Genauigkeitsevaluierung und Optimierung der Methoden zur objektbasierten Extraktion von begrünten Dachflächen in Berlin basierend auf sehr hochauflösenden UltraCamX Daten

Kurzfassung

An increase of sealing areas caused by urban pressure leads to several environmental issues such as urban heat islands, floods and air pollution. One approach to restore the ecological balance is the establishment of green areas. Due to space constraints setting up green roofs is a practical solution. In order to manage the establishment of new green roofs appropriately it is necessary to monitor their distribution in urban areas. However, monitoring of these roofs is hampered by the lack of solid methods to acquire their spatial positioning. The research object of this thesis is to optimize existing methods of object based image analysis to detect green roofs that were developed within a former project of the German Aerospace Center. The study areas are two districts in the center of Berlin, Mitte and Friedrichshain-Kreuzberg, of which very high resolution data in terms of True Orthophotos (red, green, blue, near-infrared, count-mask, normalized digital elevation model) as well as the Automated Real Estate Map are available. Reference data is acquired to evaluate the performance of the methods of the former project and the optimized methods. To enhance the object based image analysis, which is an iterative procedure of segmentation and classification steps, object attributes like layer values, extent, shape, relations to direct neighboring segments and surrounding objects are involved. The extraction steps are organized in a Rule Set. The training data set is the district Mitte. The optimization focuses mainly on a better detection of sparsely green areas, the homogeneity of shadow areas and the difficulty to exclude tree canopies above buildings from green roofs. Of special relevance are also areas with similar reflectance behavior as green roofs, vegetation on buildings that are not part of the Automated Real Estate Map and small vegetated areas. Additionally, green roofs are differentiated according to their type. Extensive green roofs are usually not maintained and mostly covered by low vegetation. In contrary, intensive green roofs, also referred to as roof gardens, are similar to parks at ground level with trees, shrubs, lawns and paths. The most important steps to optimize the extraction are refining the segmentation, integrating a slope layer calculated from the normalized digital elevation model, involving an additional vegetation index and declassifying pot plants. Differentiating between extensive and intensive areas supports the accuracy of green roofs as well. The results of this thesis show that the extraction accuracy of the proposed methods is greatly increased compared to existing methods. The attribution in extensive and intensive green roofs is also very accurate. The transferability of the methods is proven based on the district Friedrichshain-Kreuzberg.