Abstract: This study constructs a robust optimization operation model of cascade reservoirs to address the scheduling risks intensified by the non-stationarity of river runoffs under climate change, and applies it in a case study of the Xijiang River basin. Adopting the dynamic Bayesian network (DBN) method, we develop a dynamic probabilistic model for the risk propagation related to water shortage and the insufficient power output in the cascade reservoir system. A comparative analysis is made on the evolution and propagation characteristics of uncertainty risks under our new robust operation rules against those historical operation rules. The results indicate the robust rules lower the mean joint risk probability across the entire basin by 47.9%, effectively mitigating the water shortage risks in the downstream river sections and the insufficient power output of the reservoir system. Risk propagation in the system features significant spatiotemporal heterogeneity. Spatially, the primary risk pathways go through the river sections of Changzhou-Wuzhou and Baise-Xijin; Temporally, the most prominent sequential risk propagation is that between the Baise and Guangzhao reservoirs. We demonstrate a synergistic amplification effect under the upstream risks and the reservoirs’ own historical risk states, which is more sensitive in the case of annual regulation reservoirs. Our findings would help the prevention and control of uncertainty risks and the adaptive operation of cascade reservoir systems.

Key words: cascading reservoir operation, robust decision, dynamic Bayesian network, risk propagation