Forecasting evapotranspiration using artificial neural networks with the lowest meteorological data

Accurate estimating evapotranspiration is crucial for water resource management. Evapotranspiration is an important component in water balance in different areas. Knowing the amount of water consumed per product, water engineers are able to calculate evapotranspiration as the most important component of hydrological cycle. In this study, the daily evapotranspiration of Urmia Plain was calculated using meteorological data during the period of 1984-2011 using FAO - Penman - Monteith as a base method. Then, evapotranspiration was calculated with the help of MLP and RBF neural network models using different scenarios with different input parameters. The results indicated that the daily evapotranspiration could be predicted with acceptable accuracy (RMSE = 0.985 and R2 = 0.963 for MLP network and RMSE = 0.537 and R2 = 0.963 for RBF network) using only three parameters: average temperature, sunshine hours, and wind speed. In general, it can be observed that evapotranspiration equation is more depended on the sunshine hours, wind speed, and temperature. Both MLP and RBF networks could be used for calculating the amount of evapotranspiration with high accuracy, but total accuracy of MLP network is more than RBF network.