[S-SS12] Statistical seismology and underlying physical processes

[SSS12-P20] ETAS-Inspired Spatio-Temporal Convolutional (STC) Model for Next-Day Earthquake Forecasting

*Chengxiang Zhan^1,2, Shichen Gao¹, Ying Zhang³, Jiawei Li⁴, Qingyan Meng⁵ (1.China University of Geosciences (Beijing), 2.The Institute of Statistical Mathematics, 3.University of Science and Technology Beijing, 4.Southern University of Science and Technology, 5.Aerospace Information Research Institute, Chinese Academy of Sciences)

Keywords:Deep learning, earthquake forecasting, statistical seismology

Limited research has explored the integration of statistical knowledge into deep learning models for earthquake forecasting. Traditional deep learning approaches typically necessitate extensive parameter learning from scratch. This study introduces a spatio-temporal convolutional (STC) model that incorporates spatio-temporal decay prior knowledge, derived from the epidemic-type aftershock sequence (ETAS) model, directly into the convolutional kernel. This integration endows the STC model with an initial capacity to learn the mainshock-aftershock triggering patterns, requiring only four trainable parameters for subsequent fine-tuning. The STC and ETAS models were evaluated for next-day earthquake forecasting in California. Performance was assessed using the receiver operating characteristic (ROC) curve, the precision-recall (PR) curve, and the parimutuel gambling score (PGS). Results demonstrate the superior performance of the STC model compared to the ETAS model, when smaller magnitude earthquakes are included in the analysis. This suggests that incorporating earthquakes below the magnitude of completeness enhances the STC model's performance.