Many countries in Europe and North America see their natural and agricultural landscapes being replaced by a fragmented, sprawled landscape. Spatially detailed modelling of changes in land use, population and transport could help to forecast the impact of scenarios aimed at mitigating the process of urban sprawl. A common problem with land-use change models however, is the lack of historical data for proper model calibration. In this paper we describe an approach for developing historical population density maps by downdating a recent high-resolution population density raster, using a time series of sealed surface data and historical census data as an input. In the proposed approach, we hypothesise a local relationship between increasing population densities and increasing sealed surface fraction estimates, the latter obtained from remote sensing imagery. We apply the method to Flanders, Belgium, a region where population growth and improved transport networks led to a diffuse urban expansion, with ribbon development along many roads and a strong fragmentation of open space. The resulting population and sealed surface maps provide interesting data on the urban sprawl phenomenon in the past decades. By computing a densification index we observe that most urban areas witness a recent population density increase while in several rural areas the built-up area per inhabitant is still growing. The downdated time series of population maps obtained in this study will be used to set up a historical calibration for an activity-based cellular automata model for Flanders and Brussels which, among other data, needs high-resolution population maps.