On 1 March 2023, an international team of scientists from the National Aeronautics and Space Administration (NASA) and other institutions published their findings on the first all-inclusive measurements of tree carbon density in Africa in the journal Nature. The expert team utilised commercial high-resolution satellites and artificial intelligence (AI) to analyse 326,000 commercial satellite images from the QuickBird-2, GeoEye-1, WorldView-2, and WorldView-3 satellites (operated by Maxar Technologies).
Furthermore, they collected images from NASA’s Center for Climate Simulation. They leveraged its Explore/ADAPT Science Cloud to organise and prepare the images for machine learning processing. In addition, the scientists used the processed insights to chart almost 10 billion individual trees to gain insight into the amount of carbon stored outside the region’s tropical forest.
The scientists discovered that the number of trees and carbon storage in the semi-arid parts of Africa is lesser than the previously confirmed predictive models. However, the team ascertained that about 0.84 petagrams of carbon are stored within the continent’s drylands (a petagram is 1 billion metric tons).
Furthermore, understanding and gaining accurate information about the estimated tree carbon is vital to predicting climate change impacts, intensifying conservation efforts, and learning about Earth’s carbon cycle.
According to the Lead Scientist and Earth Scientist at NASA’s Goddard Space Flight Centre, Compton Tucker, “The team collected and analysed carbon data down to the individual tree level across Africa’s vast semi-arid regions had previously been done only on small, local scales. Previous satellite-based measurements of tree carbon in Africa’s dryland often mistook grasses and shrubs for trees, which led to over-predictions of the carbon there.”
Assistant Professor at the University of Copenhagen, Martine Brandt and his partner, Ankit Kariyaa, gathered AI training data from 89,000 individual trees and integrated a neural network to ensure computers would identify individual trees in the high-resolution 50-centimetre scale images of Africa’s drier regions. In addition, the scientists taught the machine learning software to estimate the number of trees per a pre-defined concept as anything green, leafy crown with an adjacent shadow during millions of hours of supercomputing on the Blue Waters supercomputer at the University of Illinois.
After the analysis, the results showed a 96.5% accuracy with human projections of the landscapes and were used to arrive at the amount of carbon in each tree’s leaves, roots, and wood. Data on African tree carbon will be an essential tool for scientists, students, policymakers, and relevant authorities working on advancing conservation efforts and farmers interested in measuring the amount of carbon stored within their farmlands.
The African tree carbon data are publicly available and can be viewed using a custom-made app developed by the team.
Deborah loves reading and enjoys philosophical discussions.