Data-driven reservoir storage and release scheme parameterized for all large dams of the United States

Large-scale hydrological and water management models (LHMs) require reservoir operations schemes to simulate the influence of human systems on natural earth systems. A typical challenge for representing hundreds of reservoirs in a single LHM is creating consistent scheme capable of simulating realistic water storage and release amidst the diversity in the type of operations. We present a new dataset of observed reservoir operations that includes daily storage, inflow, outflow, evaporation, and elevation level time series for approximately 600 large reservoirs (> 0.1km³ storage) located throughout the conterminous United States. These data have been collected through a combination of mining water agencies databases and formal Freedom of Information (FOI) requests. The newly compiled data are used to derive data-driven simulation schemes, which we compare against contemporary, generic approaches adopted in state-of-the-art LHMs. With release-centric reservoir operations scheme usually more sensitive to biases in hydrologic simulations, we introduce a parsimonious scheme that defines storage targets using harmonic functions that can also be easily extrapolated to un-measured dams. This extrapolation allows us to develop realistic reservoir operations for all 1945 large reservoirs in the conterminous United States currently represented in the Global Reservoir and Dams database.