The Forest Carbon Estimation (Force) Project: Mapping GEDI-derived Forest Structure Metrics in the U.S. and Canada with Plot-based Inventory and Multimodal Remote Sensing Data in a Hierarchical Spatial Modeling Framework

Forests take up and store carbon thereby slowing the build-up of anthropogenic greenhouse gases in the atmosphere. Scientists and managers estimate carbon in forests in order to quantify sequestration and support offset programs. Carbon measurements using traditional, ground-based forest inventories are limited by their area covered and frequency of data collection. NASAs Global Ecosystem Dynamics Investigation (GEDI) instrument on the International Space Station (ISS) overcomes these limitations by using laser pulses (LiDAR) to produce detailed datasets of vegetation structure, which offer new opportunities to better monitor forests. While the GEDI mission is greatly expanding the spatio-temporal coverage of observations, it is a sampling tool that collects LiDAR data in snapshots along the orbital track of the ISS. Spatial modeling methods need to be developed for integrating GEDI measurements with other, wall-to-wall satellite data in order to fill the gaps in the observations and produce accurate and reliable, full coverage maps of forest metrics. Our new Forest Carbon Estimation (FORCE) project demonstrates the development of such a modeling framework. The method follows a two-stage modeling approach where satellite imagery (e.g., Landsat and Sentinel-2) are used in a joint regression model to spatially predict gridded surfaces of GEDI metrics (e.g., canopy cover and vertical profiles), which themselves are in turn used to predict forest metrics (e.g., aboveground biomass) based on in situ, plot-based inventory data. The outputs of the Bayesian modeling are wall-to-wall biomass and forest structure prediction maps that can be spatially aggregated to produce statistically robust summary estimates with associated confidence intervals. Here we present the initial results of this approach applied at intensive study sites in Minnesota, Ontario, Maine, and New Brunswick where we have developed biomass and other forest structure maps based on high-resolution airborne LiDAR (e.g., G-LiHT) calibrated to high-quality ground data collected from inventory plot networks. These pilot studies will then be upscaled to map biomass carbon change over a large region in northeastern North America encompassing the managed forests of the temperate-boreal transition.