Object-based Urban Land Cover Mapping Using Multitemporal Multisensor SAR: Preliminary Results
Ban, Yifang; Niu, Xin
KTH Royal Institute of Technology, SWEDEN

In 2008, the world crossed an invisible but momentous milestone - more than half of the people on the planet - roughly 3.2 billon human beings - lived in cities. Between now and 2030, the world is expected to add an additional 2 billion urban dwellers, or 62% of the estimated global population of 8.1 billion will live in cities. Although only a small percentage of global land cover, urban areas significantly alter climate, biogeochemistry, and hydrology at local, regional, and global scales. Cities are hot spots of production, consumption, and waste generation. Recent studies have demonstrated that accurate representation of urban land use is both important and poorly captured in current models. Accurate and timely information on the spatial distribution and the temporal changes of urban areas is therefore critical to a wide array of research questions related to the effect of humans on the local, regional and global environment. The objective of this research is to evaluate Multitemporal, multi-resolution multisensor SAR data for urban land cover classification. ENVISAT ASAR, ALOS PalSAR and TerraSAR-X data were acquired over Shanghai during May to the October, 2008. The major urban land cover classes include high-density builtup areas, low-density builtup areas, major roads, parks, forest, agricultural crops and bare fields. The methodology used in this research involves image preprocessing, segmentation and classification. First, the 1.25m resolution Terra SAR-X data were resampled to 10m pixel-spacing, then segmentation were carried out using compressed and filtered TerraSAR images, since the TerraSAR data have high spatial resolution. Classification were then performed using an object-based SVM classifier. Specific rules that describe the spatial relationships between classes and the object properties of each class were then applied to further improve the classification accuracy. Preliminary results show that X+L-band and C+L-band SAR yielded better classification accuracies than X+C-band SAR. The best classification is achieved using the fusion of X-, C- and L-band SAR.