基于SOM的海表温度遥感数据集的EOF算法重构

EOF Method to Reconstruct SST Remote Sensing Dataset Based on SOM

针对海表温度数据集的数据缺失,提出了一种基于自组织映射算法(SOM)和经验正交函数算法(EOF)有机结合的重构缺失值的新方法。该方法应用了SOM的非线性估计,能够很好的反映数据集的非线性结构,并把SOM估计的结果用于EOF算法的初始化,克服了EOF对数据集初始化敏感的问题。在处理过程中,对奇异值分解使用了lanczos算子分解矩阵,提高了程序运行效率。此外,该方法还引入蒙特卡罗交叉校正集,确定最佳重构的EOF模态数,最终高精度计算出重构误差。使用AQUA遥感卫星海表温度数据进行实验,结果表明该方法能够很好地重构出缺失率高达83.23%的数据集,且重构精度高。

For the missing data of SST dataset, a method based on SOM and EOF algorithms to reconstruct the missing data was proposed, which could use the nonlinear estimation of SOM to reflect the nonlinear structure of dataset and could initialize the input of EOF algorithm utilizing the result of SOM to get rid of the sensitivity problem of EOF initialization to dataset. In the process of SVD, Lanczos operator was used to decompose matrix so as to enhance the efficiency of procedure. Monte-Carlo cross validation was introduced to assure the optimal mode EOF number and calculate the error of reconstruction. After using AQUA satellite remote sensing SST data, the result shows that this method can reconstruct the dataset with missing rate of 83.23% and have a high reconstruction precision.