中文摘要：

本研究采用A1B情景下大气环流模式BCCR输出的大气环流预报因子，同时应用两种统计降尺度模型ASD（AutomatedStatisticalDownscalingModel）和SDSM（StatisticalDownscalingModel），预测太湖流域8个站点未来两个时期（2046年-2065年和2081年-2100年）的日最高、最低气温和降水。重点针对两种模型中的预报因子选择、模型率定和验证以及建立的未来气候变化情景结果，对比分析两种模型在太湖流域的适用性。结果表明：在预报因子选择方面，依靠人为主观判断的SDSM模型和提供自动预报因子选择方法的ASD模型选择的预报因子基本一致，因此ASD相对优化了预报因子的选择过程。在率定和验证效果上，两种模型对气温的模拟效果较好，对降水的模拟效果不佳，但是ASD的模拟效果要好于SDSM。在未来两个时期，两种模型模拟的未来最高和最低气温变化并不显著，但从多年平均来看，ASD模拟的气温略有上升，而SDSM模拟的气温略有下降；相较气温，未来降水变化较为显著，ASD模拟的降水量增幅明显高于SDSM模拟的降水增幅。综合考虑模拟的结果并结合他人的研究成果，认为ASD对太湖流域未来气候的模拟优于SDSM。

英文摘要：

Global climate models （GCMs） have been used widely for climate change impact studies; however, the spatial resolution of GCMs is too coarse to resolve regional scale effects and local impacts. Downscaling techniques can improve regional or local estimates of variables from GCM outputs. In this study, two statistical downscaling methods, an automated statistical downscaling （ASD） model and statistical downscaling （SDSM） model were used to downscale daily maximum/minimum temperature and daily precipitation and to construct future climate change scenarios for 2046-2065 and 2081-2100. Outputs from the general circulation model （BCCR） under A1B scenarios and ERA-40 reanalysis data （used to test the models over the calibration period 1961-1990, and variation period 1991-2000） were compared with observed temperature and precipitation data from eight meteorological stations in the Taihu Basin. The results underline limitations to downscaling precipitation and the strength when downscaling air temperature. When selecting predictors, the quality of SDSM results depended mostly on the skill of the user, and the best VIF and bias were obtained by trial and error; however, similar results could be obtained using ASD with ease. During the calibration and validation procedures, both ASD and SDSM methods performed well, but ASD performed better than SDSM. For future precipitation and temperature, ASD and SDSM showed similar patterns, and changes in maximum and minimum temperature were not obvious. On the whole, the simulations for annual averages of ASD rose slightly, while the values of SDSM decreased slightly. The precipitation generated by these two methods varied greatly： the deviation of ASD between future periods and the baseline period （1961-2000） was much smaller than for SDSM. The comparison of SDSM and ASD models indicates that neither could perform well for all stations and seasons. However, considering our simulation results and other studies, we conclude that ASD performs better than SDSM i