Sensitivity analysis of meteorological data in estimating reference evapotranspiration with the minimum data using wavelet-neuro-fuzzy, ANN and ANFIS models

The aim of this study was to estimate the ET0 in a moderately cold semi-humid climate in a 22-year statistical period by applying a wavelet-neuro-fuzzy model with a minimum number of input parameters.The results were compared with the ANN and ANFIS models to evaluate the performance of the wavelet-neuro-fuzzy model, The sensitivity analysis of the input parameters was done in three ways: Hill method, coefficient of determination, and StatSoft. Sensitivity analysis showed that temperature (T), Rs, Ra, mean daily wind speed at 2 meters (U2) and Rn were an effective parameter. Based on the results of the sensitivity analysis, six combinations with these parameters were selected.The results indicate that the wavelet-neural-fuzzy model has a better performance than the artificial neural network model. The results also showed that the estimated ET0 value with three inputs parameters of maximum and minimum temperature and solar radiation using fuzzy-neural-wavelet model was more accurate than the neural network. Based on the coefficient of determination and the amount of calculated error for the artificial neural network and the Anfis, use of the combination of 7 input parameters (Ra, Rn, Rs, U2, Tmean, Tmin and Tmax) and four meteorological input parameters (Ra, U2, Tmean and Tmax) lead to more accurate estimates of ET0 in comparison to the FAO Penman-Monteith method. The results also showed that the highest amount of explanatory factor and the lowest error value among the different wavelets used in the fuzzy-neuro-wavelet model were for the 7 and three input parameters (Tmax, Tmin, Rs), respectively.