INTRODUCING ADAPTIVE INCREMENTAL DYNAMIC

ANALYSIS: A NEW TOOL FOR LINKING GROUND MOTION

SELECTION AND STRUCTURAL RESPONSE ASSESSMENT
Ting Lin & Jack W. Baker
Department of Civil and Environmental Engineering, Stanford University, Stanford, CA 94305-4020, USA

ABSTRACT: Adaptive Incremental Dynamic Analysis (AIDA) is a novel ground motion selection scheme
that adaptively changes the ground motion suites at different ground motion intensity levels to match hazard-
consistent properties for structural response assessment. Incremental Dynamic Analysis (IDA), a current dy-
namic response history analysis practice in Performance-Based Earthquake Engineering (PBEE), uses the same
suite of ground motions at all Intensity Measure (IM) levels to estimate structural response. Probabilistic Seis-
mic Hazard Analysis (PSHA) deaggregation tells us, however, that the target distributions of important ground
motion properties change as the IM levels change. To match hazard-consistent ground motion properties, ground
motions can be re-selected at each IM level, but ground motion continuity is lost when using such “stripes” (i.e.,
individual analysis points at each IM level). Alternatively, the data from the same ground motions in IDA can
be re-weighted at various IM levels to match their respective target distributions of properties, but this implies
potential omission of data and curse of dimensionality. Adaptive Incremental Dynamic Analysis, in contrast,
gradually changes ground motion records to match ground motion properties as the IM level changes, while
also partially maintaining ground motion continuity without the omission of useful data. AIDA requires care-
ful record selection across IM levels. Potential record selection criteria include ground motion properties from
deaggregation, or target spectrum such as the Conditional Spectrum. Steps to perform AIDA are listed as fol-
lows: (1) obtain target ground motion properties for each IM level; (2) determine “bin sizes” (i.e., tolerance for
acceptable ground motion properties) and identify all candidate ground motions that fall within target bins; (3)
keep ground motions that are usable at multiple IM levels, to maintain continuity; (4) use each ground motion
for IDA within its allowable IM range. As a result, if we keep increasing the “bin sizes”, AIDA will approach
IDA asymptotically; on the other hand, if we decrease the “bin sizes”, AIDA will approach the other end of
“stripes”. This paper addresses the challenges of changing records across various IM levels. Different ground
motion selection schemes are compared with AIDA to demonstrate the advantages of using AIDA. Example
structural analyses are used to illustrate the impact of AIDA on the estimation of structural response in PBEE.
By combining the benefits of IDA and PSHA without the omission of useful data, AIDA is a promising new
tool for linking ground motion selection and structural response assessment.

