Large-scale characterization of geologic formations for CO2 injection using Compressed State Kalman Filter

Abstract:

Carbon dioxide (CO₂) storage in deep geologic formations is gaining ground as a potential measure for climate change mitigation. Such storage projects typically operate at large scales (~km), but their performance is often governed by smaller-scale (~m) heterogeneities. The large domain sizes prohibit detailed site characterization and dense monitoring networks, leading to predictions of CO₂ migration and trapping based on rough geologic models that cannot capture preferential flow. Kalman Filtering can be used to improve these prior models by assimilating available monitoring data, thereby tracking system performance and reducing prediction uncertainty. However, for large systems with fine discretization, the number of unknowns is in the order of tens of thousands or more, in which case the textbook version of the Kalman Filter has prohibitively expensive computation and storage costs. We present the Compressed State Kalman Filter (CSKF) that can be effectively used for systems with a large number of unknowns to estimate the underlying heterogeneity and to predict the state of interest (e.g., pressure and CO₂ saturation). The algorithm’s computational efficiency is achieved by using a low-rank approximation of the covariance matrix, as well as a Jacobian-free approach. We demonstrate the estimation and computational performance of our method in a typical CO₂ storage scenario with a spatially sparse monitoring network, but with multiple datasets obtained before and during CO₂ injection. Our data assimilation framework provides an efficient and practical way to characterize geological formations intended for CO₂ injection and storage using monitoring data commonly collected in field applications, as well as to quantify the reduction in uncertainty brought by different types of monitoring data.