Modelled small area population estimates from sparsely distributed health intervention campaign data

Planning and decision-making in the 21st century heavily relies on small area population counts, traditionally obtained from population and housing census. In the case of Nigeria, the last census was in 2006, and census projections could have large uncertainties, thus, necessitating the development of alternative population estimation methods for operational uses. Here, we integrated the health campaign data from the  National Malaria Elimination Programme (NMEP) conducted in Nigeria in 2022/2023 with satellite-derived human settlements and a stack of geospatial covariates to produce population estimates in Nigeria at 100 metre spatial resolution.  We used a robust Bayesian hierarchical geostatstical modelling strategy which allowed for straightforward quantification of uncertainties in the parameter estimates, whilst systematically addressing the potential data quality issues within the input health campaign data. Compared to a more conventional approach, our methodology produced between 1.5% and 65.7% reduction in relative mean absolute error (RRMAE) over different data quality scenarios in a simulation study, and 18.3% RRMAE in the real data application. These modelled population estimates provide statistically robust evidence-base for informed decision-making, equitable resource allocation, effective humanitarian response and health intervention efforts in Nigeria.