On Improving Correlation in Ground Motion Selection

ABSTRACT Ground motion selection methods simulate target intensity measure (IM) vectors and then select best‐fit records to capture the joint distribution of IMs. These approaches perform reasonably well for cases with a large number of records, but the marginal statistics and the correlation/covariance structure often depart from the target when sample sizes are small. Further, unlike the marginal parameters, the correlation is generally not considered in the final selection in the replicate‐selection and greedy optimization approaches. Recent studies have shown that such deviations and their effect can be substantial. In this study, an “ accelerated ” simulation procedure originally developed to generate correlated demand vectors is adapted for ground‐motion selection. The method captures both marginal statistics and the correlation structure exactly in the simulated IM sets, regardless of simulation size. Further, if the record library is robust, the marginal statistics and correlation are effectively carried through to the selected suite. Sample illustrations using NGA‐West2 records demonstrate that the proposed modification substantially improves marginal statistics and correlation/covariance consistency in selected suites, without requiring any change to the rest of the established selection process.