Risk Assessment and Diagnosis Model for Real‐Time Flood Control Operation of a Multi‐Reservoir System

ABSTRACT

Real‐time flood control operation of multi‐reservoir systems is affected by multiple sources of hydrological and operational uncertainties, which increase the complexity of risk propagation and decision‐making processes. Therefore, effective risk assessment and diagnosis are essential for supporting risk‐informed flood control operation. This study develops a risk assessment and diagnosis model by coupling Copula theory with a Dynamic Bayesian Network (CDBN). The proposed model characterizes nonlinear dependencies and temporal evolution among continuous uncertainties and performs dynamic risk assessment and backward risk diagnosis at both single‐reservoir and multi‐reservoir scales. It enables the calculation of posterior exceedance probabilities of risk events and the identification of dominant risk contributors. The proposed method is applied to a multi‐reservoir system in a large river basin in China. Results indicate that inflow uncertainty is the dominant contributor to reservoir overtopping risk at the single‐reservoir scale, whereas operational discharge uncertainty plays a more significant role in downstream flood control risks at the multi‐reservoir scale. Compared with the conventional Dynamic Bayesian Network, the DBN yields lower estimates of reservoir overtopping risk than the CDBN under medium and high water‐level conditions. The CDBN improves the efficiency of risk quantification and effectively represents dependency relationships among uncertainties, providing a robust tool for risk‐informed decision‐making in real‐time flood control operation of multi‐reservoir systems.