A Fuzzy Synthetic Evaluation Approach for Land Cover Cartography Accuracy Assessment

The accuracy assessment of land cover maps is traditionally based on reference sample observations randomly selected over the study area. It is assumed that reference sample observations, representing the “real” land cover at Earth’s surface, are free of errors. However, some of these may be erroneous. These errors are sometimes due to an uncertainty in the identification of the most adequate reference land cover classes by visual interpretation of aerial images and/or field work. This uncertainty is caused by landscape fragmentation and/or presence of more than one land cover class in sampled areas. The introduction of uncertainty in thematic accuracy measures remains an issue, but ignoring this uncertainty can significantly influence the land cover maps accuracy reported to end users. In this paper we propose a very simple and understandable method for thematic accuracy assessment of land cover maps that uses reference uncertainty as input feature. This fuzzy synthetic evaluation (FSE) approach is based on the combination of linguistic fuzzy operators. Specifically, we evaluate errors magnitudes per land cover class and weight their importance in map accuracy assessment process. In the sequence, we compare our approach with most traditional accuracy assessment measures and evaluate methodological gains and disadvantages. To achieve this goal we present a case study based on a land cover map of Continental Portugal derived from automatic classification of MERIS images. We demonstrate that trough the use of the fuzzy synthetic evaluation approach we provide accuracy descriptors that are more comprehensible for map users. In fact, this approach allows end users to easier decide if a land cover map satisfies their needs and to become more conscientious about map error extension and its particular impacts.