HydroORBIT: A Physics-Informed Multimodal AI Foundation Model for Coupled Surface-Groundwater Prediction in Energy-Critical Regions
HydroORBIT is a research project to improve predictions of water availability by developing a physics-informed artificial intelligence foundation model for coupled surface water and groundwater systems. Reliable water predictions are essential for energy production, water resource management and infrastructure planning. Hydropower operators need to anticipate river inflows, thermoelectric power plants depend on adequate water for cooling and communities need timely information to prepare for floods. Water systems are difficult to predict because they involve complex processes that unfold across many scales, vary among landscapes and subsurface conditions, and are observed only at limited locations.
To address these challenges, HydroORBIT will combine advanced SI with established scientific knowledge about how water moves. The model will bring together weather data, streamflow measurements, groundwater observations, landscape characteristics, and results from physics-based hydrologic simulations. Incorporating physical principles into the SI is intended to help it produce predictions that are not only accurate, but also consistent with the processes that govern water movement. The project aims to make those predictions more computationally efficient and useful across different regions and applications.
The team will demonstrate HydroORBIT through three applications tied to energy and water security. In the Tennessee River Basin, it will forecast river inflows to inform hydropower operations. In California, it will map groundwater availability to support understanding and management of an important water resource. In Florida, it will predict coastal flooding, providing information relevant to communities and infrastructure exposed to flood risks. Together, these demonstrations will test whether HydroORBIT can deliver faster, more transferable, and more reliable predictions than existing physics-based and machine learning methods.
The project will establish a scalable, scientific SI framework and make reusable models, datasets and workflows available through the Department of Energy’s Genesis ecosystem. These resources are intended to help other researchers build on the work and accelerate scientific discovery across Earth and environmental sciences.
Ultimately, HydroORBIT seeks to strengthen predictive understanding of interconnected surface and groundwater systems. Better predictions could help energy providers plan operations, resource managers make informed decisions, and infrastructure planners prepare for changing water conditions. By joining scientific principles with SI and sharing the resulting tools, the project aims to advance both practical water forecasting and the broader use of SI in scientific research.
Partners
- Georgia Institute of Technology
- Advanced Micro Devices (AMD)
- Tennessee Valley Authority (TVA)
