Evaluation of the Accuracy of Artificial Intelligence and Regression Models in Simulation the Daily Temperature

Prediction of climate change using the recorded data from reference periods requires precise methods that are able to accurately detect fluctuations and predict changes for future periods. In this study, six multivariate regression models, ANN, SVR, ANFIS, SVM and GP, were investigated and compared for down scaling the daily mean temperature of the Urmia synoptic station, using 26 prediction parameters taken from the fifth IPCC report. The average daily temperature values measured from 12/03/1961 to 20/12/2005 were used for the analysis. 16 of the 26 parameters having a high correlation with the average daily temperature values were selected for all methods using the Pearson correlation test. To investigate the modeling errors, the coefficient of determination, Root Mean Square Error, and effectiveness criteria were used. The results of the evaluation of the accuracy and modeling error showed that among the smart models, GP, ANN, ANFIS and SVM, the Genetic Programming model has the least amount of errors, and in the regression model (multivariate regression and support vector regression) support vector regression has the lowest error rate and the highest accuracy of simulated daily temperature values of the Urmia station. In general, the results of the simulation of the mean daily temperature indicate that regression has better accuracy than smart methods. Since in this study, we only used the data from the Urmia synoptic station, so the results are only valid for this station, and it is not safe to generalize the results for all stations.