A quantitative evaluation of forest aboveground biomass density map products in Oregon, USA
| Authors: | Andrew T. Hudak, Jennifer Bakken, Andrew J. Lister, Francisco Mauro, Matthew J Gregory, Karin Riley, Barry T. (Ty) Wilson, Maurizio Santoro, Sean P. Healey, Lei Ma, Yifan Yu, Benjamin C. Bright, John C. Byrne, Roy Mita, Rachel Houtman, John D. Shaw, Patrick A Fekety, Grant M. Domke, Brian F. Walters, David M. Bell, Aaron Weiskittel, Jeff W. Atkins, Laura Duncanson, Neha Hunka, Robert Kennedy, Chad Babcock, Sassan Saatchi, Brett Raczka, Carlos Silva, Hao Tang, George Hurtt |
| Year: | 2026 |
| Type: | Scientific Journal |
| Station: | Rocky Mountain Research Station |
| DOI: | https://doi.org/10.1088/1748-9326/ae8b04 |
| Source: | Environmental Research Letters |
Abstract
Maps of aboveground biomass density (AGBD) estimated from remote sensing data with models trained from inventory data are essential for carbon monitoring. Forest inventory plots in the United States (US) are measured by the forest inventory and analysis (FIA) program of the US Forest Service (USFS). These plots are spatially representative, design unbiased, and provide estimates of AGBD and other forest attributes over large spatial domains, but do not provide spatially continuous or sufficiently precise estimates over small spatial domains (e.g. forest stands). Integration with remote sensing can fill this gap, but remote sensing cannot measure AGBD directly, so models predicting ground-measured AGBD as a function of remote sensing data are often used to generate maps. We evaluated 10 alternative types of AGBD maps that used LiDAR, multispectral satellite (Landsat) image time series, synthetic aperture radar (SAR), or alternative modeling approaches. We evaluated the agreement between mapped AGBD predictions and AGBD observations collected at the FIA plots or in stands surveyed on USFS lands. Ground observations were regressed on mapped AGBD predictions at different spatial aggregation levels: county, 64 000-ha FIA sample hexagon, stand polygon, FIA plot, and FIA subplot (unaggregated). Standardized major axis regression model fit statistics were calculated and compared to evaluate the map products across the study area, a large portion of western and central Oregon. Based on these fit statistics, we found that the Landsat-derived maps performed best at the coarser evaluation levels (county, hexagon), airborne LiDAR-based products performed best at finer evaluation levels (stand, plot, subplot), and the SAR-dominated products showed lower agreement with ground-based AGBD estimates in this high-biomass Oregon study area. When choosing between AGBD map types, we recommend map users consider the scale of their decision space as much as type of remote sensing data or modeling approach used in map production.