Estimating of evapotranspiration using remote sensing, artificial neural network and comparison with the experimental method (Penman-Monteith-FAO)

Evaporation waste of water is one of the most important factors. Because evapotranspiration is a complex phenomenon that depends on many factors and data, accurate estimation of evaporation and transpiration, is very difficult and costly. Therefore, the purpose of this study was to estimate evapotranspiration using the surface energy balance algorithm for land (SEBAL) and also evaluate the performance of artificial neural networks. To estimates the Evapotranspiration rate the method of SEBAL Algorithmby using satellite images was applied. For this purpose, four images of Landsat 8 in this study were used that by comparing the results from the two methods, Remote Sensing and Penman-Monteith- FAO Equation presented MSE and MAE as respectively 1.54 and 1.04 per day. To solve the complexity of the evaporation process, Artificial Neural Networks was used for forecasting evaporation pan based on meteorological data. Perceptron with Back-propagation algorithm was applied for training it in this study. It used daily climate data that collected during 13 years from a Safi Abad station in Dezful city for network training. The results showed that the best network was the network with all inputs along with a hidden layer and 28 Neurons in the middle layer. The implementation results of this network presented that statistical Indicators were as MSE (0.0032), MAE (0.0445), R2 (0.9609). Comparing the results from Artificial Neural Networks and Penman-Monteith- FAO as reference method showed that MSE and MAE were 1.11 and 0.52 mm per day, respectively. These results presents that the performance of Artificial Neural Networks was better than the remote sensing method in the estimation of evapotranspiration rate.