An Early Warning System for Earthquake Prediction from Seismic Data Using Batch Normalized Graph Convolutional Neural...

Area of Science:

• Geophysics and seismology
• Artificial intelligence in earth sciences

Background:

• Earthquakes pose significant risks to human populations and infrastructure.
• Effective earthquake early warning systems (EEWS) are crucial for mitigating seismic hazards.
• The accuracy and speed of EEWS depend on rapid and precise determination of earthquake parameters.

Purpose of the Study:

• To develop and evaluate a novel deep learning model for accurate earthquake prediction.
• To enhance the capabilities of earthquake early warning systems through advanced computational techniques.
• To accurately estimate earthquake magnitude and depth using seismic waveform data.

Main Methods:

• Implementation of a deep learning model based on a graph convolutional neural network (GNN).
• Integration of batch normalization for stable and efficient model training.
• Utilization of an attention mechanism to focus on critical seismic waveform features.
• Preprocessing of waveform data and feature extraction using convolutional neural networks (CNNs).

Main Results:

• The proposed GNN model achieved low Root Mean Square Error (RMSE) values: 2.8 for magnitude and 2.87 for depth in Alaska.
• The model demonstrated strong performance in Japan with RMSE values of 4.0 for magnitude and 2.66 for depth.
• The deep learning approach significantly outperformed three baseline models in predicting both magnitude and depth across different seismic datasets.

Conclusions:

• The developed deep learning model accurately predicts earthquake magnitude and depth.
• This model shows significant potential for improving the performance of earthquake early warning systems.
• The approach offers a robust method for seismic risk reduction by providing reliable estimations for various earthquake sizes.