Carbon bias from grassy tree misclassification: revealing forest structural heterogeneity by integrating GEDI and Sentinel-2

Abstract

Tropical forests store roughly half of terrestrial carbon, yet carbon estimates in regenerating and disturbed forests remain highly uncertain. A major source of bias is the prevalence of fast-growing, canopy-forming monocots--such as bamboo, palms, and bananas--that are often misclassified as trees. These "grassy trees" achieve canopy dominance but lack secondary growth, violating woody allometries used in most biomass models. Although NASAs GEDI mission has transformed large-scale biomass mapping with spaceborne LiDAR, its products rely on coarse plant functional types (PFTs), causing grassy-tree-dominated canopies to be absorbed into evergreen broadleaf tree (EBT) classes. Using a texture-based Sentinel-2 classifier, we isolated bamboo-dominated forests within GEDI EBT products in Xishuangbanna, China. GEDI observations show that bamboo canopies are structurally distinct from tree-dominated forests and lead to systematic carbon overestimation of 20-44 Mg C ha-{superscript 1} relative to empirical benchmarks. Our framework improves carbon accounting in structurally heterogeneous forests while remaining adaptable for place-based management.