Fuzzy and boolean forest membership: On the actual separability of land cover classes

Forests are among the most important habitats of the Earth for several ecological reasons and their management is a prior task when dealing with landscape conservation. Thematic maps and remote sensing data are powerful tools to be used in landscape planning and forest management; nevertheless, most of the European and Mediterranean forest monitoring and conservation programs do not take into account the continuity of the variation of habitats within the landscape but they only rely on Boolean classification methods. The utilisation of a classification method that applies a continuity criterion is fundamental because it is expected to better represent the ecological gradients within a landscape. The aim of this paper is to assess the amount of classification uncertainty related to crisp (boolean) classes, particularly focusing on forest identification uncertainty. Forest fuzzy membership of the Tuscany region (Italy) derived from a Landsat ETM+ image scene was compared with the widely used crisp datasets in European forests management and conservation practices, i.e. the European JRC Forest/Non:Forest map, the CORINE Land Cover 2000 (levels 1 and 2), as well as the Global Land Cover 2000, in order to qualitatively and quantitatively assess the separability of crisp classes with respect to forest fuzzy membership. A statistically significant (p < 0.001) forest fuzzy membership separability among the considered crisp classes was found. Despite the crisp dataset and hierarchical level taken into account, both forest and non:forest crisp classes showed a high degree of forest fuzzy membership variability. Therefore, given the intrinsic mixture of crisp land cover classes, ecological studies on forestall ecosystems should rigorously take into account the classification uncertainty related to a crisp view of ecological entities which are being mapped.