Impreciseness in land-cover classification: Its determination, representation and application

Abstract

Two types of procedure are proposed to handle land-cover information at subpixel levels. In both cases, each pixel is allowed to have more than one land-cover type. A membership value is associated with each land-cover type, and all the possible membership values for a pixel are preserved for further analyses. These analyses include the evaluation of classification procedures, land-cover area estimation, and the extraction of additional spatial information. SPOT HRV data obtained over the rural-urban fringe of metropolitan Toronto, Canada, are used in the analysis. Experiments indicate that methods proposed in this study permit more detailed and accurate analyses than can be achieved by conventional procedures for thematic mapping.