BommerAcevedo JEE 2004

See discussions, stats, and author profiles for this publication at: https://www.researchgate.
net/publication/233460611
The use of real accelerograms as input to dynamic analysis
Article in Journal of Earthquake Engineering · January 2004

DOI: 10.1080/13632460409350521
CITATIONS READS
469 3,656
2 authors:
Julian Bommer Ana Beatriz Acevedo

Imperial College London Universidad EAFIT
226 PUBLICATIONS 16,454 CITATIONS 17 PUBLICATIONS 684 CITATIONS
SEE PROFILE SEE PROFILE
Some of the authors of this publication are also working on these related projects:
Thyspunt SSHAC Level 3 PSHA View project
SARA project GEM Foundation View project
All content following this page was uploaded by Julian Bommer on 12 January 2015.
The user has requested enhancement of the downloaded file.

May 18, 2004 10:44 WSPC/124-JEE 00159
Journal of Earthquake Engineering,

Vol. 8, Special Issue 1 (2004) 43–91
c Imperial College Press
THE USE OF REAL EARTHQUAKE ACCELEROGRAMS AS

INPUT TO DYNAMIC ANALYSIS
JULIAN J. BOMMER
Department of Civil and Environmental Engineering,
Imperial College London, South Kensington campus,
London SW7 2AZ, UK
j.bommer@imperial.ac.uk
ANA BEATRIZ ACEVEDO

ROSE School, Collegio Alessandro Volta,
Via Ferrata 17, Pavia 27100, Italy
The increasing availability of strong-motion accelerograms, and the relative ease with
which they can be obtained compared to synthetic or artificial records, makes the use of
real records an ever more attractive option for defining the input to dynamic analyses
in geotechnical and structural engineering. Guidelines on procedures for the selection
of appropriate suites of acceleration time-series for this purpose are lacking, and seis-
mic design codes are particularly poor in this respect. Criteria for selecting records in
terms of earthquake scenarios and in terms of response spectral ordinates are presented,
together with options and criteria for adjusting the selected accelerograms to match
the elastic design spectrum. The application of both geophysical and response spectral
search criteria is illustrated using compatible scenarios, and the selected records are
analysed and adjusted to produce suites of acceleration time-series suitable for dynamic
analyses. The paper concludes with suggestions for making use of real records in engi-
neering analysis and design, and recommendations are given for improving the current
guidelines provided in seismic design codes.
Keywords: Strong-motion records; dynamic analysis; strong-motion databank; strong-

motion database; spectral matching; seismic design codes.
1. Introduction
For earthquake-resistant design and for seismic assessment of existing structures,
the earthquake-induced ground shaking is generally represented in the form of a
response spectrum of acceleration or displacement. The spectrum used as input to
equivalent lateral force or spectral modal methods of analysis is usually obtained by
scaling an elastic spectrum by factors that account for, amongst other phenomena,
the influence of inelastic structural response. There are, however, situations in which
the simulation of structural response using a scaled elastic response spectrum is not
considered appropriate, and fully dynamic analysis is required. These situations may
include the following: buildings designed for a high degree of ductility; structures
with configuration in plan or elevation that is highly irregular; structures for which
43
May 18, 2004 10:44 WSPC/124-JEE 00159
44 J. J. Bommer & A. B. Acevedo
higher modes are likely to be excited; critical structures, the failure of which would
cause unacceptable harm or disruption; structures with special features, such as
base isolation. Faced with these special situations, the engineer will generally have
to employ time-history analysis, for which the requirements are an appropriate non-
linear model for the structure and a suitable suite of accelerograms to represent the
seismic excitation.
A workshop on improving the characterisation of earthquake ground motion
held in 1997 [ATC, 1999] reflected the importance of the issue of defining accelero-
grams for engineering design in its first conclusion, which recommended to “develop
guidelines for generating and selecting time histories that can be used by the prac-
tising engineer in seismic analysis and design of facilities”. Nonetheless, there is
relatively little published technical literature on the subject of selecting and scaling
real strong-motion records for design, and this paper therefore attempts to present
the issues involved and offer some insights as well as some guidance for engineers.
There are three basic options available to the engineer in terms of obtaining
acceleration time-series. The first is to use artificial spectrum-compatible accelero-
grams generated using programs such as SIMQKE [Gasparini and Vanmarcke,
1979]. The approach employed in SIMQKE is to generate a power spectral den-
sity function from the smoothed response spectrum, and then to derive sinusoidal
signals having random phase angles and amplitudes. The sinusoidal motions are
then summed and an iterative procedure can be invoked to improve the match with
the target response spectrum, by calculating the ratio between the target and actual
response ordinates at selected frequencies; the power spectral density function is
then adjusted by the square of this ratio, and a new motion generated.
The attraction of such an approach is obvious because it is possible to obtain ac-
celeration time-series that are almost completely compatible with the elastic design
spectrum (Fig. 1), which in some cases will be the only information available to the
design engineer regarding the nature of the ground motions to be considered. How-
ever, it is now widely accepted that the use of such artificial records, particularly for
non-linear analyses, is problematic. The basic problem with spectrum-compatible
artificial records is that they generally have an excessive number of cycles of strong
motion and consequently they possess unreasonably high energy content. Here it
is necessary to discuss terminology, since the adjective “artificial” is also applied
(sometimes with the additional qualifier of “intelligent”) to the outcome of applying
selective adjustments to real accelerograms, using techniques that are discussed in
Sec. 4.2. In this paper, the term “artificial” is used exclusively for records such as
those shown in Fig. 1. These types of records are not considered to be suitable for
use in non-linear analyses. In addition to the problems associated with how these
artificial records are generated, there can also be difficulties that arise from match-
ing the acceleration time-series to the entire elastic design spectrum. The latter
will generally be a uniform hazard spectrum (UHS), including in seismic design
codes, obtained from probabilistic seismic hazard assessment (PSHA), and there-
fore enveloping the ground motions from several seismic sources [e.g. Reiter, 1990;
May 18, 2004 10:44 WSPC/124-JEE 00159
The Use of Real Earthquake Accelerograms 45
1.2
Average of time-series
1
Code spectrum
Spectral acceleration (g)
0.8
0.6
0.4
0.2
0
0 1 2 3 4
Period (seconds)
Acceleration (g)
0.4
0.2
0
-0.2
-0.4
0 5 10 15 20
Acceleration (g)
0.4
0.2
0
-0.2
-0.4
0 5 10 15 20
Acceleration (g)
0.4
0.2
0
-0.2
-0.4
0 5 10 15 20
Time (seconds)
Fig. 1. Artificial accelerograms generated to match the S1 soil category elastic response spectrum
from the French seismic design code; the uppermost plot compares the average ordinates of the
three spectra with the code spectrum.
May 18, 2004 10:44 WSPC/124-JEE 00159
Bommer et al., 2000]. Naeim and Lew [1995] assert that: “To generate an accelera-
tion time-history to be compatible to a PSHA-generated design spectrum is neither
reasonable nor realistic”. Certainly it is the case that if the UHS is strongly influ-
enced by more than one source of seismicity, for example by small, local earthquakes
and by distant, large magnitude events, spectrum-compatible artificial records will
tend to be particularly unrealistic.
The second category of ground-motion records available to the engineer is syn-
thetic accelerograms generated from seismological source models and accounting
for path and site effects. These models range from point source stochastic simula-
tions through their extension to finite sources, to fully dynamic models of stress
release, although the latter are still under development. Programs for some of the
many methods of ground-motion generation that have been developed [e.g. Zeng et
al., 1994; Beresnev and Atkinson, 1998; Boore, 2003] are freely available, but their
application, in terms of defining the many parameters required to characterise the
earthquake source, will generally require the engineer to engage the services of
specialist consultant in engineering seismology. The determination of the source
parameters for previous earthquakes invariably carries a high degree of uncertainty,
and the specification of these parameters — to which the resulting ground motions
can be highly sensitive — for future earthquake scenarios can involve a significant
degree of expert judgement.
The third category of records is real accelerograms recorded during earthquakes,
which by definition are free from the problems associated with artificial spectrum-
compatible records. Real strong-motion records are now easily accessible in large
numbers and their retrieval and manipulation is relatively straightforward, whence
the design engineer will often be able to prepare a suite of records without the ser-
vices of an engineering seismologist. This paper provides an overview of the issues
involved in preparing suites of real records for use in dynamic analyses, and exam-
ines different procedures for the selecting and scaling of the records. The following
section provides an overview of the sources from which strong-motion data are now
available, including some assessment of the coverage of different earthquake scenar-
ios and the ease with which each source allows the user to perform searches. The
two sections that follow deal with the issues of how records are selected and how
they can be scaled to match, in some specified sense, the elastic design spectrum.
The penultimate section of the paper then explores all of these issues through two
approximately compatible searches, one using an earthquake scenario and the other
using direct matching to a code spectrum, and through the application of differ-
ent selection and scaling procedures to the suites of accelerograms obtained from
each search. The paper closes with simple guidelines for the selection and scaling
of real strong-motion records, and discusses how these might be incorporated into
seismic design codes.
Before closing this section, mention should be made of synthetic accelerograms
generated using empirical Green’s functions, which are effectively a hybrid of the
second and third categories of acceleration time-series.
May 18, 2004 10:44 WSPC/124-JEE 00159
2. Availability of Strong-Motion Accelerograms

The operators of strong-motion recording networks generally produce reports
presenting the records obtained from their accelerographs, either on a periodic
basis or following a significant earthquake. Others have produced compendia of
records from one or more networks in a given region or country. One of the first
examples of such reports was the CALTECH (California Institute of Technology)
volumes produced after the 1971 San Fernando earthquake in California. These
volumes, which accompanied magnetic tapes on which the digitised records and
associated response spectra were distributed, were an important landmark in mak-
ing strong-motion records more widely available to both researchers and practicing
engineers. A useful distinction can be made between a collection of digitised ac-
celerograms and a catalogue of associated information about the earthquakes and
the recording stations from which the accelerograms were obtained, as well as about
the records themselves: the former is a strong-motion databank, the latter a strong-
motion database [Bommer and Ambraseys, 1992]. To facilitate the use of strong-
motion records in engineering analysis and design, the practicing engineer requires
both an extensive databank of accelerograms and access to a database of reliably
determined parameters in order to select appropriate recordings.
2.1. Global databanks

Amongst the first efforts to compile a global catalogue of earthquake accelerograms
was the databank presented in a series of reports by Leeds [1992]. The reports
listed about 400 horizontal component records from shallow earthquakes in western
USA and a similar number from earthquakes in Alaska, the Cascadia subduction
zone and the rest of the world, mainly coming from Japan and Mexico. In or-
der to facilitate selection, the accelerograms were organised into bins according to
magnitude ranges, focal depth and site classification, the latter distinguishing only
between hard and soft sites. Epicentral and hypocentral distances were reported
for each record. Another extensive catalogue of strong-motion records, almost ex-
clusively from North America (including Mexico) was presented by Naeim and
Anderson [1996]. The report presented 1470 horizontal component records and 527
vertical components, but the focus adopted by the authors was to make the records
accessible to engineers by providing listings of strong-motion parameters, includ-
ing peak amplitudes, elastic and inelastic spectral ordinates, and durations; the
only seismological parameters given were magnitude (for which the largest value
reported by the USGS was given, which resulted in a mixture of Mw , Ms and ML ,
amongst others), focal depth and hypocentral distance.
In recent years, several strong-motion databanks and databases have been is-
sued and distributed on CD-ROM, which has been another important development
in making accelerograms more widely available to end-users. Several strong-motion
network operators have produced CD-ROM collections of their own records, includ-
ing agencies in Japan and Mexico, and others have been issued with the recordings
May 18, 2004 10:44 WSPC/124-JEE 00159
from a particular earthquake, such as the 1999 Chi-Chi event in Taiwan [Lee et al.,
2001]. The USGS (United States Geological Survey) issued an important collection
of almost 1500 accelerograms from 500 earthquakes recorded by ground-level in-
struments in North and Central America between 1933 and 1986, with information
regarding the earthquake and recording station provided in the header of each com-
ponent file. The project was presented as an update and consolidation of the work
originally made available via the CALTECH volumes [Seekins et al., 1992]. The
global strong-motion databank compiled by the National Geophysical Data Center
(NGDC), which was distributed as a 3-volume CD-ROM in 1996, contains 15000
individual component records from about 1000 earthquakes around the world up
to 1994 [Row, 1996]. Although the NGDC databank makes a very large number of
records available to users, the parameters in the database have not been uniformly
re-evaluated and this limits its use as a selection tool: only epicentral distances are
given for the records and various magnitudes are reported, with a quarter of the
earthquakes having either a magnitude of zero (presumably implying no value is
available) or a value on an unspecified scale.
In this respect, some smaller collections of data may be of greater use simply
because they provide more complete and more uniformly determined source, path
and site parameters for the accelerograms. Work has been ongoing for many years to
determine uniform parameters associated with strong-motion records from Europe
and the Middle East [Ambraseys and Bommer, 1990, 1991] and in 2000 a CD-ROM
of European Strong-Motion Data was issued and distributed as a result of a Eu-
ropean Union-funded project [Ambraseys et al., 2000]. The CD-ROM includes just
over 1000 accelerograms from more than 400 earthquakes, with a database of associ-
ated parameters including uniformly calculated Joyner-Boore distances [Abraham-
son and Shedlock, 1997] for nearly all records from earthquakes of magnitude 6 or
greater; for smaller earthquakes, generally only epicentral distance is provided, but
for such events the two measures are comparable. The style-of-faulting is known for
more than half of the records, and the site classification for more than 80% of the
records, although the reliability of the information on which the latter is based is
highly variable. The CD-ROM allows the user to search records in terms of different
combinations of parameters such as magnitude, distance and site classification, and
peak ground acceleration (PGA) can also be used as a search parameter. Another
useful collection, in which the data (exclusively from soft rock and stiff soil sites in
western US) has effectively been pre-searched to be presented in magnitude-distance
bins, is available on the CD-ROMs accompanying NUREG/CR-6728 [McGuire
et al., 2001].
2.2. Internet sites

The most significant development in strong-motion data dissemination in recent
years is the creation of several Internet sites from which users can search and down-
load accelerograms in digital form [Wald, 1997]. Many operators of accelerograph
May 18, 2004 10:44 WSPC/124-JEE 00159
networks maintain their own web sites, in some cases allowing users to download
digitised records. An excellent Internet site for obtaining Japanese strong-motion
data is the K- Net site at http://www.k-net.bosai.go.jp/k-net/index en.shtml. There
are also a number of sites that provide data from several networks, although these
vary in the degree of access that they actually provide to the digital strong-motion
records. The NGDC site (http://www.ngdc.noaa.gov ) allows users to search the
database mentioned in the previous sub-section, but the records can only be ob-
tained from the CD-ROM collection. The databank of accelerograms from Eu-
rope and the Middle East, containing almost three times as many records as
were available on the CD-ROM described above, can now be searched via the In-
ternet Site for European Strong-Motion Data (ISESD) launched in March 2002
(http://www.isesd.cv.ic.ac.uk ) [Ambraseys et al., 2003]. Figure 2 shows the distri-
bution of the strong-motion records in the European Internet site with respect to
magnitude, distance and site classification; it can be appreciated that although the
databank is extensive, the majority of the records actually correspond to earth-
quakes that are unlikely to be of engineering significance, given that the threshold
magnitude considered worthy of consideration for engineering purposes is generally
taken to be about 5.
Two other important websites for accessing strong-motion data are COSMOS
and PEER. The COSMOS website (http://db.cosmos-eq.org) contains a databank
of more than 4000 freely available records from around the world, 40% of which are
from western US, 20% from Japan and about 18% from New Zealand, the main
objective of the website being to make as many records as possible available to
users [Stepp, 2000]. Simple searches can be performed in terms of ranges of mag-
nitude, distance and PGA, as well as by region. Moment magnitudes are provided
for almost half of the earthquakes in the database; distances can be searched as
hypocentral or distance from the fault rupture, but the latter is provided for a
much smaller proportion of the data. Advanced searches can be performed in terms
of several other parameters, including mechanism, rake angle, site geology, peak
ground velocity (PGV), and spectral ordinates at a few response periods, although
these parameters are not provided for all records.
The PEER databank (http://peer.berkeley.edu/smcat) includes 1557 records
from 143 earthquakes in tectonically active regions, for which the time-histories
and response spectra for different damping ratios can be downloaded. The distribu-
tion of the records in the PEER databank with respect to magnitude, distance and
site classification is shown in Fig. 3. The PEER database reports Mw , Ms and ML
for earthquakes, with 90%, 85% and 78% of the records having a value on each of
the scales, respectively. Distances are reported using three different metrics, these
being the closest distance to the fault rupture, hypocentral distance and the Joyner-
Boore distance. The proportions of the records for which each distance is given are
80% for Rrup , 15% for Rhyp and 48% for Rjb (see Abrahamson and Shedlock [1997]
for distance definitions). The site geology at the recording stations is classified ac-
cording to two different schemes, one attributed to the USGS, using four classes
May 18, 2004 10:44 WSPC/124-JEE 00159
6
Magnitude
2 rock
stiff soil
alluvium
soft soil
unknown
1 10 100
Distance (km)
Fig. 2. Distribution of European strong-motion databank with respect to magnitude, source-to-
site distance, and site classification.
whose limits are defined by Vs,30 values of 750, 360, and 180 m/s, which are the
values used by Boore et al. [1997] based on the NEHRP classification scheme, and
the other being either the Geomatrix scheme or the CWB classification for stations
in Taiwan; the Geomatrix scheme includes five categories, the stiffest with shear
wave velocities above 600 m/s, the softest those with less than 150 m/s [Abraham-
son and Silva, 1997]. 65% of the records are classified in terms of USGS scheme,
and 84% in terms of the Geomatrix or CWB schemes.
The PEER database lists some earthquakes for which the digitised records are
not actually available at the site, most of these corresponding to European events.
May 18, 2004 10:44 WSPC/124-JEE 00159
Magnitude 7
5
rock
intermediate
soft
unknown
4
1 10 100
Distance (km)
Fig. 3. Distribution of PEER strong-motion databank with respect to magnitude, source-to-site
distance, and site classification. Magnitude is assigned according to the following order of priority:
Mw , Ms , ML , and distances as Rrup , Rjb , Rhyp . Rock sites are those classified as USGS or
Geomatrix class A or CWB class 1, intermediate sites are class B or C or 2 in the CWB scheme,
and others are soft.
The COSMOS site also includes very few accelerograms from Europe, which makes
the ISESD a useful complement to the COSMOS and PEER sites.
Using the PEER database, searches can be performed in terms of magnitude,
distance, site classification, rupture mechanism, PGA, PGV and peak ground dis-
placement (PGD), or alternatively in terms of the maximum spectral acceleration
in a user-specified period range. In terms of search capabilities, provided one has
access to a large databank, the optimum approach — as illustrated in Sec. 5 —
can be to use both seismological and response spectral criteria simultaneously. In
a follow-up to the ISESD website, a new CD-ROM is to be distributed in early
2004, which will allow users to execute searches using a wide range of possible
combinations of parameters related to the characteristics of the earthquake source,
the source-to-site path, and the site itself, as well as in terms of strong-motion
parameters, including response spectral ordinates [Ambraseys et al., 2004].
A final point worthy of note with regard to the strong-motion records that can
be obtained either from CD-ROM collections or downloaded from Internet sites, is
with respect to the processing applied to the signals. The problems associated with
distortion of high-frequency components of motion due to instrument response, and
more importantly with baseline errors and long-period noise in digitised analogue
strong-motion recordings, are well known [Trifunac et al., 1973; Hudson, 1979].
May 18, 2004 10:44 WSPC/124-JEE 00159
The problems are reduced with digitally recorded accelerograms, but by no means
eliminated [e.g. Boore et al., 2002]. For any application that requires displacement
time-series, such as asynchronous analysis of bridges, and the analysis of long-
period structures, the issues of baseline errors and long-period noise can become
particularly significant. The COSMOS site offers records as contributed by net-
work operators, in uncorrected and/or corrected format, but often no details are
provided of the correction procedures applied. Important exceptions to this are
records supplied by the USGS and CSMIP (Californian Strong-Motion Instrumen-
tation Program), for which processing details are contained in the record headers.
Nearly all records on the ISESD site are available in both uncorrected and cor-
rected formats, except for those cases where only corrected records were provided
by the network operators. The remaining records have all been corrected by the
subtraction of a linear baseline and band-passed filtered using an elliptical filter
[Sunder and Connor, 1982] with cut-off frequencies of 0.25 and 25 Hz; no instru-
ment correction has been applied. Where uncorrected records are available, users
may apply their own preferred correction procedures. The accelerograms available
at the PEER web site are all in corrected format, with most of the records having
been individually processed by Dr. Walter Silva, using a causal Butterworth filter,
with cut-off frequencies based on inspection of the Fourier amplitude spectrum and
the integrated displacement time-series, and a correction for instrument response.
For applications where long-period response and ground displacements are impor-
tant, the PEER records are likely to be an attractive choice since there is some
degree of confidence in the displacement records (reflected in the fact that PGD
is offered as a search parameter), which may not be true for the corrected records
from the COSMOS and ISESD sites.
3. Criteria for Selecting Strong-Motion Records

The way in which records can be chosen is to a large extent dictated by the infor-
mation available to the engineer regarding the seismic hazard or the design ground
motions at the site of interest. Figure 4 provides an overview of the different options
that are available.
Guidance on this topic in the literature is very limited, and in seismic design
codes very little useful guidance is given on how appropriate records should be
selected. Bommer and Ruggeri [2002] considered 33 current or recent codes for the
seismic design of buildings, and identified that only eight of these specify time-
history analysis to be compulsory (under certain specified circumstances); most
codes allow dynamic analysis and even those that require it often specify the use
of spectral modal analysis rather than direct integration techniques. This partially
explains why guidelines on preparing ground-motion input for full dynamic analysis
are not well developed. Indeed, some codes seem to consider this an issue outside
their scope: the 1984 Indian code simply informs the user of the option of using
“time-history analysis based on expected ground motion for which special studies are
May 18, 2004 10:44 WSPC/124-JEE 00159
Fig. 4. Overview of the options available for selecting accelerograms to be used in engineering
analysis and design.
required”. All codes that discuss the application of acceleration time-histories allow
the use of real records, with the exception of the Portuguese code that specifies only
spectrum-compatible artificial motions. Some codes, amongst them UBC 1997 and
IBC 2000, favour real records but allow the design engineer to supplement these
with simulated motions when sufficient suitable real records cannot be found.
More than half of the design codes reviewed do not specify the critical issue of
the number of records to be selected; amongst those that do, the most commonly
encountered figure is three. This is the number specified both in UBC 1997 and
IBC 2000, but both of these codes have the provision that if only three records are
used in the analyses, the maximum structural response must be used, whereas if
seven or more are used, the average response may be used; the same specification is
made in EC8. Other variations exist, such as the stipulations presented in ISO/DIS
19901-2 [ISO, 2003] for the seismic design of offshore structures, which specifies
that a minimum of four time-histories should be used “to capture the randomness
in a seismic event” and that the structure must be demonstrated to survive under
four or half of the time-histories, whichever is the greater. The issue of the number
of records to be used in dynamic analyses is discussed further in Sec. 4.1.
May 18, 2004 10:44 WSPC/124-JEE 00159
An important implicit assumption underlying the recommendations made in this

paper is that provided similar tectonic environments are considered, strong-motion
records from one country can be selected and applied in another. The three basic
categories of tectonic regions that should be considered, for matching the design
situation and the selected recordings, are subduction zones, active crustal regions
and stable continental regions.
3.1. Selection in terms of strong-motion parameters

Guidance given in seismic design codes on how to select appropriate real records is
usually focused on compatibility with the response spectrum rather than seismolog-
ical parameters, for the simple reason that the information on seismic source zones
and activity rates that underlie zonation maps is not presented and only the uni-
form hazard spectrum (UHS) is given. In current codes, earthquakes are effectively
invisible and for this reason the engineer using the code will not easily be able to
identify scenario earthquakes. This gives rise to the use of generally rather vague
specifications such as that encountered in the 1995 Greek seismic code, which states
that the selected records “must be representative of the ground motion at the site
and must be recorded at a consistent source-site distance”. An important exception
to the general shortcomings vis-à-vis seismic design codes is represented by the
dashed arrow in Fig. 4, which corresponds to the unique situation in the United
States. Since the zonation map and uniform hazard spectrum in IBC 2000 are
closely based on the USGS hazard maps, users can obtain disaggregations — and
even suites of hazard-consistent stochastically generated acceleration time-series —
from the USGS website at http://eqint.cr.usgs.gov/eq/html/deaggint.html.
Where specific criteria for selecting records are provided in seismic codes, they
are generally based on the ordinates of the elastic design spectrum, although some,
notably the Spanish code, only specify a match with PGA. Some codes do not spec-
ify the relationship between the selected records and the elastic design spectrum,
but rather specify that the base shear obtained from dynamic analysis should not
be lower than a certain proportion — usually between 0.7 and 0.9 — of that ob-
tained using the equivalent lateral force method, which does not actually help the
engineer in making the initial selection of records.
Most of the codes that give some guidance on the preparation of suites of accel-
eration time-series to be used as input to dynamic analyses specify conditions that
the records must meet with respect to the ordinates of the elastic design spectrum.
These matching criteria are discussed in Sec. 4.1. In order to implement searches
that will produce records likely to meet the spectral matching criteria, or at least to
do so with a minimum of manipulation of the records, it is useful to have a tool
that allows records to be searched on the basis of the spectral ordinates. Such
a tool is included in the new European strong-motion data CD-ROM discussed
previously [Ambraseys et al., 2004], which allows records to be searched by match-
ing the spectral shape to the shape of the design spectrum. The search is based on
May 18, 2004 10:44 WSPC/124-JEE 00159
the average root-mean-square deviation of the observed spectrum from the target
design spectrum:
v
uN 2
1 uX SAo (Ti ) SAs (Ti )
Drms = t − , (1)
N i=1 PGAo PGAs
where N is the number of periods at which the spectral shape is specified, SAo (Ti )
is the spectral acceleration from the record at period Ti , SAs (Ti ) is the target
spectral acceleration at the same period; PGAo and PGAs are the peak ground
acceleration of the record and the zero-period anchor point of the target spectrum,
respectively. The smaller the value of Drms the closer the match between the shape
of the record and target spectrum; the value specified will depend on the extent of
the databank being accessed and the number of records required. Smaller values
of Drms can be specified if the spectral matching is being done at short rather
than longer spectral response periods. Making searches on a database of about
7000 accelerograms held in the Imperial College London strong-motion archive,
it was found that to return less than about 30 accelerograms, values of Drms of
the order of 0.15 were needed for matching ordinates in the period range of 0.4–
0.8 second, whereas values as low as 0.06–0.07 could be used for matching the
spectral ordinates from 0.1 to 0.3 second [Bommer et al., 2003a]. By simultaneously
specifying an acceptable match with the design PGA, the search then matches the
record and target spectrum in the specified period range. This procedure is superior
to matching on the basis of spectrum intensities (area below the response spectrum)
in the specified period range, because a good match in that case could easily be
obtained with the record having ordinates significantly above the target spectrum
at one period and significantly below at another. The procedure proposed effectively
limits the maximum deviation of individual peaks or troughs on the spectrum from
the target ordinates.
The most serious limitation with any selection procedure based solely on the
ordinates of the elastic spectrum is that the records obtained can have very different
durations. If the starting point for the selection is a seismic design code, in which the
earthquake actions are represented by an elastic response spectrum of acceleration,
the duration of the design ground motions will generally not be specified. Amongst
the 33 seismic design codes reviewed by Bommer and Ruggeri [2002], only six
specify duration criteria, and only two of these — the codes of France and Turkey
— actually specify how the duration is to be measured, an important issue given
that there are more than 30 different definitions of strong-motion duration in the
technical literature [Bommer and Martı́nez-Pereira, 1999, 2000]. This problem is
not easily overcome because no code currently includes a map of hazard in terms
of duration of shaking and without knowledge of earthquake magnitude, it is very
difficult to estimate this parameter. The absence of suitable criteria can also lead to
unrealistic specifications. One code that provides a great deal of information about
the criteria that the records should fulfil is the 1990 French code, although these are
May 18, 2004 10:44 WSPC/124-JEE 00159
so demanding that it is almost impossible to obtain realistic ground motions that

satisfy them; the artificial accelerograms in Fig. 1 were generated to satisfy these
requirements, but no real accelerograms could be found that also met the criteria.
One criterion in the French code that makes it so difficult to obtain compatible real
records is the specification of a minimum duration of 20 seconds for real records.
The 1988 Iranian seismic design code makes the selection of input accelero-
grams for dynamic analysis very easy, specifying use of the Naghan record of the
1977 Ardal (Ms 6) earthquake and the Tabas record of the 1978 Tabas (Ms 7.3)
earthquake, both recorded within 5 km of the seismic source and with PGA values
in excess of 0.6 g and 1.0 g respectively. Although very simplistic — since no con-
sideration is given as to whether such severe motions could be generated at the site
under consideration — this approach could usefully be adapted by suites of suitable
records being specified in codes for different hazard zones and site categories, as
discussed later.
3.2. Selection in terms of geophysical parameters

If the engineer has at his or her disposal a site-specific seismic hazard assessment,
then the possibilities for selecting suitable records are quite different. If a determin-
istic seismic hazard assessment (DSHA) has been employed, the design earthquake
scenario will be fully defined, at least in terms of the earthquake magnitude, the
distance from the site to the fault rupture, and the nature of the surface geology
at the site (Fig. 4). The search could then be performed directly in terms of these
three parameters, as well as others such as style-of-faulting. If PSHA has been used,
then the controlling earthquake scenarios need to be obtained by disaggregation,
using one of several techniques that have been developed for this purpose [Chap-
man, 1995; McGuire, 1995; Harmsen et al., 1999; Bazzurro and Cornell, 1999].
These techniques yield dominant scenarios contributing to the hazard at different
parts of the response spectrum, defined by a magnitude, distance and number of
logarithmic standard deviations above or below the logarithmic mean from the
ground-motion prediction equation used in the analysis. If the vertical component
of motion is considered important, it should be borne in mind that the control-
ling M-R (magnitude-distance) scenarios for the vertical component motion at the
fundamental period of the structure may be different from those for the horizontal
component, which creates an additional complication for performing the searches.
Once the controlling earthquake scenarios have been identified, then the searches
can be undertaken in the same way as would be the case had a DSHA been carried
out to define the design spectrum for the site. The number of standard deviations
above the median, often specified as ε, will generally not be used as a search param-
eter. The design spectrum will nearly always have ordinates above those predicted
by the median values from attenuation equations using the scenario M-R pair, even
if it is the result of a deterministic assessment since current practice in DSHA is
generally to use the 84-percentile level of motion [e.g. Krinitzsky, 2002]. As a result,
May 18, 2004 10:44 WSPC/124-JEE 00159
the selected records will generally need to be scaled or adjusted to match the design
spectrum, as discussed in the next section.
Clearly, if a search is carried out in terms of an exact match with the design
scenario, for example a normal-faulting earthquake of magnitude Ms 6.4 recorded
at 12 km on a site with a Vs,30 of 470 m/s, it is very unlikely to yield any records.
Therefore, the search must be performed with less restrictive criteria, and for this
reason it is important to decide which parameters should be included in the search
(apart from the tectonic criteria discussed earlier), and for each parameter how
much tolerance should be allowed in the degree of matching between the record
and the scenario.
3.2.1. Earthquake magnitude

Opinions differ about the importance of correctly matching parameters such as
earthquake magnitude: Shome et al. [1998] concluded that provided the records
are scaled to match the elastic design spectrum at the fundamental period of the
structure, then matching the records for the magnitude-distance combination of the
design earthquake scenario is not important. The core of the issue is the degree to
which the duration of shaking influences structural demand, an issue of on-going
debate and investigation. Most studies do specify that magnitude should be a search
parameter, indeed even Shome et al. [1998] recommend in their conclusions that
the user should use “records from roughly the same magnitude”. Others are more
adamant, such as Stewart et al. [2001], who state that it is important to select
records from events of appropriate magnitude because this parameter strongly in-
fluences frequency content and duration of the motion, going on to recommend
selecting records from events within 0.25 units of the target magnitude. Since there
is little doubt that earthquake magnitude exerts a very pronounced influence on du-
ration (or number of cycles) and on the shape of the response spectrum (Fig. 5), we
are of the opinion that it is an indispensable selection parameter, and furthermore
that the match between the record and scenario magnitudes should be close, if pos-
sible within 0.2 magnitude units either side of the target value; this is in agreement
with the proposal of Stewart et al. [2001] but given that magnitude is generally
expressed to the nearest decimal the value is rounded down rather than up. An
objection that has been raised to using such a narrow window of magnitude is that
the interval is comparable to the standard deviation associated with magnitude
determinations, which is generally of the order of about 0.2. However, it has been
pointed out that the standard deviation of the individual station determinations of
magnitude is not really a measure of the uncertainty in the published magnitude
values, and that a better measure would be the standard error of the mean [J. Dou-
glas, personal communication, 2003], obtained by dividing the standard deviation
of the observations by the square root of the number of observations. The standard
error of the mean magnitude estimate is generally an order of magnitude less than
the standard deviation of the station determinations.
May 18, 2004 10:44 WSPC/124-JEE 00159
Although techniques are available to adjust strong-motion records in a way

that allows the spectral shape to be altered, as presented in Secs. 4.2 and 4.3, it
is generally preferable to keep the degree of alteration to a minimum (although
there may be cases, such as records containing strong resonance at a particular
frequency, in which it will be desirable to change the record before using it in
analysis). For this reason, and given the pronounced effect of magnitude on the
shape of the response spectrum, a close match between the scenario and record
magnitudes should be sought.
3.2.2. Source-to-site distance

The second parameter that must be included in defining the search window is
distance. Figure 6 shows normalised spectral shapes, using the same attenuation
equations as in Fig. 5, constructed from median values predicted for rock sites
located at 5, 20 and 50 km from a magnitude 7 earthquake. The spectral shape
appears to be much less sensitive to distance than to magnitude; if the ratios of
the ordinates for 5 and 50 km were calculated from each of the equations, the
average ratio calculated from the four values would be almost invariant with period
(Fig. 7).
1.2 1.2
Ambraseys et al. (1996) Campbell (1997)
1 1
SA(T) / SA(0.2)
SA(T) / SA(0.2)
0.8 0.8
0.6 0.6
M7
0.4 0.4 M7
M 5.5
0.2 0.2 M 5.5
0 0
0 0.5 1 1.5 2 0 0.5 1 1.5 2
Period (seconds) Period (seconds)

1.2 1.2
Boore et al.(1997) Abrahamson & Silva (1997)

1 1
SA(T) / SA(0.2)
SA(T) / SA(0.2)
0.8 0.8
0.6 0.6
0.4 M7 0.4
M7
M 5.5
0.2 0.2
M 5.5
0 0
0 0.5 1 1.5 2 0 0.5 1 1.5 2

Fig. 5. Response spectral shapes (normalised to the ordinate at 0.2 s) for rock sites at 10 km
from earthquakes of magnitude 5.5, 6 and 7 using the median values obtained from the follow-
ing attenuation equations (clockwise from top left): Ambraseys et al. [1996], Campbell [1997],
Abrahamson and Silva [1997] and Boore et al. [1997].
May 18, 2004 10:44 WSPC/124-JEE 00159
Krinitzsky and Chang [1977] proposed that if scaling factors of 4 or more needed
to be applied to accelerograms, then the records should be rejected, although no
justification was given for this assertion. Subsequently Vanmarcke [1979] proposed
reduction of the limits on scaling to a factor of 2 for liquefaction analysis, although
the limit of 4 was upheld for linear elastic systems. Vanmarcke [1979] based his con-
clusions on a study of inelastic spectra and of correlations amongst different strong-
motion parameters, using a dataset of 70 accelerograms. The dataset only included
12 accelerograms that had horizontal peak accelerations of at least 0.2 g and 41 of
the accelerograms had PGA values below 0.1 g. Despite the limitations of the data
and the analyses underlying the conclusions, the recommendations from these two
studies are frequently used as a rule-of-thumb in practice: Malhotra [2003] finds a
scaling factor of 5.84 is required for one record used in his study and concludes
that this “is higher than the normally accepted upper limit of 4”. Presumably, the
rationale behind imposing limits on scaling is to avoid creating unrealistic ground
motions, since this would undermine the inherent value in using real accelerograms
in the first place. However, it is not clear that such severe restrictions on scaling
values are justified, since over the distance ranges for which spectral shapes are
depicted in Fig. 6, amplitudes of ground motion can vary significantly: from 5 to
1.2 1.2
Ambraseys et al. (1996) Campbell (1997)

1 1
SA(T) / SA(0.2)
SA(T) / SA(0.2)
0.8 0.8
0.6 0.6
0.4 0.4
0.2 0.2
0 0
0 0.5 1 1.5 2 0 0.5 1 1.5 2

1.2 1.2
Boore et al. (1997) Abrahamson & Silva (1997)

1
1
SA(T) / SA(0.2)
SA(T) / SA(0.2)
0.8
0.8
0.6
50 km
0.6 5 km
0.4
5 km
0.4
20 km 0.2
0.2 0
0 0.5 1 1.5 2 0 0.5 1 1.5 2

Fig. 6. Response spectral shapes (normalised to the ordinate at 0.2 s) for rock sites at 5, 20
and 50 km from an earthquake of magnitude 7 using the median values obtained from the fol-
lowing attenuation equations (clockwise from top left): Ambraseys et al. [1996], Campbell [1997],
Abrahamson and Silva [1997] and Boore et al. [1997].
May 18, 2004 10:44 WSPC/124-JEE 00159
20
Ambraseys et al. (1996)
SA at 5 km / SA at 50 km
Campbell (1997)
Abrahamson & Silva (1997)
15 Boore et al. (1997)
Average
10
0
0 0.5 1 1.5 2
Period (seconds)
Fig. 7. Ratios of spectral ordinates for a magnitude 7 earthquake at 5 and 50 km from the
earthquake source, calculated from the median spectra shown in Fig. 6. The mean of the four
ratios shows very little variation with period.
50 km, median values of spectral ordinates will reduce by about a factor of 7, as

shown in Fig. 7; the very rapid decay of amplitudes with distance is often not
appreciated because of the tendency to plot attenuation curves in log-space. The
duration of the motion, if measured using the significant duration concept (the in-
terval over which a specified proportion of the Arias intensity is accumulated), does
increase with distance, due to different wave propagation velocities and scattering,
but according to the equations of Abrahamson and Silva [1996] the increase is only
about 0.6 seconds for every 10 km (Fig. 8). We propose, therefore, that in making
selections of real records, the search window should be as narrow as possible in
terms of magnitude, and if it needs to be widened to capture the required number
of records, that the distance range be extended.
There are two important exceptions to this line of reasoning, the first being if
records are selected from soft soil sites, since weak distant motion would not scale
linearly for sites closer to the source due to soil non-linearity. The second excep-
tion is if near-source rupture directivity effects are to be considered as part of the
design scenario. The effect of forward directivity is to produce short-duration mo-
tions with high-energy pulses that amplify the spectral ordinates at intermediate or
long periods [Somerville et al., 1997]. Near-source directivity effects cannot easily
be artificially introduced into real accelerograms hence if this is a design criterion
the search will need to specifically identify records obtained at short distances and
in the forward directivity zone. The number of records available which correspond
to such conditions is relatively small, but a start can be made using the database
presented by Somerville et al. [1997] and near-source recordings from recent earth-
quakes including the 1999 events in Turkey and Taiwan.
May 18, 2004 10:44 WSPC/124-JEE 00159
25
5-75% Significant Duration (s)

20
8
15
10 7
6
5
M=5
0
0 20 40 60 80 100
Closest Distance (km)
Fig. 8. Predicted median values of significant duration (5–75% of Arias intensity) at rock sites
from the equation of Abrahamson and Silva [1996].
3.2.3. Site classification

The third parameter that is obviously desirable to include in the search is the site
classification, since this also exerts a strong influence on the nature of the ground
motion, affecting both the amplitude and shape of response spectra. However, spec-
ifying a close match for this parameter may not always be feasible since the geotech-
nical profile has been determined with confidence for a relatively small number of
strong-motion recording sites. Even if reliable site classifications are available, these
will generally be based on, at best, the nature of the uppermost 30 m at the site,
whereas the deeper structure can also exert an important influence [Boore, 2004].
Within any site class — and especially within sites classified simply as “rock” —
there can be considerable variation in dynamic response characteristics. Adding
site classification as a third search parameter will obviously reduce significantly
the number of records returned for any given magnitude-distance window [Bommer
and Scott, 2000]. In light of these issues, there may be cases in which it would
be advisable to relax the matching criteria for site classification in order not to
restrict too severely the number of records obtained. Clearly, if the site of interest
is characterised by hard rock, it would be advisable to exclude soft soil recordings
from the suite of records compiled for dynamic analysis, but any greater restriction
should be imposed only if there are sufficient records providing a reasonable match
to the design scenario in terms of magnitude and distance. If the number of avail-
able records matching the magnitude and distance criteria is small, we recommend
that records be considered from sites that are within one site class (e.g., NEHRP or
May 18, 2004 10:44 WSPC/124-JEE 00159
EC8) either side of the classification of the site under consideration. This approach
is far from ideal and it is not intended to discount the vital importance of site effects
in ground-motion estimation, but it reflects a pragmatic attitude towards the data
about site geology and site response generally available.
3.2.4. Additional selection criteria

In the case that a good number of records (∼ 10–20) can be obtained specifying the
site class, and using a suitable window in magnitude-distance space, a further refine-
ment could be to also consider the rupture mechanism, if this is determined as part
of the design scenario. There is no definitive evidence for systematic and significant
differences between the ground motions from normal and strike-slip faulting earth-
quakes, but there is general consensus that reverse-faulting events produce larger
amplitudes of motion. There is less agreement on the ratio of reverse to strike-slip
motions and the extent to which the ratio varies with response period [Douglas,
2003]. The best estimate of this ratio given by Bommer et al. [2003b] implies a
variation of about 12% in period range from 0.1 to 1.0 second, which would suggest
that inclusion of style-of-faulting in the record selection is not vital.
A point that is not often stated, but which is worth bearing in mind, is that
an additional criterion should also be added when setting up a small suite of real
records: the records should not come predominantly from one recording station. A
possible exception to this condition would be in the case of the recording station
being located very close to the site of interest. Another possible criterion is that any
suite of records used in dynamic analyses should not be dominated by accelerograms
from a single earthquake event.
4. Matching Selected Records to the Elastic Response Spectrum

Whether records are selected by performing searches in terms of response spectral
ordinates or in terms of seismological and geophysical parameters, there will gen-
erally be a requirement to ensure that the records conform to some specified level
of agreement with the ordinates of the design response spectrum. Figure 9 provides
an overview of the options available for adjusting the selected records, as well as
the alternative of using artificial spectrum-compatible time-series, which obviate
the need to apply subsequent adjustments to match the design spectrum.
4.1. Matching criteria

Spectral matching criteria specified in seismic design codes vary from being purely
descriptive to being highly prescriptive. An example of the former is the 1992 New
Zealand code in which the matching criterion is that “over the period range of in-
terest for the structure being analysed, the 5% damped spectrum of the earthquake
record does not differ significantly from the design spectrum”. In the 2002 version
May 18, 2004 10:44 WSPC/124-JEE 00159
Fig. 9. Overview of the options available for scaling selected accelerograms match the ordinates
of the elastic response spectrum specified for design. The box marked “selection” is expanded in
Fig. 4.
of Part 1 of Eurocode 8, the specification is that no value of the mean 5% damped

elastic spectrum calculated for all of the selected records is less than 90% of the cor-
responding value of the 5% damped elastic response spectrum; the period range over
which this criterion must be met is between 0.2T1 and T1 , where T1 is the natural
period of the structure. In UBC 1997, the average ordinates of the individual spec-
tra — calculated as the square root of the sum of the squares (SRSS) of the two
horizontal components — should not be less than 1.4 times the design spectrum
ordinates in the range from 0.2T to 1.5T , where T is the fundamental period of
vibration of the building. The factor of 1.4 is simply to make the SRSS spectrum
comparable to that from the code, and in effect the criterion is that the average
spectrum of the records should not be below the design spectrum over the period
range specified.
May 18, 2004 10:44 WSPC/124-JEE 00159
One notable feature of the specifications in seismic design codes is that it is

the average ordinates of the real spectra that have to match the target and not
the individual spectra. If matching is required over a wide period range, there may
be advantages in matching the average spectral ordinates of the scaled records to
the elastic design spectrum, but this overlooks — and indeed can conceal — the
equally important issue of the maximum exceedance of the target spectrum by
the ordinates from any individual record. The derivation of the target spectrum
should be borne in mind when using records scaled to match its ordinates: any
design spectrum that is derived through the use of PSHA will include the influ-
ence of the scatter in the ground-motion prediction equations, which represents the
aleatory variability in ground-motion parameters for given combinations of magni-
tude, distance and site conditions. The strong-motion parameters of the selected
records will also display an aleatory variability, although the standard deviation
of these values about their mean may be slightly smaller than the standard devi-
ations associated with ground-motion prediction equations [Bommer et al., 1998;
Shome et al., 1998]. Therefore, if the analysis is performed following the procedure
presented in some design codes of selecting three accelerograms, scaling their aver-
age ordinates to not fall below the design spectrum, and then using the maximum
structural response as the basis for design decisions, the variability in the ground
motion is effectively being double counted. This procedure should, therefore, not
be used. A common misconception is to assert that the selected suite of accelero-
grams should capture the variability in ground-motion amplitudes, whereas this
variability is already fully accounted for in the derivation of the probabilistically
defined response spectrum. For this reason, Stewart et al. [2001] propose that if
only three records are used, they should be adjusted with one of the techniques
presented in the next sub-section, to remove their peaks and troughs so that “the
results of structural analyses are not unduly controlled by the particular time his-
tories that are chosen”. An alternative — and unorthodox — approach is that the
hazard assessment could be performed using only median values from the ground-
motion prediction equations and then using large suites of unscaled accelerograms
to capture the aleatory variability in the ground shaking [Bommer et al., 1998]. We
do not recommend that this latter approach be adopted, but it does bring out the
issue of taking explicit and measured account of the aleatory variability in ground
motions.
From a seismological perspective, a preferable approach may be to use at least
seven records and then use the average response obtained from the structural anal-
ysis. However, we resist making inflexible recommendations on this issue since in
any design situation a balance must be found, generally driven by considerations
of time and cost, between using realistic input for dynamic analyses and reducing
the number of analyses that need to be performed. Many engineers will consider
that using adjusted time-histories, which may not be entirely realistic, is an accept-
able price to pay for being able to limit the number of complex dynamic structural
analyses performed.
May 18, 2004 10:44 WSPC/124-JEE 00159
The dispersion of the results of dynamic analyses has been shown to be inversely
proportional to the square root of the number of records used. Shome et al. [1998]
demonstrated that seven is a suitable number to produce acceptably low dispersion
in the results, although other studies have found that to obtain a stable mean in
the results of the structural analyses at least 10 records are required [A. Pecker,
personal communication, 2003]. As well as ensuring a stable mean of the results,
consideration should also be given to the maximum spectral exceedance of any
of the individual records [McGuire et al., 2001] since even if the average of the
scaled record spectra match the target spectrum, there may be individual records
imposing exceptionally high demands on the structure. Selection procedures that
include criteria such as the Drms residual, as described previously, will help to avoid
these problems. As noted in the next sub-section, spectral matching techniques can
also remove pronounced peaks and troughs from the selected records.
A final issue to also be considered is the issue of the two horizontal components of
motion from each triaxial accelerogram. For any analysis requiring two orthogonal
components of horizontal motion to be used, careful consideration must be given to
the selection and scaling of the two acceleration time-histories, an issue addressed by
Malhotra [2003]. The guidelines for seismic design of bridges (Part 2) in Eurocode 8
expressly, and quite correctly, forbids the use of the same acceleration time-history
simultaneously in both horizontal directions. When the two components of one real
accelerogram have been chosen, it is recommended that their average ordinates be
used in deriving the scaling factor by comparison with the design spectrum and
the factor then applied to the two components separately in order to conserve their
differences, particularly for those cases where there is fault-normal and fault-parallel
polarisation [Stewart et al., 2001]. The definition of the horizontal component of
motion used in deriving the design spectrum should also be kept in mind when
deriving the scaling factors, since ground-motion prediction equations use a variety
of definitions, the most popular being the larger of the two horizontal components
and their geometric mean [Douglas, 2003].
4.2. Selective manipulation of accelerograms

Techniques are available that allow the user to manipulate real records not only
to scale their spectral ordinates but also to change the spectral shape [Preumont,
1984]. The techniques make no claim to have a geophysical basis and their expressed
purpose is to obtain suites of records with low variability in order to reduce the
number of structural analyses required to obtain stable results. Some of these tech-
niques operate in the frequency domain by adding harmonic components through-
out the record; the most widely used of these techniques is that embedded in the
program WES RASCAL [Silva and Lee, 1987] which has been widely employed and
recommended [e.g. Idriss, 1993].
A time-domain method, based on earlier proposals [Kaul, 1978; Lilhanand
and Tseng, 1988], with an improved capacity to preserve the non-stationary
May 18, 2004 10:44 WSPC/124-JEE 00159
0.15
Spectral Acceleration (g)

0.3 Target (EC8)
0.1 Original
Acceleration (g)
Original
0.05 RASCAL
RSPMATCH
0.2
0
-0.05
0.1
-0.1
-0.15 0
0 10 20 30 40 50 60 0 0.5 1 1.5 2
Time (seconds) Period (seconds)
0.15 0.02
Spectral Velocity (g.s)

0.1
Acceleration (g)
RASCAL 0.015
0.05
0 0.01
Target (EC8)
-0.05 Original
0.005 RASCAL
-0.1 RSPMATCH
-0.15 0
0 10 20 30 40 50 60 0 0.5 1 1.5 2
Time (seconds) Period (seconds)
0.15 0.15
Arias Intensity (m/s)
0.1 RASCAL
Acceleration (g)
RSPMATCH
0.05 0.1
RSPMATCH
0
Original
-0.05 0.05
-0.1
-0.15 0
0 10 20 30 40 50 60 0 10 20 30 40 50 60
Time (seconds) Time (seconds)
Fig. 10. Modification of the Corralitos record (top left) of the 1984 Morgan Hill earthquake using
RASCAL (middle left) and RSPMATCH (bottom left), to match the EC8 acceleration spectrum
(top right). The resulting velocity spectra (middle right) and Husid plots (bottom right) are also
compared.
characteristics of the motion, has been developed by Abrahamson [1993] in the

program RSPMATCH. This program adds wave packages to those parts of the
time-series for those frequencies for which there is a mismatch between the record
and target spectrum; the use of wavelets for this purpose is discussed by Iyama and
Kuwamura [1999] and by Mukherjee and Gupta [2002]. Results obtained applying
the two methods to a single record are compared in Fig. 10.
These techniques are convenient and can be used selectively to satisfy match-
ing criteria for a suite of accelerograms by adjusting the ordinates of the records
causing the most problematic deviations from the target. Stewart et al. [2001] rec-
ommend that in general it is preferable to use a large number of accelerograms
without making adjustments to their spectral shapes, but again the final decision
May 18, 2004 10:44 WSPC/124-JEE 00159
should be based on the acceptable balance between the realistic nature of the input
motions and the number of analyses that can be performed. Surprisingly, despite
being matched to the elastic design spectrum over a wide period range, adjusted
records can sometimes produce lower structural demands than linearly scaled real
accelerograms, especially if the latter have been matched to the target spectrum in
log space and include significant exceedances of the target spectrum [N.A. Abra-
hamson, personal communication, 2003].
4.3. Linear scaling in time and acceleration

Accelerograms can be scaled to achieve an improved match with the target spec-
trum, and possibly other specified criteria, by applying scalar factors to the ac-
celeration and/or time axes of the record. However, if there is a large mismatch
between the duration of a selected accelerogram and the duration specified for
the design scenario, there is no acceptable procedure to close the gap. Seed and
Idriss [1969] produced an artificial accelerogram for a large magnitude (M 8.25)
earthquake by scaling and splicing records from smaller earthquakes, but this was
done at a time when the global strong-motion databank was very sparse. Scaling
the time axis of an accelerogram can increase or decrease the significant duration,
which might be acceptable to compensate for small changes associated with dis-
tance (Fig. 8), but not to compensate for any mismatch in magnitude because for
that it would also be necessary to change the number of cycles of motion. Scaling
the time axis of a record changes not only the duration of the motion but also
the frequency content of the record over the entire period range; Kramer [1996]
suggests that this procedure should be used with caution, advice with which we
strongly agree.
Procedures to obtain appropriate scaling factors, with the particular aim of
reducing the scatter in the ordinates of the scaled spectra or in the results of
inelastic analyses, have been discussed by Nau and Hall [1984], Matsumura [1992],
Shome et al. [1998], and Kappos and Kyriakakis [2000]. Martı́nez-Rueda [1998]
performed a parametric study on the response of inelastic SDOF systems with the
objective of identifying a suitable instrumental measure of ground-motion intensity
for the scaling of natural accelerograms. The scaling procedures proposed by all of
these researchers, with the exception of Shome et al. [1998], are based on the use of
variations of spectrum intensity. This is convenient, since the ordinates of a design
spectrum will invariably be available to the engineer.
Shome et al. [1998] propose that records be scaled to match the median spectral
acceleration, obtained from ground-motion prediction equations, at the fundamen-
tal period of the structure. There are potential pitfalls in only considering the
spectral ordinate at the fundamental period since there is generally uncertainty
in the estimation of this parameter. Furthermore, as damage progresses there will
generally be period elongation and if higher modes contribute to the response then
shorter periods will also be of interest. For these reasons, in the current study the
May 18, 2004 10:44 WSPC/124-JEE 00159
focus is on scaling to match the target spectrum across a period range around the
fundamental period of the structure being analysed.
The present paper is considered to be complementary to these earlier studies
rather than to supersede them or to contradict their conclusions, for the simple
reason that the focus herein is largely on the selection of the records, which was
not a key issue addressed in the previous papers. Nau and Hall [1984] used only 12
ground-level records obtained on various different sites from both crustal and sub-
duction earthquakes with a large range of magnitudes. Matsumura [1992] also used
only 12 components from US and Japanese accelerograms, the magnitude, distance
and site classifications not even being mentioned. Martı́nez-Rueda [1998] used both
horizontal components from a total of 50 accelerograms recorded at epicentral dis-
tances of up to 400 km from crustal and subduction earthquakes with magnitudes
ranging from Ms 5.4 to 8.1; soft soil, stiff soil and rock site recordings were included.
The importance of selecting records on a consistent basis is implicitly recognised,
however, in the example application presented by Martı́nez-Rueda [1998], for which
10 Californian accelerograms obtained at distances of less than 30 km from crustal
earthquakes with magnitudes in the range 6.4–7.2; all but two of the records are
obtained from stiff soil sites.
The problem of defining appropriate input to dynamic structural analyses ulti-
mately involves aspects of both engineering seismology and structural dynamics; the
studies cited above have generally placed greater emphasis on the latter, whereas
the current study primarily addresses engineering seismological aspects. Kappos
and Kyriakakis [2000] used 11 records from 11 Greek earthquakes, and another 13
records from eight US earthquakes, both data sets being approximately divided
between recordings from rock and soil sites; the authors claim that the “compi-
lation of records permits consideration of the effect of soil conditions (inevitably
in a rough way) as well as of the tectonic regime”. The Greek records are filtered
with a low-frequency cut-off at 1 Hz, whence the elastic and inelastic displacement
spectral ordinates at longer periods, discussed at some length in the paper, should
really have been neglected. Differences in the characteristics of the two data sets
are instead attributed to other factors including “the deeper deposits in some Cal-
ifornian sites (e.g. the Bay area)”, despite the fact that two of the three records
obtained in the San Francisco Bay Area are from rock sites and the third (Parking
Garage, Stanford) is not located on Bay mud. As in other studies, the dispersion
of structural responses is measured by the coefficient of variation (COV), which is
the standard deviation divided by the mean. Kappos and Kyriakakis [2000] find
that for inelastic spectral responses the COV increases with increasing ductility
factors, but this may simply be due to the large range of magnitudes — and hence
durations — in their data sets. A major conclusion of their study is that the COV
is higher for the rock data sets than for the alluvial (soil) data sets, which they
attribute to the different site classifications. However, it is also possible that the
difference was mainly due to the different ranges of magnitude in the data sets: for
the US data, the maximum differences in magnitude amongst the soil records was
May 18, 2004 10:44 WSPC/124-JEE 00159
1.4 as opposed to 1.8 for the rock data, and for the Greek records the magnitude
variation amongst the soil site recordings was just 1.3 compared with 2.1 for the
rock site accelerograms.
Amongst the studies cited above, the one which did give careful consideration
to seismological and geophysical selection criteria was Shome et al. [1998], who
used records selected in magnitude-distance bins from stiff soil sites, excluding
accelerograms with near-source forward directivity pulses. As mentioned previously,
however, the study concluded that if the records are then individually scaled to
match the predicted median elastic spectral acceleration ordinate corresponding to
the scenario at the centre of the bin, careful selection in terms of magnitude and
distance becomes unnecessary. This critically important point is re-visited in the
final section of this paper.
5. An Illustrative Example
The possibilities for using real accelerograms as input for engineering analysis and
design, and the issues involved, are best illustrated by practical examples. In this
section two approximately compatible data searches are defined, one based on the
parameters of an earthquake scenario, the other on the ordinates of the elastic
design spectrum. The suites of records obtained from both searches are examined
and from each, using careful selection and adjustment of the records, appropriate
input for dynamic analysis is prepared.
5.1. Selection criteria and recovered data sets

In order to compare and contrast the use of seismological and strong-motion param-
eters as the basis for data selection, two design situations have been defined. The
first is one that may be typical of engineering practice where the designer has ac-
cess to very little information about the underlying hazard and is simply presented
with an elastic design spectrum, in this case the Type 1 spectrum from Eurocode 8
[CEN, 2002]. The design peak ground acceleration in bedrock is taken to be 0.3 g
and the chosen site class is B, corresponding to very dense sand or gravel, or very
stiff clay, with a 30 m shear wave velocity in the range from 360 m/s to 800 m/s.
The 5% damped acceleration spectrum is anchored at 0.36 g (the product of the
bedrock PGA and the soil factor of 1.2), with a constant acceleration plateau at
0.9 g between 0.15 and 0.5 seconds (Fig. 11). The search for direct matching to
the EC8 spectrum was performed on the basis of the average Drms residual (see
Sec. 3.1) on the spectral shape being no greater than 0.09 in the period range from
0.1 to 0.4 seconds, and the PGA of the record lying in the range from 0.26 g to
0.46 g. The period range was chosen because between these limits there is close
agreement between the ordinates of EC8 spectrum and the spectral ordinates of
the design scenario described below (Fig. 11), and the maximum ordinates are the
same in both cases. At longer periods there is considerable divergence between the
May 18, 2004 10:44 WSPC/124-JEE 00159
1
Design scenario (M6.4, 10 km, 0.6 sigmas)
Eurocode 8
Spectral Acceleration (g) 0.8
0.6
0.4
0.2
0
0 0.5 1 1.5 2
Period (seconds)
Fig. 11. Response spectra corresponding to criteria used for strong-motion data searches: EC8
class B spectrum anchored to 0.3 g in bedrock (thick line), and a scenario represented by M s 6.4,
a stiff soil site at a distance of 10 km, and 0.6 standard deviations above the median motions (thin
line).
ordinates of the code and scenario spectra, which is to be expected; the spectral
shape in EC8 was calibrated to normalised spectra from European accelerograms
from earthquakes with magnitudes in the range from Ms 6 to Ms 7, but with a
strong bias towards the larger values [Rey et al., 2002]. The search performed on
the basis of matching, in an average sense, the ordinates of the EC8 spectrum in
Fig. 11 yielded 40 strong-motion records from 22 earthquakes.
The second search is performed on the basis of an earthquake scenario chosen to
be representative of the results that might be yielded from a disaggregation of the
500-year hazard in the seismically active parts of Europe. The scenario is defined
by a surface-wave magnitude Ms of 6.4, a source-to-site distance of 10 km, and an
exceedance of the median values of spectral acceleration of about 0.6 standard
deviations; this value of 0.6 for ε corresponds to the 73-percentile ground motion.
As for the Eurocode 8 spectrum, the site is characterised by stiff soil with a Vs,30
in the range from 360 to 800 m/s.
The spectral ordinates for this scenario, obtained using the prediction equations
of Ambraseys et al. [1996] — after smoothing the coefficients with a 1/4-1/2-1/4
running average — are shown in Fig. 11.
As shown in Sec. 3.2, the most important geophysical parameter for selecting
records is earthquake magnitude, and therefore the search was designed to have a
small window on magnitude and a larger window on distance. Table 1 shows the
number of records recovered from different M-R search windows, including the effect
of adding the site classification as a third search parameter, for searches performed
May 18, 2004 10:44 WSPC/124-JEE 00159
Table 1. Numbers of records obtained using different

search windows on magnitude and distance; numbers in
parentheses are records from stiff soil sites.
Distance Ms Ms Ms Ms
Ranges 6.0–6.8 6.1–6.7 6.2–6.6 6.3–6.5
d: 0–50 km 377 (121) 315 (106) 213 (76) 72 (28)
d: 0–30 km 229 (58) 186 (47) 114 (25) 40 (13)
d: 0–20 km 157 (37) 121 (30) 69 (15) 26 (10)
d: 5–15 km 94 (21) 72 (17) 37 (9) 12 (6)
on the Imperial College strong-motion data archive, which contains about 7000
records. The final search window, which yielded 55 ground-motion records from 16
earthquakes, was defined by the following limits: 6.2 ≤ Ms ≤ 6.6, 0 ≤ Rjb ≤ 40 km,
and stiff soil.
An initially surprising result of the two searches is that there was not a single
accelerogram common to the two sets. However, if the basis of the selection criteria
are carefully considered, this result is perhaps less unexpected: unless an elastic
design spectrum has been obtained from a DSHA using median values from ground-
motion prediction equations, the scenario will always include an ε term that will
be responsible for an appreciable proportion of the spectral amplitudes. For PSHA
in which no truncation is applied to the scatter in the ground-motion prediction
equations, the contribution from ε will grow with the return period [e.g. Restrepo-
Vélez & Bommer, 2003]. Only one earthquake was common to the results obtained
from the two searches (the 15 October 1979 Imperial Valley, California, main shock)
but each search picked up different records from this event. The M-R distributions,
and the site classifications for the records obtained by spectral matching, are shown
in Fig. 12.
The objective of the exercise is to produce an optimal suite of 10 accelerograms
from each dataset, for which there is a good match with the elastic spectral ordinates
specified for design, and for which there is low variability amongst the spectral
ordinates of the scaled records; an additional, but less critical objective, is to obtain
the suite with the least amount of scaling possible. The match is defined by the
average ordinates of 10 scaled spectra not being below the target spectrum in the
period range from 0.1 to 0.4 seconds. To begin with, there is as many as five times
the required number of records in each data set, so some preliminary “manual”
pruning of the selections can be applied. This is done for each data set, and then
the reduced data sets are examined in order to perform further selection and then
to scale to the design spectrum; for both cases, the ordinates of the Eurocode 8
spectrum are taken as the target.
For simplicity, it is assumed that only a single horizontal component of motion
is required for each dynamic analysis. An additional clarification is required at this
point, related to the definition of the horizontal component of motion, as noted in
Sec. 4.1. Spectral ordinates predicted by the equations of Ambraseys et al. [1996]
May 18, 2004 10:44 WSPC/124-JEE 00159
7.5
Magnitude (Ms)
7
6.5
6
unknown
5.5 soft soil
stiff soil
5 alluvium
rock
4.5
0 25 50 75
Distance to surface projection of rupture (km)
7.5
Magnitude (Ms)
6.5
5.5
4.5
0 25 50 75
Distance to surface projection of rupture (km)
Fig. 12. Magnitude-distance distributions of records recovered from searches by matching to
ordinates of code spectrum (upper ) and by use of a magnitude-distance window and specification
of the site classification (lower ).
correspond to the envelope of the two horizontal components and the calibration
of the Eurocode 8 spectrum used a similar definition. For the records obtained
from the search in terms of magnitude, distance and site classification, the larger
component is chosen on the basis of the larger spectrum intensity; in most, but not
all, cases this is also the component with the larger PGA. For the search performed
May 18, 2004 10:44 WSPC/124-JEE 00159
in terms of spectral ordinates, individual components are returned hence the issue
of selecting components does not arise.
5.2. Records selected by spectral ordinates

The search based on minimising the average SRSS residuals between the actual and
target spectra yielded 42 single component records. A few of these can be elimi-
nated on the basis of representing outlier cases or other reasoning. For example,
one of the soft soil stations is actually classified as “very soft” (Vs,30 < 180 m/s),
so this may be eliminated (although it is noted in passing that there also exists
a possibility of some of the stations that are classed as “alluvium” or “unknown”
being equally soft). Additionally, the recording obtained at a distance greater than
70 km can be dropped; this is from a site classified as alluvium and to produce
such high accelerations (the PGA is 0.27 g) at such a distance it is likely that the
soil deposits are very soft. There is a group of six recordings, all obtained at very
short distances, from the Ms 7.6 1999 Chi-Chi (Taiwan) earthquake; since the sce-
nario under study is nominally for southern European conditions, this event may
be considered excessively large and therefore these records can also be eliminated.
The recordings from smaller magnitude earthquakes can also be eliminated since
these are very unlikely to be of engineering significance by virtue of their relatively
low energy content and limited number of cycles of motion; a threshold of Ms 5.7
is applied for this criterion, only retaining records from larger events. The applica-
tion of these criteria removes 13 of the records, leaving a total of 29 records from
19 earthquakes.
A possible way forward is to simply select from the remaining records those
recorded on sites classified as stiff soil, thus matching one of the features of the
specified design scenario. This would yield 11 records, which would be easily handled
and make the optimum selection of 10 records relatively straightforward. However,
as was mentioned earlier, site conditions may not be one of the most critical selection
criteria and it is preferable to retain all 29 records while considering other criteria.
The records have been selected on the basis of a search that considered both
an approximate match to the design PGA of 0.36 g and an approximate agreement
with the spectral shape in the period range from 0.1 to 0.4 seconds. A very close
match of both the PGA value and the spectral shape is likely to indicate a record
that will require little adjustment. An additional indicator of the match between
the record and target spectrum is the ratio of spectrum intensity in the period range
0.1–0.4 s; the value of the spectral intensity in this range for the EC8 spectrum is
10.45 cm. Figure 13 shows the normalised average Drms residuals plotted against
the ratios of record-to-target PGA and record-to-target spectrum intensity for the
remaining 29 records.
The best fit to the target spectrum is defined by a very low value of the average
residual combined with a value of both PGA and spectrum intensity ratios very
close to unity. The data manipulation may be made easier if a few records with
May 18, 2004 10:44 WSPC/124-JEE 00159
0.1
Average Normalised Residual 0.09
0.08
0.07
0.06
unknown
soft soil
0.05 stiff soil
alluvium
rock
0.04
0.6 0.8 1 1.2 1.4 1.6
Record-to-Target PGA Ratio

0.1
Average Normalised Residual
0.09
0.08
0.07
0.06
0.05
0.04
0.6 0.8 1 1.2 1.4 1.6
Record-to-Target Spectrum Intensity Ratio

Fig. 13. Measures of matching between the record and target spectra for the accelerograms
selected on the basis of spectral matching.
May 18, 2004 10:44 WSPC/124-JEE 00159
relatively high average residuals and ratios significantly different from unity are
eliminated. On this basis, three records with average residuals greater than 0.078
and PGA ratios smaller than 0.8 are removed from the suite. The cluster of three
records with similarly high residuals but with PGA residuals above 1.2 are not
removed for the following reason: the ultimate goal is a suite of 10 records whose
average spectrum does not fall below the target EC8 spectrum in the period range
from 0.1 to 0.4 seconds. Since in both plots of Fig. 13 there are far more points
with ratios lower than unity than greater than one, it is decided to retain these
three records with rather high ratios of PGA and spectrum intensity, since they
may contribute to finding a good average match with the target spectrum without
scaling. This is done herein in order to reflect current code procedures.
The reduced data set now consists of 26 component records from 17 earthquakes.
The next step is to find the average spectral ordinates of groups of 10 records in
order to identify if there are combinations that will produce a mean spectrum
whose ordinates in the range 0.1–0.4 s are always equal to or greater than those
of the EC8 spectrum. It is found that in fact there are several sets of records
whose mean ordinates, in the period range of interest, are always above the EC8
spectrum ordinates, hence no scaling is required at all (in fact, scaling factors of
less than unity could be applied to reduce the amount by which the ordinates of
the design spectrum are exceeded). Figure 14 shows the mean ordinates of 10 of the
records compared with the elastic design spectrum. All of the combinations that
produce mean spectral ordinates above the target spectrum include no fewer than
six records from a suite of 10 that from amongst the 26 that have particularly high
12
EC8 spectrum
Spectral Acceleration (m/22)
Mean of 10 records
10
0
0 0.2 0.4 0.6 0.8 1
Period (seconds)
Fig. 14. Mean ordinates of a suite of 10 accelerograms compared with the EC8 target spectrum;
for this combinations of records, no scaling or adjustment is required.
May 18, 2004 10:44 WSPC/124-JEE 00159
amplitudes. All but one of these 10 records are from earthquakes of magnitude 6.6
or greater, with four coming from events of magnitude of 7.2 or above. The single
record from a smaller magnitude (Ms 5.9) earthquake was recorded at 5 km on
soft soil. In conclusion, this exercise reinforces the fact that selecting in terms of
matching to elastic spectral ordinates only is unlikely to result in accelerograms that
are consistent with the underlying design earthquake scenario (which, it is recalled,
in this case is Ms 6.4 at 10 km and a stiff soil site). This is the fundamental problem
in the specification of acceleration time-series in current seismic design codes, as
discussed previously.
5.3. Records selected by geophysical criteria

The search for records from stiff soil sites with magnitudes in the range Ms 6.2–
6.6 and distances in the range from 0 to 40 km yielded 55 accelerograms from 16
earthquakes; a point worth noting is that 20 of the component records are from a
single event, the 1983 Coalinga (California) earthquake, all recorded in the distance
range from 31 to 40 km.
There are several options for reducing the dataset of 55 records to a more man-
ageable number, but care must be taken not to remove potentially useful records
in the process.
The 10 records with distances closest to scenario distance of 10 km are from six
earthquakes, with four of the records coming from a single event. As was mentioned
earlier, it is advisable not to have any single earthquake event or recording station
excessively represented, whence it would be advisable to drop at least two of the
four records from the same earthquake, and choose others, albeit from greater dis-
tances, in their place. This procedure will be re-visited after considering alternative
strategies.
Another way to proceed with the search would be to now include the style-of-
faulting as a fourth criterion, which would generally be feasible because in most
regions of the world the dominant rupture mechanism for nearby seismic sources
is usually known with some confidence. However, there are two arguments against
doing this, as indicated previously, the first being that the style-of-faulting does not
exert such a strong influence on the ground motion, and the second being that the
inclusion of this parameter can be very restrictive in terms of the number of records
that will be retained. There are six records from two normal faulting earthquakes,
13 records from six strike-slip events, and 35 records from six reverse ruptures, plus
a single record from an event classified as oblique.
An alternative way to proceed is now to search within the selected suite of
55 records using the spectral matching criteria that were used to search the suite
of accelerograms examined in the preceding sub-section. Figure 15 shows the same
information as Fig. 13 but for the records selected on the basis of magnitude,
distance and site classification. Since all the records are from the same site class,
different symbols are used to represent the style-of-faulting of the earthquakes. The
May 18, 2004 10:44 WSPC/124-JEE 00159
0.25
normal
oblique
strike-slip
Average Normalised Residual reverse
0.2
0.15
0.1
0.05
0 0.5 1 1.5 2
Record-to-Target PGA Ratio
0.25
0.2
0.15
0.1
0.05
0 0.5 1 1.5 2
Fig. 15. Measures of matching between the record and target spectra for the accelerograms
selected on the basis of geophysical parameters.
May 18, 2004 10:44 WSPC/124-JEE 00159
larger component of each accelerogram is selected on the basis of the larger value
of spectral intensity in the 0.1–0.4 s range, which in all but a few cases is also
the component with the larger value of PGA; it is found that for those situations
where this is not so, the two PGA values are generally quite similar. The ratios
of the record PGA and SI to those of the target spectrum are plotted against the
normalised average residual, as was done previously for the other data set.
Comparison of Figs. 13 and 15 shows that the agreement with the target spec-
trum for the records selected on the basis of the earthquake scenario is much poorer,
with only two records in the residual-ratio space covered by the records selected
by spectral matching. This result, however, is hardly surprising, since most of the
selected records are from much longer distances than the target of 10 km (Fig. 12)
and since the target spectrum corresponds to the 73-percentile motions, only one-
in-four of the records obtained at the scenario distance would be expected to match
the target spectrum. Therefore, it can be concluded at this point that it is very
unlikely that a suite of records can be found whose mean ordinates will match the
elastic design spectrum without some form of scaling being applied to the records.
0.25
0.2
0.15
0.1
0.05
0 0.5 1 1.5 2

Fig. 16. Measures of matching between the record and target spectra for the accelerograms with
Ms in the range 6.2–6.6 and recorded at distances no greater than 25 km. Different symbols are
used to identify records coming from either the same earthquake or the same station.
May 18, 2004 10:44 WSPC/124-JEE 00159
Figure 16 shows the spectrum intensity ratios and average normalised residuals,
without distinction by style-of-faulting, for only those records recorded within 25 km
of the source. Comparison of Figs. 15 and 16 shows that the use of a smaller
window on distance mainly removes the very low amplitude records, although a few
records with favourable characteristics have also been lost. Nonetheless, this reduced
selection of 17 records is retained since it has been obtained in a fashion that is likely
to be followed in routine data searches. The 17 records come from nine earthquakes,
and two of them are recordings from one station of two consecutive events separated
by a few hours (each shown as an asterisk in the plot). Immediately there is the
issue that in selecting the final 10 records it is desirable not to have any one event or
station contributing excessively to the suite of accelerograms, whence the different
symbols used in Fig. 16. Application of a minimum average normalised residual of
0.18 — double that used in the spectral matching selection described previously —
reduces the number of accelerograms to 13 records from nine earthquakes, with one
station represented twice and one earthquake represented four times.
The basic characteristics of the 13 selected records are presented in Table 2.
One additional record is also brought into the selection at this stage, bringing the
total number to 14. This record is from a distance of 37 km, and therefore was
excluded by the limit of 25 km used to reduce the data set. However, as can be
appreciated from Table 2, despite the distance, the record displays high values of
PGA and SI (0.392 g and 9.88 cm, respectively); these values are close to the target
values of 0.36 g and 10.45 cm. This record is included to illustrate the point made
Table 2. Records selected by magnitude (Ms 6.2–6.6), distance less than 25 km except No. 14
(see text), and stiff soil sites.
No. Date Time Mech1 Ms Station R2jb Drms SI PGA F3
1 02/03/1987 01:42:35 N 6.5 Matahina Dam 19 0.074 8.65 0.243 1.70

2 09/02/1971 14:01:42 R 6.6 Lake Hughes 12 15 0.137 9.81 0.284 0.77
3 Pasadena JPL 17 0.158 6.68 0.218 1.80
4 09/06/1980 03:28:00 SS 6.5 Cerro Prieto 8 0.167 10.92 0.572 1.00
5 15/06/1995 00:15:51 N 6.3 OTE, Aegion 7 0.158 11.53 0.540 1.00
6 15/10/1979 23:16:55 SS 6.6 Cerro Prieto 22 0.055 6.09 0.167 1.80
7 24/05/1979 17:23:18 R 6.3 Bar — Skupstina 12 0.167 5.40 0.270 1.50
8 Tivat — Aerodorm 15 0.157 3.90 0.166 —
9 Kotor — Naselje 19 0.070 1.73 0.057 —
10 Kotor — Zovod 19 0.147 4.06 0.152 —
11 26/04/1992 07:41:44 SS 6.5 Centerville Beach 20 0.117 12.13 0.440 1.00
12 26/04/1992 11:18:28 SS 6.5 Centerville Beach 25 0.096 7.60 0.257 1.30
13 28/06/1992 15:05:31 SS 6.5 Big Bear CC 13 0.094 18.74 0.545 0.70
14 26/04/1992 07:41:44 SS 6.5 Fortuna Fire St. 37 0.160 9.88 0.392 —
Notes: 1 — rupture mechanism: N — normal, R — reverse, SS — strike-slip.

2 — Joyner-Boore distance (horizontal distance from surface projection of rupture) in km.
3 — scaling factors applied to records (see Fig. 19).
May 18, 2004 10:44 WSPC/124-JEE 00159
in Sec. 3.2 regarding the recommendation to impose strict limits of magnitude but
not necessarily on distance. This record is assigned the identifier no. 14; application
of the minimum Drms value of 0.18, discussed above, would not have removed this
record from the dataset.
The variation in spectral amplitudes amongst the selected records is very large.
The largest ordinates correspond to the recording of the 1995 Aegion earthquake in
Greece (No. 5), which is a clear case of forward rupture directivity [Lekidis et al.,
1999], and the recording of the 1992 Big Bear earthquake in California (No. 13),
which produced ground motions that were on average twice as high as expected for
an earthquake of this magnitude in California [Cramer and Darragh, 1994]. The
lowest amplitudes correspond to the recordings of the 1979 Montenegro earthquake
in Yugoslavia (Nos. 7–10), all of which were obtained in the backward directivity
zone according to the orientation of the fault rupture plane [Boore et al., 1981].
As before with the records selected on the basis of spectral ordinates, groups of
10 accelerograms were selected from amongst the 14 candidate records in Table 2
and their mean ordinates compared with the ordinates of the design spectrum.
For each grouping, the scaling factor required to bring the minimum ordinate of
the average spectrum to the level of the EC8 spectrum was calculated; although
strict limits on scaling are generally not warranted, as discussed previously, it would
nonetheless be desirable to have a scaling factor as close to unity as possible. Fig-
ure 17 shows the mean spectra of a suite resulting in a low scaling factor of just
1.158. Figure 18 shows a suite that includes record No. 14, which allows an even
lower factor (1.138) to be applied. However, in both cases this is achieved through
the inclusion of records Nos. 5 and 13, which exceed appreciably the target spec-
trum and whose ordinates will be raised even higher by the application of the scaling
factors.
In general, discussions of the issue of scaling suites of record have focused on
identifying the scaling factor required to ensure that the average ordinates do not
fall below the target spectrum and then applying this factor to each of the records.
An alternative is to find the optimum combination of records and individual scaling
factors to simultaneously consider the very important criterion of minimal disper-
sion amongst the spectral ordinates of the scaled records; an algorithm could be
developed to at least partially automate this process but for illustrative purposes
herein a suite of individually scaled records has been prepared manually. Several of
the selected records in Table 2 have response spectra that are in good agreement
with the target EC8 spectrum shown in Fig. 11, hence these are retained without
scaling. Some of the stronger records are then added to the suite, and their ordi-
nates reduced by applying scaling factors of less than unity. Finally, other records
are added in, scaled up from their original amplitude, to achieve the required match
with the target spectrum. Table 2 presents one such combination and the scaling
factors applied; Fig. 19 compares their average spectral ordinates with the EC8
target spectrum. More optimal combinations of records and scaling factors could
be identified (the search is time-consuming if not automated), but even the results
May 18, 2004 10:44 WSPC/124-JEE 00159
12
Spectral Acceleration (m/s2)

EC8 spectrum
Average of 10 records
10 Scaled (1.158)
0
0 0.2 0.4 0.6 0.8 1
Period (seconds)
Fig. 17. Combination of 10 records whose mean ordinates are close to the target spectrum, using
only records from less than 25 km.
12
Spectral Acceleration (m/s2)
EC8 spectrum
Average of 10 records
10 Scaled (1.138)
0
0 0.2 0.4 0.6 0.8 1
Period (seconds)
Fig. 18. Combination of records whose mean ordinates are close to the target spectrum, including
record No. 14 (see Table 2).
May 18, 2004 10:44 WSPC/124-JEE 00159
12
Spectral Acceleration (m/s2) EC spectrum

Average of records
10
0
0 0.2 0.4 0.6 0.8 1
Period (seconds)
Fig. 19. Comparison of average spectral ordinates with EC8 target spectrum and the suite of 10
individually scaled records identified in Table 2.
1
Coefficient of Variation
Records scaled as group

Scaled Individually
0.8 Records selected by matching spectrum
0.6
0.4
0.2
0
0 0.2 0.4 0.6 0.8 1
Period (seconds)
Fig. 20. Comparison of coefficients of variation, against period, for the suites of 10 records whose
average ordinates are presented in Figs. 14, 18 and 19.
May 18, 2004 10:44 WSPC/124-JEE 00159
in Table 2 are encouraging: as many as four of the 10 records are retained at their
natural scale, and, as noted below, a relatively low level of dispersion is achieved.
As has been mentioned previously, a key issue in compiling a suite of records
to be used in dynamic analysis is the dispersion amongst the scaled records. In
fact, the real issue is the dispersion in the inelastic structural responses, but this is
dependent upon the specific structural model employed and is therefore beyond the
scope of this study. Figure 20 shows the variation of COV (coefficient of variation)
with response period for the suites of scaled records whose average spectral ordinates
are shown in Figs. 14, 18 and 19. The lowest dispersion is clearly that of records that
were selected on the basis of matching spectral ordinates (Fig. 14). This does not,
however, militate against the use of seismological criteria for the selection of records
since the dispersion of the inelastic responses obtained using the latter records may
well be higher due to the greater variation in the strong-motion durations. However,
it would still be desirable to reduce the COV values for the records selected on the
basis of seismological and geophysical parameters. Figure 20 clearly shows that the
dispersion of the records that are individually scaled after selection by seismological
criteria is not much larger than that associated with the records selected on the basis
of matching the target spectrum, with the advantage of more consistent durations.
Of course, even lower COV’s could be obtained using the adjustment procedures
described in Sec. 4.2, but at the expense of creating less realistic ground motions.
6. Discussion and Conclusions

