Landsat-5 Thematic Mapper (TM) data of a subscene of Washington, D. C., were used to distinguish natural and man-made surfaces through analysis of their spectral response. Such information is needed to create and update geographical data bases at the national, regional and local levels. A quantitative measure of the separability and thus the ability to separate various surface features was calculated using the quantity known as divergence (D), a measure of the dissimilarity between two distributions. A general grouping of surface classes (nine) and a more specific set of classes (21), derived from the nine general classes, were studied. Discrimination between class pairs which were derived from a common class (such as, grass, mixed woods, pine woods, deciduous woods from the vegetation class) were surprisingly good. The asphalt and residential classes were the most difficult to separate/discriminate. The thermal band, TM band 6 (10#xB7;4#x2212;12#xB7;5#x3BC;m) was found to improve the separability when replacing a visible band, but decreased the separability when included with the middle-infrared bands.
展开▼
