Estimation of reference evapotranspiration using experimental models and approaches based on neural network (case study: Khorramabad City, Iran)

ABSTRACT FAO model is used as a reference to estimate reference evapotranspiration. In this method, a lot of input information is needed to estimate evapotranspiration, which is difficult to collect this data in many cases, so, the need to replace simpler models with low inputs and proper accuracy is felt. Therefore, this research aims to investigate the capability of models based on wavelet neural networks and Group Method of Data Handling, along with empirical models of reference evapotranspiration estimation, which require less data, compared to the FAO–Penman–Mantith model under urban climatic conditions. Various monthly meteorological data were used in the period of 1982–2017. Based on the correlation between input and output parameters, six input patterns were determined for modeling and Tmax was the most effective parameter on ET. Experimental ET analysis showed that Kimberly Penman has the best accuracy with RMSE = 0.39 and R = 0.98. The results of the analysis of the models showed that the combined models have a good performance in estimating evapotranspiration. The results of the evaluation criteria showed that the wavelet neural network model with the average absolute value of relative error of 0.0524 and the mean ratio of 1.001 had a better performance in estimating reference evaporation and transpiration.