Object-based Approach to Map Semi–natural Areas in Mountain Region with High Spatial Resolution Images

Semi–natural areas are characterized by cultivated and natural areas with continuous transition zones. Especially in the alpine regions, this type of land-cover is predominant. The paper presents a method to classify semi –natural alpine areas by combining the spatial detail of orthophotos with the spectral information of SPOT satellite images. As case study, an alpine region in South Tyrol (Italy) has been considered. A four step approach is applied: (1) object delimitation, (2) object information assignment, (3) data mining and (4) classification. First, a segmentation procedure on orthophotos is used for an accurate delimitation of spatial objects, namely the segment. In a second step, spectral information from SPOT data as well as textural information from orthophotos are assigned to the segments. In a third step, those information are inserted in a data mining procedure, reducing the information to the most relevant ones for the next step based on classification. The classification procedure is mainly based on a decision tree approach. The main aim of this study was to compare the results from standard pixel vs. object-based classification approach by combining and considering different attributes (spectral, textures, topography) with decision tree approach and by comparing two classification approaches (maximum likelihood, decision tree). The study shows the potential of mapping semi–natural areas by combining high geometric detail (orthophotos) with spectral information of satellites images (SPOT).