Detecting baobab trees (Adansonia digitata) in drone imagery and evaluating their anthropogenic legacy in eastern Africa

Recent research has led to a reevaluation of the scale of human impacts on tropical environments, notably through arboriculture. The extent to which this occurred in Africa is poorly understood, but the baobab tree ( Adansonia digitata) may serve as a useful proxy for measuring the impacts of long-term human land use. These trees are long-lived, produce important economic products, and function culturally as shrines and markers on landscapes. Archaeologists in eastern Africa have long suspected an association between baobab tree groves and sites, but no full-coverage studies have measured this spatial relationship. This paper explores different methods for detecting this tree species in high-resolution drone imagery, which is increasingly available. Then, the paper evaluates associations between baobab tree distributions and archaeological and historical settlements on the island of Unguja, in Zanzibar, Tanzania. Qualitative assessments and a Nearness test show positive associations between baobab groves and Swahili settlements from the 7th-18th centuries CE. Thismay reflect the anthropogenic factors of baobab dispersal and propagation, especially from the 11th c. onward. Baobabs are negatively associated with historical 19th-century settlements, possibly reflecting deforestation in areas cleared for clove plantations. These results demonstrate the potential for landscape analysis with full-coverage, high-resolution drone imagery, and shed light on the baobab tree as an anthropogenic legacy in the African tropics. • Methods for detecting baobabs in high-resolution drone imagery across Unguja Island in Zanzibar, Tanzania were tested. • Baobabs are near sites of the 7th–18th centuries CE, but far from settlements of the 19th centuries CE. • This may reflect increasing baobab-human interactions as the island was first settled, but deforestation during colonialism. • Baobab distributions are a useful proxy for measuring anthropogenic legacies in Africa. • 8507 Digitized baobab trees in imagery are a useful dataset for training detection algorithms for future research.