Investigation of Nonparametric Models Performance for Estimating Annual Evapotranspiration Using Gamma Test in Semi-arid Areas of Iran

The need for optimized exploitation of water resources has been increased due to the limited water resources and the different section´s competition. For this purpose, data of six selected synoptic stations including Mashhad, Shiraz, Tabriz, Kermanshah, Khorramabad and Urmia stations were used. Input variables consist of mean temperature (T), relative humidity (RH), sunshine hour (S) and wind speed at 2 m elevation (U2). The M test method was used to determine the length of test period. Since, both Gama index and Standard Error are closed to the axis at the end of figures, the last five-year results were used to test the models. According to the gamma test results, the best input parameters for Mashhad, Shiraz, Tabriz, Kermanshah, Khorramabad and Urmia are respectively (S, U2, RH), (T, U2, RH, S), (T, U2, RH, S), (T, U2, RH), (T, RH, S), (RH, S) under the combined conditions and in a same way, the lowest gamma are 0.005, -0.01, 0.001, -0.002, 0.008, 0.009. Local Linear Regression (LLR), Dynamic Local Linear Regression (DLLR), ANN (conjugate gradient) and ANN (BFGS) models were used to estimate the annual evapotranspiration. The R, MAE, RMSE, MBE, Jakouvidiez (t) and Sabagh (R2/t) criteria were used to evaluate the proposed models. The results showed that the best performance was obtained for the stations; Mashhad, Kermanshah, Tabriz and Shiraz using the best inputs, so that the correlation coefficients for neural network model conjugate gradient were 0.91, 0.98, 0.96 and 0.97, respectively. The general results showed that the non-parametric methods are able to estimate the annual ET, properly.