Date on Master's Thesis/Doctoral Dissertation

5-2016

Document Type

Master's Thesis

Degree Name

M.S.

Department

Geography and Geosciences

Degree Program

Geography (Applied), MS

Committee Chair

Gaughan, Andrea

Committee Co-Chair (if applicable)

Stevens, Forrest

Committee Member

Stevens, Forrest

Committee Member

Linard, Catherine

Author's Keywords

population; random forest; population modeling; urban

Abstract

Between 1990 to 2015, numerous groups used ancillary data about the environment surrounding populations to more accurately map global populations from standard census data. No comprehensive study has been undertaken to characterize the observed relationships between population density and ancillary data. Better understanding these relationships may produce more accurate population maps, focus resources on new datasets with a high probability of modelling importance, and lead to expanded end-user applications. This study examined these relationships by extracting variable importances from 36 independently run, country-specific population models from the WorldPop project’s population data. Covariate data describing urban/suburban extents were found to be the most significant predictors of population. Little difference was found in the resolution of urban/suburban data regarding their modelling importance. Further examination of the effect of different definitions of built-/urban-area, methods of quantifying input data quality, and the probability of specific variable classes as significant predictors of population is required.