Why Facebook’s Satellite Imagery Analysis Announcement is Important

Yesterday, Facebook announced that they have mapped human population in 20 countries with unprecedented spatial granularity.  It might seem odd for a social network to be diving into global population patterns using satellite imagery, but it was done for a practical reason: to figure out the best type of internet to deploy in the developing world.  This is a powerful signal to all companies that satellites are about to change how global businesses understand our planet.

Simultaneous advances in commercial satellite technology, cloud computing, and machine learning have enabled a breakthrough in our understanding of the world.  Emerging applications for this technological conjunction include real-time monitoring of global deforestation, understanding shipping traffic, and forecasting of food production.

Facebook’s new result shows Earth observations can support enterprise-scale decision making for a wide range of commercial applications. Whether that’s better understanding of where your future customers live, or monitoring key existing infrastructure, or, in this case, informing deployment of communications resources for developing communities, satellite imagery can arm your enterprise with information that was previously too costly or impossible to collect.

Though we live in a world dominated by data, there are plenty of problems that require information that isn’t currently available.  To solve these problems, we traditionally resort to statistical sampling of populations to estimate the numbers we need.  For example, to estimate agricultural production - an area where our company, Descartes Labs, is focused - the USDA uses paper surveys to collect sample data from thousands of farms, out of two million farms across the U.S., to estimate production at the county, state, and national levels.

Expensive sampling programs can be accurate, but the more granular you try to get, the harder it is to get enough data.  In the case of population density studies, even Facebook with their billion users lacks data on roughly 80% of the planet’s population.  Commercial and public satellite imaging systems can address a large class of problems that can be solved by direct observation from orbit without the need for sampling.  With daily pictures of the globe, courtesy of Planet Labs and other start-ups, we’ll have a window into even the most remote places on Earth, and be able to monitor a broad spectrum of trends as they unfold.

Of course, doing this requires a lot of ingenuity.  Facebook chose human dwellings as a proxy for population density. Residences, industrial structures, parking lots, and agriculture all suggest how many people live in an area.  High-resolution satellites take infrequent observations of any specific location, so watching cars on the roads and in parking lots is still a sampling problem, while buildings tend to stay still and can be tracked on longer timescales. Observation of visible and infrared emissions of the changes that occur daily, seasonally, and beyond lets us directly observe what’s happening to human populations, and correlate observable changes in both the physical world and in cyberspace.  

We at Descartes Labs are excited to see how Facebook and other companies are using satellite data in their decision-making process.  Facebook has shown one way that automated analysis of satellite analysis can inform decision making at the intersection of social media networks and physical space. At Descartes Labs, our customers are already beginning to use our real-time agricultural forecasts to address a range of deeper problems, such as forecasting food production around a set of physical assets, or understanding historical productivity to assess long-term risk.  It’s no longer sampling, but is instead global, persistent, direct observation, and we’re looking forward to help push this technology to its limit.