An evaluation of fuzzy approaches to mapping land cover from aerial photographs

Fuzzy approaches are of increasing importance for representing and handling geographical area and boundary delineations. This paper evaluates three different approaches that may be applied to derive fuzzy maps of land cover obtained by photogrammetry from aerial photographs. The three approaches are all spatial interpolation methods, based respectively on distances, on triangulated irregular network (TIN) models and on indicator kriging. For each approach, a set of methods for assessing the accuracies of fuzzy maps has been employed, including the overall classification accuracy, entropy, cross-entropy and divergence. A local Edinburgh suburb was used as a test site where there is a mixture of well-defined and poorly defined locations. It was found that while the TIN model-based approach produces the best empirical results in terms of the overall classification accuracy, entropy and cross-entropy, indicator kriging is theoretically the most sound approach to deriving fuzzy maps of land cover from photogrammetric data, especially when fuzziness is properly accommodated in the assumed reference data.