In partnership with the Center for International Earth Science Information Network (CIESEN) at Columbia University, we use state-of-the-art computer vision techniques to identify buildings from publicly accessible mapping services to create the world's most accurate population datasets. Our maps are available at 30-meter resolution - much more accurate than existing high-resolution maps, which are only available at a resolution of 100 meters. These high-resolution maps estimate the number of people living within 30-meter grid tiles in nearly every country around the world. Additionally, our datasets provide insights on the distribution of certain populations within each country, including the number of children under five, the number of women of reproductive age, as well as young and elderly populations, at unprecedentedly high resolutions. These maps aren’t built using Facebook data and instead rely on combining the power of machine vision AI with satellite imagery and census information. One use case for these maps are disease prevention efforts - gender and age are crucial indicators for the transmission and control of diseases. These high-resolution maps can provide the necessary insights for health organizations to allocate resources and control outbreaks.
There are many use cases for understanding the demographics of various populations – demographics can help organizations target vaccination campaigns, plan infrastructure, and distribute resources. To create maps that can help identify where these populations are, the first step is to figure out where people are generally, and the second step is to obtain the demographic profiles of the identified people. These steps are described in more detail below.
Step 1: Determine where people are
To obtain the population density of each country, we used Convolutional Neural Networks on high resolution satellite imagery to locate houses and combined these with the best census data sets available. Please refer to the methodology outlined for high-resolution population density for further details. Population estimates are based on data from the Gridded Population of the World data collection. Imagery used to identify settlements is from the DigitalGlobe Basemap +Vivid.
Step 2: Disaggregate data using the most granular demographics datasets available
We start by disaggregating data with respect to the most granular demographics datasets available. Below is a schematic representation of the level of demographic details we have for South Africa. For the source of the demographic datasets, refer to Columbia University’s Center for International Earth Science Information Network website, or you can access the direct link to the Excel sheet here.
Step 3: Zoom into a specific area
By zooming into a specific area, we can get a better feel on the granularity of the data, which we then integrate into existing models to obtain the most detailed demographic breakdowns by age and gender. For each administrative boundary, as shown in the plot below, we obtain the proportions of male vs. female, as well as the proportions by 5-year age bands. The 20 categories of age-band, combined with gender, results in an overall 40 unique groups.
Step 4: Combine 40 demographic categories with the high-resolution population density maps
By combining these 40 categories of demographic proportions with the high-resolution population density maps, we are able to obtain detailed spatial heterogeneities that exist over various regions in a given country. A schematic representation of this process is depicted in the image below.
Adjustments to match the census population with the UN estimates are applied at the national level. The UN estimate for a given country (or state/territory) is divided by the total census estimate of population for the given country. The resulting adjustment factor is multiplied by each administrative unit census value for the target year. This preserves the relative population totals across administrative units while matching the UN total.
We have identified the demographics that we hope will be most useful to humanitarian organizations.
- Overall population density
- Women of reproductive age (ages 15-49)
- Children (ages 0-5)
- Youth (ages 15-24)
- Elderly (ages 60+)