Evaluating Artificial Neural Network and Physical-Empirical Models of Reference Crop Evapotranspiration Estimation in Semi-Arid Climate

Accurate estimation of evapotranspiration (ET), which is one of the main components of hydrologic cycle, is very important in water resources management, irrigation planning and environmental studies. Accurate measurement of this component is very difficult, so the application of the models relying on available meteorological variables might be an alternative to direct measurement. This study aimed at evaluating multilayer perceptron artificial neural network as well as sixteen reference crop ET estimation models in Sanandaj. To achieve this object, lysimeteric ET values were considered as standard values for evaluating the applied models. The results showed that the FAO improved pan evaporation method provides more accurate results than other applied methods. In addition, the results indicated that the artificial neural network produces more accurate results for estimating reference evapotranspiration (ET0). Among others, the pan evaporation and artificial neural network models showed, respectively, 0.28% and 3.37% under-estimation and overestimation in the studied region.