Machine learning-based earthquake catalog construction and three dimensional velocity structure imaging in western Sichuan region

The western Sichuan region, located along the eastern margin of the Tibetan Plateau, lies at the convergence zone of three major active tectonic blocks: the Sichuan-Yunnan, Bayan Har, and South China blocks. This area is characterized by intense tectonic deformation, complex fault structures, and frequent seismic activity. Constructing high-precision seismic catalog and high-resolution three-dimensional (3D) velocity model is essential to characterize fault structures, understand the preparatory environment of strong earthquakes, and investigate deep dynamic processes. We employed data from dense seismic arrays and permanent networks in western Sichuan region. By integrating machine learning-based phase detection, earthquake relocation, and dual-grid double-difference tomography, we constructed a high-precision earthquake catalog (WS-AIcat) and a high-resolution 3D velocity model (WS-VM0.15deg) for this region, achieving a lateral resolution of 0.15°. Our results reveal discontinuous high-velocity anomalies along the Longmen Shan fault zone. At the time of the Wenchuan earthquake, the Yingxiu-Beichuan and the Guanxian-Anxian faults co-ruptured with dominant thrust motion, a phenomenon likely controlled by the high-velocity anomaly and distinctive geometry west of the Yingxiu-Beichuan Fault. Furthermore, a significant low-velocity anomaly in the deep northwestern Sichuan block extends forward to the Jinhe-Qinghe and Yongsheng-Binchuan faults, indicating that deep plastic materials in northwestern Sichuan may have crossed the Xiaojinhe-Lijiang Fault, which serves as the surface boundary of the secondary block. A distinct low-velocity anomaly is observed in the southeastern Sichuan Basin, with dense seismic activity along its margins. This correlation may be due to the propensity of high-low velocity transition zones to accumulate more strain under compression. Such a tectonic setting could thereby increase the susceptibility of shale gas extraction to induce seismicity, potentially explaining the frequent earthquakes in the Changning and Weiyuan shale gas fields. Furthermore, AI-based earthquake detection and high-precision relocation not only characterize seismicity on major faults but also reveal previously unrecognized buried faults. The high-resolution velocity model and high-precision earthquake locations obtained in this study provide key constraints on the geometry of active faults, the physics of seismic processes, and the deep seismic environment in this region.