Predicting seismic aftershocks using a neural network

Neural networks appear suited for the study of earthquakes, for which there is a vast amount of raw data but no adequate theory. Neural networks are capable of discerning intrinsic patterns in the presence of noise. By using neural network technology, earthquake data can be analyzed without knowledge of parameters such as the age, the stress and the structure of underlying geological data. The authors apply a backpropagation neural network to forecast the location and time of an aftershock. Since, at some locations, the strain due to plate movement is released in instalments, it is thought that a neural network should be able to capture the seismographical pattern leading to the aftershock using previous seismic data recorded after the initial main event. The trained neural network should then be able to predict the imminent aftershock. The data used are supplied by SCSN (Caltech). The time interval considered begins at 9 hr 42 min 10.88 sec on January 31, 1996 and terminates at 1 hr 1 min 1.53 sec on the following day. Immediately thereafter a strong aftershock of magnitude 4.0 occurred in California