Accuracy Assessment of GEDI and ICESat-2 Canopy Height Products by Excluding Geolocation Error Based on Virtual Sampling Units

The launch of two spaceborne light detection and ranging (LiDAR) spaceborne LiDAR (SL) missions, Ice, Cloud, and Land Elevation Satellite-2 (ICESat-2)/Advanced Topographic Laser Altimeter System (ATLAS) and Global Ecosystem Dynamics Investigation (GEDI), has enabled the generation of global canopy height products with unprecedented coverage and resolution. However, geolocation error remains a major source of uncertainty, hindering reliable validation and cross-comparison of these products. This study proposes a validation strategy based on a new concept, the “virtual sampling unit” (VSU), which identifies regions that support homogeneous canopy height sampling, while matching the spatial resolution and footprint geometry of both products, thereby minimizing the impact of geolocation errors on apparent height estimates. Using high-resolution airborne laser scanning (ALS)-derived canopy height models (CHMs) from multiple eco-climatic sites, 20 in the U.S. and one in China, as the baseline, we assess the performance of the latest GEDI and ICESat-2 canopy height products in terms of accuracy, consistency, and sensitivity to influencing factors. Our VSU-based framework effectively decouples error sources, revealing that nearly half of the apparent GEDI L2A canopy height error stems from geolocation uncertainty, which is larger than previously reported (1- $\sigma $ level), with many footprints deviating by over 20 m. We recommend using coarser mapping scales to mitigate its impact. In contrast, the inaccuracies of ICESat-2 ATL08 primarily arise from the photon cloud denoising and classification processes, while its geolocation remains largely unaffected. Crucially, after excluding geolocation error impact, GEDI (RMSE =3.30 m) outperforms ICESat-2 (RMSE =4.61 m) in terms of accuracy and consistency, as validated using data from all scenarios. ICESat-2 canopy height estimates display a systematic underestimation bias of ~1 m. GEDI showed higher stability regarding influencing factors, performing best in needleleaf forests (RMSE =2.88 m), whereas ICESat-2 shows higher accuracy at night (RMSE =3.84 m). Both products exhibit decreased accuracy in steep terrain and tall canopies. Overall, this study presents the first comprehensive assessment of the actual accuracy of SL canopy height products with minimized geolocation impact. The proposed VSU-based strategy enables a more objective validation means applicable to evolving SL products, supporting accurate forest structure mapping and carbon stock estimation.