Simulation of extreme flood events for risk assessment for flood control of a reservoir-lake-river system under spatially dependent uncertainties

Chaohu Basin, the lower reach of Yangtze River region, China. This study proposes an integrated framework to assess extreme flood risks in a reservoir-lake-river system under spatially dependent uncertainties and data scarcity. A Vine copula is applied to model the spatial dependencies across multiple risk sources in extreme flood events. Scenario trees are constructed using the Neural Gas Algorithm to generate the representative extreme flood scenarios. These scenarios are incorporated into a flood control optimization model to assess the flood risks under extreme flood scenarios. The results show that the proposed method can effectively simulate the spatial interdependencies of risk source variables based on limited historical data. In addition, the scenario trees effectively extract representative flood scenarios. The flood risk probabilities for the reservoir and lake under extreme compound flood events (with occurrence probabilities below 0.02) are assessed to be 0.0072 and 0.0182, respectively. Compared to random sampling and K means clustering, the proposed method achieves the lowest continuous ranked probability scores (CRPS), indicating higher reliability. This study provides an efficient and systematic approach for flood risk assessment in the Chaohu Basin and can support flood management in similar reservoir-lake-river systems. • Proposes a Vine copula-based scenario tree to generate extreme flood scenarios. • Develops an optimization model for assessing extreme flood risks in reservoirs. • Demonstrates the proposed framework’s accuracy and stability over existing methods.