Comparison betweenNeuro-Fuzzy and ASD Methods to Predict Climate Change, Case Study: Synoptic Stationof Kerman (1971-2000)

Many climate change impact studies require information at a finer resolution than that provided by Global Climate Models (GCMs). In this study, performance of two statistical models namely, ANFIS and ASD, for downscaling daily precipitation (occurrence and amount) and temperature has been compared. A combination method of Genetic algorithm and ASD was employed to identify downscaling predictors that have the most significant influence on the study variables for a 30 years period of 1971 to 2000 in Kerman Station, south east of Iran. The first 15 years of data (1971 to 1985) were used for calibration and rest was kept for evaluation. One of the main steps in downscaling is choosing the most dominant variables. The results revealed that in case of precipitation, these variables are relative and specific humidity at 500 HPa, surface airflow, strength, 850 HPa zonal velocities and 500 HPa geopotential heights. For modeling temperature, mean sea level pressure, surface vorticity and 850 HPa geopotential heights were the most dominant variables. Outputs from the third generation Canadian Coupled Global Climate Model (CGCM3 were used to test two models over the current period (i.e. 1971-2000), and comparing the results with observed temperature and precipitation in Kerman station. Results indicated that the agreement of simulations with observations depends on the GCMs atmospheric variables used as andlsquo;andlsquo;predictorsandrsquo;andrsquo; and the performance of the statistical downscaling model vary for different seasons. The results showed a slight increase in temperature in future period comparing to baseline (1971-2000). The comparison of ANFIS and ASD models indicated that they performed well for temperature with almost similar results, but ASD model performed better in projecting precipitation than ANFIS.