Application of ANN and SVM for identification of tsunamigenic earthquakes from 3-component seismic data

Two pattern classification methodologies - Artificial Neural Network (ANN) and Support Vector Machine (SVM) have been used in this study to identify shallow focus (depth<;70 km) tsunamigenic earthquakes at a regional distance. Besides location and magnitude, the root mean square (rms) amplitudes of seismic phases recorded by a single 3-component station have been considered as inputs to both ANN and SVM. The trained ANN and SVM have been found to categorize 100% and 91% respectively of the new earthquakes successfully as tsunamigenic or non-tsunamigenic. Both the methodologies have been tested using three component broad band seismic data recorded at PALK (Pallekele, Sri Lanka) station provided by IRIS (Incorporated Research Institutes for Seismology) for earthquakes of magnitude 6.0 and above originating from Sumatra region. The fair agreement between the two methodologies ensures the existence of a single station based tsunami alert system which could be very useful for the poorly instrumented regions.