Using high-resolution satellite imagery and AI, an international team of scientists from NASA has mapped 10 billion trees in the arid lands of Africa. The purpose? To assess the amount of carbon trapped in African trees.
The result is the first comprehensive estimate of tree carbon density in African regions of the Sahara, Sahel, and Sudan. The team reported their findings on March 1st in Nature (ref.) and the data is free and available to the public.
Carbon Residence Time
The researchers found that there are many more scattered trees in Africa’s semi-arid regions than previously thought, but they also store less carbon than predicted in models. In the new study, the team estimated that approximately 0.84 petagrams of carbon are trapped in African trees. A petagram is 1 billion tonnes.
Having an accurate estimate of tree carbon is essential for climate change projections. These are influenced by how long trees and other vegetation store carbon. Carbon residence time is very short for grasses and shrubs, which grow seasonally, but much longer for trees that grow for years.
Beyond the vast tropical forests scattered in the center of the continent, African landscapes range from dry grasslands with few trees to savannas with scattered trees to wetter areas with many scattered trees. This dispersed tree cover has made it difficult to estimate the number of trees in these areas, and there have often been overestimations or underestimations. Yet such measurements are essential for conservation efforts and understanding the carbon cycle on our planet.
Help of AI
“Our team has collected and analyzed carbon data down to the level of individual trees in the vast semi-arid regions of Africa. This analysis had only been done on small local scales” said Compton Tucker, a scientist at NASA’s Goddard Space Flight Center. Previous satellite estimates of tree carbon in Africa’s arid lands often mistook grasses and shrubs for trees. “This has led to overestimations of carbon”.
Carbon constantly cycles between the land, atmosphere, and ocean. Trees remove carbon dioxide from the earth’s atmosphere during photosynthesis and store it in their roots, trunks, branches, and leaves. For this reason, increasing tree cover is often suggested as a way to offset rising carbon emissions.
In the study, the team used sophisticated machine learning and AI algorithms to sort through over 326,000 satellite images. The researchers acquired the images through NASA’s Center for Climate Simulation. By leveraging the Explore/ADAPT Science Cloud, they prepared the images for machine learning.
The results and climate change
Martin Brandt from the University of Copenhagen compiled data on AI training from 89,000 individual trees. Colleague Ankit Kariyaa adapted a neural network so that computers could detect individual trees in high-resolution images at a scale of 50 centimeters, in the driest and least green landscapes of Africa.
The researchers defined a tree as anything with a green, leafy canopy and adjacent shade. From this, they trained the machine learning software to count trees during millions of hours of supercomputing on the Blue Waters supercomputer at the University of Illinois. When the team compared the results of machine learning with human landscape assessments, the computers achieved 96.5% efficacy in measuring tree canopy area.

From the measurements, scientists can derive the amount of carbon in the leaves, roots, and wood of each tree. A group led by Pierre Hiernaux from the University of Toulouse examined 30 different species to measure leaf, wood, and root mass. By evaluating the carbon masses, they established a statistical relationship with tree canopy area. The results will be useful for scientists and students studying the carbon cycle. But above all, they should be useful for policymakers seeking to combat climate change and farmers who will want to determine the carbon stored in their fields.