Rapidly estimating population using building footprints from satellite imagery

Context

Understanding the number of people living in a given geographic area is crucial for planning public health interventions. For example, inaccurate estimates of target populations for a vaccination program can result in resource shortage, waste of supplies or even campaign failure.  

Conducting a full census is time consuming and logistically difficult, especially in rural and sparsely populated or conflict affected areas. As a result, public health interventions may rely on population estimates projected from incomplete and outdated census data.

Recent demographic surveys conducted for specific interventions might be more up-to-date, but are limited to specific geographic areas. The challenge is to use all recently available data to improve a population estimate for the larger region.

Objectives

  • Represent official population estimates in 100-meter resolution geospatial layers
  • Create new estimates which are extrapolated from more recent, targeted demographic survey data
  • Rapidly refresh estimates as new data becomes available to support public health planning

13

Countries

100

Meter Resolution

<10

Days to Update new RPE

Geographic deployment

Geographic deployment

Benin, Cameroon, Chad, Democratic Republic of the Congo, Ethiopia, Liberia, Ivory Coast, Mali, Nigeria, Pakistan, Republic of the Congo, Togo, Zambia

Benin, Cameroon, Chad, Democratic Republic of the Congo, Ethiopia, Liberia, Ivory Coast, Mali, Nigeria, Pakistan, Republic of the Congo, Togo, Zambia

Population RPE
Laptop Background
  • slideshow RPE 1
  • slideshow RPE 2
  • slideshow RPE 3
  • slideshow RPE 4
  • slideshow RPE 5
Description

High-resolution satellite imagery and recent advances in machine learning algorithms make it possible to identify buildings remotely, classify them into settlement types, and use them with available demographic survey data to rapidly create straightforward population estimates useful for early planning. Public health officials can use the resulting layers in any GIS-enabled system.

The resulting population estimates are a collection of GeoTIFF raster files and point shape files for total population and children < 5 years old. The population rasters use the WorldPop grid alignment. They are optionally adjusted based on World Bank population figures and can be rapidly updated and republished as additional data becomes available. 

With 100-meter resolution, the RPE can be used to compute population estimates for nearly any size catchment and every aggregation level.