Performance evaluation of GEDI and ICESat-2 canopy height products in subtropical monsoon forests of southern China

Abstract

Accurate assessment of spaceborne LiDAR canopy height products is essential for reliable large-scale forest height mapping. Although ICESat-2/ATL08 and GEDI/L2A products have been widely validated in temperate and boreal forests, their performance in subtropical monsoon forests remains insufficiently explored. This study evaluated the canopy height accuracy of ICESat-2/ATL08 and GEDI/L2A in subtropical monsoon forests of southern China using airborne LiDAR-derived canopy height models as reference. The results show that both products capture the general variability of forest height but exhibit distinct error characteristics. ICESat- 2/ATL08 demonstrates a clear tendency toward systematic canopy height overestimation, with larger dispersion, yielding an RMSE of 10.52 m and an MAE of 8.35 m, particularly under dense vegetation, tall canopies, and complex terrain. In contrast, GEDI/L2A achieves improved performance with a lower RMSE of 7.09 m and an MAE of 5.51 m, and exhibits a more balanced distribution around the 1:1 line, indicating reduced systematic bias and greater robustness. Further analyses reveal that increasing vegetation cover fraction, canopy height, slope, and elevation significantly degrade canopy height accuracy, with ICESat-2 being more sensitive to environmental complexity than GEDI. These findings provide a region-specific quantitative evaluation of ICESat-2 and GEDI canopy height products in subtropical monsoon forests of southern China and offer practical guidance for their application in forest height assessment under dense vegetation and complex terrain conditions.