Human settlement characterization at the global scale by fusing SAR and multispectral data sets at multiple resolutions

Nowadays more than half or the Earth’s population is living in cities and the urbanization trend is going to be continued. Human settlements are in the focus of various research areas, including Remote Sensing and Earth Observation. The constant change of urban areas has a huge impact on the environment and on global climate. Therefore, visualization and quantification of urban extents and their changes are needed in order to guarantee a sustainable development. Remote Sensing offers a powerful tool to generate such information in a convenient, fast and cost- effective way. The quality of EO data, in terms of spatial, spectral, radiometric and temporal resolution, is improving extremely fast and therefore new opportunities to develop innovative methodologies to extract timely and accurate information about the Earth’s surface and its change evolve. SAR and optical data sets have proven their usefulness for urban area mapping. However, in order to be able to exploit the complementary information from these sensors, data fusion is required. This thesis is devoted to the development, application and evaluation of novel data fusion approaches, aiming at the generation of detailed urban extent maps through integration of multisensor, multi-scale, and multi-temporal EO data. The first part of this thesis deals with multi-scale SAR data fusion. A set of novel data fusion approaches for SAR data of different spatial resolution are introduced, discussed and evaluated. This task is performed in a comparative study of four test sites, each one of them a different Megacity (i.e., a city with more than 10 million inhabitants). The developed fusion methods work at the pixel-, feature-, and decision-level and are intended to cover a variety of possible data availability scenarios. Accordingly, a general framework for multi-resolution SAR data fusion for urban area extraction has been formalized and is presented. The possibility of fusing SAR and optical data is investigated in the second part. Aiming at the incorporation of complementary information from active as well as passive sensors, two decision level fusion methods using pixel-, and object- based techniques are presented. Their accuracy is quantitatively assessed against manually extracted reference data sets for several urban areas around the world. The research conducted in this thesis aimed at a better understanding of the integration of multiscale EO data for the purpose of urban area mapping on the global scale. The possibilities to fuse EO data sets are manifold and the main challenge is to take advantage of each input data set while minimizing their disturbing effects. The proposed multi-scale and multi-source data fusion techniques offer options for a variety of data availability scenarios and are well suited to be expanded in the future.