Integrating UAV Photogrammetry and High Resolution Soil Data for Spatial Prediction of Forest Productivity in Eucalyptus grandis Plantations

Abstract

Accurate estimation of dendrometric variables in mature Eucalyptus grandis plantations is fundamental for decision-making in both harvesting and silvicultural activities and the quantification of carbon stock; however, traditionally certain field activities present limitations in precision due to canopy closure and intra-stand spatial heterogeneity. This research integrates structural metrics derived from aerial photogrammetry, high-resolution soil data obtained by grid sampling and interpolation by ordinary Kriging (1 × 1 m), and field inventory data collected from 36 plots established in E. grandis plantations (9, 12, and 15 years old) in northern Uruguay, with the objective of estimating diameter at breast height (DBH), total height (Ht), volume per hectare (Vol), and dominant mean Assmann height (MDH) using Random Forest models. Prior to spatial modelling, a site-specific allometric equation was fitted to estimate stem volume using 75 destructively sampled trees (R2adj = 96.44 %, RMSE = 0.17 m3). Random Forest models, trained by iterative variable selection (Forward Feature Selection) in a 10-fold cross-validation scheme, achieved high predictive accuracy for all target variables (DBH: R2 = 0.95, RMSE = 1.57 cm; Ht: R2 = 0.98, RMSE = 1.08 m; Vol: R2 = 0.94, RMSE = 62.25 m3 ha-1; MDH: R2 = 0.98, RMSE = 1.06 m). Soil pH was shown to be the dominant soil predictor across all four models and the overall dominant variable for volume estimation, explaining 62.26 % of the relative importance in the volume model and ranking second in DBH (45.44 %), Ht (33.36 %), and MDH (32.25 %), highlighting its critical role in these models and the fact that it is usually underestimated among soil factors in regulating productivity at the microsite level. UAV photogrammetry also efficiently captured structural metrics, serving as a low-cost alternative to airborne LiDAR without compromising predictive capabilities. The resulting productivity maps are operational products for the digital inventory and sustainable management of commercial Eucalyptus plantations.