Evaluation of Data-Driven Models Based on Downscaling of Daily Temperature Values

In this study, using six models of neural network (ANN), ANFIS, support vector machine (SVM), genetic programming (GP), support vector regression (SVR) and multivariate regression (Reg), the mean daily temperature at Kerman and Bam stations, Iran were studied and simulated during the period of 1961-2005. The results showed that the mean daily temperature during the mentioned periods will increase significantly for both stations. The overall results indicate the superiority of the results of the SVR model (Kerman: RMSE = 1.105 oC and R = 0.992) and (Bam: RMSE = 1.01 oC and R = 0.99). The results showed that the SVR model improved the simulation error rate compared to the neural network (ANN), ANFIS, genetic programming (GP) and multivariate regression (Reg) models in Kerman station about 32, 42, 30 and 11 percent respectively and 62, 59, 27 and 27 percent respectively in Bam station. The results of the root mean square error showed that among the six studied models, the support vector regression model and genetic planning for Bam station and the support vector regression model for Kerman station have higher accuracy. The results also showed that estimating the mean temperature of Bam station has more efficiency and accuracy than Kerman station. In this study, although the analysis of the output results of the models did not lead to the same results, but the results of the models indicate an increase in temperature variables in the two stations of Kerman and Bam in future periods.