Multi-Fidelity AI Foundation Model for Coupled Surface–Groundwater Predictions of Water Availability and Flood Hazards Across CONUS

E. Yan, ANL

For Proposal 297423 in response to DE-FOA-0003612 (Genesis Mission, UQ Focus Area 16C)

Reliable water supplies and flood forecasting are essential for U.S. energy production, infrastructure, agriculture, and communities. Power plants, hydropower facilities, data centers, and irrigation and municipal water systems all depend on accurate information about drought, groundwater, soil moisture, streamflow, and flooding. Current large-scale hydrologic models are often too computationally expensive to provide local detail across the continental United States, and many do not fully represent human influences such as groundwater extraction.

This project will develop an artificial intelligence (AI) foundation model that learns jointly from multiple physics-based surface-water and groundwater simulations spanning resolutions from roughly one kilometer to tens of meters, together with real-time observational data. By fusing these sources within a single generative modeling framework, the model will produce rapid, uncertainty-aware forecasts of groundwater levels, soil moisture, streamflow, and flood depth under both natural and human-influenced conditions. The research will begin with a focus on Midwest water availability and flood hazards and expand toward the first continental-scale, high-resolution AI-driven hydrologic prediction capability.

The resulting tools will support analyses of drought and flood impacts on energy systems and other water-dependent infrastructure, informing both near-term operational decisions and long-term resilience planning. This work will advance scientific understanding of coupled surface-water and groundwater processes across scales and provide actionable tools to strengthen national water-energy security.