Small area population estimation from health intervention campaign surveys and partially observed settlement data

Abstract Recent and reliable small area population numbers are required for effective governance, but financial and logistical challenges mean that national censuses are typically only undertaken every ten years or more. Geospatial modelling approaches have been developed that utilise bespoke microcensus surveys linked with satellite-derived settlement maps and other spatial datasets to fill population data gaps across countries with outdated or incomplete census data. However, microcensus surveys can be complex logistically and expensive, while satellite-based settlement maps can often be incomplete in tropical rural areas where tree canopies and cloud cover can obscure them. These factors limit the wider application of geospatial modelling approaches. Here, we present a novel two-step Bayesian hierarchical modelling approach that can integrate routinely collected health intervention campaign data and partially observed settlement data to produce reliable small area population estimates. Reductions in relative error rates were 32-73% in a simulation study, and ~32% when applied to malaria survey data in Papua New Guinea. The results highlight the value of demographic data that is collected routinely through health intervention campaigns or household surveys for improving small area population estimates, and how biases introduced through satellite data limitations can be overcome.