Colombian Seismic Monitoring Using Advanced Machine-Learning Algorithms

Abstract

Abstract Seismic networks worldwide are designed to monitor seismic ground motion. This process includes identifying seismic events in the signals, picking and associating seismic phases, determining the event’s location, and calculating its magnitude. Although machine-learning (ML) methods have shown significant improvements in some of these steps individually, there are other stages in which traditional non-ML algorithms outperform ML approaches. We introduce SeisMonitor, a Python open-source package to monitor seismic activity that uses ready-made ML methods for event detection, phase picking and association, and other well-known methods for the rest of the steps. We apply these steps in a totally automated process for almost 7 yr (2016–2022) in three seismic networks located in Colombian territory, the Colombian seismic network and two local and temporary networks in northern South America: the Middle Magdalena Valley and the Caribbean-Mérida Andes seismic arrays. The results demonstrate the reliability of this method in creating automated seismic catalogs, showcasing earthquake detection capabilities and location accuracy similar to standard catalogs. Furthermore, it effectively identifies significant tectonic structures and emphasizes local crustal faults. In addition, it has the potential to enhance earthquake processing efficiency and serve as a valuable supplement to manual catalogs, given its ability at detecting minor earthquakes and aftershocks.