Rule-based generalization of satellite-derived raster thematic maps

Thematic maps are increasingly derived from satellite imagery using automatic classification methods. However, whereas most classification algorithms operate at the pixel level, map users usually are interested in information at the landscape unit with minimum mapping unit of several hectares. Consequently, most land cover mapping operational programs with satellite images (e.g. CORINE Land Cover) are still based on visual analysis. In this paper, we present a methodology developed by the Remote Sensing Group of the Portuguese Geographic Institute (IGP) to automatically generalize satellite-derived land cover maps using priority rules. Operations include exaggeration, merging, aggregation and amalgamation. The initial map may have any number of information classes, with the user assigning the priority of each class according to the purpose of the final map within a table. This priority table is used to define a cost-distance map for the aggregation of different classes. Generalization is accomplished by integrating existing spatial analysis functions using map algebra. The methodology was implemented in a well known commercial GIS software and allows the production of land cover maps at different minimum mapping units by operating on a raster map resulting from per pixel classification of satellite imagery. The methodology was successfully tested in a study area in Central Portugal.