ABSTRACT
The incorporation of vegetation classifications derived from the analysis of remotely sensed data into geographic information system (GIS) structures has become common practice. However, the usefulness of classifications derived from satellite remotely sensed data as an input may be questioned for those applications concerned with environments that display gradual change (Jupp and Mayo, 1982; Frank, 1984; Allum and Dreisinger, 1987; Griffiths et al., 1988). This is because classification results in elements being ‘true’ or ‘false’ for a class, whereas their membership is frequently uncertain between classes. Furthermore, the derivation of landcover classes from the analysis of satellite remotely sensed data by frequently used maximum likelihood methods does not provide all of the information generated during the classification process (Trodd et al., 1989). This is because relative likelihoods (posterior probabilities of class membership) are replaced by the code of the class with the highest ranked likelihood as the final output of the classification.
