictions and SEA09 is generally the
upper bound.
Figure 6. Trellis chart showing predicted PSAs for preselected SZ GMPEs for various interface
earthquake scenarios for rock site conditions. Dashed lines indicate where the scenario falls out-
side the magnitude-distance range of validity of the model. Abbreviations for these GMPEs are
defined in the legend.
28 STEWART ET AL.
Almost none of the SCR GMPEs include site terms allowing the ground motions on non-
rock sites to be predicted. Only A08′, AB06′, and RKI07 include such terms; these are shown
for NEHRP classes B–E in Figure A3 in the online Appendix. In the case of A08′and AB06′,
these were adopted from results for ACRs. In the case of RKI07, the predicted nonlinear
effects are very strong and the amplifications are not smooth, but show large period-to-period
variations, which we consider unrealistic. Most of the SCR GMPEs apply for hard rock site
conditions with reference velocities much faster than those used as the reference in typical
empirical site factors for ACRs or SZs; e.g., 760 or approximately 1;000 m∕s. Accordingly,
for those models, before site factors of the types shown in Figures 4,5, and 9can be applied,
an additional correction must be made to adjust from hard rock to approximately 1;000 m∕s.
This correction is generally not provided in the SCR GMPE documentation, nor is it well
defined elsewhere in the literature. The aforementioned weaknesses with the SCR GMPE site
terms are discussed further in the following with our recommendations.
The standard deviation terms associated with the SCR GMPEs (Figure 13) show sub-
stantial model-to-model differences. These standard deviation models are generally a direct
consequence of the simulation method and variability in input parameters, rather than the
result of statistical comparisons between data and predictions. The standard deviations
Figure 7. Trellis chart showing magnitude scaling of predicted PSAs for preselected interface SZ
GMPEs for various structural periods and source-to-site distances under rock site conditions.
Dashed lines indicate where the scenario falls outside the magnitude-distance range of validity
of the model.
SELECTION OF GMPEs FOR THE GLOBAL EARTHQUAKE MODEL 29
associated with RKI07 are much lower than those from the other models because only the
parametric component of the variability was included, rather than also including the model-
ing component. This argues against its selection. The standard deviation models of DEA06,
SEA02, and SEA09 show strong period dependencies, which are not observed in the SZ or
ACR GMPEs. We believe that this argues against selecting these models. Only three of the
ten GMPEs separate standard deviations into between- and within-event components, which
is valuable for certain analyses. E06 (EPRI 2006) proposed generic standard deviation mod-
els for SCR GMPEs, which were considered as possible replacements for those standard
deviations that were not thought to be physically realistic or are not spilt into two com-
ponents.
GMPE-DATA COMPARISONS
CRITERIA FOR ACCEPTING GMPE-DATA COMPARISON STUDIES FOR USE IN
MODEL SELECTION
GMPEs for SZs and ACRs are most often developed from the regression of empirical
strong-motion data; thus, model-data comparisons are integral to the process by which they
Figure 8. Trellis chart showing distance decay of predicted PSAs for preselected interface SZ
GMPEs for various structural periods and magnitudes under rock site conditions. Dashed lines
indicate where the scenario falls outside the magnitude-distance range of validity of the model.
30 STEWART ET AL.
are prepared. Nonetheless, GMPE-data comparisons were considered a critical component of
the selection process. GMPEs derived for SCRs are generally based on ground motion simu-
lations; therefore, model-data comparisons are even more important for these equations. The
value of these comparisons is often derived from the comparison data set being beyond the
parameter space considered for the original GMPE. For example, the data may be derived
from a different region from that used in the original model development, which can gen-
erally be useful for studies of model applicability to the data region and regional variations of
ground motions. Another significant example specific to SZs is the recent availability of
data sets from large magnitude earthquakes, such as the M8.8 Maule, Chile, and M9.0
Tohoku, Japan events, which are beyond the upper-bound magnitudes available during
GMPE development.
Most GMPE-data comparisons in the literature consist of plots of ground motion intensity
measures versus distance, along with GMPE median trend curves. Plots of this type have
limited applicability for formal analysis of GMPE performance because it can be difficult
to judge trends when the data span a very wide range on the amplitude axis and event-specific
bias, or event terms, are not taken into account. Accordingly, we restrict our literature com-
pilation for SZs and ACRs to studies that include a formal analysis of residuals into the
GMPE-data comparisons. This still leads to a considerable number of studies: eight for
SZs and 13 for ACRs. For SCRs, however, restricting our compilation to only this type
of analysis would lead to considering only one or two studies. Hence, it was decided for
SCRs to also compile those studies only showing plots of predicted versus observed ground
motions. Even with this relaxation of the criterion, only a few studies were identified.
Figure 9. Trellis chart showing VS30 scaling of the SZ GMPEs for a reference rock peak accel-
eration of PGAr¼0.1g. Amplification has been computed relative to a consistent reference velo-
city of Vref ¼1;000 m∕s, regardless of the reference condition used in the GMPE. Stepped
relationships (e.g., AB03) describe site response relative to discrete categories, whereas contin-
uous relations use VS30 directly as the site parameter. The range shown for LL08 and YEA97
occurs because these relations do not have a formal site term but alternative GMPEs for rock and
soil sites; therefore, the differences are magnitude and distance dependent.
