Innovative 3D photosynthetic trait assessment of slash pine using drone-LiDAR fusion and machine learning algorithms.

Accurate, spatially explicit quantification of the fraction of absorbed photosynthetically active radiation (fPAR) in tall conifer plantations is essential for productivity modelling and breeding, yet standard nadir-view optical UAV imagery yields only two-dimensional surface estimates. We developed an unmanned aerial workflow that fuses centimeter resolution LiDAR point clouds with five band multispectral imagery to produce a three-dimensional voxelized canopy structure in which top-of-canopy multispectral reflectance values are propagated downward within each vertical column. Ground measurements of fPAR and chlorophyll fluorescence were collected contemporaneously and used to calibrate Random Forest, XGBoost, Support Vector Machine (SVM), and Partial Least Squares Regression models built from 14 spectral indices. Random Forest explained 84 % of fPAR variance (RMSE = 0.12), outperforming alternative algorithms. Application of the trained Random Forest model to the voxelized canopy (0.01 m × 0.01 m × 2 m) across 28 ha generated three-dimensional fPAR maps that revealed a 26 ± 4 % increase from lower to upper crowns and a seasonal shift of up to 9 %. Compared with conventional plot-level inversion, the workflow significantly reduced field labour and improved prediction accuracy. The fusion pipeline provides a species-specific tool for high-throughput phenotyping, precision silviculture, and genomic selection in slash pine plantations under clear-sky conditions (solar zenith angle 20-30°); transferability to other sites, species, or illumination conditions requires further validation.