Fuzzy methods for categorical mapping with image-based land cover data

This paper presents an approach to capturing and representing the uncertainty inherent in any attempt to classify continuously varying geographical phenomena into discrete categories. This uncertainty is captured during a visual photo-interpretation and a computerised image classification process and encoded as a series of fuzzy surfaces. These store the fuzzy membership values (FMVs) of each location to all candidate classes in a desired classification scheme. These surfaces are used to explore graphically the underlying variations in the level of certainty of assigning candidate classes to individual locations. A technique is presented that analyses these FMV surfaces by applying alpha-cuts (thresholds) to derive a series of traditional categorical maps in the form of vector polygons. The relative certainty of the attribute classification is used to determine an appropriate Epsilon band width around boundary lines separating different land cover classes on the resulting categorical map. The approach is tested on the practical problem of producing categorical maps of land cover for a suburban area. Uncertainty surfaces are derived for land cover classifications created both from photogrammetric interpretation and from satellite image classification. A series of categorical maps of land cover are derived for different minimum levels of certainty in the attribute classification.