In this study, the precipitation and temperature of Tabriz and Ardabil stations in North west of IRAN were downscaled in monthly scale via the artificial neural network (ANN) and adaptive Nero fuzzy inference system (ANFIS). The time periods from 1950 to 2005 and 1977 to 2005 were considered for downscaling the general circulation models (GCMs) respectively for Tabriz and Ardabil stations parameters. The parameters from four grid points around each station were considered as inputs of downscaling. Since there are various parameters, which can be selected as inputs of modeling, nonlinear mutual information (MI) was used to select dominant inputs. The geopotential height, northward near-surface wind, relative humidity, specific humidity and air temperature were among dominant selected inputs. The performance of modeling was assessed via Root Mean Square Error (RMSE) and Determination Coefficient (DC). The comparison of the obtained results indicated that downscaling temperature led to more accurate results than precipitation, since its more linear behavior. Downscaling GCMs via ANN led to more DC than that for ANFIS. Moreover, it was concluded that Can-ESM2 could be a proper GCM for downscaling precipitation in Tabriz station but ACESS-1 led to more accurate results in downscaling Ardabil station parameters. The investigation of the results indicated that different GCMs may have various performance in different stations.
Anahtar Kelimeler: climate change, general circulation models, ANN, ANFIS
|
