Optimizing Earthquake Nowcasting With Machine Learning: The Role of Strain Hardening in the Earthquake Cycle

Area of Science:

• Geophysics
• Seismology
• Data Science

Background:

• Nowcasting, used in economics and meteorology, involves determining current conditions indirectly.
• Earthquake seismicity often appears chaotic, making direct analysis challenging.
• Understanding earthquake cycles and predicting seismic hazards are critical for public safety.

Purpose of the Study:

• To introduce a simple two-parameter data analysis for revealing order in earthquake seismicity.
• To investigate an earthquake cycle associated with major earthquakes in California.
• To optimize earthquake hazard estimation using machine learning and assess forecast skill.

Main Methods:

• Developed a two-parameter data analysis technique for earthquake seismicity.
• Applied machine learning, specifically the Receiver Operating Characteristic (ROC) skill score, as a loss function in a supervised learning mode.
• Analyzed seismic data in a 5° × 5° region centered on Los Angeles.

Main Results:

• Identified a hidden order in seemingly chaotic earthquake seismicity.
• Observed an earthquake cycle linked to major seismic events, consistent with prior hypotheses.
• Demonstrated that the state variable time series and associated spatial probability densities exhibit forecast skill.
• Showcased the ability to define current earthquake hazard probabilities rigorously using ROC and Precision (PPV) metrics.

Conclusions:

• The simple two-parameter analysis effectively reveals underlying patterns in earthquake data.
• Machine learning, particularly ROC skill, significantly enhances the optimization of earthquake hazard estimation.
• The developed methods provide a straightforward and rigorous way to define current earthquake hazard probabilities and improve forecasting capabilities.