Hybrid signal processing/machine learning and PSO optimization model for conjunctive management of surface–groundwater resources

Abstract

Conjunctive management of surface–groundwater resources systems by means of mathematical optimization–simulation techniques becomes an important issue for sustainable water resources development, namely in water-scarce regions. In this study, the particle swarm optimization (PSO) method has been coupled with a hybrid wavelet/ANFIS–fuzzy C-means (FCM) simulation model to determine the optimal agricultural irrigation water allocation in the Miandarband plain, western Iran. Firstly, the optimal amount of conveyed water (CW) from the Gavoshan Dam into the plain is determined by constrained PSO. The constraints are the long-term minimum monthly exceedance streamflows that are estimated for different exceedance probabilities—with a 70% value found to best reflect the average annual river inflow of 3.4 m3/s into the dam—using the two-parameter Weibull distribution as well as the classical Weibull nonparametric plotting position method. Then, based on the politically prioritized proportions of the dam’s allocated water for domestic, environmental and agricultural uses, as well as the share of the plain devoted to agriculture, the optimal monthly CW available for the plain (= 112 MCM/a) is obtained. However, the subsequent estimation of the irrigation water request (IWR) (= 265.8 MCM/a), calculated by the FAO-56 method and using empirical crop coefficients of the present agricultural pattern in the plain, indicates that there is an irrigation water deficit of 153.1 MCM/a that must be made up by groundwater withdrawal (GW), in a way that neither waterlogging nor severe drop conditions in groundwater levels (GL) will occur. The latter are then calculated by the hybrid wavelet/ANFIS (FCM) model, wherefore good performance indicators R2 and RMSE, equal to 0.98 and 0.21 m and 0.94 and 0.31 m in the training and testing phases, respectively, are obtained. Finally, PSO and the hybrid model are coupled to simulate the GL fluctuations—with the above GL constraints—under conjunctive use of the optimal surface (CW) and groundwater resources (GW) in the Miandarband plain. In conclusion, the innovative coupled simulation/optimization model turns out to be a very useful tool for optimal and sustainable conjunctive management of surface–groundwater resources in an irrigation area.