Integrating UAV-LiDAR and Wood Density Measurements to Characterize Phenotypic Variation in Peruvian Ceiba Populations

Phenotypic variation among tree populations along climatic gradients reflects the combined effects of phenotypic plasticity and local adaptation. However, comparative studies on ecologically important Amazonian tree species such as Ceiba sp. remain limited. This study evaluates key phenotypic traits of Ceiba populations growing under contrasting climatic conditions in the Peruvian Amazon, with a focus on tree architecture and wood density. Two study regions were selected: San Martín and Madre de Dios, which differ in annual precipitation and dry season intensity according to the 45-year PISCO climatological dataset. The structural characteristics of the trees were characterized using LiDAR data acquired with a DJI Zenmuse L1 sensor and processed in R using the lidR package: land cover classification using the CSF algorithm, generation of a Digital Elevation Model (DEM) via TIN triangulation (resolution 0.2 m), and individual crown segmentation using the algorithm by Li et al. (2012). Wood density was measured from growth core samples collected in the field. To estimate height, two individual canopy predictors (maximum height, zmax, and the 99th percentile, zq99) were compared using simple linear regression against a Random Forest model that incorporated the complete set of extracted structural metrics. Maximum canopy height was the best predictor, with R² = 0.89 and RMSE = 3.07 m, outperforming both the 99th percentile and the Random Forest model (R² = 0.82; RMSE = 4.08 m), confirming the usefulness of UAV-LiDAR for the rapid phenotypic assessment of large tropical trees. In contrast, models predicting diameter at breast height based on aerial metrics showed considerably lower performance: the multiple linear regression model achieved R² = 0.29 (RMSE = 13.96 cm), while Random Forest improved the fit to R² = 0.49 (RMSE = 12.02 cm), without ever exceeding the threshold for acceptable performance reported in the literature (R² > 0.7). This suggests that trunk-related traits are more difficult to infer from canopy structure alone. Wood density measurements from San Martín population (n = 21) averaged 0.236 ± 0.017 g cm⁻³ and showed low variability (CV = 7.3%), indicating a relatively homogeneous population. The values were consistent with published estimates for Ceiba pentandra and followed a normal distribution. These results provide an important basis for evaluating phenotypic responses to climatic seasonality. Ongoing analyses incorporating samples from Madre de Dios will allow us to assess whether the more severe conditions of the dry season are associated with divergence in wood density and crown architecture. Taken together, these findings highlight the potential of integrating LiDAR data collected using unmanned aerial vehicles (UAVs) with field measurements to study phenotypic variation in tropical forest species.