Deep Learning Models Augment Analyst Decisions for Event Discrimination. Issue 7 (2nd April 2019)

• Record Type: Journal Article
• Authors: Linville, Lisa; Pankow, Kristine; Draelos, Timothy
• Is Part Of: Geophysical research letters. Volume 46:Issue 7(2019)
• Page Start: 3643
• Page End: 3651
• Publication Date: 2019-04-02
• Subjects: Utah -- event classification -- event discrimination -- deep learning -- convolutional neural network -- recurrent neural network; Geophysics -- Periodicals; Planets -- Periodicals; Lunar geology -- Periodicals; 550
• DOI: 10.1029/2018GL081119
• Languages: English
• ISSNs: 0094-8276

Abstract

Abstract: Long‐term seismic monitoring networks are well positioned to leverage advances in machine learning because of the abundance of labeled training data that curated event catalogs provide. We explore the use of convolutional and recurrent neural networks to accomplish discrimination of explosive and tectonic sources for local distances. Using a 5‐year event catalog generated by the University of Utah Seismograph Stations, we train models to produce automated event labels using 90‐s event spectrograms from three‐component and single‐channel sensors. Both network architectures are able to replicate analyst labels above 98%. Most commonly, model error is the result of label error (70% of cases). Accounting for mislabeled events (~1% of the catalog) model accuracy for both models increases to above 99%. Classification accuracy remains above 98% for shallow tectonic events, indicating that spectral characteristics controlled by event depth do not play a dominant role in event discrimination. Plain Language Summary: Seismic events observed using sensor networks are typically reviewed manually by seismic analysts to determine which source generated each event. In Utah, two of the most common event types are tectonic, or naturally occurring earthquakes, and quarry blasts from surface quarry operations. Since analysts in Utah have been reviewing events in Utah for more than 50 years, a large catalog of events labeled by source type exists. In this work we explore methods to leverage labeled event types from part of the catalog to automate event labelling for future events. Our approach includes two neural network variations (recurrent and convolutional) to identify events as either quarry blasts or earthquakes. Both methods achieve similar classification accuracies above 99%, which rivals the accuracy of human analysts on the same task. Key Points: Neural network models achieve above 99% classification accuracy between surface mining and tectonic events in Utah Model predictions are sensitive to P, S, and coda wave energy The three‐component data are typically more accurate but not required for prediction