Abstract
Gridded Population Datasets (GPDS) are extensively used in research and public policy. However, the methods used to estimate populations in these datasets vary widely,resulting in different population distributions that can lead to varying statistical results
and conclusions. Despite the importance of this issue, prior research examining these
inconsistencies has been limited to small areas and specific case studies due to limited
validation data. This dissertation addresses this gap by utilising various techniques to
explore the inconsistencies of GPDS at global, national, and local scales.
First, I focus on characterising the distinctiveness of several GPDS for urban populations by calculating national urban shares and settlement counts. I classify populations
as “urban” using different spatial demographic criteria for three commonly used GPDS;
the Global Human Settlement Population Grid (GHS-POP), Gridded Population of the
World (GPW) and WorldPop. The most notable finding is that the urban statistics were
distinctly higher using the GHS-POP, particularly when considering the lower population
density thresholds and in lower-income countries.
Second, the mechanical uncertainties that underlie these datasets are explored by
interpolating household income for several local authorities in England. The choice of
targets for interpolation is the dominant factor for minimising areal interpolation errors,
and the novel “geobootstrap” method generally has the highest accuracy. The binary
dasymetric method used by the GHS-POP did not offer any clear accuracy or variance
improvement over the simple areal weighting method.
Finally, the level of agreement between different GPDS is explored to a global “consensus estimate”. The GPW has the highest agreement to the consensus estimate, and
the GHS-POP has the lowest agreement, particularly for the smallest cities. WorldPop
should be favoured since it has the highest agreement across the full range of city sizes.
Further research is needed to understand better the geobootstrap’s utility for areal interpolation and broader spatial analytics.
| Date of Award | 7 May 2024 |
|---|---|
| Original language | English |
| Awarding Institution |
|
| Supervisor | Levi J Wolf (Supervisor) & Sean Fox (Supervisor) |