Validating Remotely Sensed Biomass with Forest Inventory Data in the Western US

Abstract

The expansion of global carbon offset markets and the growing impacts of wildfires have increased the importance of spatio-temporal extensive, detailed, and reliable remote-sensing aboveground biomass density (AGBD) mapping. The Global Ecosystem Dynamics Investigation (GEDI) mission provides valuable spaceborne LiDAR data for AGBD mapping. However, due to GEDI’s spatio-temporal extent, sparse footprint, and the limited field data, current remote-sensing AGBD products based on GEDI are limited in spatio-temporal extent and the lack of robust fine-resolution maps validated by independent field data. Although the US Forest Service’s Forest Inventory and Analysis (FIA) program provides extensive field data, the location-perturbed plots hinder the development of a rigorous validation framework using the FIA data. We have developed a high-resolution, time-serial AGBD dataset (terraPulse AGBD) with extensive spatio-temporal coverage, by imputing GEDI, ICESat with ecologically informed covariates derived from satellite imagery. In this study, we present a rigorous and replicable validation of this annual, 30-meter resolution AGBD dataset using independent estimates from the FIA in the states of Utah and Nevada, and Washington. Because FIA blurs plot coordinates, we validated the remotely sensed AGBD at two scales—the 64,000-hectare hexagon scale and the coarser county scale. At the hexagon scale, the linear model relating the remotely sensed AGBD and FIA AGBD estimates achieved an of 0.88, a slope of 0.99 (95% confidence 𝑅2interval: 0.97 ~ 1.02), an intercept of 9.79 (95% confidence interval: 8.05 ~ 11.53), an RMSE of 26.68, a correlation coefficient of 0.94 and an index of agreement of 0.82—indicated a strong significant correspondence. At the county scale, the linear model achieved an of 0.9, a slope 𝑅2of 1.07 (95% confidence interval: 0.98 ~ 1.15), an intercept of 12.56 (95% confidence interval: 4.16 ~ 20.96), an RMSE of 32.62, a correlation coefficient of 0.95, and an index of agreement of 0.85—indicated a better alignment than the hexagon scale. The geospatial analysis reveals higher terraPulse values in non-forested areas due to FIA’s lack of sampling, leading to lower AGBD. Conversely, terraPulse underestimates AGBD in extremely high-biomass forests at the hexagon scale. While at the county scale, terraPulse performed well at high-AGBD areas compared to the FIA, possibly because the saturation was offset by estimating more non-forest biomass. Our validation not only introduces novel, reliable remote-sensing data for monitoring AGBD with wide coverage and high resolution, but it also provides a rigorous framework for validating future remote-sensing AGBD maps.