1 INTRODUCTION stability. IDA was specifically developed for seismic

assessment: the dynamic load is earthquake ground
Structural response assessment can be categorized as motion, often scaled from lower to higher intensity;
static or dynamic, linear or nonlinear. The complexity a suite of ground motions are typically applied to the
in the static regime increases from linear to nonlinear structure, to obtain statistics about the structure’s per-
to pushover, where incremental static load is applied formance, characterized by displacement and eventu-
to the structure, leading to component by component ally collapse, under a range of earthquake excitation.
failure and eventually system failure. Similarly, there The concept of IDA involves ground motions at mul-
is a parallel in the dynamic regime from linear to non- tiple intensity levels.
linear, with a dynamic analysis termed incremental Ground motion selection provides the seismic in-
dynamic analysis (IDA) by Vamvatsikos & Cornell put for structural response assessment. Ground mo-
(2002) used widely in the last decade. Vamvatsikos tion intensity is often characterized by spectral ac-
& Cornell (2002) vividly described IDA as a “dy- celeration (Sa) at the period of vibration of interest
namic pushover”, where incremental dynamic load is (T ∗ ). Probabilistic seismic hazard analysis (PSHA)
applied to the structure until it reaches dynamic in- incorporates uncertainty from earthquake sources and
 progress in hazard-consistent ground motion selection
utilizes the Conditional Spectrum (CS), a target re-
sponse spectrum to select ground motions for nonlin-
ear dynamic analysis. Computation of the CS can be
refined by incorporating multiple causal earthquakes
and ground motion prediction models (Lin, Harmsen,
Baker, & Luco 2013). Algorithms to match the mean
and variance of the target spectrum are developed as
a basis for selecting ground motions (Jayaram, Lin, &
Baker 2011). The use of the CS in ground motion se-
lection for risk-based and intensity-based assessments
is investigated and compared with alternative target
spectra (Lin, Haselton, & Baker 2013a, 2013b). Al-
ternatively, a generalized conditional intensity mea-
sure approach that considers intensity measures other
than Sa can be used if non-spectral ground motion pa-
rameters are also deemed important for predicting the
structural response of interest (Bradley 2010, 2012a,
2012b).
The performance-based earthquake engineering
(PBEE) framework starts with an intensity mea-
sure (IM), to estimate engineering demand parame-
ter (EDP), in order to quantify damage measure (DM)
and subsequently, decision variable (DV) (Cornell &
Krawinkler 2000, Deierlein 2004). Ground motion se-
lection can be viewed as the bridge between IM and
EDP, whereas structural response assessment is linked
to EDP. PSHA-consistent ground motion selection in-
volves MSA and potentially the CS as the target spec-
trum, whereas IDA is still frequently used in struc-
tural response assessment despite its lack of hazard
consistency. To combine the best of both worlds, we
propose a PSHA-consistent IDA, adaptive incremen-
Figure 1: USGS deaggregation of Sa(3s) corresponding to (a)
50% probability of exceedance in 30 years and (b) 1% probabil- tal dynamic analysis (AIDA).
ity of exceedance in 50 years in Palo Alto, California. This paper introduces AIDA, a new tool for link-
ing ground motion selection and structural response
ground motion predictions using total probability the- assessment. Section 2 answers the question “What
orem (Cornell 1968, Kramer 1996, McGuire 2004). is AIDA?”; Section 3, “What are the challenges of
Its reverse process, deaggregation, identifies the spe- AIDA”; Section 4, “How good or bad is AIDA?”.
cific contributing scenario(s) for the given ground mo- The last section then concludes with an overview of
tion intensity level using Bayes’ Rule (McGuire 1995, AIDA.
Bazzurro & Cornell 1999). Petersen et al. (2008) de-
veloped the United States national seismic hazard 2 METHODOLOGY
map using the concept of PSHA. The corresponding
online deaggregation feature by the US Geological 2.1 AIDA compared to alternative methods
Survey (https://geohazards.usgs.gov/deaggint/2008/)
provides plots such as Figure 1. As illustrated by this AIDA adaptively changes the ground motion suites
deaggregation of causal earthquakes for two different at different ground motion intensity levels to match
return periods, ground motion properties vary as in- hazard-consistent properties. AIDA evolves from the
tensity level changes. Careful ground motion selec- ideas of IDA and multiple stripe analysis, as illus-
tion needs to reflect such variation of ground motion trated in Figure 2. In this figure, EDP is plotted against
properties with intensity levels. IM, where Sa at the first-mode period of vibration
To match ground motion properties, PSHA- (T1 ) is chosen as the IM to represent the level of shak-
consistent ground motions can be re-selected at each ing experienced by the structure. When a ground mo-
intensity level. This is often termed “stripes” or multi- tion corresponding to each Sa(T1 ) is used as an seis-
ple stripe analysis (MSA). Alternatively, the data from mic input to the structural model, EDP can then be ob-
the same ground motions in IDA can be re-weighted tained by running a nonlinear dynamic (response his-
at various intensity levels to match their respective tar- tory) analysis. This EDP is typically associated with
get distributions of properties (Jalayer 2003). Recent displacement, but can also be acceleration, member
Table 1: Comparison of ground motion selection methods.
Traditional IDA Multiple Stripes Adaptive IDA
Matching properties No Yes Yes
Adaptive records No Yes Yes*
Continuity in records Yes No Yes*
* Adaptive IDA gradually changes records to maintain partial continuity.

