Abstract

Deep-learning-based surrogate models show great promise for use in geological carbon storage operations. We target an important application — the data assimilation of storage systems characterized by a high degree of prior geological uncertainty, including uncertain scenario metaparameters. In this work, we first develop a dimension-robust hierarchical Markov chain Monte Carlo (MCMC) data assimilation procedure. Using this procedure, we generate history-matched geomodel realizations and posterior estimates of the metaparameters. We then develop an efficient surrogate modeling workflow for coupled flow–geomechanics systems to predict saturation, pressure, and surface displacement for use in the data assimilation. Finally, we develop a recurrent transformer U-Net surrogate model for faulted aquifer systems and apply it for global sensitivity analysis and data assimilation, showing substantial reductions in geological uncertainty.