One, two, tree: how AI helped find millions of trees in the Sahara

Efforts to map the Earth’s trees are growing – and could change our understanding of the planet’s health

When a team of international scientists set out to count every tree in a large swathe of west Africa using AI, satellite images and one of the world’s most powerful supercomputers, their expectations were modest. Previously, the area had registered as having little or no tree cover.

The biggest surprise, says Martin Brandt, assistant professor of geography at the University of Copenhagen, is that the part of the Sahara that the study covered, roughly 10%, “where no one would expect to find many trees”, actually had “quite a few hundred million”.

Trees are crucial to our long-term survival, as they absorb and store the carbon dioxide emissions that cause global heating. But we still do not know how many there are. Much of the Earth is inaccessible either because of war, ownership or geography. Now scientists, researchers and campaigners have a raft of more sophisticated resources to monitor the number of trees on the planet.

Satellite imagery has become the biggest tool for counting the world’s trees, but while forested areas are relatively easy to spot from space, the trees that aren’t neatly gathered in thick green clumps are overlooked. Which is why assessments so far have been, says Brandt, “extremely far away from the real numbers. They were based on interpolations, estimations and projections.”

The most recent attempt at a global tally of trees was in 2015, when researchers, using a combination of satellite data and ground measurements, estimated there were just over 3tn. This was a dramatic increase from the previous estimate of 400bn in 2009, which was based on satellite imagery alone.

The research by Brandt and his colleagues in west Africa promises a more accurate picture in the future. In a collaboration with Nasa’s Goddard Space Flight Center, they were able to use satellite images from DigitalGlobe, previously available only to commercial entities, which were high enough resolution to make out individual trees and measure their crown size.

Using AI deep learning, and one of the world’s most powerful supercomputers – Blue Waters at the University of Illinois – the team was able to count individual trees from space for the first time. They manually marked nearly 90,000 across a variety of terrain, so the computer could “learn” which shapes and shadows indicated the presence of trees. This enabled them to count every tree with a crown size of at least 3 sq metres in a 1.3m sq km area comprising mostly the Sahara but also the semi-arid Sahel area along the southern edge of the desert and a sliver of the sub-humid zone beneath that. Overall, they detected more than 1.8bn trees.

Those in the Sahara tended to be clustered around human settlements. Arid areas had on average 9.9 trees per hectare, rising to 30.1 in semi-arid zones and 47 in the southernmost sub-humid rim of the patch being studied. There were just 0.7 trees per hectare in areas classified as “super-arid”.

“Most maps show these areas as basically empty,” says Brandt. “But they’re not empty. Our assessment suggests a way to monitor trees outside of forests globally, and to explore their role in mitigating degradation, climate change and poverty.”

Keeping the planet’s arboreal accounts is key to understanding the impact trees are having on our planet’s health. If the number of trees can be mapped, so can the amount of carbon they store.

The most high-profile existing world tree map is released annually by Global Forest Watch. Launched by the World Resources Institute (WRI) in 2014, it uses data from Nasa Landsat satellites (which don’t have as high resolution as commercial equivalents) to keep tabs on what it diplomatically calls “tree cover loss”.