Download as pdf or txt
Download as pdf or txt
You are on page 1of 9

Chapter 4

Record Selection for Performing


Site-Specific Response Analysis

4.1 General

Compare to simplified methods using design codes, site-specific response analysis


is a more refined method to determine ground motions. Even though the P-waves
arrive first and cause the vertical shaking of the ground, it is normally the shear
waves that cause strong horizontal ground motions and possible subsequent
structural damage. Therefore, apart from the shear waves transmitted vertically
through soil (from bedrock to ground), which normally show little amplification or
de-amplification, a site-response analysis places special focus on the estimation of
the horizontal ground and soil motions due to the transmission of horizontal shear
(SH) waves propagated vertically. In addition, to capture different types of seismic
waves (body and surface waves) propagations, and also to account for the irregular
geometric properties of the sites (Sect. 3.6) that may affect the seismic wave
propagation, a two-dimensional or three-dimensional site-response analysis, usually
with the aid of finite element analysis, may need to be performed.
In a site-response analysis, it is normally required to use a minimum of three to
seven accelerogram sets for response history analysis. Each accelerogram set
selected must consist of at least two horizontal components, and in many cases, the
vertical component may also be included. It is important to select records that are
generally compatible with the earthquake magnitude and source-to-site distance
found from deaggregation (Sect. 11.1.4). If multiple magnitude–distance combi-
nations contribute significantly to the hazard, then one should select records from
each contributing earthquake as part of the total number of records. When the
hazard is controlled by faults producing moderate to large magnitude earthquakes at
locations near to the site, an appropriate number of ground motion records should
be selected to include near-fault and directivity effects, such as velocity pulses
producing relatively large spectral ordinates at long periods [218].
Obviously, the selection of time history records is essential in a site-response
analysis to calculate the design ground motions. Ideally, recorded accelerograms

© Springer International Publishing AG 2018 167


J. Jia, Soil Dynamics and Foundation Modeling,
https://doi.org/10.1007/978-3-319-40358-8_4
168 4 Record Selection for Performing Site-Specific Response Analysis

with response spectra equal to the design spectra should be used in site-response
analyses. However, this is almost impossible because the design spectra are based
on the medians of numerous response spectra, each with varying frequency content.
The resulting design spectra are therefore smoother or more uniform than any
individual spectrum derived from recorded time histories of earthquake motions.
Moreover, the equal probability of exceedance design spectra (also called uniform
hazard spectrum, as will be discussed in Chap. 11) can result from contributions
from various earthquakes with difference sources and magnitudes at different
spectra periods, with the larger magnitude seismic events contributing more to the
higher period.
Therefore, artificial accelerograms must be generated to best approximate the
design response spectrum at bedrock. This requires selection of motion recordings
and a modification of the recordings to fit into the design spectrum at bedrock. The
obtained artificial accelerograms at the bedrock are used to perform site-response
analyses in order to calculate the ground motion time histories at the selected
locations (depth), which are used to best estimate the ground motion response
spectra at the selected locations to reach a design response spectrum. The entire
procedure above is presented in Sects. 4.2, 4.3, and 4.4.
It is noted that selecting, scaling, and matching ground motions to obtain a
compatible set of records play a key role in seismic assessment with an acceptable
level of confidence. No unanimous consensus has yet been achieved on this subject.
The current book briefly introduces a few popular and general methods for selecting
and modifying ground motions recordings in order to perform a site-response
analysis. For more elaborated information on this topic, source [299], which was
developed as part of the PEER Transportation Research Program, is recommended.
Readers should bear in mind that the general rules for selecting, scaling, and
matching ground motions discussed in this chapter can also be applied for obtaining
artificial ground motion histories that are compatible with the design spectra at
ground surface of a site, which may also be used for seismic time history analyses,
even if the site effects are roughly accounted for when using a code-based design
spectrum.

