Fine-scale population estimation: how Landsat ETM+ imagery can improve population distribution mapping

The spectral response based and land use based methods were compared for population estimation with Landsat Enhanced Thematic Mapper Plus (ETM+) imagery for Marion County, Indiana, USA. With the spectral response based method, impervious surface and vegetation fractions and land surface temperature derived from the Landsat ETM+ thermal band along with spectral bands were tested for estimating population density at the census block group level. With the land use based method, land use and land cover images were first extracted from Landsat ETM+ reflected bands. The population count for the block groups was estimated based on the area of residential land use. Population was then dasymetrically redistributed onto a 30 m grid based on a land cover and land use map of the study area. A comparative analysis shows that the spectral response method produced a larger mean relative error of 237% for population density at the block group, whereas the land use based method yielded a much better estimation, with a mean relative error of 21.4% for population counts for the block group. It is suggested that the dasymetric map provides a more accurate portrayal of population distribution than a choropleth map at small geographical scales such as that of a US county.