Вы находитесь на странице: 1из 17

EARTHQUAKE ENGINEERING & STRUCTURAL DYNAMICS

Earthquake Engng Struct. Dyn. 2016


Published online in Wiley Online Library (wileyonlinelibrary.com). DOI: 10.1002/eqe.2739

SPECIAL ISSUE PAPER

Spectral shape metrics and structural collapse potential

Laura Eads1,*,†, Eduardo Miranda2 and Dimitrios Lignos3


1
Risk Management Solutions, London EC3R 8NB, UK
2
Department of Civil and Environmental Engineering, Stanford University, Stanford, CA 94305-4020, USA
3
Department of Civil Engineering and Applied Mechanics, McGill University, Montreal, QC H3A 2K6, Canada

SUMMARY
This paper examines various parameters that provide a measure of spectral shape and studies how they relate
to the potential of ground motion records to cause the collapse of a given structure. It is shown that when
measuring the ground motion intensity by the spectral acceleration at the first-mode period of the structure,
Sa(T1), records causing collapse at low ground motion intensities typically have significantly different
spectral shapes than those that do not cause collapse until much higher ground motion intensities. A spectral
shape typical of damaging records is identified, and a metric for quantifying the spectral shape of a record
called SaRatio is proposed and evaluated. SaRatio is defined as the ratio between Sa(T1) and the average
spectral value over a period range. The ability of SaRatio to predict the collapse intensity, i.e. the minimum
intensity at which a given ground motion causes the collapse of a given structure, is compared to other
recently proposed spectral shape metrics including epsilon (ε), eta (η) and Np. The results demonstrate that
SaRatio is typically a much better predictor of collapse intensity than other spectral shape metrics. Copyright
© 2016 John Wiley & Sons, Ltd.

Received 30 June 2015; Revised 17 December 2015; Accepted 10 March 2016

KEY WORDS: structural collapse; spectral shape; record-to-record variability; collapse intensity;
performance-based earthquake engineering; ground motions

1. INTRODUCTION

As early as 1952 Housner used the response spectrum to quantify earthquake intensity and estimate the
demands earthquakes impose on structures [1], as he pointed out that peak ground acceleration ‘is not a
good measure of the intensity of shaking as regards effects on structures’ [2]. Currently the most
commonly used measure of ground motion intensity for both design and for estimating the seismic
performance of structures is through the use of response spectrum pseudo-acceleration ordinates,
which provide the peak displacement response of linear elastic single-degree-of-freedom (SDOF)
systems as a function of the period and level of damping. When estimating the seismic performance
of structures, the structural response estimates are often conditioned on a measure of ground motion
intensity such as the 5%-damped pseudo-acceleration spectral ordinate at the first-mode period of
the structure, Sa(T1). Unless stated otherwise, this paper assumes that the ground motion intensity is
measured by the value of Sa(T1). While ground motions with identical Sa(T1) values will produce
the same peak responses in a 5%-damped linear elastic SDOF system, the ground motions will, in
general, produce different peak responses in inelastic SDOF and multiple-degree-of-freedom
(MDOF) systems. Many studies by Cornell and his research associates as well as by others have
found that incorporating information about the spectral shape (i.e. information beyond the ordinate

*Correspondence to: Laura Eads, RMS, Peninsular House, 30 Monument Street, London EC3R 8NB, UK.

E-mail: laura.eads@rms.com

Copyright © 2016 John Wiley & Sons, Ltd.


L. EADS, E. MIRANDA AND D. LIGNOS

at the fundamental period) or peak ground responses improves the prediction of nonlinear structural
response (e.g. [3–12]). Kennedy et al. [3] considered the average spectral value between the
fundamental period of the structure and a lengthened period and noted that the relationship between
the average spectral value and the value of Sa(T1) influenced the nonlinear response of the structure.
In particular, they observed that when the average spectral value was greater than Sa(T1), the ground
motion would require less scaling beyond the yield level to produce a certain ductility demand
compared to a ground motion where the average spectral value was less than Sa(T1). Similarly,
Sewell et al. [4] noted that the damage potential of records was related to the ‘spectral breadth,
slope, etc. obtained from spectral amplitudes at several frequencies over the frequency range of
interest’, and that for records scaled to the same Sa(T1), the slope of the spectrum at T1 was related
to the damage potential of the ground motion record.
Several approaches have been proposed for improving the estimation of nonlinear structural
response by considering the effects of spectral shape. Cordova et al. [7] proposed using the spectral
ordinate at a period longer than T1 in addition to that at T1 and noted that the additional spectral
ordinate led to a reduction in the variability of the peak inter-story drift demand. Vamvatsikos and
Cornell [9] evaluated the use of one, two and three spectral ordinates as well as a weighted sum
over the spectral range of interest of the logarithmic ratio of spectral ordinates normalized by Sa(T1).
Baker and Cornell [10] found that epsilon (ε), which is the number of standard deviations by which
the logarithmic spectral acceleration of a record is above or below the mean logarithmic spectral
acceleration estimated by a ground motion prediction equation (GMPE), influenced peak inter-story
drift demands and proposed its use in record selection as a proxy to spectral shape. In particular
they noticed that ground motions were on average less damaging to structures as ε increased.
Subsequent studies observed that ε also has an influence on the collapse potential of structures and
proposed methods to use this proxy to spectral shape when evaluating the probability of collapse of
structures [13, 14]. Another spectral shape proxy called eta (η), which is a linear combination of the
ε based on spectral acceleration and the ε based on peak ground velocity, has been shown to be
more effective than ε (based on spectral acceleration alone) at predicting the collapse intensity of
structures [15]. Bojórquez and Iervolino [11] found that including spectral shape measure Np,
defined as the ratio of the average spectral value between T1 and 2 · T1 to the value of Sa(T1), could
reduce the dispersion in peak inter-story drift ratios by up to 70% compared to using Sa(T1) alone.
This paper examines the relationship between spectral shape and collapse intensity. A four-story
steel moment-resisting frame building is used to demonstrate that records causing collapse at low
ground motion intensities typically have significantly different spectral shapes than records that do
not cause collapse until much higher ground motion intensities. The spectral shape is quantified
through a parameter called SaRatio, which is a measure of spectral shape and is used to distinguish
the shapes of different records. This metric is shown to be strongly correlated with the collapse
intensity of a ground motion record. The correlation between collapse intensity and spectral shape
metrics SaRatio, ε and η are compared for nearly 700 moment-resisting frame and shear wall
structures of various heights.

2. SPECTRAL SHAPE OF DAMAGING RECORDS

This section focuses on identifying and quantifying the spectral shape of damaging records. Results
from a case study of a four-story office building with steel special moment-resisting frames designed
according to the US practice in highly seismic regions are used.

2.1. Structure and ground motions


The structure is a four-story office building designed for the metropolitan area of Los Angeles,
California according to the 2003 International Building Code [16] and the 2005 AISC seismic
provisions [17]. The lateral force-resisting system is a steel special moment-resisting frame (MRF)
incorporating reduced beam sections with a design base shear coefficient of V/W = 0.082. A two-
dimensional model of one of the MRFs was created in OpenSees [18]. A concentrated plasticity

Copyright © 2016 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. (2016)
DOI: 10.1002/eqe
SPECTRAL SHAPE METRICS AND STRUCTURAL COLLAPSE POTENTIAL

modeling is used with the frame members modeled as elastic elements with nonlinear rotational springs
connected in series at their ends. A modified version of the bilinear Ibarra–Medina–Krawinkler
deterioration model [19] that captures stiffness and strength deterioration, including in-cycle
degradation is used to model the hysteretic behavior of the nonlinear springs. P–Δ effects are
simulated via a leaning column carrying gravity loads that is connected to the frame by axially rigid
beams pinned at both ends. The first three modal periods of the structure are 1.33 s, 0.43 s and
0.22 s. Rayleigh damping is implemented following the approach of Zareian and Medina [20] with
2% of critical damping assigned to the first and third modes of vibration. For additional information
of the building, its design and modeling the reader is referred to [19, 21].
Ground motions used to assess the collapse consist of 137 acceleration records (each with two
horizontal components). For the purposes of this paper each horizontal component was treated as an
independent record, for a total of 274 ground motions. All records were downloaded from the PEER
ground motion database [22]. Ground motions records were selected from earthquakes with strike-
slip, reverse or reverse-oblique fault mechanisms with moment magnitudes, Mw, between 6.9 and
7.6, Joyner-Boore distances between 0 and 27 km, and from free field stations that can be classified
as NEHRP site classes C or D. These parameters were chosen based on the seismic hazard
deaggregation at the site. The ground motion set includes records with a variety of characteristics,
including pulse-like (records affected by forward directivity) as well as non-pulse-like records.
Further details on the ground motion records are provided in [21].

2.2. Collapse assessment results


Each of the 274 ground motions is scaled until it produces the collapse of the structure (i.e. lateral
dynamic instability in which the structure no longer experiences a vibratory oscillation in the
opposite direction but experiences only monotonic displacements in one direction that increase
rapidly without bounds). The minimum peak ground acceleration (PGA) value at which each record
triggers collapse ranges from as small as 0.25 g to as high as 12.3 g with a logarithmic standard
deviation (σ ln) of 0.77, illustrating the enormous record-to-record variability in collapse PGA
because of the low correlation between this intensity measure (IM) and collapse. The minimum Sa
(T1) value at which each record triggers collapse (Sa(T1)col) ranges from 0.48 g to 3.27 g, with a
median collapse intensity of Sa(T1)col = 1.03 g and σ ln = 0.39. The reduction in structural response
dispersion for Sa(T1) relative to PGA is in agreement with previous studies (e.g. [1, 2]) and
indicates that Sa(T1) is a much better parameter than PGA for describing the intensity of a ground
motion as it relates to structural collapse estimation. Although the Sa(T1)col values have significantly
less variability than the PGA values at collapse, the Sa(T1) value at which all ground motions
produce collapse of the structure is 6.8 times greater than the Sa(T1) value at which only one ground
motion produces collapse of the structure.
The displacement spectra of records associated with the 20 lowest collapse intensities (i.e. those in
the lower tail of the collapse fragility curve) and with the 20 highest collapse intensities (i.e. those in
the upper tail of the collapse fragility curve) are shown in Figure 1 where all records are scaled to Sd

Figure 1. Comparison of displacement spectra for records associated with the (a) 20 lowest; and (b) 20
highest collapse intensities, where all records are scaled to the same Sd(T1).

Copyright © 2016 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. (2016)
DOI: 10.1002/eqe
L. EADS, E. MIRANDA AND D. LIGNOS

(T1) = 10 cm. The spectra are presented in the displacement domain rather than the acceleration domain
to make it easier to observe differences in spectral shape. As shown in Figure 1, the spectral shape of
the ‘damaging’ records (those with the lowest collapse intensities) is significantly different than the
spectral shape of the ‘benign’ records (those with the highest collapse intensities). With the
exception of one record1 that is shown with a black line, the spectral displacement shapes of
damaging records generally exhibit sharply increasing spectral ordinates for periods greater than T1.
The spectral displacement shapes of benign records, on the other hand, exhibit relatively constant or
slightly decreasing spectral ordinates for periods greater than T1. For reference, the mean collapse
intensity of the records in Figure 1(a) is Sa(T1)col = 0.55 g and 2.37 g for the records in Figure 1(b),
compared to median collapse intensity of the whole ground motion set of Sa(T1)col = 1.03 g. These
Sa(T1)col values indicate that records in Figure 1(b) must be scaled up more than four times, on
average, relative to those in Figure 1(a) to cause collapse, meaning the records in Figure 1(a) can be
considered to be approximately four times as damaging as the records in Figure 1(b).
A geometric mean spectrum indicated by a thick red dashed line has been added to each plot in
Figure 1 to illustrate that, on average, there is a clear difference in the spectral shape of the two
groups of records. In particular, the displacement spectra of the damaging records in Figure 1(a),
that is those that produce collapse at lower intensities, tend to exhibit a sharper increase in spectral
ordinates for T > T1. It should be noted that simply having a record whose displacement spectral
ordinates sharply increase for T > T1 does not necessarily mean that it will be damaging to a
particular structure (i.e. produce collapse at a low Sa(T1) intensity). Sharply increasing displacement
spectral ordinates for T > T1 simply indicate that the record imposes significant displacement
demands on linear elastic SDOF systems with periods longer than T1. However, it should be noted
that the response of nonlinear systems can be quite different from that of linear systems. In
particular, if the ground motion causes the structure to undergo significant inelastic excursions in
primarily a single direction the damaging effects will be cumulative, leading to a ratcheting-type
behavior. On the other hand, if the structure experiences significant inelastic excursions in both
displacement directions some inelastic excursions might tend to re-center the structure. In this case it
will typically take a higher ground motion intensity to produce the collapse of the structure
compared to a record that primarily pushes the structure in one direction. Note that, among many
other factors, the location and type of mechanism, strong motion duration and the effects of cyclic
deterioration in the strength and stiffness of structural components can also influence how a system
responds to the ground motion, as severe deterioration may make the structure less likely to
experience a significant reduction in residual displacement.

3. MEASURES OF SPECTRAL SHAPE

3.1. Epsilon (ε)


Epsilon has been used for a long time by seismologists to measure how much a spectral ordinate differs
from the median value computed with a GMPE. The ε value is computed by taking the difference of the
natural logarithm of the spectral acceleration of the record at the period of interest and the mean of the
natural logarithm predicted by a GMPE and then dividing the difference by the logarithmic standard
deviation associated with the GMPE at the period of interest as follows

LnSaðT Þ  μLnSaGMPE
ε≡ (1)
σ LnSaGMPE

1
This record is from the 1999 Duzce, Turkey earthquake and was recorded at the Lamont 375 station (PEER NGA record
1617, first horizontal component). This record, which has been highlighted by D. Boore (http://www.daveboore.com/
daves_notes/daves_notes_whats_wrong_with_duzce.pdf) as being peculiar, has unusually high ordinates in the short
period range (e.g. Sd(0.4 s) = 31 cm as seen in Figure 1) caused by unusual high frequency content in the record and leads
to a collapse mechanism in the fourth (top) story. The vast majority of the 274 records lead to a collapse mechanism
involving the lower three stories.

