Agricultural land-cover discrimination using thematic mapper spectral bands

Abstract Multispectral scanner system data simulating the thematic mapper (TM) of LANDSAT-4 were analysed for an area near Gedney Hill, Lincolnshire, U.K. The data were found to have a three-dimensional statistical structure similar to that for the LANDSAT-4 TM of parts of the United States. Divergence analysis indicates that the optimal choice of bands for cover discrimination should include one band from the visible, one from the near-IR (infrared) and one from the middle- or far-IR. It was further shown, primarily from consideration of principal-component images, that significant discriminatory power may be lost if all bands are not used. Comparisons with LANDSAT-4 TM principal-component images are made. The role of noise factors in obscuring information, especially from highly correlated bands, is shown to be of considerable importance.