Report - PEER - University of California, Berkeley

espect to X has presumably been established, in principle one may select recordsfrom any values of X (Iervolino, 2004). In practice even in this case it is prudent touse records from the general magnitude regime of interest. In deciding which recordcharacteristics to mirror in the selection it is helpful to think in terms, primarily, ofany systematic effects on spectral shape. Systematic spectral shape deviation fromthe appropriate range can effect linear response of MDOF systems and nonlinearresponse of even SDOF systems. Hence, for example, it is prudent to avoid selectingrecords from soft soil sites or from records that may include directivity effects. If thesite should include such effects special efforts are necessary.Recent efforts (Baker, 2004b) have demonstrated that one such systematic effectis that of “epsilon”. Epsilon is the deviation of a record’s S a (at the structure’s firstmodeperiod, say) from that expected for the record’s specific values of X; in short itis the deviation or “residual” from the S a attenuation law (normalized by the “sigma”or standard error of the law.) High epsilon values are associated with peaks in therecord’s spectrum, and hence (for a fixed S a or IM level) with more benign nonlinearbehavior. (As the effective period of the structure lengthens it “falls off the peak” andinto a less energetic portion of the frequency content.) But rare, high IM levels (orlow λ IM levels) that contribute most directly to rare MIDR levels are in turn associatedwith high values of epsilon (as evidenced in PSHA disaggregations for epsilon).Therefore when selecting records for analyses at these high IM levels one shouldconsider choosing them from among records that have comparable epsilon levels(e.g., 1 to 2), in order that they do have the right, non-smooth shape near the period ofinterest. This is one reason why selecting records with shapes close to that of theUHS (or artificially matching a record’s spectrum to the UHS) may bias the responseconservatively.Spectrum-matched or “spectrum-compatible” records have the advantage ofreducing the dispersion in the response and hence of reducing the required samplesize. There is also evidence that they are unconservatively biased for large ductilitylevels (Carballo, 2000).Geophysically sound synthetics may be the only way we can obtain appropriaterecords for certain infrequently recorded cases, such as very near the source of largemagnitude events. The various empirically based schemes of record simulation (e.g.,from simple to evolutionary power spectral models, ARMA-based procedures, etc.)have the merit that one can produce from them large samples of nominally similar“earthquakes”. Care should be exercised to insure that their spectra are “roughenough” for accurate nonlinear analysis.2.6 Seeking Better IMs: Sufficiency and EfficiencyThe benefits of sufficient IMs are clearly a reduction in difficulty and reduction in thenumber of nonlinear analyses. The observation raises the subject of seeking stillbetter IMs, i.e., ones that might prove sufficient over a broader range of seismicconditions (i.e., regions of X ) and ones that might reduce the dispersion in response44