High Spatiotemporal Resolution RGB UAV Imagery: A Robust Approach for Assessing Juvenile Eucalyptus Crown Dynamics

South Africa's highly productive forestry industry relies heavily on fast-growing exotic plantations, where maximising early growth is critical for long-term yield. Unmanned aerial vehicle (UAV) platforms enable spatiotemporal monitoring of crown development at the resolution and scale required for robust modelling. This study assesses juvenile Eucalyptus crown dynamics using high-resolution red, green, and blue (RGB) UAV imagery across a gradient of planting densities and species compositions to evaluate early individual tree competition.

Weekly beyond visual line of sight flights were conducted using a DJI Dock 2 and Matrice 3D UAV from October 2025 to May 2026 over an 8-hectare trial at the IMPACT Open-Air Laboratory near Stellenbosch, South Africa. The trial monitored four Eucalyptus species (1-2 years old) across four planting densities (400, 1111, 2500 and 10000 stems/ha). Sub-1 cm and 3 cm ground sampling distance (GSD) orthomosaics and canopy height models were generated using Pix4D and lidR. Individual tree crowns (n = 3402) were segmented using the Geo-Segment Anything Model, and extracted tree heights were calibrated with terrestrial laser scanning and manual field measurements.

While the RGB Structure from Motion workflow yielded accurate crown delineations, initial sub-1 cm GSD flights significantly underestimated tree heights (R² = 0.01, MAE = 0.6 m). Optimising flight parameters at 3 cm GSD and dynamically calibrating the data delivered reliable tree height estimates (R² = 0.85, MAE = 0.3 m). Weekly time-series monitoring revealed distinct allometric plasticity across planting density treatments. Lower planting densities exhibited rapid growth recovery following a Mediterranean summer drought, while early canopy closure at 10000 stems/ha intensified light competition, constrained lateral crown expansion and increased stem slenderness. High spatiotemporal resolution RGB UAV imagery, paired with deep learning segmentation, provides a robust approach for monitoring early Eucalyptus crown dynamics, competition, and allometric plasticity.