Earthquake source characterization by machine learning algorithms applied to acoustic signals

Area of Science:

• Geophysics and Seismology
• Acoustics
• Oceanography

Background:

• Underwater seismic events generate acoustic radiation, including acoustic-gravity waves, which propagate information about the source over long distances.
• Rapid characterization of earthquake fault geometry and dynamics is crucial for effective early warning systems for earthquakes and tsunamis.

Purpose of the Study:

• To analyze hydrophone recordings of earthquakes using acoustic signal processing and classification.
• To identify earthquake types (slip type, magnitude) and estimate fault properties in near real-time.
• To validate acoustic-derived properties against established seismic catalogs.

Main Methods:

• Analysis of hydrophone recordings from 201 earthquakes in the Pacific and Indian Oceans.
• Application of acoustic signal processing and classification techniques.
• Comparison of results with the Harvard Global Centroid Moment Tensor (gCMT) catalog.

Main Results:

• Successful identification of earthquake types, including slip type and magnitude, from acoustic data.
• Near real-time estimation of effective fault dynamics and geometry properties.
• Statistically significant correlation found between acoustic properties and predicted slip/magnitude values.

Conclusions:

• Hydrophone acoustic analysis provides a viable method for rapid earthquake characterization.
• This approach can significantly enhance early warning systems for seismic and tsunami events.
• Machine learning algorithms fed with acoustic properties show promise for predicting earthquake parameters.