Are northeast and western Himalayas earthquake dynamics better "organized" than Central Himalayas: An artificial neural network approach

Abstract

The Himalaya covering 20–38° N latitude and 70–98° E longitude, is one of the most seismo–tectonically active and vulnerable regions of the world. Visual inspection of the temporal earthquake frequency pattern of the Himalayas indicates the nature of the tectonic activity prevailing in this region. However, the quantification of this dynamical pattern is essential for constraining a model and characterizing the nature of earthquake dynamics in this region. We examine the temporal evolution of seismicity (M > 4) of the Central Himalaya (CH), Western Himalaya (WH) and Northeast Himalaya (NEH), for the period of 1960–2003 using artificial neural network (ANN) technique. We use a multilayer feedforward artificial neural network (ANN) model to simulate monthly resolution earthquake frequency time series for all three regions. The ANN is trained using a standard back–propagation algorithm with gradient decent optimization technique and then generalized through cross–validation. The results suggest that earthquake processes in all three regions evolved on a high dimensional chaotic plane akin to "self–organized" dynamical pattern. Earthquake processes of NEH and WH show a higher predictive correlation coefficient (50–55%) compared to the CH (30%), implying that the earthquake dynamics in the NEH and WH are better "organized" than in the CH region. The available tectonogeological observations support the model predictions.