Ocean Seismic Parameter Estimation from Multi- Station Waveforms Using Deep Learning Method

Recently, Deep learning methods for seismic parameter estimation have gained attention. However, most deep learning models for seismic parameter estimation are constructed for land earthquakes, which are not suitable for ocean earthquakes due to sparse station distribution and low signal-to-noise ratio. In this study, we develop a deep learning model with an architecture containing both convolutional neural networks and graph neural networks, which is capable of processing waveform data from multiple stations in ocean earthquakes. The model can estimate the latitude, longitude, depth, and magnitude of seismic source by integrating waveform features with station location information. Offering the advantages of fast response time and high accuracy, this work can provide the earthquake early warning systems with a more efficient and rapid method for characterizing seismic source automatically.

I. Introduction

Seismic parameter estimation is the process of inferring characteristics of seismic events which involves the estimation of epicenter location, depth, and magnitude. Seismic parameter estimation plays a fundamental role in earthquake monitoring and seismological research, which is a critical basis for earthquake early warning, disaster mitigation, and seismic engineering.