Gridded Population Maps Informed by Different Built Settlement Products

Area of Science:

• Geographic Information Science
• Demography
• Remote Sensing

Background:

• Accurate spatial distribution of human populations is crucial for various disciplines.
• Gridded population techniques provide spatially explicit data, but constraining estimates remains a challenge.
• Remotely sensed built-area datasets offer potential for improving dasymetric disaggregation of population data.

Purpose of the Study:

• To evaluate the effectiveness of three high-resolution built-area datasets for dasymetric population mapping.
• To compare different modeling techniques for disaggregating census counts using built-area data.
• To assess the utility of these methods for studying human populations and related phenomena.

Main Methods:

• Dasymetric disaggregation of census counts using three distinct high-resolution built-area datasets.
• Implementation of three modeling techniques: binary dasymetric redistribution, random forest with a dasymetric component, and a hybrid approach.
• Application of these methods across six diverse countries: Haiti, Malawi, Madagascar, Nepal, Rwanda, and Thailand.

Main Results:

• The study assessed the performance of binary, random forest, and hybrid dasymetric models across six countries.
• Random forest and hybrid models demonstrated comparable accuracy in five out of the six studied nations.
• The effectiveness of different built-area datasets varied, influencing the accuracy of gridded population estimates.

Conclusions:

• High-resolution built-area datasets can effectively constrain gridded population estimates through dasymetric disaggregation.
• Random forest and hybrid modeling approaches show strong potential for accurate population mapping.
• The findings support the use of integrated remote sensing and census data for improved understanding of human population distributions.