Application of standardized principal component analysis of land-cover characterization using multitemporal AVHRR data

The concept of a vegetation vector has been developed to better visualize and characterize land-cover at regional scales. The vegetation vector is derived from long time-series multitemporal normalized difference vegetation index (NDVI) data sets from the National Oceanic and Atmospheric Administration (NOAA) Advanced Very High Resolution Radiometer (AVHRR) by means of principal component analysis (PCA). The vegetation vector can characterize vegetation cover based upon both the spatial variation of the magnitude of NDVI and the seasonal variation of NDVI. The PCA study showed that the area under analysis must exhibit a variety of dissimilar vegetation communities in terms of density and phenology to successfully derive these two factors. When the PCA was applied to the entire state of Arizona, these two factors were derived as the first two principal components (PCs). However, when the PCA was applied to subset areas extracted from the entire study area by overlaying a vegetation map compiled through bioclimatological and ecological studies, the first two PCs did not always represent these two factors. This indicated that these two factors were not always the major cause of variation in NDVI for some vegetation communities. In this study the vegetation vector was constructed utilizing the first two PCs derived from the entire study area. Analysis of histograms of the direction of the vegetation vector for each community found in the vegetation map could be used to characterize most of the communities in terms of photosynthetic activity and phenology. A profile of the histogram could be interpreted as characteristic of the community. In addition, the range exhibited by the histogram could be used as a measure of the homogeneity/heterogeneity of the community based upon photosynthetic activity and phenology. Graphical projection of the mean vegetation vectors could be used to visualize characteristics and relationships between communities. The position of the plot represents the mean characteristics of the community. The difference in the mean vegetation vector between communities represents the similarity/dissimilarity of the characteristic between communities. These techniques represent a simple means of visualizing many vegetation communities and should facilitate characterizing land cover at global scales.