Abstract: Detailed residential area (ResA) distribution data have significances to seismic risk analysis. In order to obtain more detailed ResA data with higher timeliness, and give their full play in indicating the spatial distribution of population and buildings, we integrated the advantages of multi-source RS images, and carried out the extraction and re-classification research of urban ResA based on hierarchical classification method. With Qinzhou District, Tianshui, Gansu Province of China as study area, we used new Landsat-8 OLI image to classify urban land-use and recognize the holistic ResA through building a decision tree classification model. Inside the urban ResA, we further used domestic ZY3 satellite image to re-classify the land-use and buildings based on object-oriented method. Finally the ResA data with different levels of detail were obtained. The overall land-use classification precision of Landsat-8 image was about 92% (the recognition rate of ResA was 86%), and the re-classification precision of buildings inside urban ResA is 81% (only consumed half of the time). It indicated that based on multi-source RS images, the hierarchical classification method is available to extract and re-classify urban ResA.

Key words: Residential area extraction, Earthquake risk analysis, Hierarchical classification, Landsat-8