Document Type : Original Article


University of Tehran



Background and purpose: Long-term and short-term weather forecasting has been one of the challenges for researchers in the field of climate and climate. Therefore, in order to overcome this challenge, several tools have been developed and presented. Atmospheric public circulation models are among the tools that have received much attention from researchers in recent years for long-term weather forecasting. Taking into account several aspects, these models produce different large-scale spatial forecast scenarios that need to be micro-scaled according to the meteorological characteristics of each region. In this study, the efficiency of exponential microscale models was investigated using different artificial intelligence approaches to predict the average daily temperature of three synoptic stations in Ardabil province.Materials and Methods: Large-Scale Predictive Parameters of the Statistical Period 1961 to 2003 from the National Environmental Prediction Centers (NCEP) Database, Large Scale Data of the HadCM3 Model A1B and A2 Predictive Scenarios in the 2001 to 2100 Statistical Period from the Center Canadian Climate Assessment and Modeling called CCCma and meteorological data of Ardabil stations have been received from the Meteorological Organization. In this study, three methods of statistical microscaling (SDSM), support vector machine least squares (LS-SVM) and multilayer perceptron (MLP) were used for exponential microscaling. The performance of various exponential microscale approaches was evaluated based on CC, MSE, RMSE, NMSE, Nash-Sutcliffe, MAE and Taylor diagrams.Results and Discussion: Based on the obtained results, the MLP model based on the average of the stations has the best result with the values (CC = 0.85), (NMSE = 0.63), (NSH = 0.73) and (MAE = 0). / 52) and are in the second and third ranks of LS-SVM and SDSM models, respectively. Based on the evaluation of the models in the Taylor diagram, the SDSM model performed much worse than the other two models, and the results of the LS-SVM and MLP models were slightly similar. Also, based on the results of exponential microscale and temperature forecast until 2100, a relative increase in temperature has been estimated.Conclusion: MLP model compared to other models in all stations of Ardabil province has a higher ability in microscaling. Also, in the exponential micro-scale of temperature using the MLP model based on the mentioned scenarios in all stations until 2100, an increase in the average daily temperature is predicted. Therefore, it is necessary to prepare the 100-year vision document for land management in this region, its temperature conditions should be taken into account to prevent possible damage.