Centroid-based Cluster Analysis of HVSR Data for Seismic Microzonation

Abstract

Horizontal to Vertical Spectral Ratio (HVSR) datasets acquired for studies of seismic microzoning in various urban centers of Sicilian towns, have been used to test clustering analysis through a nonhierarchical centroid-based algorithm. In this context clustering techniques may be useful to identify areas with similar seismic behaviour through HVSR data. Centroid-based algorithms generally require the number of clusters, k, and the initial centroid coordinates to be specified in advance. This aspect is considered to be one of the biggest drawbacks of these algorithms. The proposed algorithm doesn’t limit the number of k clusters and choose the initial centroids automatically from the data set. Azimuthal variation of the H/V peaks was also taken into account. Finally different partitions obtained using the centroid-based algorithm were superimposed on the geological map of the analyzed sites to identify possible correlations with geology and topography. The obtained results underline how the most appropriate clustering algorithm for a particular site often needs to be chosen experimentally. In fact in many cases the choice of the partition is strongly linked to the choice of parametric distance and to geological knowledge, while in other cases, the results showed similar results regardless of a priori choices.