Copyright © 2016 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. (2016)
DOI: 10.1002/eqe
SPECTRAL SHAPE METRICS AND STRUCTURAL COLLAPSE POTENTIAL

where LnSa(T) is the natural logarithm of the spectral acceleration ordinate of the record at the period
of interest, μLnSaGMPE is the mean logarithmic spectral acceleration ordinate estimated by a GMPE at
the period of interest and σ LnSaGMPE is the logarithmic standard deviation estimated by the same
GMPE at the period of interest. The numerator in Equation (1) is the residual of the spectral
ordinate, and therefore epsilon is a normalized residual of the spectral ordinate. Because the
normalization is performed with the logarithmic standard deviation, ε measures the number of
standard deviations that a spectral ordinate is above (for the case of positive values of ε) or below
(for the case of negative values of ε) the median spectral ordinate predicted by a certain GMPE. It
should be noted that both μLnSaGMPE and σ LnSaGMPE depend on the particular GMPE that is used,
and therefore ε also depends on the GMPE used in the calculation.
Baker and Cornell [10] noted that if two records were scaled to the same Sa(T1) value, the record
with higher Sa values at periods near T1 (e.g. a record whose pseudo-acceleration spectrum has a
‘valley’ at or near T1) would tend to cause larger responses in a nonlinear MDOF system.
Conversely, a record with lower Sa values at periods near T1 (e.g. a record whose pseudo-
acceleration spectrum has a ‘peak’ at or near T1) would tend to cause smaller responses in a
nonlinear MDOF system. Furthermore, they noted that if a record has a peak or a valley at the
fundamental period of the structure, then ε, which measures deviation from the expected spectral
value, may be an indicator of this condition and, if so, would be useful for predicting structural
response. Hence, they proposed using ε as a convenient implicit measure of spectral shape (or proxy
to spectral shape). It should be noted that they found that the use of ε as a proxy to spectral shape is
not significantly affected by the choice of GMPE [10].
A limitation of ε that is often overlooked is that is not effective in capturing the effects of spectral
shape for pulse-like near-fault ground motions [10], and therefore its use with this type of ground
motions is not recommended [10, 14]. This is an important limitation of ε because pulse-like near-
fault ground motions are one of the most important contributors to the collapse risk of structures
located near active seismic faults.
Several studies have indicated that records with negative ε values tend to be more damaging than
records with positive ε values [10, 13, 14]. These studies have provided valuable information but
have not compared this proxy of spectral shape to other proxies or to parameters that are more direct
measures of spectral shape. Figure 2 shows a scatter plot of ε calculated using the 2008 Boore and
Atkinson GMPE (BA08) [23] versus the natural logarithm of collapse intensity in the four-story
structure for all 274 records in the ground motion set. A linear regression and its corresponding
equation are also added to Figure 2. It can be seen that although there is a tendency of increasing
collapse intensity (i.e. the ground motion becomes less damaging) with increasing ε, significant
scatter exists above and below the regression line fitted to the data. The correlation coefficient (ρ)
between ε and the natural logarithm of collapse intensity is relatively low (ρ = 0.32). Similarly the

Figure 2. Relationship between ε and the natural logarithm of the collapse intensity for the four-story steel
building.

Copyright © 2016 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. (2016)
DOI: 10.1002/eqe
L. EADS, E. MIRANDA AND D. LIGNOS

coefficient of determination (R2) is only R2 = 0.10. In other words only 10% of the variability observed
in the collapse intensity can be explained by ε.
To further evaluate these results, the collapse intensity of each record was modified to take into
account the ε of each record following the procedure proposed by Zareian and Krawinkler [13] as
follows

LnSa′ðT 1 Þ ¼ LnSaðT 1 Þ þ mðε  εset Þ (2)

where LnSa′(T1) is the adjusted natural logarithm of the spectral acceleration ordinate of the record at
the period of interest, LnSa(T1) is the unadjusted natural logarithm of the spectral acceleration ordinate
of the record, m is the slope of a straight line fitted by regression analysis using all (ε, LnSa(T1)) data
pairs, m is the number of standard deviations that LnSa(T1) is above or below μLnSaGMPE and εset is the
mean of all ε in the ground motion set. Equation (2) adjusts the collapse intensity of each record by
taking into account the ε of the record, increasing the intensity for records with ε < εset and
decreasing the intensity for records with ε > εset . For the ground motion set of 274 records,
m = 0.149 and εset = 0.13 (an approximately ‘ε-neutral’ ground motion record set) when using the
BA08 GMPE. Even when adjusting the collapse intensity of each record by considering its ε value,
the logarithmic standard deviation of the adjusted collapse capacities becomes σ LnSa′ = 0.37, which is
only 5% smaller than that of the unadjusted collapse capacities (σ LnSa = 0.39) and indicates that
taking into account the ε of each record does not significantly reduce the variability in collapse
intensities for this building compared to the variability computed using Sa(T1) alone. Haselton et al.
[14] also found that the reduction in the variability of collapse intensities when accounting for ε is
not significant. This is because the (ε, LnSa(T1)) data pairs with ε < εset (mainly those with negative
ε) that are significantly above the regressed line or the (ε, LnSa(T1)) data pairs with ε > εset (mainly
those with positive ε) that are significantly below the regressed line are moved further away from
the regression after adjustment. Similarly, the (ε, LnSa(T1)) data pairs with ε ≈ εset (records that are
approximately ε neutral) remain largely unchanged after adjustment for ε even if they are
significantly above or below the regression. Even though adjusting for ε may reduce/correct a
possible bias in the collapse risk estimate [10, 13, 14], a large number of records are still required to
obtain a reliable estimate of the collapse intensity [24].
Literature documenting the relationship between the natural logarithm of the collapse intensity and ε
reports ρ values on the order of 0.4 to 0.6 (R2 from 0.2 to 0.4) when using record sets without near-
fault, pulse-like motions [13–15]. The higher correlation may partly be explained by the fact that
some of these studies removed several data points that were considered outliers because they fell too
far from the regression. Their removal led to an artificial increase in ρ and R2, as individual data
points whose collapse intensity is far from the regressed line have a much larger influence on ρ and
R2 than points closer to the regression, particularly in smaller ground motion sets. Another possible
explanation for the higher correlation between ε and collapse intensity in some of those studies is
that, unlike the ground motion set used in this paper, some of the ground motion sets did not
contain near-fault, pulse-like ground motions in the record set, and as it has been noted before that ε
does not work well for these types of ground motions [10]. However, even when all ground motions
identified by Shahi [25] as pulse-type were removed from the ground motion set, ρ and R2 remained
unchanged and lower than those reported in previous studies. As demonstrated later, other structures
subjected to record sets without near-fault, pulse-like ground motions can have low correlation
coefficients and in some cases even smaller than the value observed in this case study (ρ = 0.32).
The slope of the linear regression between ε and LnSa(T1)col computed in this study (m = 0.149) is
not as large as other studies have found. For example, Haselton et al. [14] studied 65 reinforced
concrete (RC) special MRFs using a far-field ground motion set and found that the average slope
between ε and LnSa(T1)col was 0.28. To investigate if the removal of pulse-like ground motions
from the 274 record set would result in an increase in slope, all ground motions included in the
Shahi pulse-like database [25] were removed and the linear regression analysis was repeated;
however, the slope of the linear fit between ε and LnSa(T1)col remained practically unchanged
(m = 0.150). Haselton et al. [14] also examined 20 RC ordinary MRFs, which are less ductile than
special MRFs, as well as 26 non-ductile RC MRFs and found that the average slopes were 0.19 and

