Probing Slow Earthquakes With Deep Learning.

Area of Science:

• Geophysics and Seismology
• Earthquake Science
• Machine Learning Applications in Earth Science

Background:

• Slow earthquakes can potentially trigger larger seismic events on adjacent locked faults.
• Observing slow earthquakes and associated tremor has been challenging due to their intermittent nature.
• Previous observations of slow earthquakes and tremor were limited to discrete bursts.

Purpose of the Study:

• To develop a method for detecting quasi-continuous tremor associated with slow earthquakes.
• To use tremor detection as a proxy for quantifying slow slip rates.
• To investigate the universality of tremor characteristics across different tectonic settings.

Main Methods:

• Trained a convolutional neural network (CNN) using seismic data from a single station in Cascadia.
• The CNN was trained to detect known tremor preceding and following larger slow slip events.
• The model's performance was evaluated by its ability to recognize tremor in diverse geological settings.

Main Results:

• Successfully isolated and identified tremor and slip associated with slow earthquakes.
• The deep neural network demonstrated the capability to detect quasi-continuous tremor.
• The trained model recognized tremor patterns in other subduction zones and along the San Andreas Fault.

Conclusions:

• Deep learning provides a powerful tool for detecting continuous seismic tremor, offering a proxy for slow slip rates.
• The findings suggest a universality in waveform characteristics and source processes of tremor across different tectonic environments.
• This research advances our understanding of slow earthquake behavior and their potential role in the seismic cycle.