Monitoring heathland habitat status using hyperspectral image classification and unmixing

Natura 2000, an EU-wide network of nature protection areas, has as main objective the achievement or maintenance of a favorable conservation status of habitats protected by the EU Habitats directives. Within this framework, this study examines a strategy to characterize the status of heathland vegetation from airborne hyperspectral AHS data in the Kalmthoutse Heide, Flanders, Belgium. A hierarchical classification scheme was set-up with the highest detail focusing on vegetation structural elements that determine the conservation status of the habitat. Although conventional classification algorithms performed very well (accuracies > 90%) in discriminating broad land cover classes and habitat types (level 1 to 3), they failed in accurately distinguishing different heather age classes which are an important indicator for the structural quality of the heathland habitat (level 4). Since all heather life stages have their specific structural characteristics, a subpixel unmixing approach succeeded by a decision tree classification was implemented to map variations in heathland morphology and as such enhance the ecological value of information derived from remote sensing data.