Analysis of Pre-earthquake Signals Using ANN: Implication for Short-Term Earthquake Forecasting

摘要

Earthquake is complex physical phenomena. The heterogeneous nature of the earth's interior is the reason for the unpredictable nature of earthquake occurrence. In recent years, scientists across the world are trying to develop a model using multiparameter earthquake precursors. In this paper, we discuss the association of abnormal irregularity in solid earth tides (SET) and anomalous transient change outgoing longwave radiation (OLR) with major earthquakes and utilizes a neural network to forecast the occurrence of notable earthquakes. We have considered the area of Simeulue, Indonesia region, and considered earthquakes of magnitude ＞5.0 takes place during the period from 2004 to 2014. Earthquake parameters for Simeulue, Indonesia region, has been taken for analysis by which anomaly date of solid earth tide and weights has been assigned for the continual anomaly days, OLR anomaly date, distance, day of OLR anomaly, latitude, longitude, anomaly index which appears before the earthquake are selected as input parameters, whereas the date of occurrence of earthquake, latitude, longitude, depth, magnitude are selected as output parameter for the neural network. We have used Elman backpropagation neural network model for forecasting the above-said output parameters. The analysis of the results given by the EBPNN have shown reasonable accuracy. Even though the results have to be tested in other regions, the results of the EBPNN have shown encouraging signs in developing an effective short-term earthquake-forecasting model.