force, or any response of interest. Each color line in

Figure 2(a) and (c) corresponds to a ground motion
that is used across a number of IM levels. For tra-
ditional IDA shown as Figure 2(a), every color line
spans the whole IM range, illustrating that the same
suite of ground motions are used across all IM levels,
simply by scaling their Sa(T1 ) up and down. IDA is
intuitively attractive, yet PSHA deaggregation tells us
that the target distributions of important ground mo-
tion properties change as the IM levels change. Fig-
ure 2(b) shows multiple stripe analysis, where PSHA-
consistent ground motions are re-selected at each IM
level to match the changing properties. Each dot in
Figure 2(b) corresponds to a nonlinear dynamic anal-
ysis with EDP as a function of IM. However, ground
motion continuity is lost when using such stripes, re-
gardless of the number of stripes. Alternatively, the
data from the same ground motions in IDA can be re-
weighted at various IM levels to match their respec-
tive target distributions of properties (Jalayer 2003),
but this implies potential omission of data and curse
of dimensionality (Baker 2007).
By combining the benefits of IDA and stripes,
AIDA makes adaptive changes to ground motions for
IDA, as illustrated in Figure 2(c). This allows us to
vary stripes-like ground motions to match the chang-
ing properties as the IM level changes, while par-
tially maintaining IDA-like ground motion continu-
ity. Figure 2(c) shows color lines that cross various
numbers of IM levels, illustrating PSHA-consistent
ground motions that are shared among some IM lev-
els but not across the entire range. The evolution of
response history analyses with various ground motion
selection strategies is also compared in Table 1.

2.2 Basic Algorithm

AIDA requires careful record selection across IM lev-
els. Steps to perform AIDA are listed as follows: (1)
obtain target ground motion properties for each IM
level; (2) determine “bin sizes” (i.e., tolerance for ac-
ceptable ground motion properties) and identify all Figure 2: (a) Incremental dynamic analysis (IDA); (b) multiple
candidate ground motions that fall within target bins; stripe analysis (MSA); (c) adaptive incremental dynamic analy-
(3) keep ground motions that are usable at multiple sis (AIDA).
IM levels, to maintain continuity; (4) use each ground
motion for IDA within its allowable IM range.

2.3 Ground motion selection criteria

Potential ground motion selection criteria include
seismological properties such as magnitude (M ) and
distance (R) from deaggregation associated with
Sa(T1 ) (McGuire 1995, Bazzurro & Cornell 1999,
Lin & Baker 2011), and/or spectral content similar
to a target spectrum such as the Conditional Mean
Spectrum (CMS) (Baker & Cornell 2006, Baker 2011,
Gulerce & Abrahamson 2011) or more recently the
Conditional Spectrum (CS) (Abrahamson & Al Atik
2010, Lin, Harmsen, Baker, & Luco 2013). The reader
is referred to these documents for relevant computa-
tion procedures related to deaggregation and target
spectrum as record selection criteria. Although the
first-mode period T1 is often used to obtain corre-
sponding Sa, any period of vibration of interest T ∗
can be applied instead. Deaggregation of M and R,
along with computation of the CMS, for sites in the
US, can be obtained directly from the USGS hazard
mapping tool or commercial seismic hazard analysis
software.
Other effects such as duration (e.g., Iervolino, Man-
fredi, & Cosenza 2006) and directivity (e.g., Shahi &
Baker 2011) can also be used as ground motion selec-
tion criteria. A generalized conditional IM that is ex-
tended from the concept of the CMS (Bradley 2010)
can be an alternative criterion for the engineering ap-
plication of interest. In principle any IM that is used
for typical ground motion selection (e.g., Shome et al.
1998, Luco and Cornell 2007, Haselton et al. 2009,
Katsanos et al. 2010) can be used as the selection cri-
terion for AIDA. The basic algorithm is then applied
to the IM of interest. Figure 3: (a) Magnitude and (b) distance distributions across IM
levels.
3 CHALLENGES
(second lowest IM level) to IM 6. It is expected that
3.1 Effect of bin size the extent of ground motion overlaps varies depend-
ing on the selection criteria.
One major step in the AIDA algorithm is to determine Bin size determines the degree of overlapping of
bin sizes. This is required to identify all candidate ground motions. To illustrate, take the deaggregated
ground motions that fall within target bins, so as to mean magnitudes in Figure 3 and vary their bin sizes
keep ground motions that are usable at multiple IM in Figure 4. Assume a bin with magnitude bounds of
levels. To maintain continuity, each ground motion is +/-0.5 is considered wide (Figure 4(a)), AIDA selects
then used for IDA within its allowable IM range. ground motions that fall within this target bin, with
PSHA deaggregation, as previously illustrated in resultant ground motion overlaps that span many IM
Figure 1, implies that causal earthquake magnitudes levels (Figure 4(b)). Altenatively, take a narrow bin
(M ) and distances (R) change as IM levels change. with magnitude bounds of +/-0.2 (Figure 4(c)), and
Figure 3 shows the distribution of M and R condi- the resultant AIDA motions then overlap fewer IM
tional on Sa(T1 ). First, mean M and R values are levels (Figure 4(d)). If we keep increasing the bin
obtained from deaggregation for the range of Sa(T1 ) sizes, the relaxed selection criteria allow for more
considered. Next, bin bounds are applied to the deag- ground motion overlaps, and AIDA will approach
gregated mean M and R respectively, in this case, IDA asymptotically. On the other hand, if we de-
M +/-0.5 and R+/-10km. Using these M and R ranges crease the bin sizes, the stringent selection criteria
as the selection criteria, ground motions that match limit ground motions to be usable across IM levels,
both criteria can be identified. The corresponding se- and AIDA will approach the other end of multiple
lected records are marked as color lines in Figure 3, stripes.
with each distinct color illustrating a unique ground
motion. Many of these ground motions are usable 3.2 Benefit of CS as the target spectrum
across multiple IM levels (up to 8 IM levels in this ex-
ample), and their respective allowable IM range is in- Using the Conditional Spectrum as the target spec-
dicated by the length of the corresponding color line. trum for AIDA application allows more sharing of
For instance, Selected Record 1, marked as a blue line ground motions across different IM levels without ar-
with diamonds at IM levels, spans the length of IM 2 bitrarily defining bin sizes. The CS removes the con-
Figure 4: Distribution of wide and narrow magnitude bins (a and c respectively) and their resulting AIDA (b and d).

