CLASSIFICATION OF MULTISPECTRAL IMAGES BASED ON FRACTIONS OF ENDMEMBERS - APPLICATION TO LAND-COVER CHANGE IN THE BRAZILIAN AMAZON

Four time-sequential Landsat Thematic Mapper (TM) images of an area of Amazon forest, Pasture, and second growth near Manaus, Brazil were cl assified according to dominant ground cover, using a new technique bas ed on fractions of spectral endmembers. A simple four-endmember model consisting of reflectance spectra of green vegetation, nonphotosynthet ic vegetation, soil, and shade was applied to all four images. Fractio ns of endmembers were used to define seven categories, each of which c onsisted of one or more classes of ground cover, where class names wer e based on field observations. Endmember fractions varied over time fo r many pixels, reflecting processes operating on the ground such as fe lling of forest, or regrowth of vegetation in previously cleared areas . Changes in classes over time were used to establish superclasses whi ch grouped pixels having common histories. sources of classification e rror were evaluated, including system noise, endmember variability, an d low spectral contrast. Field work during each of the four years show ed consistently high accuracy in per-image classification. Classificat ion accuracy in any one year was improved by considering the multiyear context. Although the method was tested in the Amazon basin, the resu lts suggest that endmember classification may be generally useful for comparing multispectral images in space and time.