Three approaches to computer assisted Landsat multispectral classifications are described. The supervised classification technique enables the analyst to focus on land cover categories of interest. The unsupervised approach uses the statistical properties of the image to identify spectrally pure classes. Guided clustering combines the characteristics of both approaches to develop the maximum number of low variance classes for each land cover category defined.

The application of guided clustering to forest land classification is explained. EDITOR software was used to merge and edit spectral statistics to produce the maximum number of low variance, statistically separate classes. Color-infrared aerial photography was used to assign meaningful forest cover labels to spectral classes of unknown vegetative composition. Classification accuracies were high (91.6% omission, 91.4% commission).

Date of this Version