Reconstructing annual 30 m forest aboveground biomass across Southwest China’s karst region from 1986 to 2024

Zezhi Yang, Xiaoxin Zhang, Xiaowei Tong

Fine-resolution reconstruction of multi-decadal forest aboveground biomass (AGB) remains challenging in karst landscapes, where steep terrain, fragmented stands, and strong environmental gradients can limit the regional transferability of national and global products. We developed a regionally adapted two-stage framework to reconstruct annual 30 m forest AGB across Southwest China’s karst region from 1986 to 2024 (KRAFT-30). First, National Forest Inventory (NFI) observations were integrated with Sentinel-2 imagery, canopy-height information, topography, and climate to construct a 2019 regional AGB baseline, with plot-level AGB allocated to 10 m pixels using an NDVI-weighted strategy. Second, the baseline was transferred to Landsat annual composites through biomass-stratified sampling, feature selection, four-model comparison, and XGBoost-based historical reconstruction, with uncertainty propagated across the two stages. The baseline agreed with withheld NFI plots (R² = 0.75; RMSE = 25.73 Mg ha⁻¹), and the selected Landsat model achieved R² = 0.95 and RMSE = 13.54 Mg ha⁻¹ on the test set. More importantly, direct comparison with the existing national 30 m annual AGB product using the same NFI-6–NFI-9 references showed lower RMSE in all four inventory cycles and a lower regional mean bias (+16.6 versus +36.4 Mg ha⁻¹), while adjacent-period biomass changes agreed in direction with NFI observations in 81.2% of province–period combinations. Airborne LiDAR, GEDI, and independent vegetation indicators further supported structural and temporal consistency. The reconstructed series reveals substantial long-term biomass and aboveground-carbon accumulation while preserving fine-scale heterogeneity across fragmented karst forests. These results demonstrate the value of regional calibration and multi-source validation for long-term quantitative biomass monitoring in environmentally heterogeneous landscapes.