Earthquake Similarity through Graphical Modeling of Arrival Time Data

We examine how earthquakes can be represented as graphical models to quickly estimate inter-event similarity, with applications for identifying anomalous data in seismic bulletins. A directed graphical model is constructed for each earthquake, where the edge weights represent relative P arrival times between seismic stations in a network. This technique utilizes the fact that the source and path properties of an earthquake are encoded in the P phase arrivals at stations, so that graphical similarity between earthquakes indicates events that are similar. A quick algorithm to compute a similarity measure between earthquakes is applied to the corresponding adjacency matrices between different earthquakes. The algorithm is based on subgraph similarity and a weighted combination of common edges between graphs. Earthquakes are then clustered into groups based on relative similarity. This method is applied to cluster earthquakes based on relative arrival time patterns using the seismic catalog of the 2018 Dynamic Networks Experiment (DNE18) and the University of Utah Seismograph Stations (UUSS) network.