Using observation data to improve simulation of man-made reservoirs in a global hydrological model

Reservoir operation modeling is challenging since it directly depends on human decision-making that varies from dam to dam. Large-scale reservoir modeling is even more difficult due to the lack of observed operation data. Therefore, generic reservoir operation models are used to model large-scale reservoir operations focusing on a specific purpose, rather than real operations. One of the well-known generic schemes for reservoir operation is Hannsaki et al. (2006) algorithm which is currently used in several global hydrological models including the Water Global Assessment and Prognosis (WaterGAP) model. This algorithm improves hydrological process modeling compared to natural lake methods; however, its performance still needs to be improved, particularly for storage simulations. In this study, a new approach is implemented in the WaterGAP model to improve Hannsaki’s algorithm by using different one-parameter linear operation rules for different reservoir storage levels i.e., above 70 %, between 40 % and 70 %, and below 40 % of reservoir capacity (in total three equations). As a result, we can model each reservoir individually. In addition, we use storage at each time step for estimating the release coefficient instead of the storage at the beginning of the operational year in Hannsaki’s algorithm. The ResOpsUS dataset (historical reservoir inflow, storage, and outflow of major reservoirs across the US) is used to estimate the best parameters for each reservoir and evaluate the results over the US. The results of the new approach show an improvement compared to Hannsaki’s algorithm e.g., when the inflow has a good quality (the Kling-Gupta Efficiency (KGE) greater than 0.50), the median of KEG for storage (outflow) of the new approach reaches 0.23 (0.49) compared to -0.71 (0.25) for Hannsaki’s algorithm.