Multispectral scanner data, obtained over Marion County (Indianapolis), Indiana at an altitude of 915 kilometers, were analyzed by computer-implemented techniques to evaluate the utility of satellite data for urban land use classification. Several land use classes, such as commerce/industry, single-family (newer) residential, trees, and water exhibited spectrally separable characteristics and were identified with greater than 90 per cent accuracy. Difficulties were encountered in the spectral separation of grassy (open, agricultural) areas and multi-family (older) housing. The confusion between these two classes was largely eliminated, however when spectral characteristics of samples (instead of individual data points) were considered. Another solution to the problem consisted of spatially dividing the data into urban and rural land uses prior to classification. Over 95 per cent accuracy of recognition may be achieved by this "pre-processing" step in an analysis.
Date of this Version