A hybrid object-based classification approach for mapping urban sprawl in periurban environment

Abstract

With the projected global increase of urbanization, pressure on land availability for urban use will increase. In areas where urbanization is not controlled, the concentration of human presence in residential and industrial settings may lead to an alteration of ecosystems patterns and processes. More specifically, urban growth points to an increasing need to continuously assess the urban flood hazard from the urban planning, water resource management and ecosystem management perspectives. The general impacts of urbanization on hydrologic processes and flood behaviour were widely studied and remote sensing has been demonstrated as an essential tool in any environmental and socioeconomic studies of urban areas. But, there is still a need to propose an operational method to map urban objects in complex landscape mainly characterized by different types of urbanization. According to these elements, the scope of the study is defined by three main objectives: to assess which data are most appropriate for a given unit of analysis (e.g. catchment vs. county level); to assess the ability of an object-based approach to classify urban objects (e.g. houses, farms, buildings) at multiple spatial scales; to propose an operational methodology for mapping urban sprawl. In order to validate the argument discussed above, and to test the efficiency of the suggested methodology, a case study of the Touch River catchment in southwest Toulouse (France) has been carried out. Urban objects were extracted at regional and local scales from two SPOT-5 images at 5 m and 10 m spatial resolution using an object-based classification method. Results were validated using ground data. The quality of the urban objects classification decreased within the density of urbanization. Only in areas characterized by spread urbanization, post-treatments are required. Nevertheless, the comparison of the results with an urban database realized using a Computer Assisted Photo-Interpretation method shows a better capacity to delineate urban extent at regional scale and to quantify urban objects at local scale. Information on internal distinction within the urban matrix is not provided. Findings of this study can be used for different applications: to produce detailed map required for the production of hazard assessment maps; to produce input data for hydrological models; to form policies on land-use planning and balanced urbanization by the city development, planning and conservation authorities and flood control agencies.