Introducing fuzzy set theory to evaluate risk of misclassification of land cover maps to land mapping applications: Testing on coastal watersheds

• Published in: Ocean & coastal management Vol. 184; p. 104903
• Main Authors: Teixeira, Zara, Roberto de Oliveira Vital, Saulo, Vendel, Ana Lúcia, Lourenço de Mendonça, Juan Diego, Patrício, Joana
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01-02-2020
• Subjects: Accuracy assessment; Brazil; Classical set theory; Classification; Error relevance; Fuzzy set theory; Paraíba state; Uncertainty; Paraíba state; Error relevance; Uncertainty; Classical set theory; Classification; Brazil; Accuracy assessment; Fuzzy set theory
• ISSN: 0964-5691; 1873-524X
• DOI: 10.1016/j.ocecoaman.2019.104903

In land cover maps, categories represent a continuum of variation and for this reason, fuzzy set theory, which accepts degrees of membership, has been suggested for land classification. Nevertheless, classical set theory, which only assumes single map categories, is still widely used. The purpose of this study is to develop a methodology to reduce the weakness of land cover maps in which classical theory has been applied. To do so, we propose adding an error relevance step after accuracy assessment, which evaluates how relevant are the classification errors to selected land applications. First, a membership matrix is built based on a linguistic scale associated to land cover rates obtained from literature. Then, two fuzzy measures are calculated and the frequency of categories, that do not pose a problem to the user in light of the land application, is determined. The methodology is demonstrated using two Brazilian tropical coastal regions and two land applications relevant for coastal watershed management. The study presents land cover maps of the Mamanguape and the Paraíba estuarine regions, their full accuracy assessment, and the relevance of the classification errors to the land applications.
The accuracy assessment step has demonstrated that the land cover maps are reliable. The error relevance step has shown that the map weakness can be reduced. Both steps show that the land cover maps produced are suitable for further land mapping applications. The results on land cover composition point to the importance of future work focused on the environmental sustainability of the studied regions. The new procedure has proven useful to decrease the degree of distrust with which land cover maps are regarded. The framework provided is suitable for virtually any land mapping application.
•We propose adding the error relevance step, based on fuzzy set theory, to classification processes.•The new methodology accepts errors that do not pose problems to the user.•The framework is suitable for virtually any land mapping application.