Comparsion Of Experimental, Regression Models and Artificial Neure Network in Estimating Net Radiation (Rs) In Synoptic Station of Zahedan

Solar radiation is one of the key inputs for most hydrological models in estimating reference evapotranspiration. Furthermore providing and making the measurement tools for this parameter is very costly. In this research, ridation (Rs ) of zahedan meteological station in 1385 to 1389 were used. Some non- linear models such as neure systemwith algorithm BFGS, and neure system with conjugate Gradient training algorithms, and locallinear regression through gamma test were developed. Then , these non- linear models and two expereimental model including Angstrom - Prescott and Glory Mac Kalut were assessed for predicting radiation. For predicting none- linear method, maximum temperature parameters, average speed of wind, surface radiation, and Sunshine were used. Result of comparing measured amounts with models with measured amount by parameter show that the neure system with BFGS algorithm has RMSE= 1.95 , MAE= 1.47 and R2=93% which are the best operation in these models. After that, neure system model with conjugate Gradient training algorithms and local regression model are in secand rank in which RMSE, MAE and R2 are 2.53 , 1.77 , 88% and 2.89 , 1.89 , 82% respectively. Angstrom and MAC colt method have RNSE = 4.38 , MAE=3.21 , R2=33% and RMSE= 4.46, MAE= 3.07, R2=50% respectivety.