GEOthermal SEISmic AI Platform (GEOSEIS-AI): P-wave First Motion Focal Mechanism Determination Module

Focal mechanism solutions serve as an effective observational tool for fracture detection in geothermal exploration and monitoring induced seismicity, aiding in the understanding of subsurface stress states. In these monitoring tasks, often involving high-density, small-scale networks, there is a critical need to generate real-time focal mechanism solutions for a large volume of microseismic events characterized by low signal-to-noise ratios. In this study, we develop an automated workflow integrating deep learning models to determine focal mechanisms. To resolve smaller seismic events (especially magnitude < 3), the P-wave first motion method is employed. Validation tests demonstrate that the workflow can rapidly provide a reliable catalog of focal mechanism solutions. The workflow includes: (1) performing signal-to-noise ratio threshold on P-waves to exclude phases with ambiguous polarities; (2) utilizing a suitable deep learning model, RPNet, to determine first-motion polarity, ensuring accurate identification even with arrival time offsets (about 0.02s), which is characteristic of deep learning-based seismic catalogs; and (3) calculating focal mechanisms using three distinct methods: HASH, FPFIT, and FOCMEC, to ensure solution stability, with the Kagan angle used to quantify consistency (smaller differences indicate higher stability). This workflow has been implemented at the Miaoli geothermal field in Taiwan. The resulting focal mechanisms are predominantly strike-slip; the P-axes exhibit a NW-SE orientation, while the T-axes show a NE-SW orientation, aligning with shear wave splitting results. This workflow has been integrated into the GEOSEIS-AI Platform, aiming to get focal mechanisms rapidly and reliably, enhancing our understanding to the seismogenic structure.

Keywords: GEOSEIS-AI; Deep Geothermal Energy; focal mechanisms; deep learning; automated workflow