期刊名称：ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences
印刷版ISSN：2194-9042
电子版ISSN：2194-9050
出版年度：2010
卷号：XXXVIII - 4/C7
出版社：Copernicus Publications
摘要：Monitoring and reporting on the status of Natura 2000 habitats is an obligation under the 1992 Habitats Directive for each member state of the European Union (EU). Satellite imagery providing up-to-date information for a large areal coverage could be an interesting source to complement conventional, but laborious, field-driven surveying methods. Quality of habitats can be assessed through their structures as represented by combinations of various life forms. If these life forms can be classified by remote sensing, then structural analysis can be applied and readily translated into useful information for conservation status assessment. Previous experiences have shown that hyperspectral imagery is more effective for detailed vegetation classification than multi-spectral images. A limitation of spaceborne hyperspectral imagery, however, is that the resolution is too coarse. In this study, we investigated the use of superresolution (SR) enhanced CHRIS/Proba imagery for the derivation of a habitat map. To obtain the final habitat map, two strategies were compared. The first strategy consists of a direct classification of the habitats (objects) from the imagery. The second strategy is an indirect classification approach, consisting of two steps. First, a detailed classification of twenty-four vegetation types was performed, while in the second step, the obtained vegetation patches were subsequently merged into habitat patches using predefined rules. Kalmthoutse Heide, a nature reserve in the North of Belgium dominated by heathland, was used as a study area. The tree-based ensemble classifier Random Forest was used for the classification, with its internal unbiased Out-Of-Bag estimation as a measure of accuracy assessment. While both strategies achieve around 62% overall accuracy, the area distributions of various habitats show large differences. Visual interpretation confirms that the indirect classification approach, which aggregates detailed vegetation patches into habitat patches, better reflects a field mapping approach. A method to combine the strengths of these two strategies could provide more valuable results for Natura 2000 habitat identification
