cientists believe that the increase in the concentration of green house gases in the atmosphere will cause climate changes. Global Climate Models (GCMs) have been developed for predicting these changes. But, the low resolution of the GCMs has made it impossible to be used for hydrologic local studies. Most of the GCMs have the resolution of more than (2*2) degrees. Due to this low spatial resolution and the elimination or simplifying of some meso scale events in the atmosphere general circulation models, these models are unable to offer a correct estimation of the weather condition of the study region in comparison with the short-term and regional models. Consequently, the model outputs should be downscaled by regional or statistical dynamical models up to 50 KMs in the spatial scale or stations. So by considering the local effects, it would involve the least errors in general circulation models. Therefore, downscaling is necessary. The objective of this study is to offer the ways for improving the downscaling models accuracy. There are four methods for downscaling: 1) Regression, 2) Weather patterns, 3) Probabilistic, 4) Dynamical. Among these methods, regression is mostly used because of its simplicity. Also among the regression methods, the linear regression is simpler than others. Hence in this study, this method would be investigated. This model works monthly as due to some limitations, the statistical models can’t work daily. For the daily running of the model, the artificial intelligent downscaling can be used. At the second section of this study, the neural network has been used for modeling and improving the models output. BF and RBF networks are used for modeling. The combination of neural networks and fuzzy logic has been considered during recent years by scientists and has been used in many cases. This model works on daily basis. The combination of neural networks and fuzzy logic has been used for the purpose of improving the models outputs. The achieved results show the improving of the offered method in comparison with the previous methods. Then by use of this optimum model, the tempreture is predicted for the next 30 years and its trend is studied and is compared with the result of the past years. Key words: green house, climate change, GCMs, meso scale, downscale, regression, statistical, Artificial Neural Networks, fuzzy logic