Clustering analysis of probabilistic seismic hazard for the selection of ground motion time histories in vast areas

We present a methodology for the selection of accelerometric time histories as input for
dynamic response analyses over vast areas. The method is primarily intended for seismic
microzonation studies and regional probabilistic seismic hazard assessments that account
for site effects. It is also suitable for structural response analyses if one would like to use
a fixed set of ground motion records for analyzing multiple structures with different (or
unknown) periods. The proposed procedure takes advantage of unsupervised machine
learning techniques to identify zones (i.e., groups of sites) with homogeneous seismic hazard,
for which the same set of earthquake recordings can be reasonably used in the numerical
simulations. The procedure consists of three steps: (1) data-driven cluster analysis to
identify groups of sites with comparable seismic hazard levels for a specified mean return
period (MRP); (2) for each zone, definition of a single, reference uniform hazard spectrum
(UHS) corresponding to the MRP of interest; (3) selection of a set of accelerometric
recordings that are consistent with the magnitude-distance scenarios contributing to
the hazard of each zone, and meet the spectrum-compatibility requirement with respect
to the reference UHS. An application of the procedure in the Po Plain (Northern Italy) is
described in detail.