Probing Slow Earthquakes With Deep Learning

• Published in: Geophysical research letters Vol. 47; no. 4; pp. e2019GL085870 - n/a
• Main Authors: Rouet‐Leduc, Bertrand, Hulbert, Claudia, McBrearty, Ian W., Johnson, Paul A.
• Format: Journal Article
• Language: English
• Published: United States American Geophysical Union (AGU) 28.02.2020; John Wiley and Sons Inc; Wiley
• Subjects: Cascadia; Computational Geophysics; Computational Seismology; deep learning; Earth Sciences; Earthquake Dynamics; Geodesy and Gravity; GEOSCIENCES; Informatics; Machine Learning; Neural Networks, Fuzzy Logic, Machine Learning; Research Letter; Research Letters; Satellite Geodesy: Results; Seismic Cycle Related Deformations; Seismicity and Tectonics; Seismology; slow earthquakes; Solid Earth; tectonic tremor; Time Variable Gravity; Transient Deformation; Cascadia; deep learning; tectonic tremor; machine learning; slow earthquakes
• ISSN: 0094-8276; 1944-8007
• DOI: 10.1029/2019GL085870

Slow earthquakes may trigger failure on neighboring locked faults that are stressed sufficiently to break, and slow slip patterns may evolve before a nearby great earthquake. However, even in the clearest cases such as Cascadia, slow earthquakes and associated tremor have only been observed in intermittent and discrete bursts. By training a convolutional neural network to detect known tremor on a single seismic station in Cascadia, we isolate and identify tremor and slip preceding and following known larger slow events. The deep neural network can be used for the detection of quasi‐continuous tremor, providing a proxy that quantifies the slow slip rate. Furthermore, the model trained in Cascadia recognizes tremor in other subduction zones and also along the San Andreas Fault at Parkfield, suggesting a universality of waveform characteristics and source processes, as posited from experiments and theory. Plain Language Summary Slow earthquakes cyclically load fault zones and have been observed preceding major earthquakes on continental faults as well as subduction zones. Slow earthquakes and associated tremor are common to most subduction zones, taking place downdip from the neighboring locked zone where megathrust earthquakes occur. In the clearest cases, tremor is observed in discrete bursts that are identified from multiple seismic stations. By training a convolutional neural network to recognize known tremor on a single station in Cascadia, we detect weak tremor preceding and following known larger slow earthquakes, the detection rate of these weak tremors approximates the slow slip rate at all times, and the same model is able to recognize tremor from different tectonic environments with no further training. Key Points Deep learning models can recognize tremor on a single seismic station New detections of tremor correlate with local geodetic displacement rate and seem to increase weeks to months before slow earthquakes A model trained to detect tremor in Cascadia can detect known tremor from Japan and California with no further training