Advancing aboveground biomass mapping in Alaska's boreal forest using UAVSAR polarimetric features and NDVI-normalized field data

Accurate estimation of aboveground biomass (AGB) is critical for quantifying carbon stocks and monitoring ecological change in Alaska's boreal forests, which are experiencing rapid environmental shifts due to warming temperatures and changing disturbance regimes. However, reliable biomass estimation in these ecosystems remains challenging due to strong spatial heterogeneity in forest structure, limited field accessibility, and persistent scale mismatches between ground observations and satellite remote sensing. In this study, we integrated field-based biomass measurements with polarimetric features from NASA's Uninhabited Aerial Vehicle Synthetic Aperture Radar (UAVSAR) to generate high-resolution AGB estimates in structurally complex mid- and late-successional boreal forest stands. We calculated plot-level biomass using species-specific allometric equations for the major tree species dominant in the study area and distributed this biomass to individual pixels using optical information from high-resolution PlanetScope imagery as a spatial weighting function. Unlike conventional SAR-optical fusion approaches that use optical data as direct predictive variables, this approach uses Normalized Difference Vegetation Index (NDVI) to guide the spatial redistribution of field-derived biomass, enabling sub-plot-scale representation of canopy heterogeneity. We used nine UAVSAR polarimetric parameters, representing different scattering mechanisms, along with a digital elevation model (DEM) as predictors in a Random Forest regression framework. Model training and validation were performed using independent datasets to assess robustness and generalizability. The resulting model achieved strong performance (R2 = 0.83; RMSE = 23.2 Mg ha−1) across independent test samples. Variable importance analysis showed that surface scattering (Ps) and elevation (DEM) were the most influential predictors, indicating that variations in ground conditions, topography, and radar-vegetation interactions strongly shape biomass patterns across the landscape, under the assumption that pixel is fully occupied by tree cover. The resulting 5 m spatial resolution biomass map captured fine-scale variability consistent with observed forest structure and vegetation patterns across the study area, demonstrating the ability of airborne SAR to resolve local-scale biomass differences in boreal environments. This study demonstrates a transferable geospatial approach for pixel-level biomass estimation using airborne SAR polarimetry combined with optical-guided spatial allocation and machine learning and provides a methodological foundation for future scaling using satellite missions such as the NASA-ISRO Synthetic Aperture Radar (NISAR).