Satellite Tree Enumeration Outside of Forests at the Fifty Centimeter Scale
Machine Learning, Deep Learning and Artificial Intelligence
TimeTuesday, 17 November 20201:45pm - 2:30pm EDT
DescriptionA large proportion of trees grows isolated outside of forests and is not well documented. These non-forest trees play a crucial role in biodiversity, carbon storage, food resources and shelter for humans and animals. We have enumerated the crown size of individual trees greater than three square meters in a land area spanning ten million square kilometers from the Atlantic Ocean to the Red Sea in the West African Sahara, Sahel, in sub-humid zones using sub-meter satellite imagery, high-performance computing and deep learning. We enumerated over 14 billion isolated trees, or 140 trees/ha, with a median crown size of 12 square meters along a rainfall gradient from zero to 1000 mm. The canopy cover increases from near zero in hyper-arid zones to 9.9 trees/ha in the arid zone, to 30.1 trees/ha in the semi-arid Sahelian zone, to 470 trees/ha in the sub-humid zone. Although the overall tree cover is low, the unexpected higher density of isolated trees challenges prevailing narratives about dryland desertification, where even the desert showed a surprisingly higher tree density than previously thought. Our machine learning approach using high-performance computing enables enumeration of isolated trees globally to identify localized areas of degradation and to quantify the role of isolated trees in the global carbon cycle.