SELECTION OF GMPEs FOR THE GLOBAL EARTHQUAKE MODEL 31
The three general methods of relatively rigorous model-data comparisons present in the
collected literature are: (1) the maximum likelihood approach of Scherbaum et al. (2004) and
its extension to normalized within- and between-event residuals distributions by Stafford
et al. (2008), which is intended to judge the overall fit of model to data; (2) the information
theoretic approach of Scherbaum et al. (2009) for model-data comparisons, which also pro-
duces various overall goodness-of-fit metrics; and (3) analysis of within- and between-event
residuals specifically targeted to investigations of GMPE scaling with respect to magnitude,
distance, and site parameters (Scasserra et al. 2009).
Figure 10. Trellis chart showing predicted PSAs for preselected SCR GMPEs for various earth-
quake scenarios under rock site conditions. Dashed lines indicate where the scenario falls outside
the magnitude-distance range of validity of the model. Abbreviations of these GMPEs are listed in
the legend.
32 STEWART ET AL.
APPLICATION FOR GMPE SELECTION IN THIS STUDY
Tables 1and 2summarize the model-data comparisons considered in this study for ACRs
and SZs, respectively. More information on each study and its individual findings are
given in a consistent format within Appendix A of Stewart et al. (2013a). The columns
in Tables 1and 2indicate the GMPEs that were tested, whereas the rows correspond to
the model-data comparison studies.
For ACRs, the most frequently tested models are the Next-Generation Attenuation
(NGA) models of AS08, BA08 (which is the single most-tested model), CB08, and
CY08. Many of these studies seek to evaluate the applicability of the global NGA models
to specific regions, including Europe, Japan, Iran, and New Zealand. The overall goodness-
of-fit approaches determine varying levels of fit for these and other models. Sometimes when
poor fits are encountered, relatively local GMPEs better fit the data, but such models are
likely to not extrapolate well to larger magnitude events. None of the NGA models or
other preselected models are clearly superior according to these studies. Slightly more useful
are the second type of model-data comparisons, in which specific GMPE attributes
are tested. These studies find certain instances of misfits in distance attenuation trends.
Figure 11. Trellis chart showing magnitude scaling of predicted PSAs for preselected SCR
GMPEs for various structural periods and source-to-site distances under rock site conditions.
Dashed lines indicate where the scenario falls outside the magnitude-distance range of validity
of the model.
SELECTION OF GMPEs FOR THE GLOBAL EARTHQUAKE MODEL 33
In some cases, when the data driving the misfit are from small magnitude events outside the
range of applicability of the original models, a model by Chiou et al. (2010) performs well,
although this was not a GEM preselected model from Task 2 because it only provides coeffi-
cients for a few oscillator periods.
For SZs, the most frequently tested models are AB03 and ZEA06, which reflect data
globally and from Japan, respectively. The studies rather clearly indicate regional variations
in subduction ground motions. Overall goodness-of-fit approaches find Japan-based models
such as ZEA06 performing better than other models for Japanese data. Those studies also find
that the AB03 model performs relatively poorly against Japanese and Greek data (Beauval
et al. 2012,Delavaud et al. 2012) and relatively well against Central and South American data
(Arango et al. 2012). The AEA15 model, although tested relatively sparsely, has generally
performed well in the tests. As with ACRs, slightly more useful were the model-data com-
parisons in which specific GMPE attributes were tested. Applications of this approach to the
Maule, Chile, and Tohoku, Japan, data (Boroschek et al. 2012,Stewart et al. 2013b) identi-
fied different distance attenuation trends from these large events. In the case of the Maule
earthquake, the relatively slow distance attenuation of the AB03 model provided a good fit to
the data, whereas the Tohoku data attenuated relatively fast with distance and was better
matched by the models of ZEA06 and AEA15.
Figure 12. Trellis chart showing distance decay of predicted PSAs for preselected SCR GMPEs
for various structural periods and magnitudes under rock site conditions. Dashed lines indicate
where the scenario falls outside the magnitude-distance range of validity of the model.
34 STEWART ET AL.
Table A2 in the online Appendix presents the model-data comparisons considered in
this study for SCRs. Most of the studies show plots of recorded data against median fit
lines, and between-event variability is not considered in the analysis or interpretation.
Very few quantitative comparisons of the type undertaken for SZs and ACRs have
been performed, and these have not provided conclusive results. The AB06′model
has been the most well-tested model for SCRs.
RECOMMENDED GMPEs
The GEM-PEER Task 3 core working group developed consensus, or near-consensus,
selections based on the criteria and information presented previously for the ACR, SZ, and
SCR regimes. These selections were carefully reviewed via an in-person meeting and written
correspondence by all of the project experts (Table A1 in the online Appendix). Based on
expert feedback, final recommendations for GMPEs to be used by GEM for hazard calcula-
tions were developed, as described below.
ACTIVE CRUSTAL REGIONS
The following three models were selected for ACRs: AB10 (Akkar and Bommer 2010),
CY08 (Chiou and Youngs 2008), and ZEA06 (Zhao et al. 2006). These models provide a
Figure 13. Trellis chart showing inter-event (between), intra-event (within), and total natural log
standard deviations of the preselected SCR GMPEs.
