Integrated AI-driven weeks-to-years prediction of water for energy in the Tennessee Valley Authority region
Water availability, river flows and water temperatures can shape the reliability and efficiency of energy production. Reservoir operators need to anticipate changing inflows, hydropower depends on dependable water supplies, and thermoelectric plants require sufficient cooling water while meeting thermal limits. This project aims to give domestic energy providers more useful forecasts, from weeks to years ahead, so they can plan for these conditions with greater confidence.
The project focuses on the Tennessee Valley Authority (TVA) region and will connect Earth system observations, simulations from the Energy Exascale Earth System Model (E3SM) and decision measures developed with stakeholders. The goal is to move beyond describing past hydrological variability, toward forecasts that inform real operational and planning decisions. Better seasonal and annual outlooks could help energy providers consider water availability when making choices about reservoir operations, fuel planning, maintenance schedules, water allocation, and cooling-water compliance.
At the center of the work is an ORBIT/LANDSIM-based SI foundation model developed at Oak Ridge National Laboratory and designed for Department of Energy high-performance computing. It will draw on information from global climate connections, local land-surface conditions, and long-term system behavior. This broad context is intended to support forecasts for specific watersheds and reservoirs at flexible lead times, including predictions of water availability, streamflow, and river temperature.
The project treats SI as a way to investigate and quantify predictability, rather than as a replacement for scientific understanding. Researchers will examine which signals contribute to forecasts, how predictive skill changes over time, and whether information learned from E3SM simulations is useful alongside observations. Forecasts will be compared with baseline methods and assessed for reliability and uncertainty. This evaluation is intended to help users understand both what the forecasts indicate and how much confidence to place in them.
The long-term goal is to provide dependable water and energy risk information at watershed and reservoir scales. This could include probabilities that streamflow, water availability or temperature will cross operationally important thresholds, as well as measures of how reliable those risk categories are. By combining Earth system science, SI, and advanced computing in a reproducible framework, the project seeks to support energy resilience and practical adaptation to changing hydrological conditions.
Partners
- Georgia Institute of Technology
