How fine can a top-down population model be? [draft]

We will study the use of very high-resolution covariates in top-down population model and its impact on population gridded estimates. We will examine building-footprint derived covariates in Sierra Leone population model. Disclaimer: This post is still being drafted.

Edith Darin
06-02-2021

Disclaimer: This post is still being drafted.

Background

To obtain high-resolution population estimates, WorldPop developed a top-down method that disagregates population totals into grid cells thanks to ancillary geospatial covariates.

The method consists in fitting a random-forest model (Breiman 2001) at the finest administrative level for which population count is available. The estimated model is then used to predict population density for every grid cell thanks to the geospatial covariates. Finally, the predicted gridded population density layer plays the role of a weighting layer to breakdown the population totals into the grid cells (Stevens et al. 2015).

Top-down disagregation model

The geospatial covariates commonly used by WorldPop initative for mapping population globally (WorldPop Research Group et al. 2018) are based1:

  1. on resampled rasters

The input rasters have a resolution ranging from 900m to 90 m at the Equateur (Lloyd et al. 2019).

Exemple of raster resampling: Slope in Sierra Leone

Figure 1: Exemple of raster resampling: Slope in Sierra Leone

  1. on Euclidean distance measurement

Rasters resulting from Euclidian distance measurement can have virtually any spatial resolution. They however might not be representative of localized informative variations for population modelling.

Exemple of Euclidean measurement: Distance to waterways in Sierra Leone

Figure 2: Exemple of Euclidean measurement: Distance to waterways in Sierra Leone

heir distribution

The purpose of this post will be to study the impact of the building-footprint derived covariates on the modelling and subsequently on the gridded estimates in Sierra Leone.

Data

Recently, WorldPop released openly gridded building patterns across entire sub-Saharan Africa at 100m resolution (Dooley and Tatem 2020). These were derived from building footprints extracted from satellite imagery with a spatial accuracy of 6m (Ecopia.AI and Maxar Technologies 2019). In top-down modelling, this data has already been used to produce an alternative global set of gridded population estimates constraining the estimates to settled grid cells2 (Bondarenko et al. 2020).

Method

Bondarenko, Maksym, David Kerr, Alessandro Sorichetta, and Andrew Tatem. 2020. “Census/Projection-Disaggregated Gridded Population Datasets for 51 Countries Across Sub-Saharan Africa in 2020 Using Building Footprints.”
Breiman, Leo. 2001. “Random Forests.” Machine Learning 45 (1): 5–32.
Dooley, Claire A, and Andrew J Tatem. 2020. Gridded Maps of Building Patterns Throughout Sub-Saharan Africa, Version 1.0. WorldPop Research Group, University of Southampton. https://doi.org/10.5258/SOTON/WP00666.
Ecopia.AI, and Maxar Technologies. 2019. Digitize Africa Data.
Lloyd, Christopher T, Heather Chamberlain, David Kerr, Greg Yetman, Linda Pistolesi, Forrest R Stevens, Andrea E Gaughan, et al. 2019. “Global Spatio-Temporally Harmonised Datasets for Producing High-Resolution Gridded Population Distribution Datasets.” Big Earth Data 3 (2): 108139. https://doi.org/10.1080/20964471.2019.1625151.
Stevens, Forrest R., Andrea E. Gaughan, Catherine Linard, and Andrew J. Tatem. 2015. “Disaggregating Census Data for Population Mapping Using Random Forests with Remotely-Sensed and Ancillary Data.” PLOS ONE 10 (2): e0107042. https://doi.org/10.1371/journal.pone.0107042.
WorldPop Research Group, Department of Geography and Geosciences, University of Louisville, Departement de Geographie, Universite de Namur, and Center for International Earth Science Information Network (CIESIN), Columbia University. 2018. “Global High Resolution Population Denominators Project - Funded by the Bill and Melinda Gates Foundation (Opp1134076).” https://doi.org/10.5258/SOTON/WP00645.

  1. They can be downloaded at: https://www.worldpop.org/project/categories?id=14↩︎

  2. Settled cells are cells containing at least one building footprint.↩︎

References

Citation

For attribution, please cite this work as

Darin (2021, June 2). Meet Edith: How fine can a top-down population model be? [draft]. Retrieved from https://edarin.github.io/thatsme/posts/2021-06-02-high-resolution-covariates-and-top-down-population-disagregation/

BibTeX citation

@misc{darin2021how,
  author = {Darin, Edith},
  title = {Meet Edith: How fine can a top-down population model be? [draft]},
  url = {https://edarin.github.io/thatsme/posts/2021-06-02-high-resolution-covariates-and-top-down-population-disagregation/},
  year = {2021}
}