SELECTION OF GMPEs FOR THE GLOBAL EARTHQUAKE MODEL 35
Table 1. Summary of studies in literature with quantitative model-data comparisons for
ACRs: rows with light gray shading are for studies using an overall goodness-of-fit
approach; rows with dark gray shading are for studies that test specific GMPE attributes
through residuals analysis
AS
2008
AB
2010
BA
2008
CB
2008 FEA10
CY
2008
KEA
2006
MEA
2006
ZEA
2006
A (Euro-Med) X X
B (Iran) X X X
C (Worldwide) X X X X X (CF08) X X X
D (CA) X X X X
E (Japan) X X X X
F (Portugal) X X X
G (Japan) X X X X (CF08) X X X
H (Italy) X X (07) X X X
I (Greece) X
J (Iran) X X X
K (Japan) X X X X X
L (NZ) X X X X
M (CA) X X X X
A=Stafford et al. (2008);B=Ghasemi et al. (2008,2009); C =Delavaud et al. (2012);D=Kaklamarios and Baise
(2011); E =Nishimura (2010);F=Vilanova et al. (2012);G=Beauval et al. (2012);H=Scasserra et al. (2009);
I=Margaris et al. (2010);J=Shoja-Taheri et al. (2010);K=Uchiyama and Midorikawa (2011);L=Bradley
(2013);M=Liao and Meneses (2013).
Table 2. Summary of studies in literature with quantitative model-data comparisons for
SZs: rows with light gray shading are for studies using an overall goodness-of-fit approach;
rows with dark gray shading are for studies that test specific GMPE attributes through
residuals analysis
AEA
2015
AEA
2010
AB
2003
GEA
2005
KEA
2006
LL
2008
MEA
2006
YEA
1997
ZEA
2006
A (S. America) X X X X X X X
B (L. Antilles) X X X X X X X
C (India-Burma) X
D (Greece) X X X X X X
E (Global) X X X X X X X X
F (NZ) X X X
G (Chile) X X
H (Japan) X X
A=Arango et al. (2012);B=Douglas and Mohais (2009);C=Gupta (2010);D=Delavaud et al. (2012);E=Beauval
et al. (2012);F=Bradley (2013);G=Boroschek et al. (2012);H=Stewart et al. (2013b).
36 STEWART ET AL.
satisfactory geographical spread: one for Europe and the Middle East, one global, and one
predominantly for Japan. Their scaling characteristics show desirable features, such as mag-
nitude and distance saturation and anelastic attenuation terms, which means that they can be
applied across the magnitude-distance range of interest to GEM. CY08 was preferred over the
other pre-selected NGA models because (1) its magnitude scaling for small and moderate
events was considered to be more appropriate than the other NGA models; and (2) it has an
anelastic attenuation term that has produced relatively favorable model-data comparisons in
past studies. The BA08 model was seriously considered for selection as an alternative or
supplement to CY08 because it has also generally compared well to international data
and has many of the desirable functional form attributes of CY08, but with simpler equations.
It was not selected because we did not want four ACR models. Recall that the NGA-West2
GMPEs were not available for consideration during the selection process.
Figures A4–A6 in the online Appendix present replots of the ACR trellis diagrams that
highlight the selected models by graying out the predictions from the nonselected GMPEs.
The figures present response spectra, magnitude scaling, distance scaling, and standard
deviation terms for ACR events and rock site conditions. These plots show how the selected
models reflect the range of behavior observed in the preselected GMPEs.
Each of the selected ACR models includes site terms, but we do not recommend applica-
tion of the linear site terms of AB10 or ZEA06. The ZEA06 and AB10 models should be used
for hard rock and rock conditions, respectively (assumed Vref ¼1;000 m∕s). The nonlinear
site amplification function from CY08 can be applied to correct the ground motions for the
VS30 of the site. The CY08 amplification function was developed relative to a reference con-
dition of 1;130 m∕s, which is sufficiently close to 1;000 m∕sthat the model can be directly
applied without significant bias.
There was some discussion about whether epistemic uncertainty in ground motions in
ACRs is sufficiently captured by these three models, because for some magnitudes and dis-
tances, the three sets of results are quite similar (Figure A4 in the online Appendix). After
some deliberation, we did not select a fourth GMPE or to scale one of the selected models
up or down. However, the within-event standard deviation terms of the selected models
(Figure A7 in the online Appendix) have significant differences, reflecting epistemic
uncertainty.
SUBDUCTION ZONES
We selected the global model of AEA15 (Abrahamson et al. 2015), also known as the BC
Hydro model; the global model of AB03 (Atkinson and Boore 2003) and the Japanese model
of ZEA06 (Zhao et al. 2006). These models were preferred because they are based on large
data sets; have desirable attributes in terms of their magnitude and distance scaling functions;
have been verified against data from well-recorded earthquakes, including 2010 Maule and
2011 Tohoku; and produce different distance attenuation trends that have been shown to
match data from different global regions, thus introducing a representation of genuine epis-
temic uncertainty into the selection.
