A0605
Title: Leveraging satellite imagery to estimate number of households at fine scale resolutions in resource-poor settings
Authors: Chibuzor Christopher Nnanatu - University of Southampton (United Kingdom) [presenting]
Abstract: Effective government policies and humanitarian response efforts require accurate knowledge of the population count and the number of households at small area scales. In settings where census data are unavailable, incomplete, or outdated, advanced statistical models have been developed to produce estimates of counts of people at fine spatial resolutions, but methods for producing estimates of number of households are generally lacking. A statistical modeling technique is presented for producing fine spatial resolution estimates of the number of households by integrating AI-powered satellite-derived human settlement maps with stacks of geospatial datasets and demographic datasets. Parameter estimation is based on a Bayesian statistical inference approach implemented via the integrated nested Laplace approximation with stochastic partial differential equation frameworks, thereby making uncertainty quantification straightforward. The methodology is evaluated using a simulation study, showing varying levels of accuracy over different magnitudes of data missingness, with lower estimation error obtained for a smaller proportion of missing samples. The approach was successfully applied to obtain estimates of the number of households along with the corresponding estimates of uncertainty at 100m grid cells across Cameroon. The methodology provides a significant advancement in the small area population estimation field for facilitating more efficient governance and resource allocation.
