Forecast-informed reservoir operations using a Bayesian approach

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 provide valuable information for streamflow forecasts and reservoir operations. A new Bayesian-based weighting approach for updating reservoir operating rules conditioned on ensemble forecasts and large-scale climate indices is proposed by optimally selecting or weighting members prior to integration into reservoir decision-making. Forecast ensemble member weights are estimated through a likelihood function conditioned on the relationship between climate state and perfect reservoir decision-making (releases), and considered in both explicit and implicit forecast-informed reservoir operations. The weighting method is easy to apply to ensemble forecast-informed reservoir operations and requires no additional parameter optimization. When no relationship exists between climate index and reservoir releases, all forecast ensemble members are weighted equally. A case study of reservoir operations for Grand Ethiopian Renaissance Dam (GERD) on the Blue Nile River demonstrates that a climate index weighting scheme can improve forecast-informed reservoir operations in terms of power generation and firm output.