There was some debate over inclusion of the AB03 model because its predicted decay
rate from large (M>8) interface earthquakes is slow, meaning that the ground motions at
great distances (>100 km) are not substantially reduced from those closer to the source. This
SELECTION OF GMPEs FOR THE GLOBAL EARTHQUAKE MODEL 37
behavior was considered physically unlikely by some members of the Task 3 expert panel,
who recommended that the model be rejected. Nevertheless, it this model was retained
because the slow decay rate has been observed in some earthquakes, such as Maule in
Chile (Boroschek et al. 2012), and the model has also been shown to work well in
model-data comparisons for smaller magnitude events as well, such as study A in Table 2.
Moreover, because variable distance attenuation rates are observed across global data sets for
interface subduction zone earthquakes, and the AEA15 and ZEA06 models have relatively
fast distance attenuation rates, the use of the AB03 model was considered desirable to capture
this important source of epistemic uncertainty. Nonetheless, we never reached full consensus
on the selection of this particular model and no strong alternative model emerged during
discussions.
Figures A8–A11 in the online Appendix present replots of the trellis diagrams for SZs
that highlight the selected models by graying out the predictions from the nonselected
GMPEs. The figures present response spectra, magnitude scaling, distance scaling, and stan-
dard deviation terms for interface subduction events and rock site conditions. Additional
similar plots for in-slab subduction events are presented in the Appendix of Stewart et al.
(2013a). These plots show how the selected models reflect the variable rates of distance
attenuation in preselected models.
Each of the selected SZ models includes site terms, but we do not recommend application
of the linear site terms of ZEA06. Instead, the ZEA06 model should be used for hard rock
conditions (assumed Vref ¼1;000 m∕s) and the nonlinear site amplification function from
AEA15 applied to these hard-rock estimates. Because the assumed reference velocity is
Vref ¼1;000 m∕sfor ZEA06 and the AEA15 site terms have period-dependent and unspe-
cified values of Vref , the appropriate site correction can be computed from the AEA15 model
as discussed in the following (where fis the site function in natural log units):
1. Compute site amplification using the appropriate VS30 for the site: fsite ðVS30;PGArÞ.
2. Compute site amplification for Vref ¼1;000 m∕s:fref ð1;000 m∕s;PGArÞ.
3. Calculate site amplification relative to Vref :fsiteðVS30 ;PGArÞfref ð1;000 m∕s;
PGArÞ.
STABLE CONTINENTAL REGIONS
Consensus was unanimous on the selection of the PEA11 (Pezeshk et al. 2011) GMPE.
Several lines of reasoning supported this selection: it is based on the hybrid empirical tech-
nique, which has desirable attributes, and uses a recent and fairly complete data set for ENA.
Furthermore, it can be considered an update of C03 (Campbell 2014). AB06′(Atkinson and
Boore 2006,Atkinson and Boore 2011) was also chosen. Arguments for this model, which is
based on finite-source stochastic simulations, include the effective calibration of input para-
meters against available data; its broad usage in previous forms (including U.S. national
hazard maps); an ability to apply the model for either very hard rock conditions or for VS30 ¼
760 m∕sconditions, thus avoiding the need for correction factors to very hard rock condi-
tions; and its position as the most prominent and well documented of the stochastic proce-
dures. An argument against its selection is that elements of the model are similar to those of
PEA11, so it can be argued that the use of PEA11 and AB06′may artificially lower epistemic
38 STEWART ET AL.
uncertainty. SEA02, a double corner model with saturation (Silva et al. 2002), was the third
model selected. The principal argument for this stochastic GMPE is its use of a point-source
double corner model for the source spectrum, which has more realistic characteristics than
single corner models with respect to long period (>1s) spectral ordinates. Single-corner
models tend to overpredict observed long-period ground motions, whereas double-corner
source spectra generally better match these motions, which may be important for applications
involving high-rise buildings and other long-period structures.
The three selected models of PEA11, AB06′, and SEA02 were all developed for applica-
tion in the central and eastern United States. To increase the geographical spread of the
selected GMPEs we considered including the Craton version of SEA09 in lieu of SEA02.
The Craton version of SEA09 applies to a SCR that is distinct geographically from ENA,
which dominates many of the other preselected GMPEs. Moreover, this GMPE was devel-
oped by using a different simulation procedure: a hybrid of stochastic simulations at high
frequencies and physics-based modeling at low frequencies. The diversity of the study region
and simulation techniques were cited as advantages of this model. However, the weaknesses
of this model were eventually considered to be too strong for its selection. These weaknesses
include: relatively poor documentation; some features of the model, such as the properties of
shallow earthquakes and large Moho bounce effect, are specific to the study region and may
not extrapolate well globally; the magnitude scaling does not saturate, but increases in slope
with magnitude at short periods (Figure 11); and the standard deviation term has an unrea-
listic step at approximately 1 s, at the interface of the stochastic and physical models
(Figure 13).
Figure A12 in the online Appendix highlights predicted spectra from PEA11, AB06′, and
SEA02 by graying out the predictions from the other GMPEs. These graphs show that the
predictions from these three models are quite similar. We felt that this similarity in predic-
tions does not accurately reflect the epistemic uncertainty in SCR GMPEs, which should be
quite large, given the lack of data. Therefore, we felt that some additional uncertainty should
be introduced into the ground motion logic tree for SCRs by adding another model. Various
ways of generating an additional model were considered. This included the idea of scaling up
or down one of the already selected GMPEs (the so-called backbone approach), but this was
considered too arbitrary a method because it was difficult to decide on a scaling factor. In the
end, it was decided to bring in an additional model.
