A seismic source characterization model of multi-station based on graph neural network

Seismic source characterization is a crucial part of earthquake early warning. With the increasing seismic stations and collected data, some deep learning methods are gradually introduced and perform well in earthquake magnitude evaluation and localization. However, how to handle the sparse and non-European multi-stations is still a problem in earthquake multi-station models. This paper designs a multi-station model based on a graph neural network to accomplish seismic source characterization. The model applies the methods of graph theory to represent earthquake data as graph structure and innovatively adds the earthquake phase picks into the edges of the graph. This method mines the potential information among multi-stations effectively. The proposed methods improve the predicting precision and perform better in real-time performance than the compared models.