4.2 Selections of Motion Recordings

The ground motion suite must include a wide range of intensity measures repre-
sentative of seismic hazard at the area of interest. Ideally, the response spectra of the
selected recordings (obtained from measured recordings at different stations) should
approximately fit or reasonably envelop the ground motion design spectra (typically
given by seismic design codes) at the bedrock or rock outcrop. However, if this is
not attainable, a modification of selected recordings or generation of artificial
bedrock seismic motions has to be performed to fulfill the requirements above, as
will be discussed in Sect. 4.3.
4.3 Modification of the Recordings to Fit into the Design … 169

4.3 Modification of the Recordings to Fit into the Design


Rock Spectrum

As mentioned in Sect. 4.2, in order to generate time histories to best approximate


the design response spectrum at bedrock, it is important to match the motion
recordings to the target design spectrum. This process to obtain the matched time
histories is typically carried out by trial and error. Note that there is no general
consensus on the best practice for the scaling/modification of the ground motion
records; it is still a subject of significant study, and the results vary with the degree
of inelastic response of structure for the chosen level of seismic hazard [300].
Generally, two methods are available for performing the matching, namely direct
scaling and spectrum matching, as will be discussed in Sects. 4.3.1 and 4.3.2.

4.3.1 Direct Scaling

Direct scaling involves applying one constant factor to individual pairs of ground
motions to make their response more closely match the target design spectrum over
a certain range of periods [301]. Because elastic response spectra correspond to
linear response of SDOF systems, the same scale factor applies to spectral accel-
erations at all periods.

4.3.2 Spectrum/Spectral Matching

Spectrum/spectral matching, also called response spectrum compatibilization, is


intended to manipulate each individual ground motion record in the time domain by
the addition of wave packets, i.e., the recordings are scaled with different factors in
different frequency ranges, using, for example, wavelets algorithm [302, 303], such
that the response spectra would approximate target design spectra at the bedrock or
rock outcrop.
A slightly different method is to scale each ground motion record so that the
average response spectrum over the periods of interests is equal to the average of
the target spectrum over the same periods.
By modifying the spectral shape of each individual ground motion record,
spectral matching essentially adjusts the frequency content of accelerograms until
the response spectrum is within user-specified limits of a target response spectrum
over a defined period range [304, 305]. One way to determine the scaling factor b
is, for example, to minimize the errors with respect to a target acceleration spectrum
at a number (Nr) of eigenperiods typically below 4 s:
170 4 Record Selection for Performing Site-Specific Response Analysis

!!
X
Nr
b  Sa ðTi Þrecord
error ¼ min 1 ð4:1Þ
i¼1 Sa ðTi Þtarget

where Sa ðTi Þtarget and Sa ðTi Þrecord are the spectral accelerations for the ith eigenpe-
riod from the target spectrum and the response spectrum of the original selected
record, respectively.
Figure 4.1 shows that three original selected recordings (Kobe earthquake,
January 17, 1995; Imperial Valley of EI Centro Site earthquake, May 18, 1940;
Trinidad earthquake, March 1988) are matched to the target design spectrum with
respect to both amplitude and frequency content, and the three matched spectra are
transferred back to the time histories as shown in Fig. 4.2.
Therefore, spectral matching generally reduces spectral variability among a suite
of records and increases the pool of available ground motions with a certain shape.
It is the former point that has made this process especially appealing to practicing
engineers [306]. Moreover, the reduction in spectral variability due to spectrum
matching generally leads to a reduction in nonlinear structural response variability
[307], which essentially implies that fewer ground motions are required to obtain an
estimate of a particular response quantity [308].
Another widely adopted spectrum scaling method is based on the 5% damped
first mode spectral acceleration, which can minimize the bias of damage estimation
statistically [309]. This is because the response spectra for time histories matched
with the first mode spectral acceleration are generally less “broadband” than those
matched with uniform hazard spectra at entire periods or over a large range of
periods, even though uniform hazard spectra usually envelope the first mode
spectral acceleration. However, for many structures, the higher order of vibration
modes also makes an important contribution to seismic responses. For example,
shear forces at the upper stories of a building or floor accelerations can be more
sensitive to higher order mode excitation than that of the first eigenmode vibrations,
even if the displacement response is more dependent on the fundamental global
vibration mode response. Therefore, a scale factor that covers a larger period range
should then be adopted.
The seismic records can be obtained from several public ground motion record
databases, which are often accompanied by a number of earthquake and site
parameters, such as earthquake magnitude, source-to-site distance, and ground type
at the corresponding recording stations. Section 4.5 provides links to some such
databases.
The target spectra can be obtained using one of the following two options:
(1) the design response spectra calculated from the relevant seismic design code
procedures, which correspond roughly to uniform hazard spectra (Chap. 11) for the
site; or (2) site-specific scenario spectra that preserve realistic spectral shapes for
controlling earthquakes and match the design spectral ordinate at different periods
of interest, also known as conditional mean spectra (CMS), as will be presented in
Sect. 11.4.
4.3 Modification of the Recordings to Fit into the Design … 171

