Forecast of seismic aftershocks using a neural network

Every significant earthquake is followed by a mostly identifiable cluster of aftershocks. To predict the occurrence of these aftershocks, we trained a neural network using seismic data from SCSN (Caltech) as input. The trained network is extrapolated recursively, using the last target as the next input. In this way we were able to reproduce the three major aftershocks with magnitude 4.0 or greater for the main shock of magnitude 5.2 on Jan. 7, 1996 in Southern California. This paradigm returns a deterministic result, but requires two adjustable parameters: the number of hidden nodes and tolerance.