Therefore, following much discussion, we selected the GMPE of TEA97′(Toro et al.
1997,Toro 2002). Although this model is also for ENA, its predictions are significantly
different from those of the other three models. In addition, its modeling of epistemic uncer-
tainty and aleatory variability is the most sophisticated of all stochastic models and it has
been used successfully in many previous projects. However, the data analyzed for this model
are now more than 20 years old; they were originally published as part of an EPRI
report (1993).
Figures A12–A16 in the online Appendix present replots of the trellis diagrams
for SCRs that highlight the four selected models by graying out the predictions from
the nonselected GMPEs. The figures present response spectra, magnitude scaling, distance
scaling, and standard deviation terms for rock site conditions. These plots show that the
SELECTION OF GMPEs FOR THE GLOBAL EARTHQUAKE MODEL 39
selected models approximately reflect the range of behavior observed in the prese-
lected GMPEs.
Of the four selected models, only AB06′includes recommendations for modeling site
amplification. The AB06’model can be applied for hard rock reference conditions of VS≥
2.0 km∕s(NEHRP Class A), or the NEHRP BC boundary, VS30 ¼760 m∕s. When the BC
model is used, a site amplification function can be applied, which was adopted from an
empirical study of site amplification for active crustal regions (Choi and Stewart 2005).
It is unknown whether these site amplification functions are applicable to the SCRs,
although this is an area of active research in the NGA-East project (http://peer.berkeley
.edu/ngaeast/).
As mentioned previously, there are problems with the standard deviation functions in
some of the selected SCR GMPEs. We recommend the application of the standard deviation
terms from AB06′, PEA11, and TEA97′in their as-published form, shown in Figure A16 in
the online Appendix. We recommend that the standard deviations of EPRI (2006) be used in
lieu of those from SEA02 because of the large increase in standard deviations for T>1sfor
SEA02, which we consider unrealistic.
SUMMARY AND CONCLUSIONS
In this article, we have presented and applied a method for selection of ground motion
models for the GEM-PEER Global GMPEs Project. This procedure aimed to be transparent,
objective, and repeatable in future projects; for example, for possible updates of the GEM
hazard assessments. The procedure consists of expert reviews of several information sources,
including (1) trellis plots showing the scaling of candidate GMPEs against period, magni-
tude, distance, and site condition, along with within- and between-event standard deviation
terms; (2) functional forms of candidate GMPEs; and (3) quantitative model-data comparison
studies in the literature.
Based on expert review of the aforementioned information sources, a set of GMPEs for
each of the tectonic regimes was proposed, as described in the previous section. These con-
sisted of three GMPEs for subduction zones, three GMPEs for active crustal regimes, and
four GMPEs for stable continental regions. For the majority of these GMPEs, their associated
standard deviation models and site terms were also selected. The only exception for the stan-
dard deviation component of the models was for stable continental regions, where a standard
deviation model by EPRI (2006) was preferred over that derived by Silva et al. (2002). In the
case of site amplification models, we do not recommend the linear site amplification func-
tions used in several of the selected GMPEs. In those cases, we recommend application of the
GMPEs for reference rock site conditions in conjunction with nonlinear site corrections from
the literature (Stewart et al. 2013a).
We emphasize that the goal of this paper is to select a set of GMPEs for global hazard
analysis; therefore, fewer GMPEs may be selected than that used for site-specific analysis
and/or development of national hazard maps. Additionally, GMPE development is a continu-
ously evolving research area, and new and/or updated GMPEs are regularly published as
more empirical and simulated data become available and our knowledge of ground motion
hazard expands. Thus, the set of GMPEs recommended here should not be viewed as a long-
term recommendation and should be reevaluated on a regular basis.
40 STEWART ET AL.
ACKNOWLEDGMENTS
This study was funded by the GEM Foundation as part of PEER’s Global GMPEs pro-
ject. Any opinions, findings, and conclusions or recommendations expressed in this material
are those of the authors and do not necessarily reflect those of the sponsors. Constructive
feedback and comments received from many international experts involved in the
GEM-PEER project and two anonymous reviewers are gratefully appreciated. We thank
Carola Di Alessandro for her assistance with the collection and synthesis of data for this
study and Emel Seyhan for her assistance with the preparation of the site amplification fig-
ures in this article.
APPENDIX
Please refer to the online appendix of this manuscript to access the Electronic
Supplement, which includes (1) a list of the GEM-PEER expert panel members and (2) var-
ious figures referenced in the main text.
REFERENCES
Abrahamson, N. A., and Silva, W. J., 2008. Summary of the Abrahamson & Silva NGA ground-
motion relations, Earthquake Spectra 24,67–97.
Abrahamson, N., Gregor, N., and Addo, K., 2015. BCHydro ground motion prediction equations
for subduction earthquakes, Earthquake Spectra, in press, DOI: 10.1193/012012EQS015M.
Akkar, S., and Bommer, J. J., 2010. Empirical equations for the prediction of PGA, PGV
andspectral accelerations in Europe, the Mediterranean region and the Middle East, Seismo-
logical Research Letters 81, 195–206.
