Application of pattern recognition techniques to long-term earthquake prediction in central Costa Rica

Abstract

The expert system GEO 2.5 has been applied to the analysis of geophysical and geological data from Costa Rica. Geophysical data consist of seismic tomography, Bouguer anomalies, seismic catalogue (RSN-ICE, 1984–1993) and seismic historical catalogue. Among the geological data, faults, lineaments, topography and a structural zonation of the region concerned have been used. Many other features have been obtained through the calculation of statistical functions of the original data such as nearness, density, length and b-value, other functions were secondary derivatives and modulus of horizontal gradient. A first analysis, mainly based on seismic tomography data and seismic catalogue, has led to the identification of a few seismogenic structures in the upper crust characterized by a high statistical correlation coefficient between seismic activity and positive velocity anomalies. Pattern recognition techniques have been applied to the long-term earthquake prediction by the definition, in the feature space, of non-linear regression functions between geophysical/geological data and a preliminary estimate of expected maximum magnitude. This preliminary estimate has been formulated on the basis of both the historical seismic catalogue and the structural zonation. This approach gives the empirical relationships among different geophysical and geological features, which are potentially related to the phenomenon of stress release. Due to the high non-uniqueness of such an approach, the results have to be physically understandable and each function has to be interpreted. The obtained multi-dimensional function has been applied to the calculation of a forecast maximum magnitude field for central Costa Rica. The results are to be considered preliminary. An improvement of such a forecast could be achieved by new data, such as heat flow, depth of Mohorovicic surface, active faults, vertical movement velocities, etc.