Copyright © 2016 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. (2016)
DOI: 10.1002/eqe
SPECTRAL SHAPE METRICS AND STRUCTURAL COLLAPSE POTENTIAL

0.18, respectively, for these buildings, which are values relatively close to the slope shown in Figure 2.
To help understand why some ground motions do not follow the expected relationship between
collapse intensity and ε (and therefore why ρ and R2 are relatively low), the pseudo-acceleration
spectra are plotted in Figure 3 along with the BA08 predictions for 12 ground motions that lie
relatively far from the regressed linear fit shown in Figure 2. These ground motions are grouped into
four types: (1) records with negative ε and relatively high Sa(T1)col (i.e. those with collapse
intensities significantly higher than the median collapse intensity) shown in Figure 3(a); (2) records
with neutral ε and relatively high Sa(T1)col shown in Figure 3(b); (3) records with neutral ε and
relatively low Sa(T1)col (i.e. those with collapse intensities significantly smaller than the median
collapse intensity) shown in Figure 3(c); and (4) records with positive ε and relatively low Sa(T1)col
shown in Figure 3(d). Each ground motion in Figure 3 has a color and symbol and is identified with
the same colors and symbols in Figure 2. The NGA record sequence numbers are also included in
Figure 3 to identify each ground motion, and the number in parentheses following the record
sequence number indicates which of the two horizontal components is shown.
As shown in Figure 3, in most cases the significant mismatch between the collapse intensity
predicted from the linear regression with ε and the actual collapse intensity for the ground motion is

Figure 3. Unscaled pseudo-acceleration spectra with predicted spectral shape from the Boore and Atkinson
2008 (BA08) model for 12 selected ground motions: (a) negative ε, high Sa(T1)col; (b) neutral ε, high Sa(T1)
col; (c) neutral ε, low Sa(T1)col; (d) positive ε, low Sa(T1)col.

Copyright © 2016 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. (2016)
DOI: 10.1002/eqe
L. EADS, E. MIRANDA AND D. LIGNOS

the result of ε failing to capture the spectral shape. For example, according to [10] records with
negative ε tend to indicate that the pseudo-acceleration spectrum has a valley at or near T1 and
therefore would tend to be more damaging records; however, the three records shown in Figure 3(a)
are actually quite benign despite having ε values below 1.5 as they need to be scaled to intensities
way above the median collapse capacity to produce the collapse of the structure. Careful inspection
of the shape of these three spectra shows that the records have a local peak at or near T1 despite
having spectral ordinates significantly below those that would be predicted from the BA08 GMPE.
Similarly, records with ε values near zero would tend to produce collapse at intensities close to the
median collapse intensity. However, the three ground motions shown in Figure 3(b) are relatively
benign records (as shown in Figure 2 in green symbols they produce collapse at intensities much
higher than the median collapse intensity). Examination of the shape of their spectra shows that they
have a local spectral peak at or near T1 despite having spectral ordinates close to the median
predicted by the BA08 GMPE. The three ground motions shown in Figure 3(c) are relatively
damaging records because (as shown in Figure 2 in cyan symbols) they produce collapse at
intensities much lower than the median collapse intensity. Their spectra show that they have local
spectral valleys at or near T1 despite having spectral ordinates close to the median predicted by the
BA08 GMPE (i.e. despite being ε neutral records).
According to [10, 14] records with positive ε values tend to indicate that the spectrum has a peak at
or near T1 and therefore would tend to be more benign records. However, contrary to this expected
tendency based on ε alone, the records shown in Figure 3(d) are very damaging records, producing
collapse at intensities well below the median collapse intensity despite having positive ε values
between 0.71 and 1.14, that is, despite having spectral ordinates at the fundamental period of
vibration significantly higher than the median spectral intensity predicted by BA08. Inspection of
their spectral shapes shows that their spectral ordinates tend to increase for periods longer than
T1 = 1.33 s despite being ε-positive records.
Because ε compares the actual and expected values of the spectral ordinate at only a single period, it
cannot directly measure spectral shape, which requires using information about the intensity of the
spectral ordinates at other periods. This means that trying to infer characteristics of the spectral shape
such as the occurrence of peaks and valleys by using ε, which provides information on the spectral
ordinate at single period relative to a reference expected median ordinate, is analogous to trying to
infer topographical prominence of sites on earth by only using information about the altitude with
respect to mean sea level. Clearly, with the exception of extreme altitudes or depths below sea level, it
is not possible to predict topographical features such as peaks and valleys with only information about
the altitude or depth of a single point on Earth. Therefore, while ε provides some implicit (indirect)
information about spectral shape for some records, as illustrated in Figure 3 and discussed in the
previous paragraphs, there are many records for which ε does not provide adequate information
regarding spectral shape and therefore the collapse potential of a ground motion. This results in
relatively low correlation coefficients between ε and LnSa(T1)col, suggesting that there might be other
parameters than one may use that lead to better predictions of the collapse intensity.

3.2. Eta (η)


This section evaluates the performance of another parameter called eta (η) that was recently proposed
as an alternative indicator of spectral shape. This parameter was proposed by Mousavi et al. [15] and is
computed as
η ≡ a1 εðSaÞ þ a2 εðPGV Þ (3)

where a1 and a2 are constants determined by regression analysis, ε(Sa) is the same ε discussed
previously and computed with Equation (1) and ε(PGV) is a parameter that measures the number of
standard deviations that the peak ground velocity (PGV) of the record is above that estimated by a
GMPE and is computed similarly to ε with the following equation

LnPGV  μLnPGV; GMPE


εPGV ≡ (4)
σ LnPGV; GMPE

Copyright © 2016 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. (2016)
DOI: 10.1002/eqe
SPECTRAL SHAPE METRICS AND STRUCTURAL COLLAPSE POTENTIAL

where LnPGV is the natural logarithm of the PGV of the record, μLnPGV,GMPE is the mean logarithmic
PGV estimated by a GMPE and σ LnPGV,GMPE is the logarithmic standard deviation of PGV estimated
by the same GMPE. The parameter η is also an indirect measure of spectral shape.
Mousavi et al. [15] studied the relationship between collapse intensity and η for 84 SDOF systems
with periods ranging from 0.1 s to 2.0 s and ductility values between 2 and 12. The SDOFs had a
tri-linear backbone curve with zero hardening slope and a capping ductility equal to 90% of the
ultimate ductility. Based on their results, they obtained the constants as a1 = 1.0 and a2 = 0.823 and
found that the median value of the correlation coefficient between η and the natural logarithm of
collapse intensity was ρ = 0.65 (R2 = 0.42) for the SDOF systems they analyzed.
Figure 4 shows the value of η computed for each record plotted against the natural logarithm of
collapse intensity. As shown in this figure the correlation between η and LnSa(T1)col of ρ = 0.59
(R2 = 0.35) is much higher than that previously observed in Figure 2 for ε and LnSa(T1)col, which is
in good agreement with results in [15] where η was found to be remarkably more efficient than ε. It
should be noted in this study the BA08 prediction was used to compute η for the case study whereas
Mousavi et al. [15] used the 2008 Campbell and Bozorgnia GMPE [26] to compute η and calibrate
the coefficients of the η equation; however, differences in the correlation coefficient because of use
of different GMPEs are expected to be small. The same 12 ground motions highlighted in Figures 2
and 3 that did not follow the expected trend between ε and LnSa(T1)col are also highlighted in
Figure 4, which shows that these ground motions are now closer to the linear regression between η
and LnSa(T1)col. Using η produces a significantly stronger correlation (85% larger) with the
logarithmic collapse intensities versus using ε (ρ = 0.59 vs. 0.32, respectively), indicating that η is
indeed a better measure of spectral shape than ε.

