Corroyez et al. (2026) Discrepancies between Sentinel-2 and LiDAR-derived LAI in temperate deciduous forests: Influence of apparent optical depth, radiative transfer priors, and horizontal heterogeneity

Identification

• Journal: Remote Sensing of Environment
• Year: 2026
• Date: 2026-10-01
• Authors: Nathan Corroyez, Jean‐Baptiste Féret, Jérôme Ogée, Jonathan Lenoir, Sylvie Durrieu
• DOI: 10.1016/j.rse.2026.115698

Research Groups

• UMR TETIS, Université Montpellier, AgroParisTech, CIRAD, CNRS, INRAE, Montpellier, France
• UMR ISPA, INRAE, Bordeaux Sciences Agro, Villenave d'Ornon, France
• UMR CNRS 7058 « Ecologie et Dynamique des Systèmes Anthropisés » (EDYSAN), Université de Picardie Jules Verne, Amiens, France

Short Summary

This study quantifies and reduces discrepancies between Leaf Area Index (LAI) derived from Sentinel-2 (S2) and airborne LiDAR (ALS) in temperate deciduous forests by accounting for sensor sampling asymmetry, optimizing radiative transfer model (RTM) parameters, and assessing canopy horizontal heterogeneity. It finds that S2 LAI is most consistent with LiDAR LAI when the latter is restricted to the upper 6–10 meters of canopy, and that RTM parameter optimization significantly improves inter-sensor agreement, with transferability across sites.

Objective

• To investigate the agreement between LAI estimated from airborne LiDAR (LAIALS) and Sentinel-2 (LAIS2) in three distinct French temperate deciduous forested areas, with particular emphasis on determining the apparent canopy thickness to which S2 reflectance is sensitive.
• Hypothesis 1 (H1): S2 primarily captures information from the uppermost layers of canopy thickness, leading to systematic underestimation of LAIS2 compared to LAIALS, which can be corrected using ALS data.
• Hypothesis 2 (H2): Adapting the RTM inversion parametrisation to account for the apparent canopy thickness sensed by S2 will improve consistency with LAIALS.
• Hypothesis 3 (H3): Forest structural characteristics, such as horizontal heterogeneity, may challenge LAI estimation assumptions and contribute to inter-sensor inconsistencies.

Study Configuration

• Spatial Scale: Three temperate deciduous forests in Metropolitan France (Aigoual, Blois, Mormal). Sentinel-2 data at 10-meter (m) resolution. LiDAR data processed to 1-m resolution DTM and CHM, with analyses at 10-m S2 pixel scale.
• Temporal Scale: LiDAR data acquired in early summer 2021 (Aigoual: 2021-07-18; Blois: 2021-06-17; Mormal: 2021-06-16 + 2021-07-19). Sentinel-2 data acquired closest in time to LiDAR flights (Aigoual: 2021-07-11; Blois: 2021-06-14; Mormal: 2021-06-14), corresponding to the leaf-on period with maximum LAI.

Methodology and Data

• Models used:
  • PROSAIL: Hybrid inversion of the PROSPECT-D leaf optical properties model coupled with the 4SAIL canopy reflectance model for Sentinel-2 LAI estimation.
  • Gap fraction method: Used to derive Leaf Area Density (LAD) profiles and LAI from airborne LiDAR point clouds.
  • Support Vector Regression (SVR): Ensemble method based on bootstrap aggregating for machine learning regression in PROSAIL inversion.
• Data sources:
  • Airborne LiDAR (ALS) data: Acquired with a VQ780 II RIEGL full-waveform LiDAR sensor, with a point density of 25–50 points per square meter (m²).
  • Sentinel-2 (S2) Level 2A reflectance product: Atmospherically corrected surface reflectance.
  • BD Forêt V2: French National Institute of Geographic and Forest Information (IGN) database for forest composition and density.

Main Results

• Initial comparison between LAIALS and LAIS2ATBD (default PROSAIL parameters) showed weak to moderate linear relationships (Pearson r = 0.11 to 0.54) and LAIS2ATBD saturation at 6–7 m²/m² while LAIALS reached up to 15 m²/m².
• Optimal canopy depth (dopt) for LAIALS integration, providing best agreement with LAIS2_ATBD, ranged from 6 m (Blois) to 10 m (Mormal), with 8 m at Aigoual, systematically lower than full canopy height. CHM-based LiDAR normalization improved agreement.
• Optimizing PROSAIL parameters (Average Leaf Angle, Leaf Brown Pigment Content, LAI distribution) through a multi-criteria Pareto framework reduced inter-sensor RMSE by 1.2 m²/m² and Bias by 1.54 m²/m² on average, relative to the ATBD baseline. Leaf Mass per Area (LMA) could be retained at default ATBD values.
• An optimized PROSAIL configuration, calibrated on two sites (excluding LiDAR-derived LAI priors), achieved 62% of the maximum attainable RMSE reduction and 57% of the maximum Bias reduction when transferred to an independent third site.
• Residual discrepancies between LAIALS and LAIS2 remained largest in stands with strong horizontal heterogeneity (high standard deviation of Digital Surface Model, DSMSD) within S2 pixels, particularly at Aigoual and Mormal.

Contributions

• Quantified the "apparent canopy thickness" (dopt, 6–10 m) to which Sentinel-2 reflectance is most sensitive in temperate deciduous forests, highlighting the asymmetry in canopy sampling between passive optical and active LiDAR sensors.
• Provided a systematic optimization of PROSAIL radiative transfer model parameters for forest environments, identifying that Average Leaf Angle, Leaf Brown Pigment Content, and LAI distribution benefit from forest-specific tuning, while Leaf Mass per Area can retain default values.
• Demonstrated the transferability of optimized PROSAIL configurations across different temperate deciduous forest sites, showing significant improvements in inter-sensor LAI consistency without site-specific LiDAR-derived priors.
• Assessed the combined influence of canopy thickness, PROSAIL parametrization, and horizontal heterogeneity on LAI retrieval accuracy, emphasizing the persistent challenges in heterogeneous stands.
• Suggested that future forest monitoring efforts could benefit from LiDAR-informed calibration of hybrid radiative transfer model inversion to enhance LAI retrieval from passive optical sensors.

Funding

• INRAE's MathNum Department
• ANR (MaCCMic project: ANR-21-CE32-0012)
• CNES (grants number 4500073693, 4500082679 and 4500082682)
• ANR JCJC, IMPRINT project (ANR-19-CE32-0005)
• “France 2030” research programme FORESTT (grant number ANR-24-PEFO-0003; MONITOR project)

Citation

@article{Corroyez2026Discrepancies,
  author = {Corroyez, Nathan and Féret, Jean‐Baptiste and Ogée, Jérôme and Lenoir, Jonathan and Durrieu, Sylvie},
  title = {Discrepancies between Sentinel-2 and LiDAR-derived LAI in temperate deciduous forests: Influence of apparent optical depth, radiative transfer priors, and horizontal heterogeneity},
  journal = {Remote Sensing of Environment},
  year = {2026},
  doi = {10.1016/j.rse.2026.115698},
  url = {https://doi.org/10.1016/j.rse.2026.115698}
}