Arango, M. C., Strasser, F. O., Bommer, J. J., Cepeda, J. M., Boroschek, R., Hernandez, D. A.,
and Tavera, H., 2012. An evaluation of the applicability of current ground-motion models to
the South and Central American subduction zones, Bulletin of the Seismological Society of
America 102, 143–168.
Arroyo, D., García, D., Ordaz, M., Mora, M. A., and Singh, S. K., 2010. Strong ground-motion
relations for Mexican interplate earthquakes, Journal of Seismology 14, 769–785.
Atkinson, G. M., 2008. Ground-motion prediction equations for eastern North America from a
referenced empirical approach: Implications for epistemic uncertainty, Bulletin of the Seismo-
logical Society of America 98, 1304–1318.
Atkinson, G. M., and Boore, D. M., 2006. Earthquake ground-motion prediction equations
for eastern North America, Bulletin of the Seismological Society of America 96,
2181–2205.
Atkinson, G. M., and Boore, D. M., 2003. Empirical ground-motion relations for subduction zone
earthquakes and their application to Cascadia and other regions, Bulletin of the Seismological
Society of America 93, 1703–1729.
Atkinson, G. M., and Boore, D. M., 2011. Modifications to existing ground-motion prediction
equations in light of new data, Bulletin of the Seismological Society of America 101,
1121–1135.
Beauval, C., Tasan, H., Laurendeau, A., Delavaud, E., Cotton, F., Guéguen, Ph., and Kuehn, N.,
2012. On the testing of ground-motion prediction equations against small magnitude data,
Bulletin of the Seismological Society of America 102, 1994–2007.
SELECTION OF GMPEs FOR THE GLOBAL EARTHQUAKE MODEL 41
Boore, D. M., and Atkinson, G. M., 2008. 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 24,99–138.
Boroschek, R., Contreras, V., Kwak, D. Y., and Stewart, J. P., 2012. Strong ground motion
attributes of the 2010 Mw8.8 Maule Chile Earthquake, Earthquake Spectra 28,
S19–S38.
Bozorgnia, Y., Abrahamson, N. A., Al Atik, L., Ancheta, T. D., Atkinson, G. M., Baker, J. W.,
Baltay, A., Boore, D. M., Campbell, K. W., Chiou, B. S.-J., Darragh, R., Day, S., Donahue, J.,
Graves, R. W., Gregor, N., Hanks, T., Idriss, I. M., Kamai, R., Kishida, T., Kottke, A., Mahin,
S. A., Rezaeian, S., Rowshandel, B., Seyhan, E., Shahi, S., Shantz, T., Silva, W., Spudich, P.,
Stewart, J. P., Watson-Lamprey, J., Wooddell, K., and Youngs, R., 2014. NGA-West2
research project, Earthquake Spectra 30,973–987.
Bradley, B. A., 2013. A New Zealand-specific pseudo-spectral acceleration ground motion pre-
diction equation for active shallow crustal earthquakes based on foreign models, Bulletin of the
Seismological Society of America 103, 1801–1822.
Campbell, K. W., 2003. Prediction of strong ground motion using the hybrid empirical method
and its use in the development of ground-motion (attenuation) relations in eastern North
America, Bulletin of the Seismological Society of America 93, 1012–1033.
Campbell, K. W., 2014. An evaluation of eastern North America ground motion prediction mod-
els developed using the hybrid empirical method, Bulletin of the Seismological Society of
America 104, 347–359.
Campbell, K. W., and Bozorgnia, Y., 2008. 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 24, 139–171.
Chiou, B., Youngs, R., Abrahamson, N., and Addo, K., 2010. Ground-motion attenuation model
for small-to-moderate shallow crustal earthquakes in California and its implications onregio-
nalization of ground-motion prediction models, Earthquake Spectra 26, 907–926.
Chiou, B. S-J., and Youngs, R. R., 2008. An NGA model for the average horizontal component of
peak ground motion and response spectra, Earthquake Spectra 24, 173–215.
Choi, Y., and Stewart, J. P., 2005. Nonlinear site amplification as function of 30 m shear wave
velocity, Earthquake Spectra 21,1–30.
Cotton, F., Scherbaum, F., Bommer, J. J., and Bungum, H., 2006. Criteria for selecting and
adjusting ground-motion models for specific target regions: Application to central Europe
and rock sites, Journal of Seismology 10, 137–156.
Cramer, C. H., 2003. Site specific seismic hazard analysis that is completely probabilistic,
Bulletin of the Seismological Society of America 93, 1841–1846.
Delavaud, E., Scherbaum, F., Kuehn, N., and Allen, T., 2012. Testing the global applicability of
ground-motion prediction equations for active shallow crustal regions, Bulletin of the Seismo-
logical Society of America 102, 707–721.
Di Alessandro, C., Bozorgnia, Y., Abrahamson, N. A., Akkar, S., and Erdik, M., 2012.
GEM-PEER global ground motion prediction equations project: An overview, in Proceedings
15th World Conference on Earthquake Engineering,24–28 September 2012, Lisbon,
Portugal.
Douglas, J., 2003. Earthquake ground motion estimation using strong-motion records: A review
of equations for the estimation of peak ground acceleration and response spectral ordinates,
Earth-Science Reviews 61,43–104.
42 STEWART ET AL.
Douglas, J., 2011. Ground Motion Prediction Equations 1964–2010, Rpt PEER 2011/102,
Pacific Earthquake Engineering Research Center, University of California, Berkeley, CA.
Douglas, J., and Mohais, R., 2009. Comparing predicted and observed ground motions from
subduction earthquakes in the Lesser Antilles, Journal of Seismology 13, 577–587.
Douglas, J., Bungum, H., and Scherbaum, F., 2006. Ground-motion prediction equations for
southern Spain and southern Norway obtained using the composite model perspective, Journal
of Earthquake Engineering 10,33–72.
Douglas, J., Faccioli, E., Cotton, F., and Cauzzi, C., 2009. Selection of Ground-Motion Prediction
Equations for GEM1, GEM Technical Report, GEM Foundation, Pavia, Italy.
Douglas, J., Cotton, F., Di Alessandro, C., Boore, D. M., Abrahamson, N. A., and Akkar, S.,
2012. Compilation and critical review of GMPEs for the GEM-PEER global GMPEs
project, in Proceedings, 15th World Conference on Earthquake Engineering,24–28
September 2012, Lisbon, Portugal.
Electric Power Research Institute, 1993. Guidelines for Determining Design Basis Ground
Motions, Volumes 1–5, EPRI TR-102293, EPRI, Palo Alto, CA.
Electric Power Research Institute, 2006. Program on Technology Innovation: Truncation of the
Lognormal Distribution and Value of the Standard Deviation for Ground Motion Models in
the Central and Eastern United States, MD: 2006. 1014381, EPRI, Palo Alto, CA and U.S.
Department of Energy, Germantown, PA.
Faccioli, E., Bianchini, A., and Villani, M., 2010. New ground motion prediction equations
for T>1s and their influence on seismic hazard assessment, Proceedings of the University
of Tokyo Symposium on Long-Period Ground Motion and Urban Disaster Mitigation,
17–18 March, Tokyo, Japan.
Frankel, A., Mueller, C., Barnhard, T., Perkins, D., Leyendecker, E. V., Dickman, N., Hanson, S.,
and Hopper, M., 1996. National Seismic-Hazard Maps: Documentation June 1996, Open-File
Report 96-532, U.S. Department of the Interior, U.S. Geological Survey.
Ghasemi, H., Zare, M., Fukushima, Y., and Koketsu, K., 2009. An empirical spectral ground-
motion model for Iran, Journal of Seismology 13, 499–515.
Ghasemi, H., Zare, M., and Fukushima, Y., 2008. Ranking of several ground-motion models
forseismic hazard analysis in Iran, Journal of Geophysics and Engineering 5,301–310.
Goulet, C. A., and Stewart, J. P., 2009. Pitfalls of deterministic application of nonlinear site fac-
tors in probabilistic assessment of ground motions, Earthquake Spectra 25,541–555.
Gupta, I. D., 2010. Response spectral attenuation relations for in-slab earthquakes in Indo-
Burmese subduction zone, Soil Dynamics and Earthquake Engineering 30,368–377.
Kaklamanos, J., and Baise, L.G., 2011. Model validations and comparisons of the next generation
attenuation of ground motions (NGA–West) project, Bulletin of the Seismological Society of
America 101, 160–175.
Kaklamanos, J., Baise, L. G., and Boore, D. M., 2011. Estimating unknown input parameters
when implementing the NGA ground-motion prediction equations in engineering practice,
Earthquake Spectra 27,1219–1235.
Kanno, T., Narita, A., Morikawa, N., Fujiwara, H., and Fukushima, Y., 2006. A new attenuation
relation for strong ground motion in Japan based on recorded data, Bulletin of the Seismolo-
gical Society of America 96, 879–897.
Liao, Y., and Meneses, J., 2013. Comparison of ground motions from the 2010 Mw 7.2 El
Mayor–Cucapah earthquake with the next generation attenuation ground motion prediction
equations, Bulletin of Earthquake Engineering 11,93–106.
SELECTION OF GMPEs FOR THE GLOBAL EARTHQUAKE MODEL 43
Lin, P.-S., and Lee, C.-T., 2008. Ground-motion attenuation relationships for subduction-
zone earthquakes in northeastern Taiwan, Bulletin of the Seismological Society of America
98,220–240.
Margaris, G. A., Mylonakis, G., Papaioannou, C., Klimis, N., Theodulidis, N., Savvaidis, A.,
Efthymiadou, V., and Stewart, J. P., 2010. The 8 June 2008 Mw6.4 Achaia–Elia, Greece
earthquake: Source characteristics, ground motions, and ground failure, Earthquake Spectra
26,399–422.
McVerry, G. H., Zhao, J. X., Abrahamson, N. A., and Somerville, P. G., 2006. New Zealand
acceleration response spectrum attenuation relations for crustal and subduction zone
earthquakes, Bulletin of the New Zealand Society for Earthquake Engineering 39,1–58.
Nishimura, N., 2010. Conformity of the attenuation relationships in Japan with those by the
NGA-project, Summaries of Technical Papers of Annual Meeting of Architectural Institute
of Japan, Hokuriku, Japan.