3.3. SaRatio
Although the spectral shape is not a perfect predictor of collapse intensity, a clear difference was
observed in Figure 1 both between the spectral shapes of individual damaging and benign ground
motion records and between the mean spectral shapes of damaging and benign ground motion
records. To quantify these distinct spectral shapes a parameter called SaRatio is used. SaRatio is
defined as Sa(T1) normalized by the geometric mean of spectral acceleration values over a period
range, that is

SaðT 1 Þ
SaRatio ¼ (5)
Saavg ðT 1 ½a; bÞ
where Saavg(T1 · [a, b]) is the geometric mean of spectral acceleration values between periods a · T1 and
b · T1, and a and b are non-negative constants with a ≤ b. Note that the value of SaRatio does not change
when the ground motion is scaled to different intensities via amplitude scaling. Although SaRatio is a

Figure 4. Relationship between η and the natural logarithm of the collapse intensity for the four-story steel
building.

Copyright © 2016 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. (2016)
DOI: 10.1002/eqe
L. EADS, E. MIRANDA AND D. LIGNOS

Figure 5. Pseudo-acceleration spectra for records scaled to a common value of Sa(T1) with (a) SaRatio < 1;
(b) SaRatio > 1 with no peak at T1; (c) SaRatio > 1 with peak at T1.

scalar measure, it is a more direct measure of spectral shape in the sense that it provides information
about how high or how low the value of Sa(T1) is relative to spectral ordinates at periods shorter
and/or longer than the fundamental period of vibration of the structure. SaRatio < 1 indicates that
the value of Sa(T1) is less than the mean value of the surrounding spectral ordinates and, depending
on the period range used, can suggest a valley in the spectrum at or near T1. SaRatio > 1, on the
other hand, indicates that the value of Sa(T1) is greater than the mean value of the surrounding
spectral ordinates and again, depending on the period range used, can possibly indicate a peak in the
spectrum at or near T1.
To illustrate the difference in spectral shape that SaRatio quantifies, the spectra of three records
from the 1999 Chi-Chi, Taiwan earthquake are shown in Figure 5. The legend in Figure 5
indicates the station name and orientation of each recording. A horizontal line indicating the value
of Saavg(T1 · [0.2, 3]) and a vertical line at T1 are also included in Figure 5.2 The record in
Figure 5(a) has SaRatio = 0.72 and has a spectral valley at T1 while the record in Figure 5(b) has
SaRatio = 2.98 but does not have a peak at or near T1. As one might expect by looking at the
difference in spectral shape, the two records produce very different responses for the case study
structure when scaled to the same value of Sa(T1). Furthermore, the record in Figure 5(a) causes
the case study structure to collapse at an intensity of only Sa(T1) = 0.51 g whereas the record in
Figure 5(b) does not cause collapse until the intensity reaches more than four times that value at
Sa(T1) = 2.17 g. Figure 5(c) illustrates another example of a relatively benign record, with
SaRatio = 2.58 and with a peak near T1.
Given that SaRatio was shown to capture differences in spectral shape very well, its potential to
predict collapse intensity is evaluated by computing the correlation between SaRatio and collapse
intensity in the log–log domain (i.e. the correlation between Ln(Sa(T1)col) and Ln(SaRatio) is
computed). The log–log domain was chosen instead of the linear–linear domain because the
correlation is slightly stronger in the log–log domain and less sensitive to the period range used to
compute SaRatio. Figure 6 presents contours of the correlation coefficient ρ as a function of the a
and b values that define the period range for SaRatio (Equation 5). The maximum correlation
coefficient between SaRatio and collapse intensity is ρ = 0.83, which is achieved when periods
between 0.2 · T1 and 3.3 · T1 are used to compute SaRatio (i.e. when a = 0.2 and b = 3.3). It is
interesting to note that the correlation coefficient is generally much more sensitive to the upper limit
of the period range (the b value) than the lower limit of the period range (the a value). Including
periods up to 3 · T1 produces very good correlations (ρ ≥ 0.80) for all a ≤ 1, including the period
range between 1 · T1 and 3 · T1. It is also interesting to note that the maximum correlation coefficient
is achieved when periods less than T1 are included. This is particularly relevant for this structure
because it is four stories and higher modes contribute to its response, although it has been
demonstrated in Eads et al. [21] that considering spectral values at periods less than T1 can also
improve collapse predictions for SDOF systems (which, by definition, do not have higher modes).

2
In this study Saavg is computed using periods uniformly spaced between a · T1 and b · T1 at intervals of 0.01 s unless
noted otherwise.

Copyright © 2016 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. (2016)
DOI: 10.1002/eqe
SPECTRAL SHAPE METRICS AND STRUCTURAL COLLAPSE POTENTIAL

Figure 6. Contour plot of the correlation coefficient (ρ) between the natural logarithm of collapse intensity
and the natural logarithm of SaRatio for the case study as a function of the period range used to compute
SaRatio.

This is because spectral ordinates both to the left and to the right of T1 contain information about the
ground motion, and therefore their inclusion provides information that indicates the potential of the
ground motion to produce collapse. For example, individual pulses in pulse-like ground motions
control many features of the intensity and shape of the ground motion spectrum both to the left and
to the right of the fundamental period of vibration of the structure, and therefore the inclusion of
these periods in an intensity measure provides implicit information about the characteristics of the
ground motion pulse that gave origin to the response spectrum. Optimum period ranges for different
structures can be found in [21].
The SaRatio parameter used here is similar to the Np parameter recently used by Bojórquez and
Iervolino [11], which they define as the geometric mean of spectral acceleration values between T1
and some period greater than T1 (they recommend 2 T1), normalized by Sa(T1). They proposed this
parameter to be used as a vector intensity measure combination with Sa(T1). SaRatio, which was
developed independently of Np, is essentially the inverse of Np except that the period range for
computing the geometric mean can be much wider and is not constrained to periods greater than T1.
As explained in the next section the motivation for studying the correlation between Sa(T1)col and
SaRatio is not aimed at proposing an alternative vector intensity measure comprised of Sa(T1)col and
SaRatio but to explain the trends between spectral shape and collapse intensity observed in Figures
1 and 3 as well as the reason for the important reduction in dispersion previously observed by Eads
et al. [27] when using Saavg. As shown in Figure 6, by incorporating information about spectral
ordinates smaller than T1 and an upper limit of 3 T1 instead of 2 T1, the correlation of SaRatio with
the natural logarithm of collapse intensity is higher than that of 1/Np (SaRatio but setting a = 1.0 and
b = 2.0), which produces a correlation coefficient of 0.76 for this building as compared to 0.83 when
a = 0.2 and b = 3.3.
A scatter plot of the natural logarithm of SaRatio for each ground motion used in the case study and
the natural logarithm of collapse intensity is shown in Figure 7 where Saavg(T1 · [0.2, 3]) is used to
compute SaRatio.3 This plot shows that all ground motions with SaRatio ≤ 1 (i.e. Ln(SaRatio) ≤ 0)
are in the lower half of the collapse fragility curve (i.e. the collapse intensities are less than the
median of 1.03 g, or equivalently less than Ln(1.03) = 0.03 as shown in Figure 7). However, not all
motions in the lower half of the collapse fragility curve have SaRatio ≤ 1 (i.e. Ln(SaRatio) ≤ 0).
Similarly, relatively benign records are characterized by having SaRatio > 1.5 (i.e. Ln(SaRatio) >
0.4). A linear regression to the data is included (with the equation also shown in the Figure 7),
which shows a relatively good fit to the data of individual ground motions.
The same 12 ground motions highlighted in Figures 2, 3 and 4 are again highlighted in Figure 7
using the same colors and symbols. Comparison of Figures 2, 4 and 7 clearly shows that SaRatio is