Fig. 4.1 Comparison between the target design rock spectrum and the response spectrum for 5%
damping of a the three selected recordings (upper) and b the matched spectra and their average
(lower)

The process of spectrum matching can be carried out by several commercial


codes, such as SeismoMatch [567] or RspMatchEDT [230], through which the
response spectrum of an individual motion record can be modified to match the
target design spectrum.
172 4 Record Selection for Performing Site-Specific Response Analysis

Fig. 4.2 Comparison


between the original motion
recordings and the time
history calculated from the
matched spectra (Fig. 4.1)
with 5% damping
4.3 Modification of the Recordings to Fit into the Design … 173

4.3.3 Pros and Cons of Direct Scaling and Spectrum


Matching

Both direct scaling and spectrum matching have merits and drawbacks.
Both methods are inconsistent with the physical conditions and can result in
characteristics that differ from the ground motion recordings [310, 311], except for
cases in which time histories are matched to a conditional mean spectrum
(Sect. 11.4).
Direct scaling in certain senses retains the original ground motion characteristics,
including peaks and valleys in the response spectrum. However, it is questionable,
since scaling a ground motion does not account for variations in ground motion
characteristics (e.g., frequency content), which change with intensity [312], i.e.,
extreme intensity ground motions have different frequency contents than
low-intensity ground motions. This is more obvious for long-period ground
motions. Moreover, natural periods for many structures are high, while ground
motion records may not have such a long period; this hence requires a large scaling
factor, which may result in unrealistic structural responses.
The drawbacks of simple scaling can be overcome by the spectrum matching
method, which usually presents a larger number of cycles than the recorded ground
motions, indicating a stronger ground motion for a wider range of structural peri-
ods, which implies a more conservative loading scenario [313]. However, since the
spectrum matching method changes the frequency content of the original records, it
may eliminate the inherent record-to-record spectral shape variability and the
variability of the structural response. This drawback of spectrum matching has been
realized by many researchers, and recent research [310, 314–318] shows a
non-conservative bias in the median nonlinear displacement of structural models
analyzed by a suite of matched ground motions compared to a suite of unmatched
ground motions representing an equivalent scenario. Note that a time-domain
analysis requires only a few ground motions to be analyzed. This may eliminate the
variability of a structure’s dynamic response if those ground motions are obtained
from spectrum matching [319]. This bias appears to result partially from the scaled
ground motions having inappropriate values of spectral shape or the parameter e
(Sect. 11.4), which is an indirect measure of spectral shape [555]. This bias is more
apparent for uniform hazard spectra, which often represent unrealistic seismic
motions, as they combine the statistical variability of several earthquakes rather
than a single one. Therefore, a number of seismic design codes discourage the use
of this method, or only permit its use with great care by fulfilling a list of strict
criteria, or if several site-response analyses are performed, with the majority of the
seismic motion time histories obtained from the real records and the rest (the
minority) from calculation by the spectrum matching method, as recommended by
the code for seismic design of buildings in China [199]. A recently proposed
method to eliminate the bias due to the inappropriate values of spectral shape, or the
parameter e is to adopt a few conditional mean spectra conditioned at a number of
periods of interests, which will be presented in Sect. 11.4.
174 4 Record Selection for Performing Site-Specific Response Analysis

Moreover, the duration of strong ground motion is dependent on magnitude and


is not explicitly represented in a spectral shape [320], though peak displacement
amplitude does tend to increase with time based on random vibration theory.
Therefore, demand parameter depending on degradation (e.g., collapse capacity)
may be biased if the duration is not included in the ground motion selection criteria.

