Keynote Speaker

This study investigates the integration of earthquake predictors with deep learning frameworks, presenting preliminary results that showcase the predictive power of individual precursors and their combination with neural networks. The analysis begins by evaluating each precursor individually, including GPS-derived crustal deformation, past seismicity, and other features, to understand their contribution to earthquake prediction. These predictors are then progressively integrated into advanced neural network architectures. Preliminary findings demonstrate that ConvLSTM (Convolutional Long Short-Term Memory) networks effectively use GPS velocity data for spatio-temporal predictions of the daily average number of seismic events in Chile, while LSTM and Transformer models excel in classifying seismic events into foreshocks, mainshocks, and aftershocks based on past seismicity and clustering techniques. Future work will focus on integrating all precursors into unified deep learning frameworks to further enhance prediction accuracy and model robustness. This research underscores the critical role of combining diverse earthquake predictors with deep learning to advance spatio-temporal prediction and event classification, contributing to improved seismic hazard assessment and disaster preparedness.