In the STEREO-II project HABISTAT (2007-2011), a methodological classification framework for hyperspectral imagery was developed and succesfully applied to reporting the conservation status of NATURA 2000 heathland habitat types. However, as is often the case with remote sensing thematic mapping, the success of such a methodology stands and falls with the availability of a rather large amount of field reference data, in order to ensure adequate training of the supervised classifier. Bearing in mind the 6-yearly recurring reporting obligations for protected areas that are part of the NATURA 2000 network (under the EC Habitats Directive (92/43/EEC)), the collection of such an extensive and reliable ground reference data set is not feasible from an operational perspective due to time and cost constraints. As a result, the often praised automation potential of remote sensing methods for monitoring purposes remains unexploited.
It has not been until recently that this issue has attracted a considerable amount of interest and research. The pattern recognition community has made recent research progress in the development of techniques that require considerably less ground reference data by including unlabeled data and/or extra expert knowledge in the classification process. This research falls mainly within the following two machine learning approaches: semi-supervised learning and active learning. Within the Relearn project, the current semi-supervised and active learning techniques are explored and further developed, with the aim of reusing reference data in space and time for vegetation mapping. The potential of the methods to enhance adaptability are tested and demonstrated in the real world example of detailed NATURA 2000 mapping. To do so, the previously developed HABISTAT mapping method is applied, but using only a very limited amount of new field reference data, for the classification of: (1) a newly acquired hyperspectral image of a heathland area that was previously studied (reuse in time); and (2) a new heathland area which has restricted access (reuse in space).