servatism from the uniform hazard spectrum (UHS),

and implicitly considers M and R through Sa. The
CS differs from the CMS only in that it additionally
accounts for the variability in the spectrum. In this
regard, instead of just matching the mean through the
CMS (or M and R) and adjusting the bin size with tol-
erance criteria, the CS automatically sets the bin size
with its spectral variability. This is illustrated through
plots of CMS (solid lines) vs. CS (solid and dotted
lines) at various IM levels (Figure 5). Note that while
there are practically no overlaps in the defined CMSs,
the CSs will have overlaps naturally. This implies that
there may be more ground motions with similar spec-
tral shapes in adjacent IM levels because the goal of
the selection is to match both the mean and the vari-
ance. To match the variance, one ground motion with
a spectral shape that is slightly above the target CS at
an IM level may be a suitable candidate for a spectral
shape that is slightly below the target CS at another
IM level (Figure 5).

3.3 Implementation schemes

Figure 5: Normalized target response spectra conditional on
Once the selection criteria are established, ground Sa(2.6s), to illustrate the change of spectral shape across IM lev-
motions that meet the criteria can be selected from els. The lower-bound IM level (IM 1) corresponds to 50% in 30
a ground motion database such as the PEER NGA years probability of exceedance, and the upper-bound IM level
database (Chiou, Darragh, Gregor, & Silva 2008). The (IM 7) corresponds to 1% in 50 years probability of exceedance.
common idea with all the selection criteria is to min-
imize the sum of squared errors (SSE) between the
selected ground motions and the target. In addition,
there should be a certain extent of ground motion
sharing in adjacent IM levels to maintain partial conti-
nuity. This requires the optimization of shared ground
motions that meet the selection criteria.
Several methods may be used here. For instance,
 Figure 7: Adaptive Incremental Dynamic Analysis using the
Figure 6: Percentage of records from lower- to upper-bound IM Conditional Spectrum as the target spectrum.
levels (corresponding to 0.04g and 0.58g respectively).
4 ILLUSTRATIVE EXAMPLE
ground motions can be selected using an optimization
algorithm at an arbitrary IM level, and this algorithm To illustrate the methodology of AIDA, we use a
can then be applied successively to the adjacent IM 20-story reinforced concrete special moment frame
levels while requiring reuse of some motions from the located in Palo Alto, California. This building was
previous level. Another practical option would be to designed for the FEMA P695 project (ATC 2009,
(1) define the upper and lower IM levels, and select Haselton & Deierlein 2007), and is denoted Building
the best-matched ground motions at those IM levels 1020 in that study. It is a 2-D model in OpenSEES
from the database; (2) for an intermediate IM level, (opensees.berkeley.edu), with strength deterioration
use the selected ground motions from the upper and (both cyclic and in-cycle) and stiffness deterioration
lower IM levels as the new candidate database to se- that is believed to reasonably capture the responses up
lect ground motions; (3) repeat (2) with updated upper to the point of collapse due to dynamic instability. Its
or lower IM levels until there are no more intermedi- first modal period of vibration is 2.6s. This structure
ate IM levels. This way, all IM levels are covered, and is analyzed using ground motions selected to match
there will be common ground motions throughout the (1) the Uniform Hazard Spectra (UHS) with IDA-like
entire IM range, while each suite of ground motions characteristics (little change in spectral shape across
meet the selection criteria at a specific IM level. IM levels resulting in similar ground motions), (2)
Using the latter approach with the Conditional Conditional Spectra at multiple stripes, and (3) Con-
Spectra illustrated in Figure 5 as the target spectra, 40 ditional Spectra using AIDA. The range of IM lev-
ground motions are selected at each IM level. Figure els correspond to 50% in 30 years to 1% in 50 years
6 shows the overlaps of ground motions through per- probability of exceedance at Sa(2.6s). The target CS
centage changes from the lowest IM 1 to the highest is illustrated in Figure 5 and the corresponding ground
IM 7. At IM 1, all ground motions satisfy the selec- motions selected for AIDA in Figure 6. The resulting
tion criteria and come from IM 1; similarly, at IM 7, AIDA with peak story drift ratio of individual ground
all ground motions satisfy the selection criteria and motions as a function of Sa(2.6s) is plotted in Figure
come from IM 7. At intermediate IM levels 2 to 6, the 7. The analysis results for median peak story drift ra-
ground motion candidates are those from IM 1 and tio and probability of collapse for the three methods
IM 7, and the selected ground motions are a subset of are shown in Figure 8.
the candidates expressed in terms of percentage IM Results from AIDA are comparable to those from
1 and IM 7. For instance, ground motions at IM 4 multiple stripes, while IDA using UHS produces
come from 35% IM 1 and 75% IM 7 motions, with higher responses. On the other hand, fewer ground
10% ground motions shared throughout the IM lev- motions can be used for the whole range of inten-
els. In this example, if 40 ground motions are selected sity levels considered in AIDA, compared to chang-
at each IM level, multiple stripe analysis at 7 IM ing ground motions at each intensity level in multi-
levels would require re-selection of ground motions ple stripes. Because of this, the structural analysis can
at each IM level totaling 280 ground motions with be further optimized by running the AIDA records at
minimal overlaps. Compared to the 280 ground mo- fewer IM levels and interpolating, or by using these
tions used in multiple stripes, fewer than 80 ground analysis results to interpolate further to intermediate
motions (only 76 in this case because of 10% com- IM levels. In addition, compared to the same suite
mon ground motions) from IM 1 and IM 7 are used of ground motions used uniformly across all intensity
and reused for AIDA. With AIDA, the number of levels in IDA, AIDA changes the ground motions at
ground motions decreases relative to multiple stripes; each intensity level gradually to reflect the change in
yet, ground motion properties are matched and partial ground motion properties according to PSHA infor-
continuity maintained. mation.
 5 CONCLUSIONS

