GSA Connects 2024 Meeting in Anaheim, California

Paper No. 53-9
Presentation Time: 4:00 PM

ENHANCING CARBON STORAGE ASSESSMENTS THROUGH THE USE OF PROBABILISTIC SIMULATION (Invited Presentation)


CALLAS, Catherine1, WEN, Gege2 and BENSON, Sally M.1, (1)Stanford Center for Carbon Storage, Department of Energy Resources Engineering, Stanford University, Palo Alto, CA 94305, (2)Earth Science Engineering, I-X, Imperial College London, London, SW7 2AZ, United Kingdom

Carbon capture and storage (CCS) is an important greenhouse gas mitigation strategy to reach net zero targets. Currently, carbon storage projects use numerical simulations to model the CO2 plume and pressure buildup required to receive a Class VI permit. However, there is a lot of uncertainty in the subsurface geology and reservoir parameters, which can result in drastically different plume areas and pressure buildups. To appropriately characterize the subsurface, hundreds of simulations should be run to get a probabilistic understanding of the plume migration and pressure buildup. However, due to the high computational cost of numerical simulation models, the number of runs to perform a sensitivity study to reservoir parameters and geology variations is limited. Machine learning methods can speed up the reservoir simulation process to quickly perform the required simulations needed to quantify the subsurface uncertainty, thus enabling a probabilistic understanding of the plume migration and pressure buildup throughout the lifecycle of the carbon storage project. CCSNet.ai (Wen et al., 2021, 2023) is a machine learning-based reservoir simulator that can perform forward simulation on detailed geological models with higher computational efficiency, allowing detailed sensitivity analyses on reservoir parameters. We found that for a conditioned permeability map with the same well log, lateral and vertical correlation, and different random seeds, the plume radius can vary 60% and the area of review up to 150%. Therefore, carbon storage projects should use probabilistic simulation approaches to better understand possible outcomes and improve decision-making.