3
Unless noted otherwise, hereafter this paper computes SaRatio using Saavg(T1 · [0.2, 3]).

Copyright © 2016 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. (2016)
DOI: 10.1002/eqe
L. EADS, E. MIRANDA AND D. LIGNOS

Figure 7. Relationship between SaRatio and the natural logarithm of the collapse intensity for case study
building.

a better predictor of collapse intensity than η and a much better predictor than ε, exhibiting a coefficient
of determination R2 = 0.68, which is 94% larger and 680% larger than those computed for η (R2 = 0.35)
and for ε (R2 = 0.10), respectively.
To further verify that the correlation with collapse intensity is generally significantly stronger for
SaRatio than for ε, the correlation coefficient was computed for 678 additional structures. These
structures are comprised of 396 generic MRF and 252 generic shear wall structures previously
studied by Zareian and Krawinkler [13] and 30 RC MRF structures studied by Haselton et al. [14].
Details about these structures, including the structural properties, modeling assumptions and the
ground motions used are described in referenced studies, but briefly the generic MRF and generic
shear wall structures range from four to sixteen stories with fundamental periods of vibration
between T1 = 0.2 s and 3.2 s while the RC MRF structures range from one to twenty stories with
fundamental periods between T1 = 0.42 s and 2.63 s.
Using collapse data generously provided by these authors, the correlation of collapse intensities with
SaRatio and ε is calculated, and the results are summarized in Table I. The ε values were calculated
using the BA08 GMPE. These additional results further demonstrate that the correlation with
collapse intensity is significantly greater for SaRatio than for ε. As seen in Table I the correlation of
the former is on the order of ρ = 0.8, on average, while the latter is on the order of ρ = 0.5, on
average. Less than 3% of the structures demonstrated a stronger correlation between collapse
intensity and ε than between collapse intensity and SaRatio. Although not shown, the average slope
of the regression between collapse intensity and ε is on the order of 0.3 for these structures, which is
consistent with the slope of the RC special MRF structures studied by Haselton et al. [14]. The
average slope of the regression between collapse intensity and SaRatio is on the order of 1.1 for the
structures presented in Table I.

3. SIGNIFICANCE OF THE RESULTS

The strong correlation between SaRatio and collapse intensity has important practical consequences
when computing the collapse fragility function and the probability of collapse of a given structure.

Table I. Summary of correlation between collapse intensity and ε or SaRatio.

ρ[ LnSa(T1)col, ε(Τ1) ] ρ[ LnSa(T1)col, LnSaRatio]


No. of
Structure type structures Mean Range Mean Range

Generic MRFs 396 0.54 0.20 – 0.75 0.80 0.33 – 0.95


Generic shear walls 252 0.52 0.27 – 0.65 0.82 0.56 – 0.96
RC MRFs 30 0.53 0.43 – 0.64 0.89 0.81 – 0.93

Copyright © 2016 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. (2016)
DOI: 10.1002/eqe
SPECTRAL SHAPE METRICS AND STRUCTURAL COLLAPSE POTENTIAL

As mentioned in the previous section, we selected the parameter SaRatio and studied its correlation
with spectral ordinates at collapse not to propose a new vector intensity measure but simply for
understanding the relationship between spectral shape and collapse intensity as measured by Sa(T1)
as well as the reasons for the reduction in record-to-record variability reported when using Saavg
instead of Sa(T1) [27]. As shown in Figure 7, a strong linear relationship exists between Sa(T1)col
and SaRatio in the log–log domain. Here a simpler relationship is examined in the linear–linear
domain to estimate the collapse capacity as a function of SaRatio as follows
SaðT 1 Þcol ≈ m · SaRatio (6)

where m represents the slope of a linear regression between SaRatio and Sa(T1)col that is constrained to
be equal to zero when SaRatio = 0 (i.e. a fitted straight line forced to pass through the origin). An
example of such linear trend between Sa(T1)col and SaRatio is presented in Figure 8 based on data
from the four-story steel MRF case study building. This figure also includes the straight line and
corresponding equation of the form shown in Equation (6) found through linear regression of the data.
As shown in Figure 8, the correlation coefficient and coefficient of determination are only slightly
lower than those found when the fit was done in the log–log domain. The scatter around this line is
related to the efficiency of Saavg as predictor of collapse intensity. Rewriting Sa(T1)col in terms of
the unscaled (i.e. as-recorded) value of Sa(T1) and using the definition of SaRatio in Equation (5) yields
 
SaðT 1 Þunscaled
m ¼ SF col  SaðT 1 Þunscaled (7)
Saavg;unscaled
where SFcol denotes the minimum factor by which the (unscaled or as-recorded) ground motion must
be scaled to trigger the structure to collapse. In other words, SFcol is the collapse scale factor and
denotes the ratio between Sa(T1)col and Sa(T1)unscaled. Equation (7) simplifies to

m ¼ SF col  Saavg;unscaled ¼ Saavg;col : (8)

Saavg is not a perfect predictor of collapse intensity, which is reflected by the scatter about the
regression line shown in Figure 8, but it is a significantly better IM than Sa(T1) for estimating the
likelihood of structural collapse. The scatter around the straight line shown in Figure 8 indicates that
the ratio between Sa(T1)col and SaRatio changes from one ground motion to another (i.e. it exhibits
record-to-record variability). The better or more efficient Saavg is at predicting the collapse intensity,
the less variability in the values of Saavg,col and the closer the data points are to the regression of Sa
(T1)col versus SaRatio.
To illustrate the large reduction in the variability (dispersion) of collapse intensities when Saavg is
used as ground motion IM, Figure 9 presents a scatter diagram of collapse intensities for each
individual record when using Sa(T1) as the IM in Figure 9(a) and when using Saavg as the IM in

Figure 8. Sa(T1)col versus SaRatio with linear regression through the origin.

Copyright © 2016 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. (2016)
DOI: 10.1002/eqe
L. EADS, E. MIRANDA AND D. LIGNOS

Figure 9. Collapse intensity for individual records for (a) IM = Sa(T1); (b) IM = Saavg. The horizontal black
line indicates the median value.

