The use of remotely sensed data for spatial disaggregation of published census population counts

Four methods were compared for spatially disaggregating or interpolating the published populations of census zones. In the absence of any other information, the areal proportions approach assumes a constant population density throughout each zone, and distributes the population of the zone uniformly. Landsat TM imagery provides additional information which can be used to distribute the population nonuniformly within each zone. Three such methods were employed, based on: classification of image pixels, regression modeling, and classification followed by regression modeling. Validity and accuracy was assessed by calibrating each method on a set of source zones within the study area (census wards within the County of Leicestershire, UK) and comparing the population estimates produced for a set of smaller target zones (census enumeration districts within the census wards). All three remote sensing methods produced substantially better estimates for the populations of the target zones than the benchmark areal proportion method, indicating more accurate spatial disaggregation of the population counts. The lowest estimation error obtained was less than half that obtained using the areal proportion method.