| Abstract |
Real-time seismic monitoring is critical for the sustainable development of Enhanced Geothermal Systems (EGS). Accurate knowledge of event locations and induced seismic patterns is essential for effective seismic hazard mitigation. However, acquiring high-resolution seismic data in real-time is challenging, as temporary geophone arrays, while providing detailed records, are deployed for limited durations. To address this limitation, we applied a deep neural network (DNN)-based workflow to far-field sensors, allowing enhanced seismic information extraction and reduced reliance on dense sensor coverage. In this study, the DNN-based approach was used to process seismic waveforms recorded during the 2024 Utah FORGE stimulation stages using permanent seismic stations located at a distance from the site. Despite the low spatial resolution of these stations, our results show a good agreement with the seismic catalog generated from high-resolution downhole geophones and Distributed Acoustic Sensing (DAS) systems and with the seismic catalog that was obtained from surface geophone temporary dense array records. The event locations, magnitude distributions, and detected fault structures derived from the DNN-based method closely align with those obtained from temporary geophone deployments, demonstrating the effectiveness of deep learning in enhancing seismic catalog generation from low-resolution seismic data. These findings highlight the potential of DNN-based methods for real-time, long-term seismic monitoring and fault detection, providing a cost-effective and scalable solution for seismic hazard mitigation in EGS operations. |