Adaptive Incremental Dynamic Analysis matches

ground motion properties at various Intensity Mea-
sure levels, and maintains continuity by overlapping
some ground motions across multiple IM levels. The
basic algorithm includes (1) obtain target ground mo-
tion properties at each IM level; (2) determine bin
sizes, i.e., tolerance for acceptable ground motion
properties; (3) identify candidate ground motions us-
able at multiple IM levels; (4) use each ground mo-
tion for Incremental Dynamic Analysis within its al-
lowable IM range. Ground motion selection criteria
can vary from causal earthquake properties such as
magnitudes and distances to a target response spec-
trum such as the Conditional Spectrum. The bin size
of the selection criteria determines the degree of over-
lapping of ground motions. As a result, if we keep
increasing the bin sizes, AIDA will approach IDA
asymptotically; on the other hand, if we decrease
the bin sizes, AIDA will approach the other end of
Multiple Stripe Analysis. Using the CS as the target
spectrum allows natural sharing of ground motions
across different IM levels without arbitrarily defin-
ing bin sizes. Examples were used to illustrate appli-
cation of AIDA using various selection criteria and
a practical implementation scheme. With a 20-story
reinforced concrete frame located in Palo Alto, Cal-
ifornia, it is shown that AIDA produces similar peak
story drift ratio and probability of collapse as its MSA
counterpart, while IDA produces higher responses
due to discrepancies in the spectral shapes of the
IDA ground motions relative to the target spectrum
(which changes with spectral acceleration amplitude).
In addition to producing comparable results as MSA,
AIDA uses fewer ground motions. AIDA combines
IDA and Probabilistic Seismic Hazard Analysis with-
out the omission of useful data, and hence an im-
provement over IDA for ground motion selection.
With its advantages over other ground motion selec-
tion methods, AIDA is a promising new tool for link-
ing ground motion selection and structural response
Figure 8: Comparison of (a) median peak story drift ratio and (b) assessment.
probability of collapse for IDA (UHS), MSA (CS), and AIDA
(CS).
6 ACKNOWLEDGEMENTS

We thank Curt Haselton for providing the structural

model used in the example here. This work was sup-
ported by the State of California through the Trans-
portation Research Program of the Pacific Earthquake
Engineering Research Center (PEER). Any opinions,
findings, and conclusion or recommendations ex-
pressed in this material are those of the authors and
do not necessarily reflect those of the funding agency.

