•Establishing a stochastic programming model for floodwater utilization of a complex multi-reservoir system.•Decomposing the system into four basic subsystems and identifying the excess storage for each subsystem.•Integrating risk constraints into the aggregation-decomposition method.•Introducing risk hedging to offset the risks between reservoirs. Floodwater utilization through dynamic operating flood limited water levels (FLWLs) of reservoirs plays an important role in flood control and water conservation. This study proposes a risk-based aggregation-decomposition method for floodwater utilization of a complex multi-reservoir system. The original system is decomposed into four basic configurations. For the subsystem with hydraulic connections, the upper bound of FLWL of each reservoir is identified based on a given risk level for the subsystem and risk hedging between reservoirs. A stochastic programming model to maximize hydropower generation is then established for joint operation of the multi-reservoir system to find and update the optimal operation strategy during flood seasons. The mixed cascade reservoirs in the Pi River Basin are selected as a case study. The application results are compared with those of the original design model and deterministic operation model without considering risk, which shows that joint operation and dynamic control of FLWLs: 1) significantly increases the power generation and improves the utilization rate of floodwaters compared to those of the original design model; 2) power production rate increases with the acceptable risk but the increasing rate between them decreases with higher levels of risk; 3) results in lower utilization rate of floodwaters compared to that of the deterministic model, but the difference is not significant. The proposed method provides a framework for risk-based decision-making in floodwater utilization to enhance the comprehensive benefits of a multi-reservoir system.