Ground motion selection for seismic risk analysis of civil infrastructure - Seismic Risk Analysis and Management of Civil Infrastructure Systems - page 93

used. This includes the specifi c fi ltering applied to a ground motion record
(since, for example, excessive fi ltering of long period ground motion content
will produce reduced spectral amplitudes at long periods, resulting in the
ground motion having a high residual when compared with a realized IM
vector). A total of 30 ground motions are selected using the fi rst 30 realiza-
tions presented in the previous section.
Firstly, consider the contemporary procedure of ground motion selection
based solely on the response spectral ordinates of a ground motion. As such,
the weight vector considered was:

19
0

if
otherwise

IM = SA

⎩

i

( ) = wIM [4.16]

i i

That is, a weight of 1/9 was given for each of the nine different SA ordi-
nates in IM , and zero weight for all other IM i s. Figure 4.4 illustrates the
details of the ground motion set which was selected based on the realiza-
tions in Fig. 4.3 using the weight vector given by Equation (4.16). Firstly, it
can be seen that the response spectra of the selected ground motions,
depicted in Fig. 4.4a provide a good representation of the GCIM distribu-
tion as indicated by the similarity between the median, 16th and 84th per-
centiles of SA values of the selected ground motion set and the GCIM
distribution. The observation that the selected ground motions provide an
adequate representation of SA ordinates from the target GCIM distribu-
tion is not surprising, given that ground motions whose spectral ordinates
(at the nine periods considered) most closely match the individual realiza-
tions were selected. This can be explicitly appreciated by comparing the
similarity of the response spectra of the single ground motion in Fig. 4.4a
which was selected based on the single realization presented in Fig. 4.3a.
Figure 4.4b-d illustrate the EDFs of PGV , CAV and Ds 595, respectively, of
the selected ground motions (as well as the random realizations) in com-
parison with the GCIM distributions. It should be reiterated that none of
these three IM i s were considered in selecting the ground motions (i.e. the
'GM selection weight' in the fi gure text for each IM i is 0.0). Despite this, it
can be seen in Fig. 4.4b that the PGV distribution of the selected ground
motion records, for this particular example, is consistent with the GCIM
distribution, an observation which was also the case for PGA , ASI , SI , and
DSI . This observation can possibly be attributed to the fact that SA ordi-
nates over a wide period range have been explicitly considered (i.e. Equa-
tion (4.16)), and that PGV exhibits a moderately strong correlation of
approximately 0.8 with SA over the period range 0.5-4.0 s (Bradley, 2011d).
In contrast, Fig. 4.4c and d illustrate that the ground motions selected based
on the weight vector of SA ordinates alone have biased distributions of
CAV and Ds 595 (indicated by the EDF of the selected ground motions
intersecting the KS rejection bounds). The specifi c reasons for this result