Improving uncertainty quantification of geological carbon storage with data assimilation

Abstract

Geological Carbon Storage (GCS) is an important component of strategies to reduce atmospheric CO2 concentrations. The long-termsecurity of stored CO2, however, depends on a deep understanding of the subsurface. The rock formations used for storage are complex and varied, and ourmeasurements are sparse, whichmakes it difficult to predict how the CO2 plume will migrate and how underground pressures will change. These uncertainties create real risks: the stored CO2 could leak, the injection could trigger small earthquakes, or the ground could move enough to damage surface infrastructures. To manage a GCS project safely and effectively, we need models that can predict these coupled flow and geomechanical effects and, more importantly, to understand and quantify the uncertainties present in each estimate of the process. The quality of our forecasts of the CO2 plume behavior depends on how well we can define the model’s uncertain parameters with the measurements available. This thesis presents amethodology that integrates physics-based simulation, data assimilation, and machine learning to improve uncertainty quantification for GCS. The work aims to deliver practical procedures that help quantify uncertainty in model predictions, guide the design of effective monitoring programs, and increase confidence in the long-termsecurity of stored CO2…