EdgePhase: A Deep Learning Model for Multi‐Station Seismic Phase Picking. (28th October 2022)

• Record Type: Journal Article
• Authors: Feng, Tian; Mohanna, Saeed; Meng, Lingsen
• Is Part Of: Geochemistry, geophysics, geosystems. Volume 23:Number 11(2022)
• Page Start: n/a
• Page End: n/a
• Publication Date: 2022-10-28
• Subjects: deep learning -- seismic phase picking -- graph neural network -- edge convolution -- Samos earthquake -- EdgePhase; Geochemistry -- Periodicals; Geophysics -- Periodicals; Earth sciences -- Periodicals; 550.5
• DOI: 10.1029/2022GC010453
• Languages: English
• ISSNs: 1525-2027

Abstract

Abstract: In this study, we build a multi‐station phase‐picking model named EdgePhase by integrating an Edge Convolutional module with a state‐of‐the‐art single‐station phase‐picking model, EQTransformer. The Edge Convolutional module, a variant of Graph Neural Network, exchanges information relevant to seismic phases between neighboring stations. In EdgePhase, seismograms are first encoded into the latent representations, then converted into enhanced representations by the Edge Convolutional module, and finally decoded into the P‐ and S‐phase probabilities. Compared to the standard EQTransformer, EdgePhase increases the precision (fraction of phase identifications that are real) and recall (fraction of phase arrivals that are identified) rate by 5% on our training and test data sets of Southern California earthquakes. To evaluate its performance in regions of different tectonic settings, we applied EdgePhase to detect the early aftershocks following the 2020 M7.0 Samos, Greece earthquake. Compared to a local earthquake catalog, EdgePhase produced 190% additional detections with an event distribution more conformative to a planar fault interface, suggesting higher fidelity in event locations. This case study indicates that EdgePhase provides a strong regional generalization capability in real‐world applications. Plain Language Summary: Identifying seismic phases from continuous waveforms is an important task for earthquake monitoring and early warning systems. Traditional phase recognition methods include visual inspection and detections based on mathematical functions (e.g., STA/LTA, kurtosis, AIC). Recently, machine learning technology has been applied to this task because of its fast operation speed and complete automation. A variety of neural‐network‐based models take the waveforms of a single station as input and predict the P‐phases and S‐phases. In this study, we improve the model performance by taking into account the mutually consistent features in multiple stations. We incorporate a Graph Neural Network module to exchange information relevant to seismic phases between neighboring stations. Compared to the standard single station model, our multi‐station model performs better on seismic data in Southern California in terms of the precision and recall rate. We also tested our model on the 2020 M7.0 Greece, Samos Earthquake and found that it detected significantly more aftershocks compared to local catalogs in the first month after the mainshock. Key Points: We developed EdgePhase, a multi‐station phase‐picking model, by fine‐tuning EQTransformer with Graphic Neural Networks Compared to the standard EQTransformer, EdgePhase increases the F1 score by 5% on the Southern California Seismic data set Performance evaluation of EdgePhase shows its strong generalization ability in real‐world applications