The safety of critical traffic and energy infrastructure is often threatened by surrounding vegetation. We compare the accuracy of six canopy height models (CHMs) created by combining UAV-borne digital leaf-off and leaf-on surface/terrain models with nationwide sparse airborne laser scanning (ALS) data across six different study sites. We conducted the statistical evaluation at three levels for all involved samples, distinguishing, among others, between trees at the edge and inside the forest, as well as between conifers and deciduous trees. We hypothesised that combining UAV-borne leaf-on and leaf-off data or a combination of fine-scale UAV data with broader-scale ALS may benefit specific tasks associated with vegetation dynamics or precise inventory. However, the UAV-borne CHM using leaf-on imagery yielded the best overall accuracy (MAE 1.77 m), performing best both for trees at the forest edges (MAE 1.59 m) and inside the forest (MAE 2.12 m). This dataset also performed best for deciduous trees (MAE 1.84 m) while for conifers, UAV-borne CHM using leaf-off imagery performed best (MAE 1.58 m); the differences between these two models were, however, quite small and the model based on the combination of leaf-on and leaf-off imagery performed similarly well. We conclude that UAV-based CHMs are of sufficient accuracy and adding low-resolution ALS-based terrain data does not enhance their performance. Considering the simplicity, the leaf-on UAV is sufficient for everyday forestry practice where it could replace time-consuming and laborious field surveys.

 https://www.sciencedirect.com/science/article/abs/pii/S1618866724000268