REFERENCES
Abrahamson, N. A. & L. Al Atik (2010). Scenario spectra for de-
sign ground motions and risk calculation. In 9th US National
 and 10th Canadian Conference on Earthquake Engineering, Lin, T. & J. W. Baker (2011). Probabilistic seismic hazard deag-
Toronto, Canada. gregation of ground motion prediction models. In 5th Inter-
ATC (2009). Quantification of building seismic performance national Conference on Earthquake Geotechnical Engineer-
factors, FEMA P695. Technical report, Applied Technology ing, Santiago, Chile.
Council, Redwood City, California. Lin, T., S. C. Harmsen, J. W. Baker, & N. Luco (2013). Condi-
Baker, J. W. (2007). Probabilistic structural response assessment tional Spectrum computation incorporating multiple causal
using vector-valued intensity measures. Earthquake Engi- earthquakes and ground motion prediction models. Bulletin
neering & Structural Dynamics 36(13), 1861–1883. of the Seismological Society of America 103(2A), 1103–
Baker, J. W. (2011). Conditional mean spectrum: Tool for ground 1116.
motion selection. Journal of Structural Engineering 137(3), Lin, T., C. B. Haselton, & J. W. Baker (2013a). Conditional-
322–331. Spectrum-based ground motion selection. Part I: Hazard con-
Baker, J. W. & C. A. Cornell (2006). Spectral shape, epsilon sistency for risk-based assessments. Earthquake Engineering
and record selection. Earthquake Engineering & Structural & Structural Dynamics (in press).
Dynamics 35(9), 1077–1095. Lin, T., C. B. Haselton, & J. W. Baker (2013b). Conditional-
Bazzurro, P. & C. A. Cornell (1999). Disaggregation of seis- Spectrum-based ground motion selection. Part II: Intensity-
mic hazard. Bulletin of the Seismological Society of Amer- based assessments and evaluation of alternative target spec-
ica 89(2), 501–520. tra. Earthquake Engineering & Structural Dynamics (in
Bradley, B. A. (2010). A generalized conditional intensity mea- press).
sure approach and holistic ground-motion selection. Earth- Luco, N. & C. A. Cornell (2007). Structure-Specific scalar in-
quake Engineering & Structural Dynamics 39(12), 1321– tensity measures for Near-Source and ordinary earthquake
1342. ground motions. Earthquake Spectra 23(2), 357–392.
Bradley, B. A. (2012a). A ground motion selection algorithm McGuire, R. K. (1995). Probabilistic seismic hazard analysis and
based on the generalized conditional intensity measure ap- design earthquakes: Closing the loop. Bulletin of the Seismo-
proach. Soil Dynamics and Earthquake Engineering 40(0), logical Society of America 85(5), 1275–1284.
48–61. McGuire, R. K. (2004). Seismic hazard and risk analysis. Oak-
Bradley, B. A. (2012b). The seismic demand hazard and impor- land, CA: Earthquake Engineering Research Institute.
tance of the conditioning intensity measure. Earthquake En- Petersen, M. D., A. D. Frankel, S. C. Harmsen, C. S. Mueller,
gineering & Structural Dynamics 41(11), 1417–1437. K. M. Haller, R. L. Wheeler, R. L. Wesson, Y. Zeng, O. S,
