Forest Inventory and Analysis (FIA) data are collected nationally in the USA following a systematic design, making them spatially unbiased but too sparse for most forest managers working at the scale of most lidar-assisted forest inventory projects. However, the recent availability of spaceborne Global Ecosystem Dynamics Investigation (GEDI) footprints, which are spatially unbiased and approximately the same size as a standard (0.04 ha), fixed-radius forest inventory plot, provide an opportunity to correct for spatial sampling bias at the project level. Our objective was to estimate the sample bias in project area(s) in Oregon and Idaho with existing airborne lidar coverage, such that the covariance between the GEDI and project field plot datasets collected within the airborne lidar collection(s) can be calculated using a co-regionalization approach. We used multiple linear regression to predict aboveground biomass (AGB) from the airborne lidar metrics at the project-level field plots using the standard operational, small-area based method. We compared precision and accuracy (before and after bias correction using GEDI data) of our estimates to FIA plot-based estimates collected within the same lidar project extent(s). We hypothesized that the temporal uncertainty inherent in the FIA plots (10% visited per year on a rolling 10-year frequency) exceeds the spatial uncertainty in the project-level inventory plots.
Results/Conclusions
Based on preliminary analysis in north central Idaho using simulated GEDI footprints, separated by ~600 m across track and ~60 m along-track, GEDI footprints provided spatially discrete measurements of the forest canopy similar in size to forest inventory plots, but at a denser spatial sampling frequency than project-level forest inventory plots of fixed-radius, established following a stratified random landscape sampling design. This translated into a 95% or better probability that an even-aged, managed forest stand was intersected at least once by a GEDI footprint during its expected 2-year operational period. Availability of canopy height information from GEDI should therefore ease (but not eliminate) the number of traditional field plots required for unbiased forest inventory at more local scales. Models that incorporate co-regionalization approaches are optimal to generate wall-to-wall maps from the forest structural information collected at the GEDI footprint level. Given the broad extent of GEDI sample data within +/- 51 degrees latitude, there is great potential for unbiased forest inventory estimates globally, even in countries where national forest inventory (NFI) datasets such as FIA may be unavailable.