显著经验正交函数分析及其在淮河流域暴雨研究中的应用

经验正交函数分解（EOF）是气候特征研究中常用的分析方法，但由于方法本身的原因，EOF 主要模态不一定都能有效揭示资料场包含的气候模态。利用中国基本站和基准站1950—2009年逐日降水资料，运用显著经验正交函数分解（Disˉ tinct EOF，DEOF）方法研究了淮河流域暴雨的统计特征。结果表明 DEOF 第1模态呈现了淮河流域暴雨量在南北方向上存在相反的变化，即流域中部、南部偏多（偏少）时，北部则偏少（偏多），第1主成分具有显著的16—17 a 周期性变化，表明流域南北的旱涝变化存在年代际振荡；第2模态表现了淮河流域中部暴雨量的异常变化，第2主成分有明显的线性趋势，说明近50年来流域中部地区暴雨量有明显的上升趋势，并且在1990年前后由偏少转为偏多。对比 DEOF 和 EOF 的分析结果，发现DEOF 能排除资料场中与随机扩散模型相关性较高的空间特征，能抓住与随机扩散模型有显著差异的分布特征并凸出显示出来，能从较强的背景噪声中凸出物理信号，因而能更好地估计真实的气候模态。

DEOF analysis and its application to the research on the rainstorms in the Huaihe River Basin

The empirical orthogonal function (EOF) is a commonly analytical tool in climate study. But because of the constraints of the method itself, not all the leading EOF modes can reveal the true climate mode from the climatological data in some cases. Based on the daily precipitation datasets from the basic stations over the Huaihe River Basin from 1961 to 2009, the climatological statistical characteristics of the rainstorms in the Huaihe River Basin are studied by using the Distinct EOF (DEOF) method. The results show that DEOF-1 displays contrary changes in the south-north direction on the rainstorm precipitation in the Huaihe River Basin, which means that when the rainstorm precipitation of the central and southern region is more (less) than normal state, the northern region is less (more) than normal state. The first principal component has obvious periodic oscillations of 16-17 years, suggesting that the drought and flood in southern and northern region show a decadal oscillation; DEOF-2 displays the abnormal changes of the rainstorms in the central region of the Huaihe River Basin, and the second principal component has a obvious linear trend showing a upward trend in the last 50 years and a convertion from less to more than normal at about 1990. Comparing with the EOF analysis, DEOF can effectively exclude the spatial characteristics related hiqhly to the stochastic diffusion model, catch the features having significant differences with the model and display the features more prominently, and detect physical signals from a strong background noise, and therefore it should be a better estimate for the true climatic mode.