Validation of the vertical plant area index profile product derived from GEDI over global forest sites

Knowledge of the vertical plant area index (PAI) profile is critical for understanding the forest structural and functional characteristics. Vertical PAI profile has been retrieved by the Global Ecosystem Dynamics Investigation (GEDI) spaceborne LiDAR. However, large-scale validation of the GEDI PAI profile products is limited, and their performance has yet to be clearly established. This study aims to systematically assess the performance of GEDI PAI profile product and investigate the impact factors on PAI profile estimates. The digital hemispherical photography (DHP) of vertical measurement and airborne laser scanning (ALS) data were collected to derive the reference PAI profiles. The results indicate that adjusting footprint geolocation before GEDI validation is essential for enhancing product assessment. The GEDI PAI profile moderately agrees with the DHP and ALS (R² = 0.84 and 0.58, respectively) but underestimates the reference (bias = −0.14 and −0.28, respectively). The needleleaf forest exhibits the highest agreement with ALS (R² = 0.60 and bias = −0.16), while shrubland shows the lowest agreement (R² = 0.38 and bias = 0.21). The agreement between GEDI and ALS increases with the canopy height but decreases with the canopy cover. Low vegetation height and steep slopes affect the GEDI PAI accuracy owing to the difficulty in decomposing the mixed ground and canopy returns. Additionally, the limited penetration of GEDI in dense vegetation with high canopy cover contributes to the underestimation. The performance of GEDI PAI profile can be improved by applying a specific canopy and ground reflectance ratio (${\rho }_v$/${\rho }_g$) value derived from the linear regression of return energy. The discrepancies between GEDI and ALS PAI profiles were partially attributed to the sub-optimal waveform processing algorithm settings and differences in LiDAR specifications. Further improvement to the GEDI PAI product may be achieved by implementing a customized waveform processing algorithm and using realistic ${\rho }_v$/${\rho }_g$ values.