How Satellite Data And Artificial Intelligence Could Help Us Understand Poverty Better

New technology lets computers understand what they see in an image–or a million images.

How Satellite Data And Artificial Intelligence Could Help Us Understand Poverty Better
[Photo: Flickr user Rodrigo Carvalho]

Data analytics firm Orbital Insight is partnering with the World Bank to test technology that could help measure global poverty using satellite imagery and artificial intelligence.


Governments and development organizations currently measure poverty levels by conducting door-to-door surveys. The new partnership will test the use of AI to supplement these surveys and increase the accuracy of poverty data. Orbital said its AI software will analyze satellite images to see if characteristics such as building height and rooftop material can effectively indicate wealth.

The pilot study will be conducted in Sri Lanka. If successful, the World Bank hopes to scale it worldwide. A recent study conducted by the organization found that more than 50 countries lack legitimate poverty estimates, which limits the ability of the development community to support the world’s poorest populations. 

“Data depravation is a serious issue, especially in many of the countries where we need it most,” says David Newhouse, senior economist at the World Bank. “This technology has the potential to help us get that data more frequently and at a finer level of detail than is currently possible.”

A satellite image of Sri Lanka; the bottom version highlights agricultural areasPhoto courtesy Orbital Insight/Digital Globe

The announcement is the latest in an emerging industry of AI analysis of satellite photos. A growing number of investors and entrepreneurs are betting that the convergence of these fields will have far-reaching impacts on business, policy, resource management and disaster response.

Wall Street’s biggest hedge-fund businesses have begun using the technology to improve investment strategies. The Pew Charitable Trust employs the method to monitor oceans for illegal fishing activities. And startups like San Francisco-based Mavrx use similar analytics to optimize crop harvest.

The commercial earth-imaging satellite market, valued at $2.7 billion in 2014, is predicted to grow by 14% each year through the decade, according to a recent report.


As recently as two years ago, there were just four commercial earth imaging satellites operated in the U.S., and government contracts accounted for about 60% of imagery sales. By 2020, there will be hundreds of private-sector “smallsats” in orbit capturing imagery that will be easily accessible online. Companies like Skybox Imaging and Planet Labs have the first of these smallsats already active, with plans for more.

The images generated by these companies will be among the world’s largest data sets. And recent breakthroughs in AI research have made it possible to analyze these images to inform decision-making

Eyes On The Prize

For many years, a central AI challenge was figuring out how to teach computers to identify objects in photographs. That changed over the last two years.

“Computers are now on par with humans at determining what they are looking at,” says Jimi Crawford, CEO of Orbital Insight. “So we can take a million images and have the computer identify and count particular objects to figure out what they collectively tell us.”

In the bottom version of this image of Shanghai, shadows of buildings–a proxy for construction rates–are highlightedPhoto courtesy Orbital Insight/Digital Globe

Engineers start by sifting through images one at a time and marking each car, tree or house. By tracking the characteristics of each object, the computer learns to identify them automatically. Once the algorithms are locked in, computers can count and track these objects without human intervention.

Orbital Insight recently showed that using AI to count cars in 50 retailers’ parking lots can be more accurate than U.S. census data at predicting the retailers’ quarterly earnings. They’ve also used the method to measure construction in China, global oil storage, and cornfield yields in the Midwest. The partnership with the World Bank will be their second humanitarian project, following an announcement last spring to work with the World Resources Institute to track deforestation.

This project marks the first time that the World Bank will experiment with AI to improve its work. Analysts predict that more and more organizations will follow suit.

“We’re moving towards a world where we’ll be able to see daily images of every spot in every city at high resolution. We think we can use that to give transparency to all aspects of the economy,” Crawford says.


About the author

Maya Craig is an integrated producer based in San Francisco. She has been producing digital and broadcast content at creative agencies and production companies for the past five years and is a student at UC Berkeley's Graduate School of Journalism, focusing on next-generation technology.