Modeling of potential evapotranspiration by Artificial Neural Network from minimum climatic variables in Mashhad synoptic station

Potential Evapotranspiration (ETo) determination is a key factor for irrigation scheduling، water balance، irrigation system design and management and crop yields simulation. Many variables needed in estimating evapotranspiration and lack of access to some of the required data has made problems in accurate estimation of ETo in these areas. This paper examines the ability of artificial neural networks (ANNs) in estimating ETo from minimum climatic variables. In this study، Multi Layer Percepteron (MLP) Neural Network used with Back-Propagation learning rule (BP) and the weather parameter data used as input were the mean of monthly maximum temperature، minimum temperature، mean relative humidity، wind speed and actual sunshine duration that were available at Mashhad synoptic weather station، from 1951 to 2005. ETo was estimated by Penman-Monteith and used as output. For evaluation of Neural Network utility under situations of absence of some parameters، a comparison was made between the estimations provided by the Neural Networks models and the following empirical modeles: Hargreaves- Samani، Blaney-Criddle، Makink. Neural Networks have obtained better results than the empirical methods for modeling potential evapotranspiration. Based on these results، under situations of limited climatic parametrs، the Neural Networks was able to estimate ETo properly even with the minimum temperature، maximum temperature and extraterrestrial radiation in the study of area (RMSE=0. 286).