Estimation of daily pan evaporation using neural networks and meta-heuristic approaches

Abstract

ABSTRACT The present study attempts to integrate a bio-inspired optimization algorithm, the firefly algorithm (FA), into a neural network model to develop efficient models capable of estimating daily pan evaporation at two weather stations (Anzali and Astara) in northern Iran. The relative importance of input variables was determined using Gamma test. The proposed hybrid model was compared with multilayer perceptron (MLP), self-organizing feature map neural network (SOMNN), and support vector machine (SVM) models. The models were evaluated using performance criteria such as correlation coefficient, root mean square error (RMSE), and the Nash-Sutcliffe coefficient (NS). The density functions of the estimates were also evaluated. Results showed that for Anzali weather station, SOMNN model estimated pan evaporation better than other models, but for Astara weather station, the hybrid model estimated better. It was also observed that in the case of reproducing the standard deviation and the density functions of daily pan evaporation, MLP, SOMNN, and SVM models were quite comparable, but the hybrid model edged all models. It may be concluded that converting the simple MLP with FA makes it a powerful hybrid model for estimating pan evaporation.