Item

Abstract

Detecting and monitoring earthquakes remains a persistent challenge in seismology. The ever-increasing volume of seismic data provides us with a great opportunity to develop more robust phase picking and association methods. While the application of machine learning models for earthquake monitoring has gained significant attention, there remains a gap in validating the efficacy of these models for unseen data. This study reviews a practical workflow that leverages machine learning techniques for phase picking and association following the September 8th, 2017, Tehuantepec earthquake (M8.2). The September 8th earthquake ranks among one of the strongest intraplate events recorded in the history of México, causing significant damage and deaths in the states of Oaxaca, Chiapas, Tabasco, and México City. The Servicio Sismólogico Nacional (SSN) alone has located over 30,000 aftershocks in six months following the mainshock. The volume of seismic recordings gathered from permanent and temporary networks deployed from collaborations of the University of Texas at El Paso, Universidad Autónoma Cuidad Jaurez, and SSN and the tectonic complexity of this rupture have made it strenuous to use traditional automated detection methods. Using PhaseNet, a convolutional neural network developed for picking arrival times of P and S waves, and the Gaussian Mixture Model Association, GaMMA, an unsupervised association method based on a Bayesian Gaussian Mixture Model, we aim to review a practical workflow for earthquake detection and a catalog of relative earthquake locations for this large number of aftershocks, contributing to studying the broader geotectonic setting of the Tehuantepec region.