Statistical Downscaling Methods In Regional Climate Change Research In China

Climate change is an important issue which was concerned by the governments and scientists all the while. Nowadays, discussions of future climate change generally depend on the global climate models (GCMs), but because of the resolution of current GCMs are relatively low (usually around 150-300km), they have difficulty in predicting the regional-scale climate signals. During recent decades, regional climate models (RCMs) and statistical downscaling models have been developed.At first, based on the daily temperature data of east China and the sea-level pressure field and aerial variables from the National Centers for Environmental Prediction and the National Center for Atmospheric Research (NCEP/NCAR) re-analysis, using the statistical downscaling methods based on multiple linear regression(MLR) and three kinds of principal component analysis (PCA) methods, specifically empirical orthogonal function(EOF), expanded EOF(EEOF) and common EOF, we studied the monthly mean surface air temperature from 1959 to 2008,and compared four kinds of statistical downscaling methods. Our results show that compared with three kinds of PCA methods, the statistical downscaling method based on MLR has some advantages in the stimulation of surface air temperature's inter-annual variation in east China. While applying the statistical downscaling methods that based on PCA to surface air temperature, the predictor domain is an important factor influencing the downscaling result. When we use statistical downscaling methods to predict surface air temperature, the use of variance inflation may improve the predicting technique. It is necessary to include temperature factors when downscaling surface air temperature. These four kinds of statistical downscaling methods all have more skill in the simulation of average annual values than spatial averaged values.Based on the daily temperature data obtained at nearly 600 observation stations in China and the sea-level pressure field and aerial variables from the NCEP/NCAR re-analysis, we studied the surface air temperature of January and July from 1970 to 1999, using the statistical downscaling method based on MLR. Then, we compared the results derived from the statistical downscaling model with observations. We also applied the statistical downscaling method to the future climate condition, in attempt to predict the surface air temperature of three periods (2010-2039, 2040-2069,2070-2099) of the 21st century. Furthermore, we analyze the results based on the statistical downscaling method under several GCMs from Intergovernmental Panel on Climate Change (IPCC) Special Report on Emissions Scenarios (SRES) A2 scenario. Our results show that the statistical downscaling method based on MLR is fine in the stimulation of the surface air temperature in China. For the present climate condition, the statistical downscaling method can improve the GCMs'ability of simulating surface air temperature, with better results in the eastern part of China than in the west, and in the plain than in the mountainous region, especially in July. When GCMs'results differ a lot from the observed results, we can still get good results through the statistical downscaling, and thus all the GCMs are more consistent with the observation. For the 21 st century, the monthly mean surface air temperature has a significant increase at almost all the stations in both January and July. The estimated mean temperature increase is larger in the north than in the south, and in the west than in the east.When we concerned about the climate change, precipitation is one of the most important meteorological elements. This paper then based on the output of EHAM5, used statistical and dynamical downscaling methods to analyze the surface air temperature and precipitation in China, and to predict the changes of these two elements in mid-21 century. The results show that Statistical Downscaling Model (SDSM) have real skill at the representation of seasonal average surface air temperature, it still has a certain capacity in the simulation of abnormal-distribution variables such as precipitation, with better results in the eastern part of China than in the west, which may results from the complicated topography of the west; we compared the changes of these two variables between SRES A1B experiment and 20C3M experiment for the two periods of 2041-2060 and 1981-2000, all the statistical downscaling, dynamical downscaling and GCM output show significant increase in the surface air temperature in all seasons, and the amplitudes of increase largen from south to north, and are bigger in winter half year than in summer half year. But the change of precipitation is completely dissimilar. In the result of statistical downscaling the precipitation mainly decrease while the dynamical downscaling result shows obvious increase. In addition, statistical downscaling can represent more local climate characteristics than dynamical downscaling and GCM outputs in both temperature and precipitation.