Evaluation of error and uncertainty in downscaling SDSM and ANN

In the last decades, greenhouse gases in atmosphere have increased as a result of natural and human activities and thus, earth temperature has increased. Rising global temperature, in turn, leads to significant changes in related fields, especially water resources and agriculture. So, investigating and modeling climate changes can be considered as a very important factor in water resources management planning. Different studies have been done in the field of climate change issues in the world, but, at the moment, AOGCM model is the most reliable tool to generate climate scenarios. It is necessary to downscale AOGCM data using different techniques in station scale and compare linear and nonlinear downscaling models. In liner method SDSM and in nonlinear method ANN programming were used in MATLAB. For investigating the amount of error, mean biomass monthly and annual and for extreme data, variance and for analyzing uncertainty Man-Witney test were used in 95 percent level. Results showed the amount of mean monthly errors are 0.75, 12, 11 and 7 mm in Ghaemshahr, Babolsar, Ghoran Talar and Bandpey in SDSM model and 3, 2, 26 and 4 mm in ANN model and the amount of mean annual errors are 9, 146, 141 and 87 mm in SDSM model and 45, 32, 321 and 48 mm in ANN model (increased or decreased), respectively. Examining the performance of variance showed that ANN model was somewhat better than SDSM model. Also, results of uncertainty for 12 months in Ghaemshar, Babolsar, Quran Talar and Bandpey stations showed 8, 3, 6 and 4 in SDSM model and 4, 2, 2 and 3 in ANN model, respectively. In general, this study showed that in studies on climate change effects on runoff, uncertainty, and when limited data are available, SDSM model should be used and when the aim is investigating the flood and its minimum and maximum estimation, it is better to use ANN model.