Developing the Long-term Runoff Forecasting Scheme for the Three Gorges Reservoir

The accurate prediction of the streamflow is an important issue in hydrology which serves as the basis for flood control and water resources management. Particularly, predicting the inflow of reservoirs at long leading time is essential for hydropower to increase the revenue while decreasing the flood risk. In this paper, we present the study of developing the medium to long-term flood forecasting for Three Gorges Reservoir (TGR) which controls the upstream area of 100 million km². We divided the upstream basin into several sub-basins based on geophysical location as well as hydrological regimes. Then for each sub-basin we analyzed the hydrological regime and adopted a novel machine learning-based framework to predict the runoff at leading time up to 18 months based on atmospheric variables. Finally, we synthesized the runoff prediction based on the topology of sub-basins to get the inflow of TGR.