Abstract:
Discusses the interpretation of LANDSAT-1 multispectral scanner data obtained over the semi-arid farming district of Verlorenvlei on the west coast of South Africa. An unsupervised iterative clustering algorithm was employed to produce a classified image containing twelve land -based classes. Ground-truth data were collected using standard botanical and soil survey techniques. All field data were subjected to similarity analysis using the unrelativised CZEKANOWSKI coefficient. In this way data were rendered suitable for correlation with spectral classes. The area studied has little topographic relief but is spectrally and spatially complex due to the fragmentary distribution of varying soil and vegetation types. As a result of this complexity, although the interpretation technique employed accurately described the land-cover in terms of the spectral classes defined on the LANDSAT image, most classes could not be translated into discrete cover types.