Self-evolving artificial intelligence framework to better decipher short-term large earthquakes.

Area of Science:

• Geophysics
• Artificial Intelligence
• Machine Learning

Background:

• Predicting large earthquakes (EQs) remains challenging due to data limitations and complex physics.
• Traditional and current machine learning (ML) methods struggle with earthquake prediction.

Purpose of the Study:

• To develop a novel artificial intelligence (AI) framework for transforming raw earthquake data into ML-friendly features.
• To enable the self-evolution of AI models for improved short-term prediction of large EQs.

Main Methods:

• Utilizing basic physics and mathematics to convert observational earthquake data into new features.
• Implementing an advanced reinforcement learning (RL) architecture for self-evolving AI models.
• Incorporating transparent ML models to predict magnitude and spatial location of large EQs (≥ 6.5).

Main Results:

• The AI framework successfully transforms raw EQ data into usable ML features.
• Transparent ML models reproduced the magnitude and spatial location of large EQs weeks before occurrence.
• Verifications using 40 years of Western U.S. earthquake data showed promising results compared to existing methods.

Conclusions:

• The developed AI framework offers a novel approach to large earthquake research.
• This AI technology can establish a new database of earthquake features for continuous self-evolution.
• The system demonstrates potential for improving short-term earthquake forecasting.