Earthquake Source Characterization with Graphical Deep Learning

Abstract

Accurate earthquake location and magnitude estimation play critical roles in seismology. Recent deep learning frameworks have produced encouraging results on various seismological tasks, e.g., earthquake detection, phase picking, seismic classification, and earthquake early warning. Most existing machine learning earthquake location methods utilize waveform information from single station. Multiple stations provide potential to extract more information for better characterizing earthquake source parameters. Available multiple-station location methods usually require stations have fixed number and order. Recently, a graph neural network has been proposed for source parameter determination, which simply incorporates geographic locations of seismic stations into a convolutional neural network (van den Ende and Ampuero, EarthArXiv, 2020). In this study, we develop a neural network model to construct graphs automatically and dynamically by an adaptive feature integration process. We design a graphical neural network with EdgeConv layers and incorporate it into the convolutional neural network for estimating earthquake locations and magnitudes. Given input waveforms collected from multiple seismic stations, the neural network constructs different graphs based on the waveform feature and station distance and thus fuses spatial-temporal consistency effectively from different stations. We apply our neural network model to earthquakes cataloged by Southern California Seismic Network from 2000 to 2020. The result shows that our model yield more accurate earthquake locations and magnitudes than that obtained with previous model. Our work demonstrates a great potential of using EdgeConv neural networks and multiple stations for better automatically estimating earthquake source parameters.