Performance Assessment of GEDI and ICESat-2 Laser Altimeter Data for Terrain and Canopy Height Retrievals

Abstract:

Access to information on terrain height and vegetation structure plays a key role in understanding the distribution of carbon stocks and fluxes and in improving terrain models and habitat mapping. The Ice, Cloud and Land Elevation Satellite-2 (ICESat-2) launched in September 2018 and Global Ecosystem Dynamics Investigation (GEDI) laser altimeter placed in December 2018 both provided valuable data for large-scale, accurate terrain and vegetation observations. Combined, these two missions will allow for enhanced coverage and improved accuracy by leveraging the advantages of the individual instrument technologies and spatial coverage. To evaluate the performance of ICESat-2 and GEDI data in terrain and canopy height retrievals, we used DTM and CHM generated from airborne Lidar data from 40 observatory sites provided by the National Ecological Observatory Network (NEON) as references. All ICESat-2 ATL08 products and GEDI L2A products within the study area were then collected and processed. Both reference and test data were acquired in 2019 to reduce the uncertainty due to the temporal variability. The results show that GEDI performs weaker than ICESat-2 for terrain height retrieval, with RMSE of 4.93 and 1.91 m, respectively. For the canopy top height retrieval, GEDI performs similarly to ICESat-2, with %RMSE of 45.34 and 42.40 and R² of 0.53 and 0.54, respectively. The mean error indicates that both GEDI and ICESat-2 underestimate canopy height (-2.55 and -3.19 m) and that GEDI tends to overestimates topographic height (0.93 m). We also explored the effects of various factors (e.g., slope, canopy height, canopy cover, land cover type, whether strong beams or night-acquired data, etc.) on the accuracy of terrain and canopy height retrievals. The retrieval error increases with increasing slope, canopy height, and canopy cover. Among the different land cover types, forests and shrubs significantly reduced the accuracy of terrain retrieval, while grasslands and crops have less influence, with bareland and tundra (for ICESat-2 only) yielding the most accurate terrain heights.