From top to bottom: gridded human population estimates in data-poor situations

Where disease risks are heterogeneous across population groups or space, or dependent on transmission between individuals, spatial data on population distributions – human, livestock and wildlife – are required to estimate infectious disease risks, burdens, and dynamics. As a result, large-scale, spatially-explicit, high-resolution human population data are being increasingly used in a wide range of animal- and public-health planning and policy-development scenarios. Official census data, aggregated by administrative unit, provide the only complete enumeration of a country’s population. While census data from developed countries are generally up-to-date and of high quality, in resource-poor settings census data are often incomplete, out of date, or only available at the country- or province-level. The challenges associated with producing accurate population estimates in regions that lack high-quality census data have led to the development of census-independent approaches to small-area population estimations. Known as bottom-up models, as opposed to the census-based top-down approaches, these methods combine microcensus survey data with ancillary data, to provide spatially disaggregated population estimates in the absence of national census data. This review highlights the need for high-resolution gridded population data, discusses problems associated with using census data as top-down model inputs, and explores census-independent – or bottom-up – methods of producing spatially-explicit, high-resolution gridded population data, together with their advantages.