Hierarchical upscaling forest biomass mapping by fusing UAV-LiDAR, calibrated GEDI, and ALOS-2 PALSAR-2 data

First, a high-precision local reference layer was generated using UAV LiDAR and field plots to serve as reliable ground truth. Next, this layer was used to correct GEDI L4A products using machine learning models, and then the approach with best performance was selected for footprint-level estimation. Finally, corrected GEDI footprints were combined with PALSAR-2 structural metrics and Sentinel-2 spectral features to produce a wall-to-wall AGB map with resolution of 30 m.

Results showed that UAV-based correction significantly improved GEDI accuracy, yielding an <i>R</i>² of 0.78 and a root mean square error (RMSE) of 19.97 Mg/ha, the total uncertainty with pixel-level was controlled within 36.26 Mg/ha. In conclusion, the proposed framework, leveraging physical mechanism feature guidance, offers a novel and robust strategy for the high-precision inversion of forest AGB across large-scale with complex topography and high spatial heterogeneity.</p>