6.1. Strong-motion data
Real earthquake accelerograms are clearly a viable option for providing input to
dynamic analysis of structures, being more realistic than spectrum-compatible ar-
tificial records and easier to obtain than synthetic accelerograms generated from
seismological source models. Real accelerograms are increasingly available to engi-
neers through CD-ROM collections and Internet sites, although the available search
tools associated with many of these data sources could be improved. Depending
upon the specific application, and particularly the sensitivity to displacements or
to long-period motions, the engineer may need to pay particular attention to the
signal processing to which the records have been subjected. In cases where displace-
ments or long-period spectral ordinates are important, the preferred options may be
either to access records from the PEER databank or to obtain uncorrected records
and apply an appropriate correction technique for baseline errors and long-period
noise.
This paper is not, however, intended to present a case against the use of syn-
thetic accelerograms but rather to address the issues related to the use of real
accelerograms and to provide some guidelines on selection and scaling. We view
synthetic records as a complement to real accelerograms and each will have its own
merits for different applications. Certainly synthetics are an attractive choice for
May 18, 2004 10:44 WSPC/124-JEE 00159
scenarios (combinations of magnitude, distance and site classification) not covered

by the existing empirical database, although by definition synthetic accelerograms
have not been validated for these situations.
The geographical distribution of the global strong-motion databank is very un-
even, and for projects located in regions of sparse or indeed no strong-motion data
it is necessary to make the assumption that regional differences in ground-motion
characteristics are sufficiently small to allow records obtained in one country to
be used for design or analysis in another. Clearly distinction needs to be made
between crustal and subduction earthquakes, and within the latter category, be-
tween in-slab and interface events, and caution must be exercised if accelerograms
from seismically active areas are to be used in low seismicity stable continental re-
gions. However, amongst seismically active regions, the implicit assumption made
in this study is that regional differences amongst the ground motions from crustal
earthquakes in different regions are, for similar combinations of magnitude, depth,
style-of-faulting, distance and site classification, sufficiently small to be ignored.
This is a topic that is clearly worthy of further research, but there is currently no
convincing evidence to invalidate the assumption.
6.2. Selection criteria

In order to make use of real accelerograms in dynamic analysis, selection criteria
are required to obtain a suitable ensemble of records. These criteria will clearly de-
pend on the information available to the engineer regarding the underlying seismic
hazard, but it is preferable to base the search on a specific earthquake scenario.
The authors recommend that the search be based on achieving a good match (to
within 0.2 units) with the design earthquake magnitude, but that, if necessary,
reasonably large mismatches with the target distance can be tolerated, since these
can be compensated for by application of linear scaling factors, which do not nec-
essarily need to be limited to the often cited ranges that are part of the “folklore”
surrounding this subject. A match with the site classification of the project is also
highly desirable but not necessarily essential and this criterion can be relaxed if few
records with reliable site classifications are available; an alternative formulation of
the problem is to exclude records from sites with very different classification from
that of the project site. The inclusion of style-of-faulting as a fourth search param-
eter, with the inevitable reduction of available records, is not recommended at this
time, unless the search in terms of magnitude, distance and site classification has
yielded a sufficiently large number of records.
There are studies that suggest that close matching between the earthquake
magnitude of the design scenario and the records is not necessary, most notably
the award-winning study by Shome et al. [1998]. Our opinion is that it has yet
to be shown conclusively that earthquake magnitude — and therefore by impli-
cation strong-motion duration — has a negligible influence on inelastic structural
response beyond controlling the spectral acceleration at the fundamental period of
May 18, 2004 10:44 WSPC/124-JEE 00159
the structure. Such a conclusion will depend on large numbers of studies considering
a wide range of realistic structural models. Martı́nez-Rueda [1998] urges caution in
extrapolating his conclusions, which are based on inelastic analysis of SDOF models,
to MDOF structures. Kappos and Kyriakakis [2000] examine more realistic MDOF
structural models, but as has been pointed out in this paper, their analysis has
some shortcomings in terms of the characterisation of the input. The conclusions
made by Shome et al. [1998] are also based on analysis of MDOF structural models,
but only one model is used, representing a steel structure — which can be expected
to be less affected by duration than a reinforced concrete structure [e.g. Jeong and
Iwan, 1988] — of 5 storeys, and for the damage metric based on dissipated energy
their conclusion of duration exerting a negligible influence did not hold. Bommer et
al. [2004] studied the inelastic response of a series of masonry structures to a large
suite of strong-motion accelerograms and correlated the damage, measured in terms
of the loss of initial strength, with the average ordinate of the elastic acceleration
spectrum from initial period of the structure to a period about three times greater.
The study showed that some of the scatter in this correlation could be explained
by differences in the strong-motion duration of the records. However, that study
was focused on the assessment of existing vulnerable building stock rather than the
earthquake resistant-design of new constructions.
6.3. Matching records to the target spectrum

Once an initial search in terms of magnitude, distance and site classification has
been performed, depending on the number of records retrieved, further pruning then
needs to be carried out to acquire the number of records deemed necessary to obtain
stable results from the inelastic dynamic analyses. If there are far more records than
actually needed, the obvious choice would be to apply a second sweep of the search
using more restrictive criteria, such as a smaller distance range or insisting on a close
match with the site classification. The authors recommend that as an alternative
the user should consider performing a search within the results of the first sweep
through the database in terms of matching spectral shapes and perhaps, but less
importantly, spectral amplitudes. At this stage it is also advisable to pay attention
to the numbers of records coming from individual recording stations, and ensuring
that none is over-represented in the final suite of acceleration time-series.
The appropriate number of real accelerograms required to obtain stable mea-
sures of inelastic response is a subject on ongoing research and debate, but estimates
generally fall in the range from seven to ten. The practice of using small numbers
of records and then taking the maximum inelastic response should be abandoned;
if only a small number of records is to be used, these should be adjusted using
programs such as RASCAL or RSPMATCH in order to obtain a closer match with
the target spectrum. For larger number of records, the current practice of scaling
records so that their mean ordinate matches or exceeds the target elastic spectrum
is a reasonable approach, but it has been shown that careful manual selection of
May 18, 2004 10:44 WSPC/124-JEE 00159
sub-sets of records, and the application of individual scaling factors, can result in
a good match with the target spectrum and reduced dispersion amongst the scaled
records, as well as allowing many of the records to be used either at natural scale or
with only limited adjustment of their original amplitudes. It must be pointed out,
however, that this will not hold for situations with very high amplitudes of design
motions that may arise for the low annual frequencies of exceedance specified for
critical projects.
Executing fully dynamic non-linear structural analyses is time-consuming (and
therefore costly) in engineering practice and the use of spectrum-compatible records,
which allow fewer runs to be made, will often be preferred by design engineers. Our
recommendation is that in such cases use should be made of “intelligent artificial”
records — obtained by adjusting real accelerograms (see Sec. 4.2) — rather than
artificial time-series generated from white noise. In such cases, the guidelines pre-
sented in this paper are still applicable for selecting the seed accelerograms from
which the intelligent artificial records will be produced.
6.4. Code applications

In many cases, searching records in terms of earthquake scenarios will not be pos-
sible because the engineer will only have access to the design response spectrum
without any knowledge of the underlying hazard calculations. Matching PGA and
the spectral shape has been shown to be a superior approach to matching spectral
intensities, but regardless of the search parameters there is likely to be little control
on the duration — especially if the matching is focused on short-period spectral
ordinates — and hence there may be a large dispersion in the results of inelastic
analyses.
In nearly all current seismic design codes the earthquake actions are represented
by an approximation to a uniform hazard spectrum (UHS) obtained from a PSHA,
hence if accelerograms are to be selected in terms of magnitude and distance the
codes will need to present, in some form, the controlling earthquake scenarios to the
users. The disaggregated hazard can be displayed as supplementary maps showing
controlling scenarios for spectral ordinates at different periods [e.g. Harmsen et
al., 1999]; if the code drafters were prepared to surrender the UHS representation,
then an alternative would be to actually replace current zonation maps in codes
with maps showing contours of hazard-consistent magnitude and hazard-consistent
distance, from which the spectral ordinates can then be calculated [Bommer, 2000].
In view of the degree of approximation in the specification of earthquake actions
in seismic design codes, it would probably be feasible to simply present, in tabular
form, the magnitude-distance pairs defining the controlling earthquake scenarios for
each seismic zone. A useful extension to this would be for the code to provide a list of
suitable records identified to match each of these scenarios; the records themselves,
perhaps pre-scaled to ensure matching with the design elastic spectrum, could be
made available either on a CD- ROM distributed with the code or else accessed
May 18, 2004 10:44 WSPC/124-JEE 00159
through an Internet site maintained by the code authority or another approved

agency. The complementary relationship between current US codes and the USGS
disaggregation web site is a good model for such a solution.
Further work is needed to establish definitive criteria for selecting and scaling
real accelerograms, systematically exploring the influence of different selection crite-
ria and scaling factors on the inelastic response of realistic multi-degree of freedom
structures. In the meantime, engineers will need to continue to make judgements
regarding the degree to which the acceleration time-histories used in structural
analyses should reflect the characteristics of recorded motions, and to balance this
with the time and cost constraints on the number of inelastic dynamic analyses
that can be performed for an engineering project.
Acknowledgments
The authors firstly wish to express their gratitude to Dr John Douglas for his
interest in the work and for providing us with the current statistics of the European
strong-motion database. Additional thanks are due to Dr Douglas for carrying out
the data searches using the beta-version of his CD-ROM based search tool; the
opportunity to employ this facility before its general release is greatly appreciated.
We also express our thanks to Dr Rui Pinho and Dr Alain Pecker who read and
commented constructively on an early version of the manuscript. The second version
of the paper was further improved by very helpful comments from Edmund Booth
and Jonathan Hancock; particular thanks are due to Juliet Bird and Luis Fernando
Restrepo-Vélez, who both provided useful reviews of two different versions of the
manuscript. Very thorough reviews by Dr David Boore, Dr Norm Abrahamson and
an anonymous reviewer, all of which significantly improved the paper, are also noted
with special gratitude. The first author also wishes to acknowledge the insights
obtained from discussions of the issues addressed herein with Dr Paul Somerville.
We are also grateful to Dr Walt Silva, for providing us with the RASCAL
computer code, and to Dr Norm Abrahamson, for providing the RSPMATCH code.
We also extend our thanks to Melinda Squibb for providing us with information
about the COSMOS database.
References
Abrahamson, N. A. [1993] “Spatial variation of multiple support inputs,” Proceedings of
the First US Symposium on Seismic Evaluation and Retrofit of Steel Bridges, Univer-
sity of California at Berkeley, 18 October.
Abrahamson, N. A. and Shedlock, K. M. [1997] “Overview,” Seismological Research Letters
68(1), 9–23.
Abrahamson, N. A. and Silva, W. J. [1996] “Empirical ground motion models,” Report to
Brookhaven National Laboratory.
Abrahamson, N. A. and Silva, W. J. [1997] “Empirical response spectral attenuation
relations for shallow crustal earthquakes,” Seismological Research Letters 68(1),
94–127.
May 18, 2004 10:44 WSPC/124-JEE 00159
Ambraseys, N. N and Bommer, J. J. [1990] “Database of European strong-motion records,”

