Application of Principal Component Analysis by Using Different Vegetation Index Derived from Multitemporal AVHRR Data
遥感植被指数对多时相AVHRR数据主成分分析的影响

Based on the 12 months' 1km AVHRR data in China, this paper computes four kinds of vegetation index (VI), that are ratio vegetation index ( RVI ), normalized vegetation index ( NDVI ), soil adjusted vegetation index ( SAVI ) and modified soil adjusted vegetation index ( MSAVI ). Then, we make the same principal components analysis ( PCA ) to them, and find that the PCA transformed first four principal components ( PCA1, PCA2, PCA3, PCA4 ) contribute about 88% cumulative variance, and PCA1 represents VI cumulation of whole year, PCA2 represents VI difference of winter and summer, PCA3 represents VI difference of spring and summer, PCA4 represents VI difference of spring and autumn. In other words, for multitemporal vegetation index of one year, PCA not only compresses the information to the first four principal components, but also extracts the key change information. The PCA1 expresses the basic land cover information, the others extract the seasonal change information of vegetation. However, the outcome of different vegetation index has some differences. As to the cumulative variance of the first four eigenvectors, the biggest is NDVI , 89.28%, the second is SAVI , 88.40%, and the smallest is RVI , only 87.44%. As to the correlation matrix of four vegetation index, SAVI and MASVI are the most similar, NDVI is much similar with the first two vegetation indices, and RVI is the least similar. Although the primary purpose of VI is to indicate the biomass of vegetation, due to the different features of VI, such as different correlation with leaf area index, different sensitivity to vegetation and different anti disturbance of soil and atmosphere, different VI indicates different biomass for the same vegetation, that is, when we use the same PCA to different VI, the result is not uniform.