ABSTRACT
Development of operational applications of remote sensing demands results that satisfy certain quality criteria. Choice of spectral and geometric resolutions must be made according to the type and phenomena to be detected. Date of acquisition must be selected so as to facilitate ready detection either because the phenomena are directly observable or because those not directly observable can be interpreted using simple relationships. Choice of visual interpretation instead of computer interpretation also depends on the quantum of data to be processed, the precision expected for the final map and the time and financial resources available. The desired level of precision should also be defined. In a number of cases, for example classification of regional units, it is not strictly necessary to classify the data with full resolution; a sampling of every 2 or 3 pixels is better. The data may have statistical parameters corresponding to none of the statistical hypotheses of the classification methods.
