Detecting and Locating Aftershocks for the 2020 Mw 6.5 Stanley, Idaho, Earthquake Using Convolutional Neural Networks

Abstract

Abstract Our study is to build an aftershock catalog with a low magnitude of completeness for the 2020 Mw 6.5 Stanley, Idaho, earthquake. This is challenging because of the low signal-to-noise ratios for recorded seismograms. Therefore, we apply convolutional neural networks (CNNs) and use 2D time–frequency feature maps as inputs for aftershock detection. Another trained CNN is used to automatically pick P-wave arrival times, which are then used in both nonlinear and double-difference earthquake location algorithms. Our new one-month-long catalog has 4644 events and a completeness magnitude (Mc) 1.9, which has over seven times more events and 0.9 lower Mc than the current U.S. Geological Survey National Earthquake Information Center catalog. The distribution and expansion of these aftershocks improve the resolution of two north-northwest-trending faults with different dip angles, providing further support for a central stepover region that changed the earthquake rupture trajectory and induced sustained seismicity.