The Baobab Foundation is committed to supporting the next generation of baobab researchers. One of those students is Chafika Phiri from the University of Exeter, who is exploring an exciting question: can satellite images and machine learning—often called AI—help us map baobab trees across vast landscapes?
The challenge is greater than it may seem. Tree size matters, because small baobabs are much harder to spot than large ones. Leaf cover matters too: in summer, many baobabs blend into the surrounding vegetation and can be difficult to distinguish from other trees. In winter, when their branches are bare, they may even be mistaken for rocky outcrops. The landscape adds another layer of complexity. Baobabs are easy to see on flat, sparsely vegetated ground, but they stand out far less in thick vegetation or mountainous terrain.
To train his model, Chafika has been working in two contrasting areas: Mapungubwe National Park and Kremetart Game Farm in the Soutpansberg. He began by marking the locations of several hundred baobabs in each landscape and teaching the model to recognise them. He also marked other trees so the model could learn what a baobab is not. The next step was to test how accurately it could find baobabs on its own.
Last week, Chafika, his supervisor Dr Andy Cunliffe, Alan Nare, Steven Khosa from SANParks, and I went into the field to evaluate the model. We were delighted to find that it performed very well in open landscapes. As expected, it was less accurate in more densely vegetated areas. The next round of data will help train and improve the model so that it can work across a wider range of environments.
Chafika Phiri measuring baobab tree
Chafika’s results will be published over the coming year. His work promises to deepen our understanding of baobabs and strengthen our ability to map these remarkable trees across much larger areas.
Reliable maps will allow us to track how baobab populations change over time. They could reveal where elephant damage has the greatest impact, identify areas with unusually high or low numbers of trees, and provide valuable evidence to guide conservation efforts.
