Thread by @HCIBremen

nature.com/articles/s4158…
Our article is now online @nresearchnews Read more on our work on using #AI to count trees 🤖🤓🌳🌴 in #africa to understand soil degradation & #ClimateChange Thanks to all collaborator @unibremen @uni_copenhagen #GSFC @NASA @WWU_Muenster @KU_Leuven @CNRS

1/8 Our paper in a #nutshell 🌰The prevailing view is that dryland areas like the Sahara or Sahel are largely free of trees and shrubs 🌴🌳. However, we find that’s not the case!😯🤓🤖the relatively high density of isolated trees challenges prevailing narratives #desertification

2/8 Although the overall canopy cover is low, the relatively high density of isolated trees challenges prevailing narratives about dryland desertification (see photo below) 🌴👉

3/8 To segment individual trees, we trained a #DeepLearning model based upon the #UNet architecture on high-resolution satellite imagery. The model was trained on 89,899 manually delineated tree crowns, and it was evaluated on independent test sets, and data from field studies.

4/8 We detected over 1.8 billion individual trees (>3m2), or 13.4 trees ha-1, with a median crown size of 12 m² along a rainfall gradient from 0 to 1000 mm in an area of 1.3 million km2 (👉📺)

5/8 Our assessment lays the foundation for a comprehensive data-base of all individual trees outside forests. This will constitute a robust basis for understanding #dryland ecosystems and the role of human agency and climate change on the distribution of #dryland trees

6/8 In the longer-term perspective, it might be an important #baseline for policy-makers and stakeholders, as well as initiatives aiming at #protecting and 🌳🌴restoring #trees in arid and semi-arid lands in relation to mitigating degradation, poverty, and climate change. 😎🤓👍

7/8 Tree detection framework based on #UNet and the derived products produced by this study are available: the crown area shape file available, and the satellite data metadata provided
👉 #Code: doi.org/10.5281/zenodo…
👉 #Data: doi.org/10.3334/ORNLDA…

8/8 We are very proud of a part of this project 🎉🌳🌴🌳🌴🎉🤖❤️ Also, thanks to @nresearchnews for publishing it! Thanks to all collaborators and founders! 🙏@unibremen @uni_copenhagen #GSFC @NASA @WWU_Muenster @KU_Leuven @CNRS @VolkswagenSt Great work by @ankitky1

@threadreaderapp unroll