A Gis Approach to Population Estimation in a Complex Urban Environment Using Spot Multispectral Images

Population estimation was carried out using high-resolution (20 metres) multispectral satellite data (SPOT) in the Kowloon metropolitan area of Hong Kong, which is characterised by high population density and multipurpose high-rise buildings. The smallest areal unit for census data Tertiary Planning Unit (TPU) -was used to aggregate population. With the aid of a raster GIS (IDRISI) and a vector GIS (pcARC/INFO), the satellite image data were registered with the TPU boundaries. A supervised image classification was carried out to extract high-density and lowdensity residential uses. Spectral radiance values of each band for each TPU were also extracted. Based on a sample of 12 representative TPUs, three-variable and one-variable linear regression models were developed, linking spectral radiance values with residential population densities. Number of living quarters estimation was also carried out in the same way. Residential land use area extraction from classified image was also employed to estimate population at the TPU level. The results showed over-estimation of population in all cases, but corrections to these estimates based on the proportion of high-density residential use in TPU, could be applied to those predominantly non-residential TPUs to improve the results of the estimates. It was concluded that the linear regression modelling and residential land use extraction could be efficiently combined under a raster GIS to obtain population estimates from high-resolution satellite image data at the very fine areal unit level. Further improvement of the accuracy of the population estimates requires research on methods of image enhancement, image classification, and the development of a knowledge-based expert system for a more realistic determination of population density for each census unit.