Extended Data Fig. 3: Mapping individual tree crowns.
From: An unexpectedly large count of trees in the West African Sahara and Sahel
a, Before training the model, the spaces between labelled tree crowns (light blue) were filled (red) and given a higher weight. During training, the model was penalized more strongly for wrongly classifying gap pixels compared to other misclassifications. As a result, tree crowns that touch or are close to each other could be reliably separated. b, Examples of predicted trees (green), showing that most trees standing close to each other were mapped as individuals trees.