Figure 9(b). The significant reduction in scatter around the median value (the horizontal black line) is
apparent when using Saavg. The colors of the dots in Figure 9 correspond to the value of SaRatio for
each ground motion. Figure 9(a) shows that there is strong relationship between Sa(T1)col and
SaRatio, with small values of SaRatio (i.e. dark blue dots) generally indicating small values of Sa
(T1)col (damaging records that trigger collapse at low levels of Sa(T1)) and large values of SaRatio
(i.e. orange and red dots) generally indicating large values of Sa(T1)col (more benign records that
will not produce collapse unless they are scaled to very high levels of Sa(T1)). This indicates that
the spectral shape of a record as captured/measured by SaRatio affects the collapse intensity of a
record when Sa(T1) is used as an IM, and hence using Saavg, which from previous equations is
related to SaRatio, leads to a significant reduction in record-to-record variability and therefore is a
better measure of ground motion intensity when estimating the probability of collapse.
This has important implications when computing the collapse fragility curve and the collapse risk.
Because using Saavg instead of Sa(T1) leads to a significantly smaller dispersion in the collapse
fragility curve, a smaller number of nonlinear response history analyses are required to estimate the
probability of collapse of the structure [24]. Figure 10 compares the collapse fragility curves using
these two different IMs for the case study building. Figure 10(a) uses Sa(T1) while Figure 10(b) uses
Saavg. As shown in this figure, the use of Saavg as an IM results in a significant decrease in the
dispersion of the collapse intensities compared to the value computed when Sa(T1) is used as IM:
σ lnSaAvg = 0.22 versus σ lnSa(T1) = 0.39, which corresponds to a 44% reduction in dispersion. The
significant reduction in the dispersion of collapse intensities when using Saavg instead of Sa(T1) has
been observed for a variety of structures [27]. As discussed in [27], Saavg was observed to be good
IM for collapse risk assessment as it was generally also predicted with less variability from a GMPE
perspective and produced collapse risk estimates that were less sensitive to the particular ground
motions used in nonlinear response history analysis compared to Sa(T1). From Figure 8 and

Figure 10. Collapse fragility curve for (a) IM = Sa(T1); (b) IM = Saavg. The circles are the collapse intensities
of individual ground motions, and the solid lines are the lognormal fragility curves fitted to the data using the
method of moments.

Copyright © 2016 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. (2016)
DOI: 10.1002/eqe
SPECTRAL SHAPE METRICS AND STRUCTURAL COLLAPSE POTENTIAL

previous equations, it is clear then that the reason why Saavg leads to significantly smaller dispersions
when used as an IM compared to Sa(T1) is because Saavg is approximately proportional to SaRatio, and
therefore it inherently incorporates information of a parameter strongly correlated with collapse
intensity and reduces the number of nonlinear response history analyses required for a given level of
confidence by using a scalar intensity measure (i.e. without having to use a more complicated vector
formulation).

4. SUMMARY AND DISCUSSION

The relationship between spectral shape and the minimum ground motion intensity producing
structural collapse (i.e. the collapse intensity) was examined. The spectral shape of damaging
records (those that produce collapse at relatively low levels of ground motion intensity as measured
by Sa(T1)) was compared to the spectral shape of benign records that do not produce collapse until
relatively high levels of intensity. Results from a four-story steel MRF building show that the
spectral shape of damaging records is significantly different from that of benign records and that the
spectral shapes of damaging records typically exhibit sharply increasing spectral displacement
ordinates for periods greater than T1.
Four measures of spectral shape were evaluated for their ability to capture these characteristics and
explain the variability of collapse intensities and, in particular, for their ability to predict the intensity at
which a given record will produce the collapse of a structure. The first parameter evaluated was epsilon
(ε), which measures the number of standard deviations that a spectral ordinate is above or below the
median value predicted by a GMPE. It is found that although there is a tendency for the collapse
intensity to increase (i.e. for the ground motion to be less damaging) with increasing ε significant
variability exists as reflected by a relatively low correlation coefficient (ρ = 0.32) between ε and the
collapse intensity of the four-story steel MRF building when subjected to 274 recorded ground
motions. Furthermore, results indicate that the change in ε from one record to another only accounts
for about 10% of the variability in collapse intensities for this building as measured by the
coefficient of determination. This relatively low level of correlation is because of the fact that ε is
only an indirect measure of spectral shape as it contains only information about the intensity of the
record relative to the predicted median intensity at a single period of vibration, and therefore it fails
to provide an adequate measure of the spectral shape of a record, which would also require
information about the spectral ordinates at other periods of vibration. It is shown that in many cases
ε fails to anticipate the occurrence of spectral shape features such as spectral peaks or valleys, which
have been shown to be important indicators of the response of nonlinear systems. These cases tend
to coincide with ground motions where the collapse intensity is poorly predicted by the ε value of
the record.
The second measure of spectral shape evaluated in this study is a parameter called eta (η) [15], which
is a linear combination of the number of standard deviations that a spectral ordinate is above or below the
median value predicted by a GMPE and of the number of standard deviations that the peak ground
velocity of the record is above or below the one predicted by a GMPE. Even though η is also an
indirect measure of spectral shape, results presented in [15] indicate that on average collapse
intensities have significantly higher (85% higher) correlation with η (ρ = 0.65, R2 = 0.42) than ε, and
therefore η is a better proxy to spectral shape. The improved performance of η with respect to ε
evaluated here for the four-story building agrees with observations of the proponents of this parameter
who had previously studied its performance to predict the collapse of SDOF systems [15].
The third measure of spectral shape that was evaluated in this study is the parameter Np [11], which
is defined as the ratio of the geometric mean value of pseudo-acceleration spectral values over a period
range from T1 to 2 T1, to Sa(T1). This parameter provides a measure how high or low a spectral ordinate
is relative to spectral ordinates averaged over a range of periods longer than the fundamental period of
vibration. Results for the four-story building show that Np is a better predictor of the collapse intensity
than η and a much better predictor than ε, having a coefficient of determination of R2 = 0.54.
The fourth measure of spectral shape that was evaluated in this study is a parameter called SaRatio,
which is defined as the ratio of Sa(T1) to the geometric mean value of pseudo-acceleration spectral

Copyright © 2016 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. (2016)
DOI: 10.1002/eqe
L. EADS, E. MIRANDA AND D. LIGNOS

values over a period range. SaRatio is very similar to Np, but the former recognizes the improvement in
correlation achieved by including periods less than the fundamental period. Although both Np and
SaRatio are scalar measures, they are more direct measures of spectral shape in the sense that the
parameters measure how high or low a spectral ordinate is relative to spectral ordinates averaged
over a range of periods shorter and/or longer than (or strictly longer than in the case of Np) the
fundamental period of vibration. It is found that SaRatio is much better correlated with the collapse
intensity than when using either Np, η or ε. Results for the four-story building show that SaRatio is a
better predictor of the collapse intensity than Np and a much better predictor than either η or ε as
SaRatio exhibits a coefficient of determination of R2 = 0.68, which is 1.2 times larger, 1.9 times
larger, and 6.8 times larger than those computed for Np (R2 = 0.58), η (R2 = 0.35) and for ε
(R2 = 0.10), respectively. The better performance of SaRatio in predicting the collapse intensity was
corroborated by using results from 396 generic MRF and 252 generic shear wall structures
previously studied by Zareian and Krawinkler [13] and 30 RC MRF structures studied by Haselton
et al. [14].
Finally it is concluded that because of the close relationship between SaRatio and Saavg and the
strong correlation between the former and the collapse intensity, Saavg is a good scalar to use as a
ground motion intensity measure for estimating the collapse intensity of a structure as it already
contains good information about the spectral shape of the ground motion that is relevant to the
structural response.

