One of the main requirements for investigation of health risks to environmental pollution is accurate information on population distribution. Although detailed data do exist in many parts of the world, based on national censuses at household or census tract level, in many others the level of spatial aggregation of these data is too coarse to allow reliable population estimates to be derived at the small area scale. Nor do census data typically provide accurate, small-area estimates of changes in population distribution in inter-censual years. One important, but as yet little-used source of information on population distribution is night-time satellite data, such as those provided by the Defense Meteorological Satellite Program (DMSP). These provide complete Earth coverage every night (subject to cloud cover) at a nominal spatial resolution of ca. 500 to 2700 metres. Previous studies have demonstrated strong associations between light emissions and population density at broad (e.g. national) scales, and the capability to use the data to detect and map changes in urban area over periods of a few years. This study assessed the ability to extend the use of these data to give local estimates of population distribution and to detect short-term variations in population as inputs to epidemiological studies. In a European-wide analysis, DMSP data were first linked to detailed land cover data within a Geographical Information System, to enhance their spatial resolution, using a combination of kriging and inverse distance weighting techniques. Associations between light emissions from relevant land cover classes and population numbers at different levels of aggregation were then analysed, both across the European Union and by country. Results showed that country-specific models typically explained 70–90% of variation in population distribution at the regional (e.g. county) scale and 50–85% of variation at the small-area (census tract) scale. DMSP data were also analysed to explore seasonal and year-to-year variations in population in sub-areas of the EU. Clear differences could be detected in both seasonal patterns of activity and long-term urban growth. Further exploration and use of these data as part of epidemiological investigations is therefor warranted.
(1) Imperial College London
(2) Centre for Application of Computer Science in Agriculture, Florence
(3) National Observatory of Athens