Leaf-on and leaf-off UAV LiDAR data for stand structure classification of Apennine beech forests

Traditional coppicing, followed by progressive abandonment and/or high-forest conversion are shaping the Apennines (Italy) beech forests, frequently exhibiting structural mosaics even at very small scales, making their ground assessment uncertain. Current forest planning requires spatially precise information of their respective stand attributes to set management priorities. In this study, we tested whether UAV-borne LiDAR scanning can accurately map stand attributes and detect the appropriate structures directly from 3D point clouds. Datasets from leaf-on and leaf-off flights were compared and analysed, together with data from ground surveys. The experimental site is a ~35 ha European beech (Fagus sylvatica, L.) previously coppiced forest at 1200 m asl in the Central Apennines (Frontone, Marche region, Italy). We set up 30 circular sampling plots on the ground, where we carried out a full stem inventory and derived plot-level dendrometric variables, including mean tree height, mean DBH and standing timber volume. Plots were clustered into three groups (stored coppice, transition to high forest and high forest) supported by ground-based observations, Principal Component Analysis and k-mean clustering. We also collected two UAV LiDAR datasets (leaf-on in July 2024 and leaf-off in March 2025) using a DJI Matrice 350 RTK equipped with a DJI Zenmuse L2 sensor. We normalized the point clouds heights and different LiDAR predictors were derived from vertical canopy profiles built with 1-m height bins for each inventory plot. We combined standard area-based metrics (height point density percentiles and return fractions) with structural descriptors that quantify canopy stratification, rugosity, openness/continuity and vertical filling. Preliminary results showed that stored coppice and high forest structures are easily distinguished, whereas the diverse stages of coppice-high forest transition are often confused. The UAV-LiDAR area-based regression models achieved solid performance, with a small subset of LiDAR metrics already capturing most of the variance in observed mean tree height (R² = 0.872; RMSE = 1.74 m), mean DBH (R² = 0.845; RMSE = 4.86 cm), and standing timber volume (R² = 0.768; RMSE = 41.67 m³ ha⁻¹). Leaf-off results classified with better accuracy the transition-to-high-forest structure, the mean DBH and standing timber volume, while the mean tree height was better estimated by leaf-on results. The LiDAR leaf-off and leaf-on data fusion slightly improved the stand attribute regression. The study suggests that the canopy-top texture of these beech forest mosaics can be better assessed using leaf-on UAV-borne LiDAR data. Conversely, structural changes and other stand attributes can be more accurately detected using leaf-off data, providing a deeper penetration into the understory and down to the ground. Multi-season UAV-borne LiDAR is a promising approach to accurately map structural mosaics and stand attributes at a spatial resolution relevant for forest management. Future work will focus on refining the data fusion strategy, identify the most informative LiDAR predictors for each classification target, quantify prediction uncertainty and evaluating model transferability across similar beech landscapes. Such developments will support the generation of repeatable, decision-support products, enabling evidence-based forest planning and management.