Climate change scenarios generated by using GCM outputs and statistical downscaling in an arid region.

Two statistical downscaling models, the non-homogeneous hidden Markov model (NHMM) and the Statistical Down-Scaling Model (SDSM) were used to generate future scenarios of both mean and extremes in the Tarim River basin, which were based on nine combined scenarios including three general circulation models (GCMs) (CSIRO30, ECHAM5, and GFDL21) predictor sets and three special report on emission scenarios (SRES) (SRES A1B, SRES A2, and SRES B1). Local climate change scenarios generated from statistical downscaling models was also compared with that projected by raw GCMs outputs. The results showed that the magnitude of changes for annual precipitation projected by raw GCMs outputs was greater than that generated by using statistical downscaling model. The difference between changes of annual maximum air temperature projected by statistical downscaling model and raw GCMs outputs was not as significant as that for annual precipitation. In total, the magnitude of these increasing trends projected by both statistical downscaling models and raw GCMs outputs was the greatest under SRES A2 scenario and the smallest under B1 scenario, with A1B scenario in-between. Generally, the magnitude of these increasing trends in the period of 2081 to 2100 was greater than that in the period of 2046 to 2065. The magnitude of standard deviation changes for daily precipitation projected by raw GCMs outputs was greater than that generated by statistical downscaling model under most of combined scenarios in both periods. [ABSTRACT FROM AUTHOR]