Extended Data Fig. 8: Evaluation of the tree crown mapping omission and commission errors.

The performance of the tree crown predictions was evaluated for 1,028 randomly selected 512 × 256-pixel areas from UTM zones 28–37, with approximately 100 areas for each UTM zone. The patches were extracted with an 80% bias towards precipitation⁴⁶ above 200 mm year⁻¹, as most tree identifications are above the 200 mm year⁻¹ isoline. 178 areas were excluded from evaluation because the rainfall was >1,000 mm year⁻¹ or the areas were devoid of trees. A total of 50,740 trees were evaluated. a, Class breakdown of our 3,765 omission and commission errors. b, Class numbers of all the trees evaluated. c, Summarizes the number of trees, errors of commission and omission and the number of trees by error classes. The highest percentage of binned-error classes, from 15% to >25%, occurred in areas with few trees and resulted from mixed tree and bush confusion for <8% of all trees evaluated. By contrast, the lowest binned-error classes, from 0% to 15%, had 92% of all trees evaluated. In the 850 patches, the number of trees ranged from one tree to 326 trees, with a total of 50,740 trees evaluated and 3,765 errors identified. Overall, the commission and omission error rates were 4.9% and 2.7%, respectively, a net uncertainty of 2.2%.