Seminars

Accurate forecasting of subsurface flow migration and trapping remains a central challenge in geoscience. Despite well-established physical understanding and advanced numerical simulators, predictive uncertainty persists due to geological heterogeneity, data scarcity, and the prohibitive cost of high-fidelity simulations. This talk presents a scientific machine learning framework to enhance forecasting capabilities for subsurface flows across scales. Starting from lab-scale experiments, we develop digital twins that combine experimental data, physics-based models, and machine learning surrogates to reproduce and predict multiphase flow dynamics under both pressure-driven and capillary–buoyancy–dominated flow regimes. The framework enables quantitative studies of energy storage retention and flow transport, while identifying key geological and barrier properties that control storage/operation performance. Through collaborations, this research aims to establish a scalable and interpretable SciML foundation for digital twins of subsurface energy storage, bridging experimental observables with predictive modeling of subsurface flow in realistic heterogeneous formations.