New Mexico Geological Society Annual Spring Meeting — Abstracts

Recent increases in seismicity due to anthropogenic activities in southeast New Mexico, USA have demonstrated the need for timely earthquake detection and updated catalogs of seismicity in this region. Previously, earthquake detection and location in this region was performed manually; automated detection methods can improve efficiency, accuracy, and completeness of earthquake catalogs, which is essential in understanding and reacting to induced seismicity. This study tested several existing automated earthquake detection tools from general waveform template matching technique to more sophisticated machine learning algorithms to assess the efficiency of these tools in event detection within our seismic network in southeast New Mexico. The study incorporated continuous waveform data of multiple seismic stations from multiple seismic networks such as New Mexico Tech seismic network, USGS, Nanometrics research network, and the Texas seismological network and evaluated the detections from multiple stations to generate an earthquake detection catalog. Standard template matching results are compared with global deep learning tools, the EQTransformer (Mousavi et., 2020) and the PhaseNet auto picker (Zhu and Beroza, 2019). The automated detections from the three methods are compared with an earthquake catalog derived for the study period through manual analyst review. We used these catalog comparisons, specifically the numbers of missed and false detections, in conjunction with ability to implement the automated tool within our routine network operations to select an appropriate automated tool for the network. EQTransformer had the lowest false detection rate of 13%, whereas PhaseNet detected the most number of events amongst the three methods. Considering the ease of implementation, false detection rate and computational resources required, EQTransformer was preferred as the automated earthquake detection tool for our real-time earthquake monitoring workflow. The chosen automated tool was further applied to continuous dataset for the year 2020 to build more detailed and complete catalogs for the seismically active southeast New Mexico. This automation will advance our ability to provide the timely earthquake detection and location that is needed in this significant region of the Delaware basin.

References:

1. Mousavi, S. M., Ellsworth, W. L., Zhu, W., Chuang, L. Y., & Beroza, G. C. (2020). Earthquake transformer—an attentive deep-learning model for simultaneous earthquake detection and phase picking. Nature communications, 11(1), 3952.
2. Zhu, W., & Beroza, G. C. (2019). PhaseNet: a deep-neural-network-based seismic arrival-time picking method. Geophysical Journal International, 216(1), 261-273.