Extended Data Table 1 Performance in relation to image quality

From: An unexpectedly large count of trees in the West African Sahara and Sahel

  1. To study the effect of image quality on the model, manually labelled tree crowns (n = 89,899) from our training dataset are compared against predictions. The comparison is done for 432 plots of varying sizes (plots without trees have been removed). Here the density (the number of trees per plot) and the canopy area (in m2) is compared between the plots. Extended Data Figure 8 shows the spatial distribution of the classes. The slope can be used as a measure of under- or over-estimation. A slope of 1 is a perfect fit; values below 1 indicate underestimation in the prediction. We conclude that satellite data from November and December with off-nadir view angles <15° are preferable.
Back to article page