A forecast-informed reservoir operation framework incorporating climate indices

Incorporating streamflow forecasts into reservoir management can often lead to improved operational efficiency. Large-scale climate variables and indices – in addition to local hydrologic variables – may also provide valuable information for reservoir operations given their limitate relationship with streamflow. A new tree-based machine learning approach for updating reservoir operating rules conditioned on large-scale climate indices is proposed by selecting the most suitable reservoir decision-making pattern for each year. Multiple types of reservoir operating rules can be extracted from the historical streamflow data with different hydrological (e.g., wet and dry) conditions. Their performance can be recorded and correlated with climate indices by using a decision-tree classification model, and then the rules with the best performance conditioned on a given climate index value can be selected for reservoir operations. A case study of reservoir operations for the Grand Ethiopian Renaissance Dam (GERD) on the Blue Nile River demonstrates that the proposed tree-based reservoir operation framework can accurately select suitable decision-making rules both for normal and forecast-informed reservoir operations. Notably, incorporating May Nino 4.0 values into GERD reservoir operations can increase power generation during flood seasons, especially in extreme years.