Estimating denominators: satellite-based population estimates at a fine spatial resolution in a European urban area

Abstract

Background: There is a need for alternative approaches to obtain population denominators when census information is unavailable, unreliable, or not available at the appropriate spatial resolution. The aim of this study is to develop an exportable population model, based on a single satellite-derived indicator, for estimating fine-scale population data and characterizing high-incidence areas in an urbanized area.

Methods: A Landsat 7 enhanced thematic mapper plus image was processed to generate population density indices at the block and block-group levels, using both an unsupervised pixel-based and a supervised classification. Spatial disaggregation was used to calculate population estimates, distributing the total population of the city of Besanon (France) into census areas by means of their respective population density indices. Accuracy assessment was performed through comparisons with census counts.

Results: At the block-group level, the simplest model produced relatively accurate and reliable population estimates within the range of observed counts. A strong agreement was found between observed and estimated incidence rates for non-Hodgkin lymphoma (intraclass correlation coefficient [ICC] = 0.73), but not for female breast cancer (ICC = 0.40). Withdrawing the sprawled block groups improved the agreements considerably (ICC = 0.84 and 0.71, respectively).

Conclusions: This apportioning procedure offers a way to obtain estimated population sizes (or at least densities) for areas with no accurate census, but does not substitute for censuses where good census data exist. Because it is rapid, relatively cheap, and computationally easy, it should be of special interest to epidemiologists, environmental scientists, and public health decision makers.