Investigating Injection Pressure as a Predictor to Enhance Real-Time Forecasting of Fluid-Induced Seismicity: A Bayesian Model Comparison

Abstract Fluid-induced seismicity is now a growing concern in the spotlight and managing its risks entails a probabilistic forecast model suited to real-time applications, which commonly relies on the operational parameter of injection rate in a nonhomogeneous Poisson process. However, due to potential injectivity change, gas kicks, and other processes, injection rate may not provide as robust a proxy for the forcing process as injection pressure, which embodies fluid–rock interactions. Hence, we present a Bayesian approach to prospective model comparison with parameter uncertainties considered. We tested nine geothermal stimulation case studies to comprehensively demonstrate that injection pressure is indeed the main physical predictor of induced seismicity relative to injection rate, and when combined with the latter as predictors, can give the best-performing model and robustly enhance real-time probabilistic forecasting of induced seismicity. We also discussed the implications of our results for seismic risk management and potential directions for further model improvement.