New Data-Driven Algorithm Improves Reservoir Simulations for National-Scale Hydrology
Dams control nearly all rivers in the United States. This means that models used for predicting river flow responses to climate change and extreme events must simulate dam operations and capture the storage and release of water at individual reservoirs. Until now, this representation in national-scale models has been based on a general theory of reservoir operations rather than knowledge of actual decisions. Researchers developed a new algorithm (“Storage Targets and Release Functions Inference Tool”—STARFIT) for inferring reservoir operating policies from data and extrapolating the policies to approximately 1350 data-scarce reservoirs (“STARFITx”). These efforts created the first national-scale inventory of data-driven reservoir operating policies for the United States.
This national-scale inventory of data-driven reservoir operations has the potential to significantly improve the ability to simulate river flow behaviors at a continental scale, including how reservoirs might mitigate extreme flood and drought risk. The reservoir operating policies also enable new metrics for better understanding droughts in the United States. The next steps will be to integrate these reservoir operating policies into a continental U.S.-scale hydrological and water management model to study water storage drought.
Large-scale hydrological and water resource models (LHMs) require water storage and release schemes to represent flow regulation by reservoirs. Given the past lack of observational data on reservoir operations, LHMs have deployed generic reservoir schemes that often fail to represent local operating behaviors. The current research introduces a new algorithm (STARFIT) that generated a dataset of water storage and release policies for 1,930 reservoirs in the contiguous United States (CONUS). The Inferred Storage Targets and Release Functions (ISTARF-CONUS) dataset relies on a new inventory of observed daily reservoir operations to generate reservoir operating rules for 595 data-rich reservoirs. Through STARFITx, researchers extrapolated operating schemes to 1,335 data-scarce reservoirs. This leads to the first inventory of empirically derived reservoir operating policies for all large CONUS reservoirs documented in the Global Reservoir and Dams database. Evaluating the new schemes in daily simulations for the data-rich reservoirs using observed inflow produces a substantial and robust improvement for both release and storage relative to the popular Hanasaki method. The extrapolation approach for data-scarce reservoirs offers, on average, modest gains over the Hanasaki method. ISTARF-CONUS may be readily adopted in any LHM featuring large CONUS reservoirs.