中文摘要：

采用小波变换,Lanczos时间滤波器,相关分析研究了南海夏季风爆发日期的多尺度特征及与海温场的关系,并研制了南海夏季风爆发日期的短期气候预测新方法——多尺度最优子集回归预测方法。结果表明,1958—2008年南海夏季风爆发的平均日期为5月18日,标准差约10d,具有弱的偏早爆发趋势,存在15.4a的年代际变化和5.3a的年际变化周期。在年际变化尺度上,南海夏季风爆发日期与前期冬季（12—2月）海温场有4个显著的相关区,分别为热带南印度洋,菲律宾以东的热带西北太平洋,赤道中东太平洋,热带东太平洋。在年代际尺度上,南海夏季风爆发日期与前一年（前一年3月至当年2月）海温场的显著相关除有与年际尺度一致的热带西北太平洋海区外,还存在5个显著不同的相关区,即西南印度洋,东南印度洋,北太平洋,热带东南太平洋和东南太平洋。1999—2008年的独立样本检验表明,多尺度最优子集回归预测方法的最大预测误差为8.5d,预测值与实际值相差在5d以内的占50%,且对爆发日期的异常迟早具有较好的预测能力,它明显好于最优子集回归预测与单,多变量的线性回归预测。多尺度最优子集回归预测可以为南海夏季风爆发日期的短期气候预测业务预报提供重要参考。

英文摘要：

Using wavelet transform,Lanczos filter,correlation analysis,Multi-scale characteristics of onset date of South China Sea summer monsoon（SCSSM） and its relationship with sea surface temperature（SST） on interannual and decadal time scale are explored.The optimal subset regression prediction method with the view of multiple time scales is developed as the new method of short-term climate prediction for onset date of SCSSM.The results show that the mean onset date of SCSSM is about 18 May and the mean square deviation is about 10 days during 1958—2008.The SCSSM seems to break out little early trend.It exhibits decadal variation with 15.4 years period and interannual variation with 5.3 years period.On interannual time scale,there are four regions with significant correlation between SSCSM onset dates and SST during previous winter,which occur in tropical southern Indian Ocean,tropical northwestern Pacific east of Philippine,equatorial central and eastern Pacific,and tropical eastern Pacific.On decadal time scale,besides tropical western Pacific,which is consistent with the result on interannual time scale,there are additional five regions with significant correlation between onset dates of SSCSM and SST during previous year（from last March to February）,which occur in southwestern Indian Ocean,southeastern Indian Ocean,north Pacific,tropical southeastern Pacific and southeastern Pacific.The test of independent samples from 1999 to 2008 shows that the maximal prediction error of multi-scale optimal subset regression prediction is 8.5 days.The probability of difference less than 5 days between prediction and actual value is accounted for 50 percent.The anomalous early and late onset date of SSCSM can be better predicted.The multi-scale optimal subset regression prediction is significantly better than optimal subset regression prediction and single/multiple variable linear regression prediction.So,the prediction from multi-scale optimal subset regression prediction can offer important reference to operationa