Seismological models based on a hybrid deep-learning strategy reveal tectonic features and earthquake risk in the Sichuan-Yunnan region

Ziyi Li, Lianqing Zhou, Mengqiao Duan, Yanru An, Zheng Wu, Lisheng Xu, Shiyong Zhou

The Sichuan-Yunnan region is a critical tectonic zone for understanding continental deformation and seismic hazards. We developed a hybrid deep-learning strategy that integrates multi-scale phase picking and high-precision first-motion polarity identification, significantly improving regional seismic monitoring. This approach yielded a unified 2013-2022 high-resolution dataset, including -180,000 relocated events, 2,524 focal mechanisms, and a 0.25° × 0.25° stress field model. Our results reveal pronounced stress segmentation along the Longmenshan fault zone, driven by the interplay between regional tectonic compression and local fault geometry. Integrated analysis suggests that the structural complexity of the Dayi gap likely hinders gap-spanning ruptures. In contrast, the Zheduotang gap faces significantly elevated hazard due to persistent stress loading from the 2014 Kangding and 2022 Luding earthquakes. This unified high-resolution dataset provides a robust seismological framework for assessing geodynamic processes and future seismic risks in the region.