LAPIS (Learning And Predicting Induced Seismicity)

Induced seismicity is a severe hazard for geoengineering projects in oil and gas, geothermal energy, underground hazardous waste disposal, energy storage (e.g., hydrogen), and carbon capture and storage. In the case of geothermal energy, induced earthquakes, such as the 2017 Mw 5.5 Pohang, South Korea earthquake, have slowed the planning and development of new geothermal projects, which threatens US energy independence and green-energy goals. Induced seismicity can be mitigated (e.g., by modulating fluid pressure), but existing methods for determining when to apply these measures are inaccurate and unreliable because they fail to account for subsurface complexities. We are developing data-driven, AI-based methods and software to determine the current status of a fault and the likelihood of future seismic events. Unlike other methods, ML approaches allow us to incorporate existing knowledge about fault physics while also “learning” about unknown or difficult-to-measure features. ML methods have been successfully applied to detect faults and probe their current and future slip characteristics in prototype studies. We intend to further develop ML approaches for fault detection and seismic hazard characterization by analyzing seismic data and incorporating equations representing governing seismic processes in the ML process. We approached this problem at both a small scale (a single fault) and a larger scale (geothermal reservoir scale). On the small scale, we extensively studied data from the Mont Terri underground research facility in Switzerland and performed several ML analyses. At the large scale, we obtained seismic, fluid, and wellbore data from Utah FORGE and inserted it into a standardized, publicly-accessible SQL database. We used this data to develop a ML framework for seismic hazard forecasting based on point processes. On the small scale, we will incorporate ML-based denoising and event detection and expand our ML forecasting methods to Distributed Acoustic Sensing (DAS) data. At the large scale, we will expand our SQL database with additional geoengineering sites and data types. We will improve our ML methods by adding new physics-informed components and incorporating more input data types to create a generalized fluid injection simulator. LAPIS will increase the viability and safety of geoengineering activities while building public trust through rigorous seismic monitoring and assessment. LAPIS will (1) evaluate induced seismic risks based on available geological, geophysical, and seismic data, (2) identify subsurface activities causing induced earthquakes, (3) pinpoint data attributes and datasets informative for seismic risk evaluation, and (4) propose mitigation activities such as pressure management to reduce the induced seismicity risks.