Aboveground woody biomass regression and uncertainty estimation from spaceborne lidar and optical datasets

Woody plant encroachment continues to reshape rangeland ecosystems by altering forage availability, water balance, and carbon storage. Yet producing spatially explicit woody biomass maps that capture this fine‑scale heterogeneity remains a major challenge, particularly in regions with sparse field data. Here, we develop a cloud-based workflow that integrates the structural sensitivity of ICESat-2 canopy height measurements with the spatial continuity of multispectral imagery and environmental predictors to estimate aboveground woody biomass at 30 m resolution. To address limitations in field sampling, we derived training labels from post-fire cumulative net primary productivity (NPP), converted these values to biomass using empirical allocation factors, and trained a Random Forest model. Our approach outperforms GEDI L4A and L4B biomass products when evaluated against U.S. Forest Service FIA plots (RMSE = 5.66 vs 6.99 and 8.07 Mg/ha) and provides continuous, higher-resolution estimates across grassland, savanna, and woodland strata. While this framework enables reproducible wall-to-wall biomass mapping in data-poor regions, we emphasize that such maps may introduce spatial biases and should not be used directly for official carbon accounting without further validation. Instead, the workflow is well suited for applications requiring higher-resolution structural information, including spatial planning, habitat assessment, fuel load estimation, and local-scale fire risk modeling. These biomass surfaces may also support investigations of ecosystem functions that are tightly linked to woody cover. As new missions such as NISAR deliver L‑band SAR data sensitive to woody vegetation and soil moisture, integrating these datasets with ICESat‑2 derived datasets and optical imagery will further enhance biomass estimation and disturbance monitoring in rangeland systems.