Probabilistic Forecasting of Hydraulic Fracturing Induced Seismicity Using an Injection-Rate Driven ETAS Model

Abstract:

The development of robust forecasts of seismicity induced by human activities is highly desirable to mitigate the effects of disturbing or damaging earthquakes. In this study, we use the well-established Epidemic-Type Aftershock Sequence (ETAS) model to investigate the efficiency of statistically forecasting the seismicity observed during and after hydraulic fracturing operations. In particular, we consider the microseismicity catalogs recorded during unconventional shale gas development at the Preston New Road (UK) site in two adjacent wells, namely PNR-1z treated in 2018 (~38,000 events, including a M_L= 1.5) and PNR-2 treated in 2019 (over 55,000 earthquakes, including a M_L= 2.9). While 17 sleeves were hydraulically fractured at PNR-1z with a total volume of ~4600 m³ of slick water, at PNR-2 operations were completed at only 7 sleeves with ~2600 m³ of injected fluid before the occurrence of the largest event. Considering that ETAS was originally developed for tectonic clustered seismicity, we generate three modified ETAS models to account for externally forced seismicity due to the pumping of pressurized fluid. To mimic operational conditions, where real-time data are not yet available for model parameterization, we also evaluate the performance of out-of-sample forecasts. By formally comparing the ETAS seismicity forecasts to the observations by means of log-likelihood scores, we find that the standard ETAS captures well the low seismicity rates between and after injection periods, but it is substantially outperformed by the modified models during critical periods of high induced seismicity. Given the non-unique and heterogeneous seismic response to fluid injection, the injection-rate driven ETAS forecasts improve further with well-specific parameter estimates and a calibration of the seismic response to sleeve-specific pumping data. Furthermore, at PNR-2 the out-of-sample modified ETAS forecast, parameterized on the PNR-1z seismicity, ranks lower than the other in-sample modified models, but encouragingly outperforms all the standard ETAS versions. The insights from this study contribute towards providing more informative probabilistic seismicity forecasts to be used for real-time decision making and risk mitigation techniques during unconventional shale gas development.