4.4 Performing the Site-Response Analysis Using


Modified/Matched Recordings

After selecting the ground motion recordings, a modification or matching of those


recordings may be performed to fit them into the design response spectrum at
bedrock/rock outcrop as has been presented in Sects. 4.2 and 4.3. By inputting the
modified/matched motion histories at the bedrock/rock outcrop and carrying out
site-response analysis, response time histories and response spectra at seabed can
then be calculated. As mentioned before, most of the codes and standards, such as
Eurocode 8 [198], Norsok N-003 [628], and China’s building code [199], require
using at least three bedrock motion histories to compute the ground motion time
histories and the response spectra. ISO 19901-2 [289] requires a minimum of four
and recommends that seven-time history analyses be used to capture the random-
ness in a seismic event.
As discussed in Sect. 3.5.1, as rock is generally not perfectly rigid, the motions
at the base of the soil profile (rock) are not identical to that of the rock outcrop,
while stations for measuring recordings are generally located on rock outcrop (or
even soil layers) rather than the bedrock [220]. Therefore, the recordings are
preferably assigned at a hypothetical bedrock outcrop (point 1 in Fig. 3.2) rather
than the base of the soil profile (point 2 in Fig. 3.2). Moreover, it is noticed that,
historically, seismic records have been measured on a variety of soil conditions,
ranging from rock outcrop to very soft soil, but rarely at bedrock due to technical
challenges. More recently, researchers have become more careful in selecting the
locations for installing the accelerometers and tried their best to place the
accelerometers on rock outcrops or very stiff soil.
It is recommended that one performs a sensitivity analysis. Accounting for the
variation of soil properties from lower bound, average (best estimate) to
upper-bound value may detect some significant impact of soil properties on the
ground motions. If the ground motions are not sensitive to the variation of soil
properties, typically, the best-estimate value of soil properties can be used to
compute the ground motion time histories. Thereafter, a site-response analysis can
be carried out typically by either equivalent linear analysis (Sect. 2.2) or by non-
linear analysis. Various computation codes are available to perform this task, such
as SHAKE 2000 or D-MOD2000.
The set of time histories of ground motions computed through a site-response
analysis can then be directly used for calculating structural responses in time
4.4 Performing the Site-Response Analysis Using … 175

domain. Alternatively, the set of response spectra obtained from the time histories
of ground motions may be interpreted in some means (e.g., by taking the mean
value of the all response spectra) and smoothed to develop one design response
spectrum at seabed/ground surface, which can account for the contribution from all
individual ground motion histories of the set of time histories, and be applied for
calculating structural responses using response spectrum analysis, as will be pre-
sented in Chap. 11. Figure 3.38 shows an example of response spectra of three
computed horizontal acceleration time-histories on the seabed/ground surface for
earthquake events with a return period (Sect. 11.1.3) of 10,000 years. Using the
average values of the three computed spectra, a design response spectrum can be
obtained as shown in Fig. 3.39.

4.5 Sources of Ground Motion Recording Data

Records of seismic ground motions are essential for the development of earthquake
engineering, for purposes such as to study the attenuation relationship, to calculate
the site-specific ground motions, to develop the seismic hazard map. Fortunately,
several databases of ground motion records have been made publicly available,
accompanied by a number of earthquake and site parameters, such as earthquake
magnitude, source-to-site distance, and ground type at the recording station. The
best-known ones are listed below:
• The Pacific Earthquake Engineering Research Center, PEER (http://peer.
berkeley.edu/smcat/)
• US Geological Survey (USGS) (http://nsmp.wr.usgs.gov for time histories of
ground motion and http://earthquake.usgs.gov/eqcenter/ for earthquake data)
• SeismicPortal (http://www.seismicportal.eu/)
• The National Geophysical Data Center, NGDC (http://www.ngdc.noaa.gov/
nndc/struts/form?t=101650&s=1&d=1)
• The China Earthquake Data Centre (www.smsd-iem.net.cn)
• The Broadband Seismograph Network of Japan (www.fnet.bosai.go.jp)
• Kyoshin Network K-NET (at http://www.k-net.bosai.go.jp/)
The time history records in the databases above are normally corrected by the
standard processing techniques that remove low- and high-frequency noise from
them. This is because raw records may contain also nonstandard errors, and
therefore, uncorrected records should not be used for the analyses [321]. The
nonstandard errors include insufficient digitizer resolution, transversal wave trigger,
insufficient sampling rate, multiple baselines, spikes, multiple shocks, early ter-
mination during recording and clipping [322].

You might also like