百分位统计降尺度方法及在GCMs日降水订正中的应用

在格点观测的逐日降水量数据基础上,采用百分位统计降尺度方法对全球气候模式(GCM)输出的日降水量进行了订正处理。5种订正方案的比较结果表明,取12个百分位数进行日降水量订正是合理的。观测资料与3个GCMs订正前后全国平均年、季降水量空间分布以及主要流域平均年、月和日降水序列多年平均、变化趋势及概率密度的对比分析表明:(1)统计降尺度处理可在一定程度上降低GCMs模拟的降水量偏差,特别是中国中部、长江以南和东北部分地区,对德国马普研究所的海气耦合模式(MPI/ECHAM5)模拟的降水量订正效果最显著;(2)GCMs统计降尺度处理的降水量季节分布特征与观测更为接近,所有流域MPI/ECHAM5订正的降水量优于或接近直接输出结果;(3)与GCM直接输出的降水相比,部分流域经统计降尺度处理后降水量变化趋势与观测的一致性有所增加,但不明显;(4)当日降水量<30mm时,订正的降水量与观测的偏差明显减小;当日降水量>30mm时,部分流域由负偏差转为正偏差。由于GCMs结构和降尺度方法的局限性,在用于具体流域未来气候变化预估及气候变化影响评估时,应选择尽可能多的、模拟能力强的GCMs数据,以包含尽可能多的模拟气候情景。

Percentile Statistical Downscaling Method and Its Application in the Correction of GCMs Daily Precipitation in China

The bias of simulated rainfall by GCMs is corrected using daily percentile scaling method.This method relates each percentile of GCMs daily rainfall to the observed ’natural’ rainfall of the same percentile.Using this relationship,the projected daily rainfall by GCMs for future climate periods is converted to the ’natural’ rainfall in the same periods.It is shown that twelve percentiles is optimal in the terms of bias magnitude and time consuming compared with other four schemes.By analyzing multi-year averaged annual,monthly rainfall and their temporal tendency,and the probability distribution of daily rainfall for observational,simulated and corrected data,the following conclusions are drawn:(1) The bias of simulated rainfall by GCMs can be reduced in some degree over China,especially over the south of the Yangtze River,central China and parts of Northeast China.The most significant reduction of the biases among the three GCMs analyzed is for MPI-EHCAM5.(2) The difference of seasonal cycle pattern between observational and corrected data is smaller than that between observational and simulated data,and this is the case for MPI-ECHAM5 in all of the basins.(3) More corrected data series than the simulated data series have the same trends with observations in some basins.(4) The biases are reduced significantly in most basins when daily rainfall is less than 30 mm.However,the biases become positive from negative in some basins when daily rainfall is greater than 30 mm.If more accurate projections of precipitation change and its impact on water resources are expected,as many models as possible should be used due to the uncertainties and limitations of GCMs simulations and downscaling.