Local Earthquake Detection and Location From Continuous Seismic Waveforms in Xichang Seismic Array With U-Net

Session: Leveraging Advanced Detection, Association and Source Characterization in Network Seismology [Poster]
Type: Poster
Date: 4/30/2020
Time: 08:00 AM
Room: Ballroom
Description: 

In the past two years, the use of deep learning (DL) methods to detect earthquakes has developed rapidly and has been applied in California, USA and Wenchuan, China. However, whether DL can be used for seismic detection in continuous waveforms, how accurate the arrival time is and whether it can be used for phase association and earthquake location has not been systematically studied.

On May 16, 2018, an M4.3 earthquake occurred in Shimian, Sichuan. The Xichang Seismic Array (XCSA) recorded this earthquake sequence very well. In this study, the Phasenet model trained on the 230,000 local earthquake waveforms of the Northern California Seismic Network was used to detect the continuous waveforms of XCSA from May 16 to June 16, 2018. The number of aftershocks processed by analysist was 133 and the magnitude distribution range was ML-0.2 ~ 4.3. The earthquakess detected by Phasenet was 245 after phase association and earthquake location. The matched catalogs between Phasenet and manual analysts is 113 (the difference in occurrence time is ≤ 1.0 s and the difference between the epicenters is ≤ 4.0 km). In addition, Phasenet detected 112 more earthquakes than analysts. The time difference between Phasenet and manual picks less than 0.2 s accounts for about 85% in total arrivals with the average value of 0.12 s. The average residual after location using the 3-D velocity model is 0.18 s. The horizontal and vertical error is about 1.6 km and 4.0 km, respectively.

This study shows that DL model trained using massive seismic waveform data can be used for seismic event detection and phase picking in continuous waveforms. The probability threshold, signal-to-noise ratio, phase association and earthquake location can be used to screen for more reliable seismic phase arrivals. The Phasenet model can determine accurate phase arrivals, small location residuals and location errors and can provide important input data for accurate earthquake location and body wave travel time tomography.

Presenting Author: Lihua Fang

Authors