Geological carbon storage (GCS) can significantly reduce emissions from hard-to-abate industries. However, large-scale deployment is hindered by the risk of induced seismicity, which has led to multiple project shutdowns. Mitigating such events would require unprecedented reliable seismicity forecasting. We present an adaptive near-real-time workflow that forecasts seismic variables. Using a random forest trained on sequentially partitioned data, we forecast seismicity rate and maximum magnitude. Applied to the Illinois Basin Decatur Project, the model receives operational parameters and the seismic catalog as input. Our results identify well pressure, past event counts, and prior maximum magnitudes as key predictors. However, accuracy declines over time as fluid pressure diffuses into the far-field, highlighting the need to incorporate spatiotemporal pressure diffusion for improved long-term forecasts. Our framework is the first to offer near-real-time forecasting from the operation start with uncertainty quantification, providing a foundation for next-generation seismic hazard mitigation in geo-energy operations.
Walter et al. (2026) studied this question.