EARLY-STAGE HEIGHT ASSESSMENT IN Eucalyptus sp. PLANTATIONS USING UAV-BASED PHOTOGRAMMETRY

Monitoring tree height during early stand development is essential for silvicultural decisions, yet conventional inventories rarely cover the period before canopy closure. Unmanned Aerial Vehicles combined with Structure-from-Motion photogrammetry offer a flexible, low-cost alternative to field surveys, but their reliability for young stands remains insufficiently validated because the low tree height amplifies errors in terrain modelling and canopy-top reconstruction. This study evaluated UAV-derived tree height estimates from RGB imagery and their calibration for young Eucalyptus sp. stands. The study comprised four commercial Eucalyptus sp. stands in Minas Gerais, Brazil, including 12 sampling units and 786 trees monitored at 3, 6 and 9 months after planting (2,358 observations). Images were processed in Agisoft Metashape to generate point clouds, from which Canopy Height Models (CHMs) were derived in R. UAV-derived heights were compared with field-measured heights and calibrated using robust linear regression, and model performance was evaluated using Monte Carlo cross-validation. Statistical equivalence was assessed using the Two One-Sided Tests procedure, and agreement was evaluated using Bland–Altman analysis. Raw UAV-derived estimates consistently underestimated field-measured height (bias -10.98 to -58.85 cm), a pattern attributed to terrain-modelling and canopy-top reconstruction limitations. Calibration reduced RMSE, MAE, MAPE, and bias (e.g., -58.85 to 0.04 cm at nine months), eliminated bias, and achieved statistical equivalence at all ages while preserving Spearman's correlation. Accuracy improved progressively with stand age, with the regression slope converging towards unity (from 0.65 to 1.01). Short-term growth projections showed agreement with field measurements throughout the first year. UAV-derived height estimates in young Eucalyptus sp. stands systematically underestimated field measurements, and calibration was essential for bias correction and statistical equivalence. Direct application of UAV-derived height estimates is unsuitable for individual-tree analyses at the earliest stages; however, when combined with calibration, this approach is a viable tool for monitoring stand height and supporting precision silviculture.