DERIVATION AND APPLICATIONS OF PROBABILISTIC MEASURES OF CLASS MEMBERSHIP FROM THE MAXIMUM-LIKELIHOOD CLASSIFICATION

The maximum-likelihood classification of remotely sensed data involves considerable computational effort, in the process calculating a large amount of information on the class membership characteristics for each case (e.g., pixel). Little of this information, however, is made available in the conventional output, which consists simply of the most likely class of membership for each case. More of the information generated in the classification can be output, specifically the a posteriori probabilities and typicalities of class membership. Each of these measures conveys different information on the class membership characteristics of a case. They may therefore be used to improve substantially the value of the classification. They also indicate the quality of the classifier's allocations on a per-case basis, a valuable supplement to the classification accuracy statement. Two case studies are discussed which show examples of how the probability measures can enhance the value of a classification. First, for a classification of agricultural crops where the probability measures can be used to identify potentially misclassified cases. By directing ground surveys to these areas and, if necessary, modifying the class allocations accordingly, the classification accuracy can be increased. The second case study focuses on a considerably different environment, heathland. In this case some of the semi-natural heathland vegetation classes do not exist in relatively discrete classes as do the crops but instead lie along continua. The probability measures here were related to the composition of the vegetation canopy and could be used to model more realistically the manner in which these vegetation classes inter-graded than would be apparent from the conventional maximum-likelihood classification ou ut.