Flow-Regime-Oriented Ecological Reservoir Operation Under Inflow Uncertainty: A Bayesian-Network Framework for Supporting Aquatic Ecological Risk Management

Reservoir regulation changes downstream flow regimes, which affects pollutant dilution, residence time, transport, and ecological exposure in aquatic ecosystems. This study developed a Bayesian-network-assisted ecological reservoir operation framework to support monthly decisions under inflow uncertainty while balancing water-supply reliability and flow-regime protection. A genetic-algorithm optimization model was used to minimize a weighted objective combining the water-supply shortage index and the hydrological alteration degree. Optimized monthly release, water supply, and storage samples were then used to train a Bayesian network, with inflow and initial storage as input variables and release and water supply as output variables. The framework was applied to the Tanghe Reservoir, China, using the monthly inflow records from 1950 to 1969; the records from 1950–1964 were used for training and those from 1965–1969 for validation. Under the equal-weight baseline scenario, the optimization produced a shortage index of 0.060 and a hydrological alteration degree of 0.276. The Bayesian network reproduced the main optimized operating patterns and gave clear release probabilities under different inflow–storage combinations. These results show that the method can support reservoir operation decisions that balance the water supply and flow-regime protection. The method is most suitable for reservoirs where ecological risk is strongly affected by flow alteration and direct pollutant observations are limited.