An Ensemble Approach to Characterizing Trailing-Induced Seismicity

Earthquakes caused by human activities can pose significant risks, and lingering seis-micity that trails the stopped anthropogenic operation is a particular challenge for risk management. To address this concern, we characterize cases in which induced seismicity stops. Five competing models are fit to 56 trailing seismicity cases that span injection operations including: hydraulic fracturing, enhanced geothermal systems, wastewater disposal, and gas storage. Models are ranked based on a suite of statistical performance metrics. We find that the Omori and stretched exponential models are typically the best fitting; however, because there are cases in which each model is best, we advocate for the use of an ensemble. We discuss a framework for a weighted ensemble that updates based on model performance and then demonstrate with a post hoc "forecast" of trail-ing seismicity. We also find some cases ( similar to 23%) that misfit all the models. Residual analysis of these outlier cases shows common themes, including productive trailing sequences that abruptly cease. Such outliers suggest room for more physically moti-vated models that can encompass phenomenon such as operator mitigation, stress shadows, or poroelasticity. The results of our study provide a quantitative framework for quantifying trailing seismicity, including both forecasting, and observable mitiga-tion criteria.