Practical management of induced seismicity risk and effective mitigation approaches are crucial to oil and gas operations. Effective risk management procedures benefit from an accurate forecast of the largest potential magnitude event in near real-time, allowing the adjustment of operational parameters to reduce the probability of a felt or damaging event. Many models have been proposed to estimate the magnitude of the strongest possible event. Some of these models rely solely on statistics of recorded seismicity while others account for the relation of event size with operational parameters. There are also models that relate the maximum magnitude with existing geological and tectonic conditions.
In this study, we discuss different published seismicity forecasting models and evaluate their performance on a number of datasets. To understand the sensitivity and reliability of the various models to data quality, we compare observed and forecasted seismicity for local and regional arrays. The results show that a high-quality catalog is essential to accurately forecast seismicity and drive a reliable risk management application.
Next, we evaluate the forecasting models by playing back 30+ datasets that were generated during hydraulic fracturing operations to simulate real-time monitoring conditions. We use three prediction models to estimate the maximum magnitude and one to evaluate the number of events stronger than a threshold magnitude. Our findings show that, in general, maximum magnitude estimates from different models are nearly identical and in good agreement with the observed seismicity. We also discuss the limitation of the models by examining a few cases where the seismicity forecasts were not successful. Finally, we show that over time, the forecasts lose their sensitivity to the injection volume.