Forecasting and characterizing induced seismicity at the Utah FORGE EGS site

Managing induced seismicity risk is an absolute must to enable the widespread use of deep geothermal technologies, and thus contribute to the transformation towards a sustainable and low-carbon energy sector. A full-scale application of real-time monitoring and forecasting of induced seismicity was tested in April 2022, during a three-stage hydraulic stimulation in a deep granite heat reservoir of low permeability at the Utah FORGE EGS site. During the stimulation, a total of ~1600 m³ pressurized fluids were injected into the target reservoir of ~2.4 km depth to generate fracture networks and improve reservoir permeability for heat extraction. Stage 3 had the most complete monitoring network and thus was used to test the components of an Adaptive Traffic Light System (ATLS). The test produced very positive results, although the data stream was lost early into the stimulation.

Here, we further characterize and perform a retrospective forecasting of post-processed data related to the induced seismicity recorded during stage 3 of the 2022 stimulation at FORGE site. We first investigate the geometrical distribution of the seismicity and discuss it in the context of the stress field at FORGE injection site. The statistical inference indicates that the distribution of the seismicity lies on a plane sub-parallel to the S_v - S_Hmax and orthogonal to S_Hmin, and with strike orientation rotated 10^o counterclockwise with respect to the N25^oE average orientation of S_Hmax. The analysis seems to indicate that seismicity is induced by a tensile fracture, although we cannot completely rule-out seismic activity on a pre-existing fault. Gutenberg-Richter b-value variations in space are compatible with large magnitude events occurring at the edges of the earthquake propagating front. We further investigate the possible fracturing mechanisms triggered by the injection operation by fitting three plausible physical models: (1) a high-pore pressure diffusion model, (2) an aseismic crack model, and (3) a penny-shaped tensile crack model as the causative process of the recorded seismicity. The analysis of the seismicity evolution alone allows us to fit the three considered scenarios to the data independently, however we cannot exclude a combination of the three processes acting together. Through pseudo-retrospective forecasting, we then replay the induced seismicity as it was happening in real-time. We demonstrate that even if the physical processes are complex and likely difficult to disentangle using the seismicity alone, a simple empirical statistical seismicity rate forecasting model has stable predictability of hydraulic fracturing.