Pan evaporation estimation and derivation of explicit optimized equations by novel hybrid meta-heuristic ANN based methods in different climates of Iran

Abstract

Pan evaporation (Ep) estimation is important in scheduling and computing irrigation water requirement. This study evaluated the ability of novel meta-heuristic optimization algorithms including Genetic Algorithm (GA), Grey Wolf Optimization (GWO), and Whale Optimization Algorithm (WOA) that hybridized with artificial neural networks (ANNs) in estimating daily Ep and optimizing explicit predictive equations. Five meteorological stations in five climate areas (Humid, sub-humid, semi-arid, arid, hyper-arid) located in Iran, and each station having ten scenarios of available input data, were used. The Subset Selection of Maximum Dissimilarity (SSMD) was used for pre-processing of data set and automatic selection of the training and testing subsets. The modeling results were compared based on the global performance indicator (GPI), root mean square error (RMSE), coefficient of determination (R2), and Nash-Sutcliffe Efficiency (NSE) criteria in addition to Taylor diagram and Box-plots. The results indicated that the performance of models of ANN-GA-4 (I4: Tmean, RH; R2=0.83, RMSE=0.95, NSE=0.82, GPI=0.68) for Astara station, ANN-GA-1 (I1: Tmax; R2=0.79, RMSE=1.39, NSE=0.78, GPI=0.68) for Gorgan station, ANN-GA-5 (I5: Tmean, n; R2=0.86, RMSE=1.98, NSE=0.86, GPI=0.96) for Tabas station, ANN-6 (I6: Tmean, U2) for Esfahan (R2=0.77, RMSE=1.57, NSE=0.77, GPI=0.6) and Hamedan (R2=0.69, RMSE=1.43, NSE=0.68, GPI=1.06) stations were superior than the other models with limited climatic data. After selecting best model in each station, the explicit optimized equations to calculate the daily Ep in each climate are provided as another major contribution of the present study.