Hybrid predictor for ground-motion intensity with machine learning and conventional ground motion prediction equation

Area of Science:

• Geophysics
• Machine Learning
• Seismology

Background:

• Machine learning models trained on biased data exhibit significant distortions.
• Predicting earthquake-generated ground-motion intensity using machine learning is challenging due to data limitations.

Purpose of the Study:

• To address the underestimation of strong ground motions in machine learning models.
• To develop a hybrid approach combining machine learning and physical models for improved earthquake intensity prediction.

Main Methods:

• Developed a machine learning predictor for earthquake-generated ground-motion intensity.
• Proposed a hybrid approach integrating machine learning with conventional ground-motion prediction equations.
• Evaluated the performance of the hybrid model against individual approaches.

Main Results:

• The machine learning predictor showed good fit for weak ground motions but underestimated strong motions.
• The hybrid approach successfully reduced the underestimation of strong ground motions.
• The hybrid model demonstrated superior prediction accuracy compared to standalone machine learning or physical models.

Conclusions:

• A hybrid approach of machine learning and physical models is effective in mitigating underestimation issues for strong earthquake motions.
• This integrated methodology enhances the reliability and accuracy of ground-motion intensity predictions.
• The study highlights the benefit of combining data-driven and physics-based models in seismological applications.