Optimal Drought Management Using Sampling Stochastic Dynamic Programming with a Hedging Rule

Abstract

This study develops procedures that calculate optimal water release curtailments during droughts using a future value function derived with a sampling stochastic dynamic programming model. Triggers that switch between a normal operating policy and an emergency operating policy (EOP) are based on initial reservoir storage values representing a 95% water supply reliability and an aggregate drought index that employs 6-month cumulative rainfall and 4-month cumulative streamflow. To verify the effectiveness of the method, a cross-validation scheme (using 2,100 combination sets) is employed to simulate the Geum River basin system in Korea. The simulation results demonstrate that the EOP approach: (1) reduces the maximum water shortage; (2) is most valuable when the initial storages of the drawdown period are low; and (3) is superior to other approaches when explicitly considering forecast uncertainty.