A new approach to the land cover and habitat classification in support of biodiversity monitoring.
Research undertaken through the EU FP7-funded Biodiversity Multi-Source MOnitoring System: From Space To Species (BIO_SOS) project has led to the development of the Earth Observation Data for Habitat Monitoring System (EODHaM). This system uses multi-source remote sensing data to first classify landscapes according to the Food and Agricultural Organisation (FAO) Land Cover Classification Scheme (LCCS) scheme. Classification of the broad LCCS categories of terrestrial, aquatic, vegetated and non-vegetated is achieved within the EODHaM 1st Stage, with this based on spectral data and derived indices alone. The EODHaM 2nd Stage integrates both spectral and contextual features and focuses first on the classification of LCCS Level 3, which establishes whether the landscape is either cultivated, managed or artificial or natural or semi-natural. Classes beyond Level 3 are then formed by first generating over 30 separate ‘layers’ relating to key descriptors (e.g., vegetation life form, cover and height, water characteristics) within the LCCS classification and assigning codes to each of these. These codes are then combined to generate a summary class code and an associated descriptor. In the EODHaM 3rd stage, the LCCS classes are translated to General Habitat Categories (GHCs) which can then be combined to map Annex I habitats. Change is detected on the basis of a change in the codes (e.g., height) used to form a class (e.g., tall broadleaved deciduous forests) and also in continuous surfaces relating to biophysical attributes such as vegetation productivity, surface moisture and amount of woody material. Methods for assessing classification accuracy have also been developed with these combining accuracies in the classification of the component layers used to generate the LCCS classes. The method was developed on Natura 2000 sites and their surrounds in Wales, the Netherlands and Italy and is being evaluated at other locations in Greece, Portugal, India and Brazil. The EODHaM System is implemented within open source software and uses a diversity of remote sensing and also ancillary datasets. The System can also be readily adopted by managers of conservation sites and their surrounds for land cover, habitat and ultimately biodiversity monitoring. Consistent classifications can also be generated within and between sites.