Estimation of the spatial distribution of urban population using high spatial resolution satellite imagery

Research and applications in geography often resort to census data for information on population distribution. The arbitrary nature of the areal-unit partitioning associated with census data presents many problems when using the raw census data for geographic research. This dissertation presents an areal-interpolation method which allocates the total population of a census enumeration unit to the land-use zones within it using high-resolution satellite imagery and geostatistics. The method simulates the population density of a land-use zone using two components: an expected population density determined by land use and image texture, and a residual population density interpolated by cokriging.; Compared to previous studies of using remote sensing to perform dasymetric mapping, this research differs in a few aspects. First, previous studies use the census enumeration unit or image pixel as the elementary spatial unit. In contrast, the areal unit in this research is the homogenous urban patch (HUP) which has clear meaning in terms of land use. Secondly, although population density is a continuous variable, previous dasymetric mapping typically treats it as a categorical variable. This research recognizes its continuity and is thus semantically sound. Thirdly, few areal-interpolation methods have the so-called volume-preserving property. This research showed that cokriging has this property and demonstrated it with a real world dataset.; The contributions of the project are methodological and applied. From the methodological perspective, the study of different measurements of image texture and their utility to differentiate land uses provide a potentially new avenue to analyze high-spatial-resolution satellite imagery. The cokriging method and the cross-validation method to assess its accuracy also provide a new pycnophylactic method for areal interpolation. From an empirical viewpoint, the dasymetric map created from this study provides a more accurate description of population distribution than the raw census data and can potentially be used for applications in urban planning and management. Combining the two aspects together, the research illustrates the new opportunities brought by fine-resolution satellite imagery to urban application, which would not be possible with previous remotely sensed data.