Abstract
NASA’s Global Ecosystem Dynamics Investigation (GEDI) has proven successful in monitoring forest structure and biomass across national and regional scales, but whether these products provide useful estimates at the scale of individual forest properties—where management decisions are made and where a substantial portion of forest carbon, habitat, and disturbance risk remains under-assessed—remains largely untested. This study evaluates three GEDI Level 4 biomass products — L4A footprint (25 m diameter), L4D imputed 30 m grid, and L4B modeled 1 km grid — against an airborne laser scanning (ALS) baseline at the University of Idaho Experimental Forest (UIEF), a 3,117 ha temperate mixed-conifer forest in northern Idaho. Field biomass from 127 plots (2,447 trees) was measured during a 2024 forest carbon inventory compliant with American Carbon Registry, California Air Resources Board, Verra, and BioCarbon Fund standards (Corrao et al., 2026). Biomass was estimated using the USDA Forest Service National Scale Volume and Biomass (NSVB) system (Westfall et al., 2024), with Jenkins et al. (2003) equations applied in parallel to characterize allometric uncertainty. Primary results are reported using NSVB, with Jenkins estimates included for comparison. GEDI L4A showed moderate agreement with ALS NSVB (r = 0.68, R² = 0.35, RMSE = 84.2 Mg/ha, bias = +1.3 Mg/ha; n = 4,352) — consistent at the property level but imprecise at the stand level. At the 30 m pixel scale, L4D performed no better than predicting the property-level mean for every pixel (R² = −0.10 NSVB; −0.08 Jenkins). However, its aggregated property-level mean fell within 1% of the ALS NSVB equivalent (129.0 vs. 127.8 Mg/ha); against ALS Jenkins, L4D underestimated by 6.7% (129.0 vs. 138.3 Mg/ha). L4B overestimated biomass by 5.3% against ALS Jenkins and 14.3% against ALS NSVB at the 1 km landscape scale, while showing the lowest RMSE of any GEDI comparison (32.9 Mg/ha against Jenkins; 34.7 Mg/ha against NSVB; n = 29). Across 127 plots, L4A underestimated field NSVB biomass by 4.7% (143.1 vs. 150.2 Mg/ha) and field Jenkins biomass by 15.5%, with plot-level agreement degrading at greater plot–footprint separations consistent with UIEF's structural heterogeneity. Jenkins estimates were 7–33% higher than NSVB across strata, comparable in magnitude to GEDI-to-ALS disagreement at this site, demonstrating that the equation chosen can matter as much as the sensor used for property-scale biomass accounting at UIEF. These results demonstrate that GEDI products cannot match the spatial precision of ALS or field surveys at fine scales but do provide property-mean estimates closely aligned with both — a capability that could lower the information and cost barriers currently excluding much of the roughly 269 million acres of U.S. family forestland from carbon programs, conservation enrollment, and informed management.