Ground motion selection for seismic risk analysis of civil infrastructure - Seismic Risk Analysis and Management of Civil Infrastructure Systems - page 91

The fact that the GCIM methodology and the proposed GCIM-
consistent ground motion selection algorithm can account for a large
number of (explicit) intensity measures of a ground motion, means that it
is possible to select ground motions without the need for ad-hoc causal
parameter criteria. For example, ground motions resulting from small mag-
nitude ruptures will likely provide a large residual when fi t to a single
realization of a GCIM distribution which is dominated by larger magnitude
events because, even if they happen to have a similar response spectral
shape (which is not affected by amplitude scale factor), they are likely to
have other intensity measures, such as signifi cant duration, which signifi -
cantly differ from those which would be expected from larger magnitude
events (Bradley, 2010a) (recall that the GCIM distributions are computed
considering the deaggregation of the seismic hazard, i.e. P Rup | IM j ). It is worth
bearing in mind that the previous statement implies that GCIM distribu-
tions and ground motion selection are conducted using a suffi cient number
of IM i s; further details are given in the subsequent applications.

4.6 Application of the ground motion

selection methodology

4.6.1 Realizations from the GCIM distributions
Figure 4.3 illustrates 50 IM nsim realizations obtained from simulation of the
distribution IM | IM j for a value of IM j 0.0827 g , which has an

exceedance probability of 10% in 50 years for the case study site consid-
ered. In Fig. 4.3b-d, the realizations of PGV , CAV and Ds 595 are depicted
in the form of their EDF (Ang and Tang, 2007), while in Fig. 4.3a the indi-
vidual realizations of multiple SA ( T ) ordinates are shown. In each of the
plots in Fig. 4.3, the theoretical GCIM distribution (Bradley, 2010a), from
which the realizations have been simulated, is also shown. In Fig. 4.3b-d, it
can be seen that the simulated realizations are representative of the analyti-
cal distributions provided by the GCIM approach, as evident from the EDF
of the realizations lying 'inside' the KS rejections bounds (for a confi dence
level of 0.1). The similarity of the median, 16th and 84th percentiles of the
realizations and GCIM distribution illustrated in Fig. 4.3a also demon-
strates the representativeness of the realized SA ordinates (explicit com-
parisons can be made for individual SA ( T ) ordinates similar to Fig. 4.3b-d,
although this is not shown for brevity).
Figure 4.3a also illustrates the specifi c values obtained from a single
realization, IM nsim . For this single realization, it can be seen that for T

= SA (3.0) =

3 s
the realized SA ordinates are similar to the 16th percentile of the GCIM
distribution, while for T

<

3 s the ordinates are similar to the median of
the GCIM distribution. Figure 4.3b-d also illustrate that for this single

>