Adapting reservoir control policies can help manage climate uncertainty within the limits of existing infrastructure. However, it has not yet been investigated how frequently, and with what record of recent observations, control policies should be dynamically reoptimized as new observations become available. Here we analyze a dynamic reoptimization approach in which control policies are adapted with a fixed frequency (f) using a recent window (w) of observations from a climate scenario ensemble combined with land use scenarios. We evaluate reservoir performance for different combinations of frequency (1 to 20 years) and historical window (5 to 50 years) using a case study of nine reservoirs in the Sacramento–San Joaquin River basin, California. We identify the best-performing parameter combination and fit a regression model mapping hydrologic properties to policy parameters over time to support computationally efficient implementation. Results show that dynamic reoptimization outperforms the calibrated historical policy, with the best-performing policy having a frequency of 15 years and a historical window of 50 years. We also find the long-term peak flow as the most influential predictor of adaptations in the regression model. The findings highlight the importance of the adaptation frequency and observation record in planning when and how to adapt control policies in response to changing hydrology.
Sunkara et al. (Fri,) studied this question.