European Earthquake Engineering 5(2), 18–37.
Ambraseys, N. N and Bommer, J. J. [1991] “Uniform magnitude re-evaluation for the
strong-motion database of Europe and adjacent areas,” European Earthquake Engi-
neering 4(2), 3–16.
Ambraseys, N. N., Douglas, J., Rinaldis, D., Berge-Thierry, C., Suhadolc, P., Costa, G.,
Sigbjörnsson, R. and Smit, P. [2004] “Dissemination of European strong-motion data,
Vol. 2,” CD-ROM Collection, Engineering and Physical Sciences Research Council,
United Kingdom.
Ambraseys, N. N., Simpson, K. A. and Bommer, J. J. [1996] “The prediction of horizontal
response spectra in Europe,” Earthquake Engineering & Structural Dynamics 25,
371–400.
Ambraseys, N. N., Smit, P., Berardi, D., Cotton, F. and Berge, C. [2000] “Dissemina-
tion of European strong-motion data,” CD-ROM Collection, European Commission,
Directorate-General XII, Environmental and Climate Programme, ENV4-CT97-0397,
Brussels, Belgium.
Ambraseys, N. N., Smit, P., Douglas, J., Margaris, B., Sigbjörnsson, R., Ólafsson, O.,
Suhadolc, P. and Costa, G. [2003] “Internet site for European strong-motion data,”
Bolletino di Geofisica Teorica ed Applicata, in press.
ATC [1999] “Proceedings: Workshop on improved characterization of strong ground shak-
ing for seismic design,” ATC-35, Applied Technology Council, Redwood City, Cali-
fornia, 70 pp.
Bazzurro, P. and Cornell, C. A. [1999] “Disaggregation of seismic hazard,” Bulletin of the
Seismological Society of America 89, 501–520.
Beresnev, I. A. and Atkinson, G. M. [1998] “FINSIM — A FORTRAN program for simu-
lating stochastic acceleration time histories from finite faults,” Seismological Research
Letters 69(1), 27–32.
Bommer, J. J. [2000] “Seismic zonation for comprehensive definition of earthquake ac-
tions,” Proceedings of the Sixth International Conference on Seismic Zonation, Palm
Springs, California.
Bommer, J. J. and Ambraseys, N. N. [1992] “An earthquake strong-motion databank and
database,” Proceedings of the Tenth World Conference on Earthquake Engineering,
Madrid 1, 207–210.
Bommer, J. J. and Martı́nez-Pereira, A. [1999] “The effective duration of earthquake
strong motion,” Journal of Earthquake Engineering 3(2), 127–172.
Bommer, J. J. and Martı́nez-Pereira, A. [2000] “Strong-motion parameters: Definition,
usefulness and predictability,” Proceedings of the Twelfth World Conference on Earth-
quake Engineering, Auckland, Paper No. 206.
Bommer, J. J. and Ruggeri, C. [2002] “The specification of acceleration time-histories in
seismic design codes,” European Earthquake Engineering 16(1), 3–17.
Bommer, J. J. and Scott, S. G. [2000] “The feasibility of using real accelerograms for
seismic design,” In: Implications of Recent Earthquakes on Seismic Risk, eds. A.S.
Elnashai and S. Antoniou, Imperial College Press, pp. 115–126.
Bommer, J. J. Acevedo, A. B. and Douglas, J. [2003a] “The selection and scaling of
real earthquake accelerograms for use in seismic design and assessment,” Proceedings
of ACI International Conference on Seismic Bridge Design and Retrofit, American
Concrete Institute.
Bommer, J. J., Douglas, J. and Strasser, F. O. [2003b] “Style-of-faulting in ground-motion
prediction equations,” Bulletin of Earthquake Engineering 1(2), 171–203.
May 18, 2004 10:44 WSPC/124-JEE 00159
Bommer, J. J., Magenes, G., Hancock, J. and Penazzo, P. [2004] “Influence of strong-
motion duration on the seismic response of masonry structures,” Bulletin of Earth-
quake Engineering 2(1).
Bommer, J. J., Scott, S. G. and Sarma, S. K. [1998] “Time-history representation of
seismic hazard,” Proceedings of the Eleventh European Conference on Earthquake
Engineering, Paris.
Bommer, J. J., Scott, S. G. and Sarma, S. K. [2000] “Hazard-consistent earthquake sce-
narios,” Soil Dynamics & Earthquake Engineering 19, 219–231.
Boore, D. M. [2003] “Simulation of ground motion using the stochastic method,” Pure
and Applied Geophysics 160, 635–676.
Boore, D. M. [2004] “Can site response be predicted?” this volume.
Boore, D. M., Stephens, C. D. and Joyner, W. B. [2002] “Comments on baseline correc-
tion of digital strong-motion data: Examples from the 1999 Hector Mine, California,
earthquake,” Bulletin of the Seismological Society of America 92(4), 1543–1560.
Boore, D. M., W. B. Joyner and Fumal, T. E. [1997] “Equations for estimating horizontal
response spectra and peak acceleration from western North American earthquakes: A
summary of recent work,” Seismological Research Letters 68(1), 128–153.
Boore, D. M., Sims, J. D., Kanamori, H. and Harding, S. [1981] “The Montenegro, Yu-
goslavia earthquake of April 15, 1979: Source orientation and strength,” Physics of
the Earth and Planetary Interiors 27, 133–142.
Campbell, K. W. [1997] “Empirical near-source attenuation relationships for horizon-
tal and vertical components of peak ground acceleration, peak ground velocity, and
pseudo-absolute acceleration response spectra,” Seismological Research Letters 68(1),
154–179.
CEN [2002] Eurocode 8: Design of structures for earthquake resistance. Part 1: Gen-
eral rules, seismic actions and rule for buildings. Draft No. 5, May 2002, Document
CEN/TC250/SC8/N317, Comité Européen de Normalisation, Brussels.
Chapman, M. C. [1995] “A probabilistic approach to ground-motion selection for engi-
neering design,” Bulletin of the Seismological Society of America 85, 937–942.
Cramer, C. H. and Darragh, R. B. [1994] “Peak accelerations from the 1992 Landers and
Big Bear, California, earthquakes,” Bulletin of the Seismological Society of America
84(3) 589–595.
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.
Gasparini, D. A. and Vanmarcke, E. H. [1979] “Simulated earthquake motions compatible
with prescribed response spectra,” Evaluation of Seismic Safety of Buildings Report
No. 2, Department of Civil Engineering, MIT, Cambridge, Massachusetts, 99 pp.
Harmsen, S. D., Perkins, D. and Frankel, A. [1999] “Deaggregation of probabilistic ground
motions in the Central and Eastern United States,” Bulletin of the Seismological
Society of America 89, 1–13.
Hudson, D. E. [1979] “Reading and interpreting strong motion accelerograms,” EERI
Monograph, Earthquake Engineering Research Institute, Oakland, California.
Idriss, I. M. [1993] “Procedures for selecting earthquake ground motions at rock sites,”
NIST GCR 93-625, National Institute of Standards and Technology, Gaithersburg,
MD, 35 pp.
ISO [2003] Petroleum and natural gas industries — Specific requirements for offshore
structures — Part 2: Seismic design procedures and criteria. ISO/DIS 19901-2, In-
ternational Organization for Standardization.
Iyama, J. and Kuwamura, H. [1999] “Application of wavelets to analysis and simulation of
earthquake motions,” Earthquake Engineering & Structural Dynamics 28, 255–277.
May 18, 2004 10:44 WSPC/124-JEE 00159
Jeong, G. D. and Iwan, W. D. [1988] “The effect of earthquake duration on the damage
of structures,” Earthquake Engineering & Structural Dynamics 16, 1201–1211.
Kappos, A. J. and Kyriakakis, P. [2000] “A re-evaluation of scaling techniques for natural
records,” Soil Dynamics & Earthquake Engineering 20, 111–123.
Kaul, M. K. [1978] “Spectrum-consistent time-history generation,” ASCE Journal of the
Engineering Mechanics Division 104(ME4), 781–788.
Kramer, S. L. [1996] Geotechnical Earthquake Engineering, Prentice-Hall.
Krinitzsky, E. L. [2002] “How to obtain earthquake ground motions for engineering design,”
Engineering Geology 65, 1–16.
Krinitzsky, E. L. and Chang, F. K. [1977] “Specifying peak motions for design earth-
quakes,” State-of-the-Art for Assessing Earthquake Hazards in the United States, Re-
port 7, Miscellaneous Paper S-73-1. US Army Corps of Engineers, Vicksburg, Missis-
sippi.
Lee, W. H. K., Shin, T. C., Kuo, K. W., Chan, K. C. and Wu, C. F. [2001] “CWB free-field
strong-motion data from the 21 September Chi-Chi, Taiwan, earthquake,” Bulletin
of the Seismological Society of America 91(5), 1370–1376.
Leeds, D. J. [1992] “Recommended accelerograms for earthquake ground motions,” State-
of-the-Art for Assessing Earthquake Hazards in the United States, Report 28, Miscel-
laneous Papers S-73-1, US Army Corps of Engineering, Vicksburg.
Lekidis, V. A., Karakostas, C. Z., Dimitriu, P. P., Margaris, B. N., Kalogeras, I. and
Theodulidis, N. [1999] “The Aigio (Greece) seismic sequence of June 1995: Seismo-
logical, strong motion data and effects of the earthquakes on structures,” Journal of
Earthquake Engineering 3(3), 349–380.
Lilhanand, K. and Tseng, W. S. [1988] “Development and application of realistic earth-
quake time histories compatible with multiple-damping design spectra,” Proceedings
of the Ninth World Conference on Earthquake Engineering, Tokyo-Kyoto 2, 819–824.
Malhotra, P. K. [2003] “Strong-motion records for site-specific analysis,” Earthquake Spec-
tra 19(3), 557–578.
Martı́nez-Rueda, J. E. [1998] “Scaling procedure for natural accelerograms based on a
system of spectrum intensity scales,” Earthquake Spectra 14(1), 135–152.
Matsumura, K. [1992] “On the intensity measure of strong motions related to structural
failures,” Proceedings of the Tenth World Conference on Earthquake Engineering,
Madrid 1, 375–380.
McGuire, R. K. [1995] “Probabilistic seismic hazard analysis and design earthquakes:
Closing the loop,” Bulletin of the Seismological Society of America 85, 1275–1284.
McGuire, R. K., Silva, W. J. and Constantino, C. J. [2001] “Technical basis for revision
of regulatory guidance on design ground motions: Hazard- and risk-consistent ground
motion spectra guidelines,” NUREG/CR-6728, US Nuclear Regulatory Commission,
Washington D.C.
Mukherjee, S. and Gupta, V. K. [2002] “Wavelet-based generation of spectrum-compatible
time-histories,” Soil Dynamics & Earthquake Engineering 22, 799–804.
Naeim, F. and Anderson, J. C. [1996] “Design classification of horizontal and verti-
cal earthquake ground motion (1933–1994),” A Report to the USGS, JAMA Report
No. 7738.68-96, John A. Martin & Associates, Inc., Los Angeles.
Naeim, F. and Lew, M. [1995] “On the use of design spectrum compatible time histories,”
Earthquake Spectra 11(1), 111–127.
Nau, J. M. and Hall, W. J. [1984] “Scaling methods for earthquake response spectra,”
ASCE Journal of Structural Engineering 110(7), 1533–1548.
Preumont, A. [1984] “The generation of spectrum compatible accelerograms for the de-
sign of nuclear power plants,” Earthquake Engineering & Structural Dynamics 12(4),
481–497.
May 18, 2004 10:44 WSPC/124-JEE 00159
Reiter, L. [1990] Earthquake Hazard Analysis: Issues and Insights. Columbia University
Press.
Restrepo-Vélez, L. F. and Bommer, J. J. [2003] “An exploration of the nature of the
scatter in ground-motion prediction equations and the implications for seismic hazard
assessment,” Journal of Earthquake Engineering 7(Special Issue 1), 171–199.
Rey, J., Faccioli, E., and Bommer, J. J. [2002] “Derivation of design soil coefficients (S) and
response spectral shapes for Eurocode 8 using the European strong-motion database,”
Journal of Seismology 6, 547–555.
Row, L. W. [1996] “An earthquake strong-motion data catalog for personal computers —
SMCAT,” User Manual (version 2.0), National Geophysical Data Center, Colorado.
Seed, H. B. and Idriss, I. M. [1969] “Rock motion accelerograms for high magnitude earth-
quakes,” EERC Report No. 69-7, Earthquake Engineering Research Center, University
of California at Berkeley.
Seekins, L. C., Brady, A. G., Carpenter, C. and Brown, N. [1992] “Digitized strong-motion
accelerograms from North and Central America through 1986,” US Geological Survey
Digital Data Series DDS-7.
Shome, N., Cornell, C. A., Bazzurro, P. and Carballo, J. E. [1998] “Earthquakes, records
and nonlinear responses,” Earthquake Spectra 14(3), 469–500.
Silva, W. J. and Lee, K. [1987] “WES RASCAL code for synthesizing earthquake ground
motions,” State-of-the-Art for Assessing Earthquake Hazards in the United States,
Report 24, Miscellaneous Paper S-73-1. US Army Corps of Engineers, Vicksburg,
Mississippi.
Somerville, P. G., Smith, N. F., Graves, R. W. and Abrahamson, N. A. [1997] “Modification
of empirical strong ground motion attenuation relations to include the amplitude
and directivity effects of rupture directivity,” Seismological Research Letters 68(1),
199–222.
Stepp, C. J. [2000] “Coordination of strong-motion programs and strong-motion data
dissemination,” Proceedings of the Twelfth World Conference on Earthquake Engi-
neering, Auckland, Paper No. 2600.
Stewart, J. P., Chiou, S.-J., Bray, J. D., Graves, R. W., Somerville, P. G. and Abrahamson,
N. A. [2001] “Ground motion evaluation procedures for performance-based design,”
PEER Report 2001/09, Pacific Earthquake Engineering Research Center, University
of California, Berkeley.
Sunder, S. S. and Connor, J. J. [1982] “A new procedure for processing earthquake strong-
motion signals,” Bulletin of the Seismological Society of America 72(2), 643–661.
Trifunac, M. D., Udwadia, F. E. and Brady, A. G. [1979] “Analysis of errors in digitized
strong-motion accelerograms,” Bulletin of the Seismological Society of America 63(1),
157–187.
Vanmarcke, E. H. [1979] “Representation of earthquake ground motion: Scaled accelero-
grams and equivalent response spectra,” State-of-the-Art for Assessing Earthquake
Hazards in the United States, Report 14, Miscellaneous Paper S-73-1. US Army Corps
of Engineers, Vicksburg, Mississippi.
Wald, D. [1997] “Surfing the Internet for strong-motion data,” Seismological Research
Letters 68(5), 766–769.
Zeng, Y., Anderson, J. G. and Yu, G. [1994] “A composite source model for comput-
ing realistic synthetic strong ground motions,” Geophysical Research Letters 21(8),
725–728.
View publication stats

BommerAcevedo JEE 2004

Uploaded by

Copyright:

Available Formats

BommerAcevedo JEE 2004

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

BommerAcevedo JEE 2004

Uploaded by

Copyright:

Available Formats

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

The use of real accelerograms as input to dynamic analysis

Article in Journal of Earthquake Engineering · January 2004

Julian Bommer Ana Beatriz Acevedo

SEE PROFILE SEE PROFILE

Thyspunt SSHAC Level 3 PSHA View project

SARA project GEM Foundation View project

The user has requested enhancement of the downloaded file.

Journal of Earthquake Engineering,

THE USE OF REAL EARTHQUAKE ACCELEROGRAMS AS

ANA BEATRIZ ACEVEDO

Keywords: Strong-motion records; dynamic analysis; strong-motion databank; strong-

44 J. J. Bommer & A. B. Acevedo

The Use of Real Earthquake Accelerograms 45

46 J. J. Bommer & A. B. Acevedo

The Use of Real Earthquake Accelerograms 47

2. Availability of Strong-Motion Accelerograms

2.1. Global databanks

48 J. J. Bommer & A. B. Acevedo

2.2. Internet sites

The Use of Real Earthquake Accelerograms 49

50 J. J. Bommer & A. B. Acevedo

The Use of Real Earthquake Accelerograms 51

52 J. J. Bommer & A. B. Acevedo

3. Criteria for Selecting Strong-Motion Records

The Use of Real Earthquake Accelerograms 53

54 J. J. Bommer & A. B. Acevedo

An important implicit assumption underlying the recommendations made in this

3.1. Selection in terms of strong-motion parameters

The Use of Real Earthquake Accelerograms 55

56 J. J. Bommer & A. B. Acevedo

so demanding that it is almost impossible to obtain realistic ground motions that

3.2. Selection in terms of geophysical parameters

The Use of Real Earthquake Accelerograms 57

3.2.1. Earthquake magnitude

58 J. J. Bommer & A. B. Acevedo

Although techniques are available to adjust strong-motion records in a way

3.2.2. Source-to-site distance

Period (seconds) Period (seconds)

Boore et al.(1997) Abrahamson & Silva (1997)

Period (seconds) Period (seconds)

The Use of Real Earthquake Accelerograms 59

Ambraseys et al. (1996) Campbell (1997)

Period (seconds) Period (seconds)

Boore et al. (1997) Abrahamson & Silva (1997)

Period (seconds) Period (seconds)

60 J. J. Bommer & A. B. Acevedo

50 km, median values of spectral ordinates will reduce by about a factor of 7, as

The Use of Real Earthquake Accelerograms 61

5-75% Significant Duration (s)

3.2.3. Site classification

62 J. J. Bommer & A. B. Acevedo

3.2.4. Additional selection criteria

4. Matching Selected Records to the Elastic Response Spectrum

4.1. Matching criteria

The Use of Real Earthquake Accelerograms 63

of Part 1 of Eurocode 8, the specification is that no value of the mean 5% damped

64 J. J. Bommer & A. B. Acevedo

One notable feature of the specifications in seismic design codes is that it is