Fawcett et al. (2025) Hyperspectral indicators of vegetation vitality across scales: From trees to forests

Identification

• Journal: Science of Remote Sensing
• Year: 2025
• Date: 2025-11-20
• Authors: Dominic Fawcett, Arthur Geßler, Katrin Meusburger, Christian Ginzler, David N. Steger, Ansgar Kahmen, Petra D’Odorico
• DOI: 10.1016/j.srs.2025.100336

Research Groups

• Swiss Federal Institute for Forest Snow and Landscape Research WSL, Birmensdorf, Switzerland
• Institute of Terrestrial Ecosystems, ETH Zürich, Zürich, Switzerland
• Department of Environmental Sciences – Physiological Plant Ecology, University of Basel, Basel, Switzerland

Short Summary

This study assesses how hyperspectral vegetation indices (VIs) resolve vitality-related differences between tree crowns across very high (0.1 m), high (1 m), and moderate (30 m) spatial resolutions. It finds that most VIs effectively resolve crown variations at 1 m resolution, and at 30 m, VIs sensitive to photochemistry, chlorophyll, and water content are responsive to available water capacity, though significantly influenced by vegetation structure and functional type.

Objective

• To investigate the ability of vegetation indices (VIs) sensitive to stress and vegetation vitality to represent differences between trees at increasingly coarser spatial resolutions.
• To assess the ability of VIs to resolve differences between crowns and forest areas when data is acquired at different spatial grains.
• To determine the spatial correlation of VIs with drivers including available water storage capacity (AWC), functional type, and other structure-related variables over Swiss forest landscapes.

Study Configuration

• Spatial Scale:
  • Very high resolution: 0.1 meter (drone-based, subplot-scale, 1–3 hectares)
  • High resolution: 1 meter (aircraft-based, landscape scale, 7–16 square kilometers)
  • Moderate resolution: 30 meters (simulated satellite data, landscape scale)
  • Analysis performed at individual tree crown level and aggregated to pixel level.
• Temporal Scale:
  • Data acquired during the peak of the growing season in 2023 and 2024.
  • Hyperspectral acquisitions: August 22, 2023, and August 28, 2024 (aircraft); within three days of aircraft data for drone.
  • Auxiliary data (ALS) from 2020 and 2021.

Methodology and Data

• Models used:
  • Ordinary Least Squares (OLS) regression
  • Multilinear models
  • Linear mixed effects model (with random effect for site and Matérn correlations for spatial autocorrelation)
  • Random forest (for functional type classification)
  • Gaussian low-pass filter (for spatial resampling to 30 meter resolution)
  • Radiometric and geometric correction (CaliGeoPro software)
  • Atmospheric and BRDF correction (ATCOR-4)
  • Empirical line method (for drone data reflectance)
  • Savitzky-Golay filtering (spectral smoothing)
  • Individual tree segmentation (local maximum sliding window, region growing algorithm in lidR package)
• Data sources:
  • Hyperspectral imaging data:
    • Drone-borne Specim AFX10 sensor (224 bands, 400–1000 nanometers, 1.3–1.4 nanometers FWHM, 0.08 meter Ground Sampling Distance)
    • Aircraft-borne Specim AisaFENIX sensor (87 bands VIS-NIR 400–970 nanometers, 277 bands SWIR 970–2500 nanometers, 5.5–6.9 nanometers FWHM, 1 meter Ground Sampling Distance)
  • Simulated satellite data (30 meter spatial resolution, derived from aircraft data)
  • Airborne Laser Scanning (ALS) data (0.5 meter resolution Canopy Height Model)
  • In-situ environmental measurements: Soil water potential (SWP) sensors (up to 0.4 meters or 2 meters depth)
  • Plant available water storage capacity (AWC) data (25 meter spatial resolution, derived from static soil properties, resampled to 30 meters)
  • Manually digitized tree crowns and georeferenced point-data for species assignment.

Main Results

• Strong linear relationships (R² = 0.46–0.86) were observed between crown-averaged VI values from 0.1 meter drone and 1 meter aircraft data, with higher correlations for European beech than for conifers (spruce, fir).
• VIs using shorter visible wavelengths (e.g., PRIm, NPCI) and those sensitive to red-edge wavelengths (REP, CHLred) showed lower correspondence or slopes different from 1 between scales.
• Within-crown shadows significantly influenced non-masked high-resolution VI values, particularly for MTVI, NPCI, PRI, and PRIm, with higher sensitivity for silver fir.
• Moderate to very high correspondence (R² = 0.57–0.89) was found between 1 meter aircraft data (crown-area weighted average) and simulated 30 meter satellite pixel values, with SWIR-based water content indices showing the strongest correlations.
• Multilinear models explained 63–90% of the 30 meter pixel VI variance, with shadow fraction having the greatest influence for MTVI, NPCI, PRIm, and PRI. Crown cover had a smaller influence, mainly on SWIR indices and NDVI.
• Landscape-scale VI variations were explained by AWC and structural parameters (functional type, crown cover, tree height, shadow fraction) with conditional R² values ranging from 0.10 to 0.88.
• AWC effects on VIs were larger and positive in 2024 compared to 2023, and more pronounced for broadleaf canopies than conifers.
• Functional type had a very strong effect on water content VIs. Tree height had a negative impact on most VIs.
• PRInorm showed potential as a less structure-sensitive PRI index, scaling similarly to PRI but less susceptible to within-crown shading and pixel shadow fraction.

Contributions

• Provides a comprehensive multi-scale assessment of hyperspectral VIs, bridging very high-resolution drone data to moderate-resolution simulated satellite data for forest vitality monitoring.
• Quantifies the significant impact of confounding structural effects (e.g., shadow fraction, functional type, tree height) on VI performance and interpretation across different spatial scales in structurally diverse forest ecosystems.
• Demonstrates the potential of 1 meter resolution aircraft-acquired hyperspectral data for mapping tree-level crown vitality differences, particularly for broadleaf species.
• Highlights the sensitivity of photochemical, chlorophyll, and water content VIs to available water storage capacity at the landscape scale, while emphasizing the necessity of accounting for vegetation structure in mixed forests.
• Offers crucial insights for leveraging current and future satellite hyperspectral missions for large-scale forest health assessment by understanding scale-related limitations and structural influences.

Funding

• Swiss State Secretariat for Education, Research and Innovation, SERI (22.00419, REF-1131-52104) for the EU Project FORWARDS.
• Swiss National Science Foundation (SNF) project FORDROUGHT (315230_208197).
• Swiss Long-term Forest Ecosystem Research programme LWF (part of UNECE Co-operative Programme on Assessment and Monitoring of Air Pollution Effects on Forests ICP Forests).

Citation

@article{Fawcett2025Hyperspectral,
  author = {Fawcett, Dominic and Geßler, Arthur and Meusburger, Katrin and Ginzler, Christian and Steger, David N. and Kahmen, Ansgar and D’Odorico, Petra},
  title = {Hyperspectral indicators of vegetation vitality across scales: From trees to forests},
  journal = {Science of Remote Sensing},
  year = {2025},
  doi = {10.1016/j.srs.2025.100336},
  url = {https://doi.org/10.1016/j.srs.2025.100336}
}