← Back to Sessions

Detecting an Enormous Number of Small-Magnitude Earthquakes Using EQCCT

Description: 

We propose to apply a compact convolutional transformer (CCT) to pick the P- and S-wave arrival times of the earthquakes (EQCCT). The proposed algorithm consists of two branches to pick the arrival time of the P and S phases, respectively. Each branch of the EQCCT is responsible for P- and S-wave arrivals, respectively. We train the proposed EQCCT using an augmented version of the STEAD dataset. The augmentation strategy includes: adding Gaussian noise, randomly shifting the waveforms, adding a second earthquake to the input window, and dropping one or two channels from the seismogram. We split the augmented STEAD data into 85% for training, 5% for validation, and 10% for testing. As a result, our EQCCT model outperforms both EQTransformer and Phasenet, which are the two most popular deep-learning-based phase-picking methods. We consider the picked phases within 0.2s as a true positive (TP). For P and S picks, the EQQCT achieves the lowest mean absolute error (MAE) and standard deviation error (sigma) compared to the EQTransformer and Phasenet methods. Besides, our EQCCT network shows the highest precision, recall, and F1 score. Afterward, we apply the pre-trained model to three independent datasets (not included in the training set), i.e., the Japanese, Texas, and Instance datasets. The proposed method shows promising results in terms of picking accuracy and the missing rate. The real-time application of EQCCT in TexNet demonstrates its production-ready robustness in terms of detection and phase-picking accuracies. Specifically, we applied the EQCCT to one-month continuous data of 23 stations in western Texas. We picked the P- and S-wave phases using EQCCT, and associated and located the picked phases using Seiscomp and NonLinloc, respectively. As a result, we detected and located a total of 11687 events, which is more than 50 times the number of catalog events in the same period (215). Among them, 11105 events have a high location quality. This test indicates that with a much shorter monitoring period, EQCCT can detect far more small-amplitude earthquake events than traditionally based on manual picking.

Session: Advances in Characterizing Seismic Hazard and Forecasting Risk in Hydrocarbon Systems

Type: Oral

Date: 4/18/2023

Presentation Time: 02:00 PM (local time)

Presenting Author: Yangkang Chen

Student Presenter: No

Invited Presentation: 

Authors