Conventional and Fuzzy Accuracy Assessment of Land Cover Maps at Regional Scale

摘要

The accuracy of a regional-scale thematic map of land cover was assessed using conventional and fuzzy set methods. The mapping process integrated Landsat Thematic Mapper (TM) data with expert knowledge to map land cover for a 12 million hectare area. Accuracy assessment was based on a stratified random sample of 113 plots. Paired observed and predicted land cover types for 9,745 sample points for conventional accuracy and 933 sample points for fuzzy accuracy were collected. Conventional map accuracies were 42%, 56%, and 74% for the Super-alliance, Subclass, and Revised Subclass maps, respectively. Fuzzy-based map accuracies were assessed at the Subclass and Super-alliance levels resulting in an improvement in map accuracy of 19% and 23%, respectively. The nature, magnitude, and frequency of errors associated with mapping land cover types at Subclass level are reported.