Modelagem da biomassa arbórea: abordagem baseada em métricas deriváveis de LiDAR para contextos florestais e urbanos

Abstract

Accurate estimation of tree aboveground biomass (AGB) is essential for understanding carbon stock, evaluating the ecological quality of urbanized landscapes, forest management, urban planning, and supporting climate policies. Remote sensing technologies such as airborne LiDAR (Light Detection and Ranging) have expanded methodological possibilities for biomass estimation across different ecosystem contexts. However, equations developed for closed-canopy forests do not represent adquately trees growing under the open and isolated conditions typical of urban areas. This study aimed to develop and compare allometric models based on total height (H) and crown area (Ca), derived from airborne LiDAR, for two contrasting tree typologies in Lavras, southeastern Brazil: trees in a seasonal semideciduous forest fragment and isolated or spaced urban trees. Field data included diameter at breast height (DBH), total tree height (H), and species identification, while wood density (WD) values were obtained from the Global Wood Density database (Chave et al., 2009; Zanne et al., 2009). UAV-LiDAR data were used to derive maximum tree height (H) and crown area (Ca) metrics. Reference AGB was estimated using the pantropical allometric equation (Eq. 4) from Chave et al. (2014) based on DBH, H, and WD. We tested 27 candidate equations separately for each typology and evaluated their fit (adjusted R²), parsimony (AIC), and predictive performance (RMSE, BIAS, PBIAS, and R²) with AGB estimates obtained from the pantropical equation. In addition, we performed agreement analyses (Bland–Altman analysis and concordance correlation coefficient - CCC) and statistical equivalence (TOST) in comparison with published allometric equations. In the forest fragment, the selected semi-logarithmic model explained 87% of the variation in AGB in the training dataset and showed an RMSE of 0.227 t and R² of 0.724 in relation to the reference values. For spaced and isolated urban trees, the second-order polynomial model explained 89% of the biomass variation and presented an RMSE of 0.376 t and R² of 0.901 in the same comparison, reflecting the higher structural heterogeneity of this context. In both cases, closer agreement was observed with equations calibrated for structurally analogous forest types, reinforcing the importance of allometric transferability. Overall, the results demonstrate that H e Ca captured substantial variation in AGB within the sampled structural domains, but the appropriate mathematical form differed between typologies. The models should therefore be interpreted as context-specific calibration equations rather than universally transferable models. Independent validation across sites and broader structural gradients is still required before operational extrapolation.