A deep learning prediction model for on-site earthquake early warning system in India
摘要
The increased success of earthquake early warning systems (EEWS) worldwide has warranted the need to explore the same for earthquake prone regions such as the active Himalayas of India. While network-based EEWS has been successfully implemented, the scope for on-site EEWS in the region has yet to be investigated. The current study presents a deep learning model, which is aimed at enabling the on-site EEWS in India. The model is developed for predicting broadband spectral acceleration of the expected ground motion using the initial P-wave portion of the seismic recording. By utilizing ground motion parameters derived from the first 3 s of the P-wave, the model can provide rapid and accurate predictions of the associated spectral acceleration between periods 0.01 s and 4.0 s. The research leverages data from multiple seismic networks in the Western Himalayas, encompassing 689 ground motion records from 124 earthquakes with magnitudes ranging from Mw 3.0 to 7.8. The model's performance is evaluated through rigorous sensitivity analyses and validation procedures, demonstrating its potential for real-time application in EEWS.