Evaluation of combined models Efficiency for improving the evaporation modeling

In this research by combining individual models outputs, the strengths of each single model are used to make a new model that has better performance than each single model. In this study the efficiency of nonparametric K nearest neighbor and BMA model were compared with BGA , GRA,  AICA, BICA, equal weights averaging and lasso methods. For this purpose, the evaporation rate from the water level was simulated at three stations of Marwa Tappeh, Gonbad and Gorgan, using the individual models of the US Land Development Organization, Tchymyorov, Ivanov, Hanfer, Shahten, Marciano and Meyer. Then each combination of models was implemented to combine the outputs of each single model. The results showed that during the calibration and validation period, the best performance was related to the experimental model of the US Land Development Organization. The best performance of the combination models for the calibration period and validation period at the Marah-e-Tappeh station in GRA in Gonbad and Gorgan stations. Regarding the error indices, the best performance of calibration period of GRA method and validation data is related to the GRA, BICA and AICA methods and the worst the function is related to the EWA method. For the validation period, the nearest KNN neighbor method has a better performance than other combination methods.  Bayesian average results showed that in all three stations, in the case of gamma distribution, different modes of its variance modeling were similar and in all stations of normal distribution modes, they had better performance than gamma distribution modes, as well as the range of uncertainty in normal distribution mode, it was smaller than gamma distribution. On the other hand, in the estimation of the point, Bayesian average modification with normal distribution after gra method performed better than other combination models.