Application of hyperspectral DAIS data for differentiation of urban surfaces in the city of Dresden, Germany

Hyperspectral data of the Digital Airborne Imaging Spectrometer (DAIS 7915) were analyzed for the 72 reflective bands in order to develop an image processing approach for differentiating urban surface cover types within a 10 x 4.5 km study area in the city of Dresden, Germany. The definition of ecologically relevant surface cover types was based on categories used in urban biotope mapping. Besides the main classification in vegetated and non-vegetated areas, further differentiation was oriented towards assessment of the degree of artificial surface sealing and the ecological characterization of main plant community types. For that purpose a new classification approach was developed which explicitly accounts for spectral mixing effects as a widespread phenomenon in the image data which includes small-sized urban structures of high spectral variability. The approach starts with statistical image classification followed by a region growing procedure which is based on iterative linear spectral unmixing. Visual inspections of the results showed that sensible endmember combinations were identified during unmixing. The result is a classification of urban surface cover types at a sub-pixel level which form the basis for a statistical analysis of the average percent area covered by each class within the ecological base units which were defined as the spatial reference of this study.