Moreover, the model performed tree density predictions with higher
r2 in tropical forests, coniferous forests, and
tropical dry forests. Notwithstanding having the highest
r2 in tropical forests, it also was the forest
ecosystem with the highest prediction error (Fig 3b).
On average, predicted tree height ranged between 4 to 9 m in all forest
ecosystems (averaged from all pixel values). Cloud forest, arid and
semi-arid zones had smaller r2 for both target
variables, which could be related to the smaller amounts of sampled data
in these forest types. However, arid and semi-arid zones seemed to have
the smallest error in both tree height and tree density predictions (Fig
3). A Taylor diagram is a graphical approach that quantifies how closely
the predicted values match the observed values and uses correlation
(r ), standard deviation of the error (SDE) and standard deviation
of observed (σz) and predicted (σẑ)
values as evaluators (Wadoux et al., 2022). According to Taylor
diagrams, the model had a better predictive performance for tropical dry
forest and broadleaf forest when predicting tree height as stated by its
correlation and RMSE together (Fig 3a). The model seemed to have the
best predictive performance for tropical forest when predicting tree
density, nonetheless, all forest types had a similar performance (Fig
3b).