Potential of hyperspectral remote sensing for characterisation of urban structure in Munich

Abstract

Urban planning requires up-to date spatial information of various aspects of the city. Most of this information is collected by digitising from aerial imagery or time consuming field surveys. (Semi-) automatic multi sensor remote sensing provides advantages for area-wide and fast data collection. For remote sensing of urban areas, high spatial and spectral resolution is needed. Therefore hy-perspectral HyMap data with a resolution of 4 m is selected. Urban structure is identified as an important indicator for urban planning. The urban structures are characterised by urban objects, which can be identified by surface material. The objective of this study is to evaluate the potential of the selected sensors for automatic mapping of urban structures for a test area in the Munich region. Urban structures are identified per building block. The surface material is derived from the hyper-spectral image using an unmixing approach. This is carried out with and without the support of a building mask. Many urban objects that are characteristic for urban structures can be recognised in the identified surface material map. Problems occur in case of small buildings and streets, because trees are overshadowing them. Using a building mask is an advantage for reducing the confusion between roof and ground materials, but it also causes an underestimation of the building area, due to inac-curate co-registration. Further research will focus on first, the improvement of surface material identification using additional thermal information and second, on automating urban structure mapping, assisted with additional information of a digital surface model.