An integrated simulation-optimization framework to optimize the reservoir operation adapted to climate change scenarios

In order to optimize the operation rule curve of the Shahrchay reservoir in the north-west of Iran under climate change, a new integrated simulation-optimization framework under different climate change scenarios was introduced in the current study. The proposed framework contains several steps: First, the more accurate general circulation models (GCMs) were chosen among several available GCMs and the predictors screening based on decision tree model (M5) was applied to select the most important predictors among enormous number of potential large-scale climate variables of GCMs. Then, the artificial neural networks (ANNs) were trained by the observed temperature and precipitation time series and the selected dominant predictors to downscale the precipitation and temperature parameters of GCMs for the base period (1951-2000) over the case area and the future monthly predictions of temperature and precipitation were implemented by the trained ANNs under A1B, B1, RCP4.5, and RCP8.5 climate change scenarios to assess the future changes of precipitation and temperature between 2020 and 2060. Thereafter, the estimated precipitation values were imposed to the hybrid Wavelet-M5 model to simulate inflow (runoff) to the Shahrchay reservoir and finally, genetic algorithm (GA) was used in order to optimize the operation rule curves of the reservoir system using the predicted average monthly precipitation, evaporation, inflow and considering water supply for agricultural and municipal targets and minimization of total squared deficiencies for future. The Wavelet-M5 model could result in a reliable performance in the one-step-ahead runoff prediction as the obtained correlation coefficient (CC) for training and verifying data sets were 98% and 97.7%, respectively. The results showed that the average long-term annual runoff volume may be decreased between 0.08% and 2.27% in the future with regard to the base period, and the simulation results for present and future conditions indicated a decrease in water availability. Also, the estimated optimal rule curves for the reservoir showed a different shape for all scenarios employed in this study.