An Object-based Approach to Quantity and Quality Assessment of Heathland Habitats in the Framework of Natura 2000 Using Hyperspectral Airborne Ahs Images

Straightforward mapping of detailed heathland habitat patches and their quality using remote sensing is hampered by (1) the intrinsic property of a high heterogeneity in habitat species composition (i.e. high intra-variability), and (2) the occurrence of the same species in multiple habitat types (i.e. low inter-variability). Mapping accuracy of detailed habitat objects can however be improved by using an advanced approach that specifically takes into account and exploits these inherent patch characteristics. To demonstrate the idea, we developed and applied a multi-step mapping framework on a protected semi-natural heathland area in the north of Belgium. The method consecutively consists of (1) a 4-level hierarchical land cover classification of hyperspectral airborne AHS image data, and (2) a kernel-based structural re-classification algorithm in combination with habitat patch object composition definitions. Detailed land cover composition data were collected in 1325 field plots. Multi-variate analysis (Ward’s clustering; TWINSPAN) of these data led to the design of meaningful land cover classes in a dedicated classification scheme. Subsequently, the data were used as reference for the classification of hyperspectral AHS image data. Linear Discriminant Analysis in combination with Sequential-FloatingForward-Selection (SFFS-LDA) was applied to classify the hyperspectral images. Classification accuracies of these maps are in the order of 74-93% (Kappa= 0.81-0.92) depending on the classification detail. To subsequently obtain habitat patch (object) maps, the land cover classifications were used as input for a kernel-based spatial re-classification process, in combination with a rule-set that relates specific Natura 2000 habitats with a composition range of the land cover classes. The resulting habitat patch maps illustrate the methodology’s potential for detailed heathland habitat characterization using hyperspectral image data, and hence contribute to the improved mapping and understanding of heathland habitat, essential for the EU member states reporting obligations under the Habitats Directive.