Discrepancies between Sentinel-2 and LiDAR-derived LAI in temperate deciduous forests: Influence of apparent optical depth, radiative transfer priors, and horizontal heterogeneity

Accurate estimation of forest leaf area index (LAI) from satellite images remains challenging due to limited sunlight penetration and intricate scattering through structurally complex canopies. We compared LAI estimated from radiative transfer model PROSAIL hybrid inversion applied to Sentinel-2 (S2) images with LAI derived from airborne LiDAR across three temperate deciduous forests in France. We assessed three factors driving discrepancies between the two estimates: (1) the attenuation of the optical signal reducing the apparent canopy depth contributing to S2 reflectance, (2) the parametrisation of the PROSAIL inversion, and (3) the local horizontal heterogeneity in canopy height. Sentinel-2 LAI estimates most closely matched LiDAR-derived LAI when the latter was restricted to the upper 6–10 m of canopy thickness. Optimising the range of PROSAIL parameters through a multi-criteria Pareto framework reduced the reported inter-sensor RMSE values in LAI by 1.2 m 2 /m 2 and the Bias by 1.54m 2 /m 2 , on average, across sites and relative to the algorithm theoretical basis document (ATBD) baseline. Yet, residual discrepancies remained largest in stands with strong horizontal heterogeneity within S2 pixels at Aigoual and Mormal. Leaf mass per area could be retained at default values, while average leaf angle, leaf brown pigment content and the LAI distribution all required forest-adapted priors. Most of that improvement transferred between forested areas. A configuration based on two sites, with no LiDAR-derived prior, achieved 62% of the maximum attainable RMSE reduction and 57% of the Bias reduction when transferred to the third site. The LiDAR-derived LAI prior contributed to a maximum of 17% of the overall RMSE reduction across the three study sites. Our findings highlight the importance of jointly accounting for canopy thickness and horizontal heterogeneity when retrieving LAI from passive optical sensors, and suggest that future forest monitoring efforts may benefit from LiDAR-informed calibration of hybrid radiative transfer model inversion.