Learning the Low Frequency Earthquake Activity on the Central San Andreas Fault

Abstract

Gradient boosted tree regression model estimates the number of low frequency earthquake 6 events per hour using seismic waveform statistical features 7 • Machine learning model reproduces bursts of LFE activity and predicts more events 8 occurring than previously cataloged 9 • Averaged daily estimates of LFE count is similar to catalog values and a useful 10 monitoring technique to track deep creep on the fault 11

Mapping glacier basal sliding with machine learning

“…The application of machine learning using continuous seismic records continues to show success in describing physical processes of complex natural systems. While the glacier motion model predictions are not as robust as those for laboratory stick-slip studies [37,63,64,65,42], slow slip in Earth [66], future prediction [67,68], or stick-slip processes in Earth [38], they are nonetheless predictive, especially when describing the long period behavior.…”

“…Numerous theoretical simulations and laboratory experiments contributed to the determination of frictional characteristics (e.g., [31,32,33,34,35,36]). Recently, analyses of seismic signals from laboratory faults [37] and faults in earth [38] applying machine learning have yielded remarkable results indicating that the seismic waves contain information about the fault characteristics at all times.…”

“…We use this analogy to guide the choice of research methodology to monitor the physical state of the glacier. So far, direct and continuous quantification of fault friction cannot be achieved using conventional geophysical approaches, whereas supervised machine learning models are suitable to directly quantify instantaneous fault friction in laboratory experiments and fault properties in tectonic environments [37,39,40,38,41,42].…”

Detection of earthquake precursors using neural networks

“…Although this approach provides real-time access to the physical state of the slowly slipping portion of the megathrust, it has not successfully been applied to seismogenic earthquake prediction. Systematic precursors to seismogenic earthquakes are yet to be identified in the continuous signal applying machine learning (Mignan & Broccardo, 2020;C. W. Johnson & Johnson, 2021).…”

“…Because CNNs can detect features of different scales (Zhao et al, 2017), we may expect that variations of the seismic signal over a wide interval of frequencies and amplitudes may be detected. CNNs have frequently been applied to earthquake detection, generating improved earthquake catalogues by efficiently analysing large quantities of seismic data (Van Quan et al, 2017;Perol et al, 2018;Mousavi et al, 2019;C. W. Johnson & Johnson, 2021).…”