GIS and Artificial Intelligence Used to Build Facebook’s World Population Map

| |

Whether or not you “like” this post, or use the new reactions built into the social media platform and get angry or cry, the Facebook map of the world is coming, and no matter what else you may think of the SEO giant, the technology they used to create the map, and the resulting data are fantastic, provided they prove useful.

Not all of the maps will be released until this summer. But the artificial intelligence they used to create it has astounding potential. So how did they do it, and what does it mean?

In the United States, we take population density information for granted. In fact, there is almost too much data, and it is overused, over analyzed, and often overshared. However, population distribution maps have several applications.

Urban planners use that information to place and improve roads and other infrastructure. After a disaster, population density and crisis mapping can be used to direct aid and aid workers. Epidemiologists use them to map outbreaks of disease, predict its spread, and develop strategies to contain them.

Weekly Influenza Map. Map: CDC
Weekly Influenza Map. Map: CDC

Facebook of course has its own interests, as do Google and other social networks: get those populations who do not yet have access to the internet a way to connect, and use its services. Connectivity Labs and Free Basic are part of an international and mostly for profit corporate initiative designed to spread the reach of the web.

“The unique ability that we have through social media is to reach to people and for them to communicate back to us,” says Jeanine Guidry of George Washington University in a webinar about using social media to accomplish your communication goals. “And if we don’t use that we’re using a large part of the functionality.” Facebook wants everyone, everywhere in the world, to be able to join that conversation on its platform of course.

Not that Facebook has been shy about its wish to expand its influence. acquiring several companies along the way, most recently WhatsApp and Oculus Rift. While its Free Basics program was just blocked in India, the spread of Facebook’s influence continues, and the social media company has even been accused of practicing a new form of colonialism.

But how did Facebook make a better map than anyone else’s? By using a ton of computing power at its disposal, developers took existing data, and rather than driving around the countryside taking photos, combined a couple of different maps to create an entirely new product.

Gridded Population of the World

The most logical place to start was the Gridded Population of the World dataset from Columbia University. It’s an agglomerated set of local census data, normalized to the same year, and the best population map available globally. However, the grid squares can vary from a few square kilometers in urban areas to tens of thousands of square kilometers in some rural areas, so it’s fairly low resolution.

Raster map:
Existing Population Distribution (Gridded Population of the World Dataset (GPW) of a coastal region in Kenya. Source:

Enter DigitalGlobe

Facebook bought literally millions of kilometers of high resolution photos from DigitalGlobe, the company that owns and operates the majority of the privately owned Earth imaging satellites in orbit. If you are looking at an image of your town from space like in ArcGIS Earth, you are probably looking through one of DigitalGlobe’s four orbiting lenses.

These photos are extremely high resolution, what’s called submetric. Instead of the photo covering meters of area, each is covers a square about 50 centimeters (a little less than 20 inches) to a side. Then developers taught their neural-net algorithms to recognize what a building looked like from above.

Dense settlements (left) can take advantage of short-range wireless hotspots. Sparse settlements (right) require long-range cellular technology. Images: Digitalglobe.
Dense settlements (left) can take advantage of short-range wireless hotspots. Sparse settlements (right) require long-range cellular technology. Images: DigitalGlobe.

Urban Density and Population Data

The program went to work. Urban density was estimated by the number of buildings in an area. The best population data was then applied over the area. The software assumed if it saw a building, there were people there.


This is a very simple approach that just requires a tremendous amount of computing power. During the process, 21.6 million square kilometers in 20 countries was analyzed. This was done using 14.6 billion images, ten times more than the images analyzed by Facebook on a daily basis. The information is explained in more detail here.

The Problem of Big Data

It remains to be seen how effective this effort really was, and how useful the maps will prove to be. But even making the effort is expensive and time consuming. As rockets become cheaper and more and more companies send satellites into orbit, more and more imagery will become available, and it will be more affordable.

As innovative as all of this big data will allow businesses to be, without an automated way to process it, it will be nearly useless. Skybox (owned by Alphabet) and Descartes Labs say they have been able to gather data from pictures without the need for human oversight. If true, and if the Facebook map works, the implications for financial industries, tech companies, urban planners, emergency responders and other industries are far reaching.

The Facebook map of the world is just the beginning. Whether or not you like the social media giant spearheading this effort, the innovative way they have combined existing data, new imagery, and used artificial intelligence to tie the two together promises new and better mapping in the future. What’s not to like about that?


Connecting the world with better maps. By Gros, A. and Tiecke, T. (2016, February 21).

Connecting the World With Better Maps: Data-assisted Population Distribution Mapping. (n.d.).

State of Connectivity 2015: A Report on Global Internet Access. (2016, February 22).

See Also

Enter your email to receive the weekly GIS Lounge newsletter: