Impact analysis of stochastic inflow prediction with reliability and discrimination indices on long-term reservoir operation

Abstract

Long-term stochastic inflow predictions can potentially improve decision making for reservoir operations. However, they are still not widely incorporated into actual reservoir management. One of the reasons may be that impacts of various types of uncertainty contained in stochastic inflow predictions have not been sufficiently clarified, thus enabling reservoir managers to recognize the advantages of their use. Impacts of uncertainties of stochastic inflow prediction on long-term reservoir operation for drought management are therefore investigated in order to analyze the kind of uncertainty that most affects improvements in the performance of reservoir operations. Two indices, namely reliability and discrimination, are introduced here to represent two major attributes of a stochastic prediction's uncertainty. Monte Carlo simulations of reservoir operations for water supply are conducted, coupling with optimization process of reservoir operations by stochastic dynamic programming (SDP) considering long-term stochastic inflow predictions, which are artificially generated with arbitrary uncertainties controlled by changing the two uncertainty indices. A case study was conducted using a simplified reservoir basin of which data were derived from the Sameura Reservoir basin in Japan with finer discretization settings for SDP. The results demonstrated the additional implication of the effect of stochastic inflow prediction's uncertainty on the authors’ previous work.