Answers are grounded in this paper and cite their sources.
Accurate estimation of individual-tree biomass is important for forest inventory, carbon assessment, and fuel management. In this study, we proposed a mask-based multimodal deep learning framework for estimating diameter at breast height (DBH) and allometry-derived tree biomass (TB) using unmanned aerial vehicle (UAV) RGB imagery. Field measurements and UAV imagery were collected from two Korean red pine plots in South Korea. Crown metrics obtained from Mask R-CNN delineation were combined with crown-specific image features in a fully connected prediction head. Leave-one-plot-out evaluation was used to compare the framework with image-only and metric-only neural networks, least squares regression, random forest, and XGBoost. In the baseline comparison, the multimodal model showed the lowest test relative root mean square error (rRMSE) in both evaluation directions: 10.93% and 11.09% for DBH and 26.24% and 25.90% for allometry-derived TB. However, paired comparisons did not consistently support lower multimodal error against every model. Model performance also varied with image resizing and alternative masks. These results support combining crown metrics and image features for individual-tree estimation within the conditions examined. Further evaluation is required for other forest conditions.
No takes yet. Share an insight, caveat, or question.
Kwon et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: