Comparison of RBF and MLP Neural Networks Performance for Estimation of Reference Crop Evapotranspiration

Evapotranspiration is one of the most important components of the hydrologic cycle. This complex phenomenon is related to several climatological factors. Over the last decades, Artificial Neural Networks (ANNs) have shown a good ability for modeling complex and nonlinear systems. In the present research, the ability of Radial Basis Function (RBF) and Multi Layer Perceptron (MLP) neural networks in estimation of reference crop evapotranspiration (ETo) was studied. First, using meteorological dataset of 1951-2004 years for Tabriz Station, the mean values of monthly reference crop evapotranspiration were calculated by Penman-Monteith (PM) method. Then, using these calculated values as target outputs various networks with different structures were defined and trained. Finally, the capabilities of these networks for estimation of evapotranspiration were analyzed using some values of dataset that were not used in the training of neural networks. The obtained results showed that, the value of reference crop evapotranspiration might acutely be estimated (RMSE<0.4081 and R2>0.976 for validation dataset) when the parameters of average temperature and wind velocity were used as the inputs of model. Also, comparison of these two neural network results specified that MLP neural networks had a relatively more accuracy than RBF neural networks in estimation of ETo, and the only advantage of RBF neural networks was their much less time of training.