Operation rules and patterns for 450+ large reservoirs in the Contiguous United States

The human dimension has long been underrepresented in large-scale hydrological and water resource models (LHM). For example, reservoir operation represents a major and complex anthropogenic process that considerably alters surface hydrology while it is absent or oversimplified in LHMs. A generic data-driven reservoir operation model (GDROM) is developed to simulate the operator’s decision on daily release. This study presents the application of the GDROM to 450+ large reservoirs in the Contiguous United States (CONUS), including a dataset of empirical operation rules for these reservoirs and the discussion of patterns and insights derived from historical operation records. For each reservoir, the hidden Markov model coupled with the Decision Tree (DT) model is employed to extract a small number (1-8) of representative operation modules, using inflow and storage as inputs. Then a DT-based classification model is applied to determine the module use sequence under different hydroclimatic conditions. The DT-based operation modules and module application conditions are converted to “if-then” statements. The input data and derived GDROMs have been published via an established HydroShare repository (www.hydroshare.org). The GDROMs in “if-then” rules can be linked with the channel routing modules of an LHM. In addition, six typical operation modules are identified for the 450+ reservoirs, and patterns of the module application and transition crossing months are analyzed with reservoir use, size, and location. Diagnostic information is provided to improve the representation of reservoir operation in LHMs and to improve reservoir operation in the future.