Assessment and improvement of GEDI canopy height estimation in tropical and temperate forests

Abstract

The Global Ecosystem Dynamics Investigation (GEDI) relative height product provides 25 m diameter footprint information that can be used to estimate canopy height but with variably reported accuracy. A methodology is presented to improve the GEDI canopy height accuracy using coincident Airborne Laser Scanner (ALS) data. The GEDI canopy height for each footprint is surface adjusted by adding the residual difference between the GEDI ground elevation estimate and coincident footprint ALS digital terrain model (DTM) value, and then calibrated using site-level additive offsets derived by Theil-Sen regression with coincident footprint ALS canopy height model (CHM) data. The approach is demonstrated at a tropical evergreen lowland forest site in the Democratic Republic of Congo (MNDP), a temperate pine and hardwood forest site in Alabama (TALL), and a temperate mix-species forest site in Maryland (SERC). The GEDI canopy height accuracy is first quantified by comparison with ALS CHM data to provide a baseline. The methodology improved the root mean squared error (RMSE) from 7.5m to 3.4m (MNDP), 5.6m to 3.5m (TALL), and 6.7m to 4.3m (SERC), and improved the relative RMSE from 33.0 % to 14.8 % (MNDP), 27.4 % to 17.3 % (TALL), and 24.9 % to 15.8 % (SERC), for GEDI beam sensitivity ≥0.9, with similar improvements demonstrated for beam sensitivity ≥0.95 and ≥ 0.98. Further accuracy improvements were demonstrated for footprints over homogenous canopies where the simulated impact of GEDI geolocation errors were small, underscoring the need for improved GEDI geolocation. Extrapolation of the methodology to regional or national scale merits further research and is discussed.