UAV-based phenotyping and spatial modeling of genotypic responses to environmental covariates in eucalyptus clonal trials

Accurate phenotyping is essential for genetic evaluation in forest breeding, but repeated measurements in large clonal trials are labor-intensive and can be confounded by within-site environmental heterogeneity. Unmanned aerial vehicles (UAVs) equipped with red–green–blue (RGB) sensors provide a scalable approach for repeated structural and spectral phenotyping. We hypothesized spatial correction and RGB covariates may improve prediction precision, while reaction norms reveal genotype-specific environmental sensitivity. Therefore, we evaluated UAV-derived canopy height and RGB vegetation indices in two Brazilian Eucalyptus clonal trials. The site from São José dos Campos (SP) included pure E. grandis and E. grandis × E. urophylla genotypes (45 clones, 2,576 trees; five campaigns at 419, 470, 530, 645, and 956 days after planting [DAP]). The site from Três Lagoas (MS), included pure E. urophylla and E. urophylla × E. grandis genotypes (72 clones, 6,048 trees; four campaigns at 272, 341, 403, and 463 DAP). Four mixed models were compared: a randomized complete block model (M1), M1 plus a spectral covariate (M2), a two-dimensional spatial P-spline model (M3), and the spatial model plus a spectral covariate (M4). UAV canopy height was validated against 777 field measurements (R² = 0.74, r = 0.86, RMSE = 1.61 m, relative error = 23.33%). Adding VARI slightly increased the model fit (AIC) in both locations. Conversely, broad-sense heritability increased substantially from 0.56 (M1) and 0.58 (M2) to 0.84 (M3) and 0.85 (M4), reaching 0.91 in later campaigns using spatial models. Overall, UAV-based RGB phenotyping provided repeatable, high-resolution information across Eucalyptus clonal trials, while spatial modeling produced the largest gains in prediction precision. However, RGB covariates provided limited additional gains after spatial correction. Finally, reaction norms revealed clone-specific responses to environmental factors, highlighting the potential of integrating UAV phenotyping, spatial models, and environmental covariates for high-throughput clonal evaluation.