The utility of digital Thematic Mapper data for natural resources classification

Abstract

Abstract Landsat Thematic Mapper data, collected over central Michigan in the U.S.A., in October 1982, were digitally analysed to determine qualitatively and quantitatively their utility and potential to classify nine natural resources categories (e.g. red pine, jack pine, scotch pine, low conifers, hardwoods, grassland, water, wetland and other). Supervised classification with a maximum likelihood decision rule was employed to 22 especially selected single-, two-, three-, four- and six-band combinations (thermal band was excluded). Analysis of the six-band combination indicated an overall classification accuracy of 92.4 per cent. The producer's classification accuracy of individual categories was 79.7 per cent (scotch pine), 80.7 per cent (lowland conifers), 80.8 per cent (red pine), 88.7 per cent (jack pine), 92.8 per cent (wetland), 96.3 per cent (grassland), 96.6 per cent (hardwoods), 78.5 per cent (other) and 1000 per cent (water). After aggregation of pine categories the accuracy of the new category...