CONTRACT/GRANT SPONSOR

George E. Brown, Jr. Network for Earthquake Engineering Simulation program of the National
Science Foundation under award number CMS-0936633

ACKNOWLEDGEMENTS
The authors would like to thank Professors Farzin Zareian, Curt Haselton and Greg Deierlein for sharing
their collapse analysis results for generic moment-resisting frame and shear wall structures and for modern
reinforced concrete moment-resisting frame structures, respectively. The authors also acknowledge the com-
ments and suggestions of Professor Jack Baker on this investigation and thank the anonymous reviewers of
the manuscript for their constructive comments. This research was funded by the US National Science
Foundation (NSF) under Grant No. CMS-0936633 within the George E. Brown, Jr. Network for Earthquake
Engineering Simulation Research (NEESR) Consortium Operations. The financial support of NSF is
gratefully acknowledged. Any opinions, findings and conclusions or recommendations expressed in this
paper are those of the authors and do not necessarily reflect the views of NSF.

REFERENCES
1. Housner GW. Intensity of ground motion during strong earthquakes. 2nd Technical Report, under contract N6onr-
244, Task order 25. Earthquake Research Laboratory, California Institute of Technology: Pasadena, California, 1952.
2. Housner GW. Intensity of earthquake ground shaking near the causative fault. Proc. 3rd World Conf. Earthquake
Eng, 1, 1965. p 94–111.
3. Kennedy RP, Short SA, Merz KL, Tokarz FJ, Idriss IM, Power MS, Sadigh K. Engineering characterization of
ground motion. Task I: Effects of characteristics of free-field motion on structural response. Report No. NUREG/
CR-3805, U.S. Nuclear Regulatory Commission, Washington, DC, 1984.
4. Sewell RT, Toro GR, McGuire RK. Impact of ground motion characterization on conservatism and variability in
seismic risk estimates. Report No. NUREG/CR--6467, U.S. Nuclear Regulatory Commission, Washington, DC,
1996.
5. Shome N, Cornell CA. Probabilistic seismic demand analysis of nonlinear structures. Report No. RMS-35, Reliability
of Marine Structures Program, Stanford University, Stanford, CA, 1999.
6. Carballo JE, Cornell CA. Probabilistic seismic demand analysis: spectrum matching and design. Report No. RMS-41,
Reliability of Marine Structures Program, Stanford University, Stanford, CA, 2000.
7. Cordova PP, Deierlein GG, Mehanny SSF, Cornell CA. Development of a two-parameter seismic intensity measure
and probabilistic assessment procedure. Proc., The Second U.S.–Japan Workshop on Performance-Based Earth-
quake Engineering Methodology for Reinforced Concrete Building Structures, Sapporo, Hokkaido, Japan,
187-206, 11-13 September, 2000.
8. Jalayer F. Direct probabilistic seismic analysis: implementing non-linear dynamic assessments. PhD Dissertation,
Department of Civil and Environmental Engineering, Stanford University, Stanford, CA, 2003.

Copyright © 2016 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. (2016)
DOI: 10.1002/eqe
SPECTRAL SHAPE METRICS AND STRUCTURAL COLLAPSE POTENTIAL

9. Vamvatsikos D, Cornell CA. Developing efficient scalar and vector intensity measures for IDA capacity estimation
by incorporating elastic spectral shape information. Earthquake Engineering & Structural Dynamics 2005;
34(13):1573–1600.
10. Baker JW, Cornell CA. Vector-valued ground motion intensity measures for probabilistic seismic demand analysis.
Report No. 150, The John A. Blume Earthquake Engineering Center, Stanford University, Stanford, CA, 2005.
11. Bojórquez E, Iervolino I. Spectral shape proxies and nonlinear structural response. Soil Dynamics and Earthquake
Engineering 2011; 31(7):996–1008.
12. Modica A, Stafford PJ. Vector fragility surfaces for reinforced concrete frames in Europe. Bulletin of Earthquake
Engineering 2014; 12(4):1725–1753.
13. Zareian F, Krawinkler H. Simplified performance-based earthquake engineering. Report No. 169, The John A. Blume
Earthquake Engineering Center, Stanford University, Stanford, CA, 2009.
14. Haselton CB, Baker JW, Liel AB, Deierlein GG. Accounting for ground-motion spectral shape characteristics in
structural collapse assessment through an adjustment for epsilon. Journal of Structural Engineering 2011;
137(3):332–344.
15. Mousavi M, Ghafory-Ashtiany M, Azarbakht A. A new indicator of elastic spectral shape for the reliable selection of
ground motion records. Earthquake Engineering & Structural Dynamics 2011; 40(12):1403–1416.
16. ICC. International Building Code. International Code Council (ICC): Falls Church, VA, 2003.
17. AISC. Seismic Provisions for Structural Steel Buildings, including Supplement No. 1. American Institute of Steel
Construction (AISC): Chicago, IL, 2005.
18. McKenna F. OpenSees: a framework for earthquake engineering simulation. Computing in Science & Engineering
2011; 13(4):58–66.
19. Lignos DG, Krawinkler H. Sidesway collapse of deteriorating structural systems under seismic excitations. Report
No. 177, The John A. Blume Earthquake Engineering Center, Stanford University, Stanford, CA, 2012.
20. Zareian F, Medina RA. A practical method for proper modeling of structural damping in inelastic plane structural
systems. Computers & Structures 2010; 88(1-2):45–53.
21. Eads L, Miranda E, Lignos DG. Seismic collapse risk assessment of buildings: effects of intensity measure selection
and computational approach. Report No. 184, The John. A Blume Earthquake Engineering Center, Stanford Univer-
sity, Stanford, CA, 2014.
22. Power MS, Youngs RR, Chin C-C. Design ground motion library. Report of PEER-LL Program Task 1F01, Pacific
Earthquake Engineering Research (PEER) Center, University of California, Berkeley, CA, 2007.
23. Boore DM, Atkinson GM. Ground-motion prediction equations for the average horizontal component of PGA, PGV,
and 5%-damped PSA at spectral periods between 0.01 s and 10.0 s. Earthquake Spectra 2008; 24(1):99–138.
24. Eads L, Miranda E, Krawinkler H, Lignos DG. An efficient method for estimating the collapse risk of structures in
seismic regions. Earthquake Engineering & Structural Dynamics 2013; 42(1):25–41.
25. Shahi SK. A probabilistic framework to include the effects of near-fault directivity in seismic hazard assessment,
PhD Dissertation, Department of Civil and Environmental Engineering, Stanford University, Stanford, CA, 2013.
26. Campbell KW, Bozorgnia Y. NGA ground motion model for the geometric mean horizontal component of PGA,
PGV, PGD and 5% damped linear elastic response spectra for periods ranging from 0.01 to 10s. Earthquake Spectra
2008; 24(1):139–171.
27. Eads L, Miranda E, Lignos DG. Average spectral acceleration as an intensity measure for collapse risk assessment.
Earthquake Engineering & Structural Dynamics 2015; 44(12):2057–2073.

Copyright © 2016 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. (2016)
DOI: 10.1002/eqe

Вам также может понравиться