Postdoc Position: Geological Carbon Storage Modeling and Uncertainty Quantification with Deep Learning

Geological Carbon Storage (GCS) is an important geo-solution to mitigate the increasing global carbon emission in order to meet the goal of Paris Climate Agreement. For the purpose of GCS management, numerical simulators with coupled physics (flow/reactive/geo-mechanics/thermal) have been developed and often applied to predict the long-term fate of the injected carbon dioxide to ensure its secure containment. The prohibitively high computational cost of such simulations necessitates the development of efficient and robust surrogate models for general GCS modeling tasks, especially when inverse modeling tasks require high-frequency evaluations of forward models, in order to quantify the uncertainties of rock and fluid properties in the subsurface formations.

Therefore, the objectives of this research project include two aspects: (1) based on the cutting-edge technologies from deep learning, computer vision or physics-informed machine learning, develop robust surrogate forward models to predict the coupled physical process of GCS, such that we can efficiently forecast the spatial-temporal patterns of the subsurface response variables, e.g., pressure, saturation, minerals etc.; (2) integrate the surrogate forward models with a Bayesian inverse modeling framework to achieve real-time or near-real-time uncertainty quantification, such that we can efficiently resolve the uncertainties rising from rock and fluid, and thus improve our understanding about of GCS systems.