The size of the population, the denominator of many statistical indicators, is crucial for public policy. National statistical offices organize the collection of this information, most often through a census. But what happens when parts of a country are not accessible to census enumerators? Today, spatial data extracted from satellite imagery offer high-resolution geographical information with complete coverage. When combined with a partial population count, they offer an unprecedented opportunity to estimate the size of the population in inaccessible areas. The spatial precision of these data also makes possible the production of a high-resolution gridded population estimate, an innovative data format at the intersection of geography and demography. Based on the case of Burkina Faso, this article analyses how, by dividing a country into 100 m by 100 m cells, a Bayesian hierarchical model can be used to estimate the population of areas with security challenges which could not be enumerated during the 2019 census. This gridding allows the resulting counts to be disaggregated using a statistical learning model, yielding unparalleled spatial precision in population estimates.

geospatial data, census, hierarchical model, Bayesian statistics, gridded population, building footprint, remote sensing