Chiou, B., R. Darragh, N. Gregor, & W. Silva (2008). NGA D. M. Perkins, N. Luco, E. H. Field, C. J. Wills, & K. S.
project Strong-Motion database. Earthquake Spectra 24(1), Rukstales (2008). Documentation for the 2008 update of the
23. United States national seismic hazard maps: U.S. Geological
Cornell, C. A. (1968). Engineering seismic risk analysis. Bulletin Survey Open-File Report 2008–1128. Technical report.
of the Seismological Society of America 58(5), 1583–1606. Shahi, S. K. & J. W. Baker (2011). An empirically calibrated
Cornell, C. A. & H. Krawinkler (2000). Progress and challenges framework for including the effects of near-fault directivity
in seismic performance assessment. PEER Center News 3(2). in probabilistic seismic hazard analysis. Bulletin of the Seis-
Deierlein, G. G. (2004). Overview of a comprehensive frame- mological Society of America 101(2), 742–755.
work for earthquake performance assessment. In Interna- Shome, N., C. A. Cornell, P. Bazzurro, & J. E. Carballo (1998).
tional Workshop on Performance-Based Seismic Design Earthquakes, records, and nonlinear responses. Earthquake
Concepts and Implementation, Bled, Slovenia, pp. 15–26. Spectra 14(3), 469–500.
Gulerce, Z. & N. A. Abrahamson (2011). Site-specific de- Vamvatsikos, D. & C. A. Cornell (2002). Incremental dy-
sign spectra for vertical ground motion. Earthquake Spec- namic analysis. Earthquake Engineering & Structural Dy-
tra 27(4), 1023–1047. namics 31(3), 491–514.
Haselton, C., J. Baker, Y. Bozorgnia, C. Goulet, E. Kalkan,
N. Luco, T. Shantz, N. Shome, J. Stewart, P. Tothong,
J. Watson-Lamprey, & F. Zareian (2009). Evaluation of
ground motion selection and modification methods: Predict-
ing median interstory drift response of buildings. Technical
Report 2009/01, Pacific Earthquake Engineering Research
Center, Berkeley, CA.
Haselton, C. B. & G. G. Deierlein (2007). Assessing seismic col-
lapse safety of modern reinforced concrete moment frame
buildings. Technical Report 2007/08, Pacific Earthquake En-
gineering Research Center, Berkeley, CA.
Iervolino, I., G. Manfredi, & E. Cosenza (2006). Ground motion
duration effects on nonlinear seismic response. Earthquake
Engng Struct. Dyn. 35, 21–38.
Jalayer, F. (2003). Direct Probabilistic Seismic Anaysis: Im-
plementing Non-linear Dynamic Assessments. Ph. D. thesis,
Stanford University.
Jayaram, N., T. Lin, & J. W. Baker (2011). A computationally ef-
ficient ground-motion selection algorithm for matching a tar-
get response spectrum mean and variance. Earthquake Spec-
tra 27(3), 797–815.
Katsanos, E. I., A. G. Sextos, & G. D. Manolis (2010). Selection
of earthquake ground motion records: A state-of-the-art re-
view from a structural engineering perspective. Soil Dynam-
ics and Earthquake Engineering 30(4), 157–169.
Kramer, S. L. (1996). Geotechnical earthquake engineering.
Prentice-Hall International Series in Civil Engineering and
Engineering Mechanics. Upper Saddle River, N.J.: Prentice
Hall.