自组织神经网络模态振幅算法及其在南海海面高度异常年际变化中的应用  

An algorithm for calculating the pattern amplitude extracted from the Self-Organizing Map and its application in the study of interannual variations of the sea surface height anomalies of the South China Sea

作　　者：金宝刚[1] 张韧[1] 王辉赞[1] 范磊 栾毅 

机构地区：[1]解放军理工大学气象学院海洋与空间环境系 [2]解放军61741部队气象中心 [3]解放军96631部队

出　　处：《海洋学研究》2010年第4期76-82,共7页Journal of Marine Sciences

基　　金：国家重点基础研究发展计划资助项目(2007CB816000)

摘　　要：作为一种非监督分类方法,自组织神经网络在气象和海洋学科中的应用越来越广泛,但该方法无法给出要素场典型模态随时间变化的振幅。基于所输入要素场与典型模态场的相似程度能够反映出典型模态在该时刻强弱程度的思想,提出了一种简便易行的自组织神经网络模态振幅算法,并应用于南海海面高度异常研究。根据该算法计算出了1993—2008年南海海面高度异常年际变化空间分布模态相应的振幅,根据振幅时间序列进一步正确计算出了Nio 3指数与这些模态的延迟相关关系,表明本文模态振幅算法有效。As an unsupervised classification method,Self-Organizing Map（SOM） has been increasingly used in climatology and meteorology.But one thing bothering for the method is that it can not give the time varying amplitude of SOM patterns.To solve this problem,an easy but effective pattern amplitude algorithm is proposed based on the fact that pattern amplitude at certain time is correlated with the degree of similarity between the SOM pattern and the corresponding input element field of this time.The amplitude of typical patterns reflects the strength of these patterns,and it is related to two factors.One is the degree of the similarity between the SOM pattern and the input element field,the greater the degree of similarity,the stronger the typical pattern.The other one is the population value of the input element field,bigger element value indicate stronger typical pattern.Taking into account these two factors,the definition of the pattern amplitude at certain time is the correlation coefficient between the input field vector and SOM weight vector multiplied by the ratio of the total absolute value of the two vectors.The above method is applied to the study of the sea surface height anomaly（SSHA） fields of the South China Sea from 1993 to 2008.Four spatial distribution patterns of SSHA interannual variation are extracted by SOM from the monthly anomaly SSHA fields.Pattern 1 is dominated by a cyclonic circulation over the whole basin,and pattern 4 almost mirrors pattern 1;pattern 2 is dominated by a cyclonic circulation in the west of the basin with a higher SSHA along the eastern boundary of the basin,and pattern 3 almost mirrors pattern 2.The time series of the amplitude of the four patterns are calculated with the proposed algorithm.In order to verify the reliability of the results,the pattern type time series is reconfirmed with the pattern of the largest amplitude and compared with those resulted from SOM.Totally 62.5% pattern numbers during the whole time series are consistent with the original pattern numb

关 键 词：自组织神经网络 模态振幅 南海 海面高度异常 

分 类 号：P731.23[天文地球—海洋科学]