Zhou et al. (2025) Global uncertainty assessment of vegetation indices from NASA's Harmonized Landsat and Sentinel-2 Project

Identification

• Journal: Remote Sensing of Environment
• Year: 2025
• Date: 2025-10-17
• Authors: Qiang Zhou, C. S. R. Neigh, Junchang Ju, Margaret Wooten, Zhe Zhu, Tomoaki Miura, Petya Campbell, Madhu Sridhar, Bradley W. Baker, Rodrigo Vieira Leite
• DOI: 10.1016/j.rse.2025.115084

Research Groups

• NASA Goddard Space Flight Center (including Science Systems and Applications, Inc. and NASA Postdoctoral Program Fellows)
• Earth System Science Interdisciplinary Center, University of Maryland, College Park
• Department of Natural Resources and the Environment, University of Connecticut
• Department of Natural Resources and Environmental Management, University of Hawai’i at M¯anoa
• Joint Center for Earth Systems Technology (JCET), University of Maryland, Baltimore County
• NASA Marshall Space Flight Center
• University of Alabama in Huntsville

Short Summary

This study globally assessed the between-sensor uncertainties of 21 Vegetation Indices (VIs) derived from NASA's Harmonized Landsat and Sentinel-2 (HLS) version 2.0 surface reflectance data. It found high consistency (R² > 0.94) for most VIs, but uncertainties increased with large view azimuth angle differences, high solar zenith angles, terrain shadows, and elevated aerosol levels, particularly at the extreme ends of VI value ranges.

Objective

• To evaluate the between-sensor discrepancies of 21 HLS V2.0-derived Vegetation Indices (VIs) from Landsat 8–9 (L30) and Sentinel-2 (S30) surface reflectance data.
• To investigate the underlying drivers of these discrepancies, including view geometry, solar zenith angle, terrain effects, and atmospheric conditions (aerosol levels), to inform future algorithm improvements.
• To quantify the uncertainties associated with VIs across their applicable value ranges and provide recommendations for robust scientific applications.

Study Configuration

• Spatial Scale: Global, covering 73°N to 40°S latitude, with a 30 meter spatial resolution. Analysis was based on a random sample of over 136 million cloud-free observations from 545 same-day L30 and S30 image pairs representing diverse landscapes.
• Temporal Scale: Data acquired during 2021 and 2022. HLS V2.0 provides revisit observations every 1.6 days globally and every 2.2 days in the tropics when all four satellites (Landsat 8–9 and Sentinel-2 A/B) are operational.

Methodology and Data

• Models used:
  • 6S-based Land Surface Reflectance Code (LaSRC) for atmospheric correction.
  • Bidirectional Reflectance Distribution Function (BRDF) correction for nadir BRDF-adjusted reflectance.
  • Inter-sensor bandpass adjustment to align Sentinel-2 MSI with Landsat 8 OLI.
  • Hampel filter for temporal outlier detection in VI time series.
  • Reduced Major Axis (RMA) regression for assessing overall consistency (R²).
  • LightGBM regression model to quantify the relative contributions of uncertainty factors (view azimuth angle difference, solar zenith angle, aerosol levels) to VI discrepancies.
• Data sources:
  • NASA’s Harmonized Landsat and Sentinel-2 (HLS) version 2.0 Landsat 8–9 30 meter (L30) and Sentinel-2 30 meter (S30) surface reflectance data.
  • HLS quality assessment (QA) layer for cloud, cloud shadow, water, and snow/ice masking, and qualitative aerosol level information.
  • European Space Agency (ESA) WorldCover 2021 product (10 meter resolution) for land cover type stratification.

Main Results

• Overall Consistency: Most VIs showed high between-sensor consistency (R² > 0.94), with the exception of the Chlorophyll Vegetation Index (CVI, R² = 0.5). Most VIs had low discrepancies (RMSDIQR < 0.1), but CVI (0.5822), CIG (0.2206), FCVI_VIS (0.2147), and RVI (0.2144) showed higher discrepancies.
• Impact of View Azimuth Angle Difference (VAD): A slight increase in Mean Absolute Difference (MAD) (≤0.01) was observed for most VIs when VAD exceeded approximately 125°, indicating potential limitations in BRDF adjustment.
• Impact of Solar Zenith Angle (SZ): High SZ (> ~60°), prevalent during winter, increased MAD by less than 0.07 and RMSEIQR by less than 0.2 for most VIs. CVI and CIG showed the highest increases (MAD of 0.386 and 0.203, respectively; CVI RMSEIQR increase of 0.477).
• Impact of Terrain Shadow: Significant discrepancies (relative difference > 20 %) were found in terrain shadow areas, particularly over dense vegetation, attributed to extremely low visible band surface reflectance and atmospheric correction challenges.
• Impact of Aerosol Level: VIs derived from moderate-level aerosol conditions closely aligned with the low-level aerosol baseline. However, high aerosol levels introduced evident discrepancies (e.g., NDVI MD up to ±0.0478, EVI MD up to ±0.0676), highlighting increased uncertainty.
• Uncertainty across VI Values: Even for low-level aerosol observations, uncertainties (MAD) increased at both the low and high ends of VI value ranges.
• Relative Contributions of Uncertainty Factors: VAD was the most influential factor for most VIs, followed by SZ. The influence of aerosols was consistently less significant.
• Recommended Valid Value Ranges: Specific value ranges for consistent between-sensor VI retrievals were proposed (e.g., NDVI: 0.1–0.94, EVI: 0.09–0.95).

Contributions

• Provided the first systematic, global assessment of between-sensor uncertainty for 21 HLS V2.0-derived Vegetation Indices, including the nine indices in the HLS VI suite.
• Quantified the impact of key environmental and geometric factors (view azimuth angle difference, solar zenith angle, terrain shadow, and aerosol levels) on HLS VI uncertainties.
• Identified specific VI value ranges where between-sensor consistency is robust, offering practical guidance for scientific applications.
• Delivered standard deviations of discrepancies, stratified by aerosol level and VI value, enabling users to incorporate uncertainty quantification into their analyses.
• Informed the ongoing refinement of HLS production algorithms, particularly for BRDF adjustment and atmospheric correction.

Funding

• NASA’s Satellite Needs Working Group (SNWG) Harmonized Landsat and Sentinel-2 project
• NASA’s Land Cover Project Science Office

Citation

@article{Zhou2025Global,
  author = {Zhou, Qiang and Neigh, C. S. R. and Ju, Junchang and Wooten, Margaret and Zhu, Zhe and Miura, Tomoaki and Campbell, Petya and Sridhar, Madhu and Baker, Bradley W. and Leite, Rodrigo Vieira},
  title = {Global uncertainty assessment of vegetation indices from NASA's Harmonized Landsat and Sentinel-2 Project},
  journal = {Remote Sensing of Environment},
  year = {2025},
  doi = {10.1016/j.rse.2025.115084},
  url = {https://doi.org/10.1016/j.rse.2025.115084}
}