Probing Slow Earthquakes With Deep Learning

Abstract

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.

An exponential build-up in seismic energy suggests a months-long nucleation of slow slip in Cascadia

“…We interpret the observed build-up in seismic noise within tremor frequency bands preceding failure as the signature of an increase in the number and intensity of low-amplitude tremors, too small to appear on several stations and be cataloged with array-based techniques. In our recent work 40 , we showed that a neural network can detect many more tremors compared to the catalog it has been trained on. The number of tremors detected on surface stations in the same area by neural network seems to accelerate around 100 days before the rupture as well (see Supplementary), although the behavior preceding failure is not as clear as in the continuous seismic noise (from borehole stations).…”

Machine Learning Modelling and Feature Engineering in Seismology Experiment

“…This leads to the idea of including autocorrelation calculation as a core operation of the deep learning algorithm instead of or in conjunction with the usual convolutional operation. As shown in [33], deep learning modelling can be adapted to recognize tectonic tremors. One of the possible directions of future research may be incorporating autocorrelation of acoustic (tectonic) data into existing algorithms of artificial intelligence in the field of seismology.…”

Machine learning and fault rupture: a review

“…These masks were used to clean new waveform data; on a few examples, retaining only the masked signal is shown to drastically improve the STA/LTA functions associated to earthquakes (Figure 1.14), a demonstration of the potential of combining ML based waveform denoising with standard seismology tools. A different approach for the same problem has been developed by Rouet-Leduc et al [127], who showed that neural network interpretation techniques can be used to denoise tremor waveforms and produce much cleaner recovered signals of interest.…”

“…Indeed, this has been demonstrated by Nakano et al, who showed showed in [131] that a CNN could reliably discriminate between earthquakes and tectonic tremor using a catalog of tremors recorded near the Nankai trough in Japan. Rouet-Leduc et al showed in [34] that a CNN trained to distinguish tectonic tremor from background noise could be used to extract the tremor signals to improve event detection, using seismic records from Vancouver Island. In [130], Hulbert et al showed that this deep learning-based extraction of tectonic tremor signals enables the location of many more tremor events in Cascadia.…”

“…The study of material failure and rupture in geophysics is an extremely broad field [1], involving observation and analysis of geophysical data from failure simulations at laboratory and field scale [2,3,4,5,6,7,8,9], laboratory experiments [10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26? ] and in the solid Earth [1,27,28,29,30,31,32,33,34,35]. Despite the vast apparent scope of this field, one approach to distill its essence is based on the following question: given a seismic signal received by a sensor or a set of sensors, what information can be gleaned concerning the process generating the signal?…”