Hypocentre location by pattern recognition

Abstract

A novel approach to hypocenter location is proposed on the basis the concept of pattern recognition. A new data misfit criterion for location is introduced which measures discrepancies between the observed arrival times of an event and those of "nearby" previous events. In the arrival pattern misfit measure, travel times predicted by an Earth model are effectively replaced by information from an ensemble of previous observations. Thin-plate spline interpolation and generalized cross validation are applied to interpolate and smooth the resulting misfit function which may then be used in standard location algorithms. Synthetic experiments show that in certain circumstances, it is possible to achieve locations with errors smaller than those in the underlying database. It is suggested that the arrival pattern approach exploits information on lateral heterogeneous Earth structure contained in the database to constrain locations. The arrival pattern approach is illustrated by relocating 395 ground truth events from the Nevada Test Site, 482 earthquakes from the Marianas subduction zone, and 457 earthquakes from the Atlantic mid-ocean ridge. It is shown that picking errors and unmodeled, small-scale lateral heterogeneity are the most significant sources of event mislocation and that errors in the original locations of the database events make a much smaller contribution.