Pezeshk, S., Zandieh, A., and Tavakoli, B., 2011. Hybrid empirical ground-motion prediction
equations for eastern North America using NGA models and updated seismological para-
meters, Bulletin of the Seismological Society of America 101, 1859–1870.
Raghu Kanth, S. T. G., and Iyengar, R. N., 2007. Estimation of seismic spectral acceleration in
peninsular India, Journal of Earth System Science 116,199–214.
Scasserra, G., Stewart, J. P., Bazzurro, P., Lanzo, G., and Mollaioli, F., 2009. A comparison of
NGA ground-motion prediction equations to Italian data, Bulletin of the Seismological Society
of America 99, 2961–2978.
Scherbaum, F., Delavaud, E., and Riggelsen, C., 2009. Model selection in seismic hazard ana-
lysis: An information‐theoretic perspective, Bulletin of the Seismological Society of America
99, 3234–3247.
Scherbaum, F., Cotton, F., and Smit, P., 2004. On the use of response spectral reference data for
the selection and ranking of ground motion models for seismic hazard analysis in regions of
moderate seismicity: The case of rock motion, Bulletin of the Seismological Society of America
94, 2164–2185.
Seyhan, E., and Stewart, J. P., 2012. Site response in NEHRP Provisions and NGA models,
in Geotechnical Engineering State of the Art and Practice, Volume of Keynote Lectures
from GeoCongress 2012, Oakland, CA, ASCE Geotechnical Special Publication No. 226
(K. Rollins and D. Zekkos, Eds.), ASCE, Reston, VA, 359–379.
Seyhan, E., and Stewart, J. P., 2014. Semi-empirical nonlinear site amplification from NGA-West2
data and simulations, Earthquake Spectra 30, 1241–1256.
Shoja-Taheri, J., Naserieh, S., and Ghofrani, H., 2010. A test of the applicability of NGA models
to the strong ground motion data in the Iranian Plateau, Journal of Earthquake Engineering 14,
278–292.
Silva, W. J., Gregor, N., and Darragh, R., 2002. Development of Regional Hard Rock Attenuation
Relations for Central and Eastern North America, Tech. Rep., Pacific Engineering and
Analysis, El Cerrito, CA.
Somerville, P., Graves, R., Collins, N., Song, S. G., Ni, S., and Cummins, P., 2009. Source and
ground motion models of Australian earthquakes, in Proceedings, 2009 Annual Conference of
the Australian Earthquake Engineering Society,3–5 April 2009, Christchurch, New Zealand.
Stafford, P. J., Strasser, F. O., and Bommer, J. J., 2008. An evaluation of the applicability of the
NGA models to ground motion prediction in the Euro-Mediterranean region, Bulletin of
Earthquake Engineering 6,149–177.
44 STEWART ET AL.
Stewart, J. P., Douglas, J., Javanbarg, M. D., Di Alessandro, C., Bozorgnia, Y., Abrahamson,
N. A., Boore, D. M., Campbell, K. W., Delavaud, E., Erdik, M., and Stafford, P. J., 2013a.
GEM-PEER Task 3 Project: Selection of a global set of ground motion prediction equations,
Report No. 2013/22, Pacific Earthquake Engineering Research Center, Berkeley, CA.
Stewart, J. P., Midorikawa, S., Graves, R. W., Khodaverdi, K., Kishida, T., Miura, H., Bozorgnia,
Y., and Campbell, K. W., 2013b. Implications of Mw 9.0 Tohoku-oki Japan earthquake
for ground motion scaling with source, path, and site parameters, Earthquake Spectra 29,
S1–S21.
Toro, G. R., 2002. Modification of the Toro et al. (1997) Attenuation Equations for Large
Magnitudes and Short Distances, Technical Report, Risk Engineering, Inc., Louisville,
CO, available at http://www.riskeng.com/downloads/attenuation_equations.
Toro, G. R., Abrahamson, N. A., and Schneider, J. F., 1997. Model of strong ground motions
from earthquake in central and eastern North America: Best estimates and uncertainties,
Seismological Research Letters 68,41–57.
Uchiyama, Y., and Midorikawa, S., 2011. A study of the applicability of NGA models to strike-
slip earthquakes in Japan, Summaries of Technical Papers of Annual Meeting of Architectural
Institute of Japan, Kanto, Japan.
Vilanova, S. P., Fonseca, J. F. B. D., and Oliveira, C. S., 2012. Ground-motion models for seismic
hazard assessment in western Iberia: Constraints from instrumental data and intensity obser-
vations, Bulletin of the Seismological Society of America 102, 169–184.
Youngs, R. R., Chiou, S.-J., Silva, W. J., and Humphrey, J. R., 1997. Strong ground motion
attenuation relationships for subduction zone earthquakes, Seismological Research Letters
68,58–73.
Zhao, J. X., Zhang, J., Asano, A., Ohno, Y., Oouchi, T., Takahashi, T., Ogawa, H., Irikura, K.,
Thio, H. K., Somerville, P. G., and Fukushima, Y., 2006. Attenuation relations of strong
ground motion in Japan using site classification based on predominant period, Bulletin of the
Seismological Society of America 96, 898–913.
(Received 30 January 2013; accepted 11 March 2013)
SELECTION OF GMPEs FOR THE GLOBAL EARTHQUAKE MODEL 45
