Detection of tectonic faults by spatial clustering of earthquake hypocenters

Identification of the structure of tectonic faults from seismic data is mainly performed with clustering and principal curves techniques. In this paper we follow an approach based on the detection of the ridges of kernel densities estimated on earthquake epicenters. We use an iterative method based on the mean-shift algorithm for mode seeking, in which each step is made orthogonal to the principal direction of the local Hessian matrix. We carry out an extensive application to the historical data of San Francisco Bay area, and we compare the performance of similar methods with simulation experiments.