A hybrid hydrologic modelling framework with data-driven and conceptual reservoir operation schemes for reservoir impact assessment and predictions

Abstract

Reservoirs have been built worldwide to address the water-related issues. To fully understand their potential impacts on the hydrologic regime, explicitly parameterizing reservoir operation in hydrologic models is often required. In this study, two data-driven reservoir operation schemes based on extreme gradient boosting (XGBoost) and artificial neural network (ANN) are respectively developed to predict the reservoir release and storage in hydrologic models for reservoirs with historic in-situ inflow, storage, release data. Then, a hybrid modelling framework is proposed by coupling a high-resolution (3 km) hydrologic model with (1) the developed data-driven reservoir operation schemes and (2) a calibration-free conceptual reservoir operation scheme designed for data-scarce reservoirs. This allows quantitative assessment of the cumulative impacts of dam operation on the hydrologic regime under different reservoir data availability. The framework is applied to the Upper Yangtze River Basin (UYRB) in China that is one of the most regulated river basins across the country due to extensive reservoir construction. Results indicate that both data-driven reservoir operation schemes can well reconstruct the reservoir releases and storage in the UYRB (daily NSE of ∼ 0.9), and the XGBoost performs slightly better than ANN. By coupling reservoir operation, the model shows a remarkably improved performance in reconstructing the daily streamflow of the basin. The major reservoirs in the UYRB can redistribute excessive water in the wet season to the dry season and attenuate the high and low flows, leading to enhanced water security along the river. Our approach provides a practical framework for reservoir impact assessment and predictions.