Metrics, clustering and simulations to evaluate seismic signals

This thesis presents an overview on seismic signals analysis and its related activities to clustering. The real applications require the use of metrics, algorithms and data to test hypothesis or to infer them. Hypocenter and focal mechanism of an earthquake can be determined by the analysis of signals, named waveforms, related to the wave field produced by earthquakes and recorded by a seismic network. Assuming that waveform similarity implies the similarity of focal parameters, the analysis of those signals characterized by very similar shapes can be used to give important details about the physical phenomena which have generated an earthquake. Recent works have shown the effectiveness of cross-correlation and/or cross-spectral dissimilarities to identify clusters of seismic events. In this thesis we propose a new dissimilarity measure between seismic signals whose reliability has been tested on real seismic data by computing external and internal validation indices on the obtained clustering. Results show its superior quality in terms of cluster homogeneity and computational time with respect to the largely adopted cross correlation dissimilarity.