Towards advancing the earthquake forecasting by machine learning of satellite data

Area of Science:

• Geophysics and Seismology
• Machine Learning Applications
• Remote Sensing in Earth Science

Background:

• Earthquakes are a major natural hazard, causing significant fatalities globally.
• Accurate earthquake forecasting remains a significant scientific challenge due to elusive precursors.
• Remote sensing offers unique capabilities for monitoring environmental changes related to seismic activity.

Purpose of the Study:

• To develop a novel machine learning method for short-term earthquake forecasting.
• To investigate physical and dynamic changes in seismic data using satellite observations.
• To assess the performance of the new method against existing state-of-the-art techniques.

Main Methods:

• Utilized satellite data from 1371 earthquakes (magnitude 6.0+) between 2006-2013.
• Developed and applied a novel machine learning algorithm: Inverse Boosting Pruning Trees (IBPT).
• Compared IBPT performance against six baseline machine learning methods using infrared and hyperspectral data.

Main Results:

• The proposed Inverse Boosting Pruning Trees (IBPT) method demonstrated superior performance.
• IBPT significantly outperformed all six selected state-of-the-art baseline methods.
• The framework showed a strong capability in improving earthquake forecasting likelihood across diverse databases.

Conclusions:

• The novel IBPT machine learning approach shows significant promise for enhancing short-term earthquake forecasting.
• Satellite-based analysis of environmental parameters is crucial for detecting subtle seismic precursors.
• This study highlights the potential of advanced machine learning in mitigating earthquake-related risks.