Assessment of the data driven models in downscaling of the daily temperature in Birjand synoptic station

Abstract

In this study, seven models such as multivariate regression, Contemporaneous Autoregressive‐Moving Average (CARMA), CARMA-ARCH (Autoregressive‐Conditional Heteroskedastic), Support Vector Regression (SVR), Adaptive Neuro-Fuzzy Inference System (ANFIS), Support Vector Machine (SVM) and Genetic Programming (GP) were investigated to down scaling the max daily temperature of Birjand synoptic station using 26 predictor’s parameters that resulting from the fifth Intergovernmental Panel on Climate Change (IPCC) report and compared. The max daily temperature values measured from 12/03/1961 until 20/12/2005. In all the above-mentioned methods from 26 predictive parameters using the Pearson correlation test, 15 parameters were selected that have a high correlation with the max daily temperature values. To investigate the values of modeling error, coefficient of determination, root mean square error and criteria effectiveness were used. The results of evaluating the accuracy and model error indicated that from the artificial models such as GP, ANFIS and SVM, the Genetic Programming model has the least amount of errors (about 4 °C) and in the regression models (multivariate regression and support vector regression), support vector regression have been lowest error rate (about 1 °C) and the highest accuracy in simulated max daily temperature values of Birjand station. The results of the investigation the error rate of the mentioned data indicated that after the support vector regression model, two CARMA and CARMA-ARCH stochastic models have high and acceptable accuracy about 97 percentages. In general, the results of simulation the max daily temperature indicates the best accuracy of regression toward smart methods