Skip to main navigationSkip to main content
The University of Southampton
Geography and Environmental Science

A Novel Bayesian Geostatistical Model-Based Approach for accounting for Partially Observed Settlement data in Population Modelling Seminar

Time:
12:00 - 13:00
Date:
9 March 2023
Venue:
Teams

Event details

Geography & Environmental Science Seminar

The design and implementation of several health services and intervention programmes often rely heavily on the availability of reliable population numbers at the required administrative/geographical levels. However, in several countries, estimates of population numbers are provided by population projections that are based on censuses conducted several years ago which are in most cases incomplete and outdated. The availability of high-quality satellite imagery as well as increased computational power provides an opportunity to combine satellite observed settlement information with population data (e.g., household survey, administrative records, etc) and geospatial covariates to provide accurate modelled estimates of population numbers.

However, satellite observations can be obscured due to factors such as forest canopy (tree cover) such that the key settlement information required for the construction of the statistical population model is only partially observed. Several analytical challenges such as extremely biased population density are usually encountered when settlement data input is partially observed. In this talk,  an alternative 2-step model-based solution to partially observed settlement data is provided. The new approach implemented within a Bayesian Geostatistical Hierarchical regression modelling framework provided accurate and more efficient estimates of population numbers.

Speaker Information

Dr Chris Nnanatu - WorldPop

Link to join in

Privacy Settings