Pullanagari et al. (2025) Mapping of sun-induced fluorescence (SIF) in kiwifruit canopy using a 3D radiative transfer modeling and airborne hyperspectral imaging

Identification

• Journal: Remote Sensing Applications Society and Environment
• Year: 2025
• Date: 2025-12-14
• Authors: Reddy R. Pullanagari, Mohammad Hossain Dehghan-Shoar, Junqi Zhu, Alvaro A. Orsi, Ian J. Yule
• DOI: 10.1016/j.rsase.2025.101840

Research Groups

• Australian Plant Phenomics Network (APPN), The Plant Accelerator, School of Agriculture. Food and Wine, Adelaide University, Waite Campus, Urrbrae, SA, Australia
• Institute of Environmental Science and Research Limited (ESR), Mt Albert Science Centre, Auckland, New Zealand
• New Zealand Institute for Bioeconomy Science Limited, Marlborough Research Centre, Blenheim, New Zealand
• Stoneleigh Consulting Limited, Whakamarama, New Zealand

Short Summary

This study developed a hybrid 3D radiative transfer model (LESS-KRR) to accurately map sun-induced fluorescence (SIF) in complex kiwifruit canopies using airborne hyperspectral and LiDAR data, demonstrating superior performance over empirical methods for precision agriculture applications.

Objective

• To investigate sun-induced fluorescence (SIF) in kiwifruit orchards by employing multiple radiative transfer models (RTMs) and a hybrid 3D RTM-Kernel Ridge Regression (KRR) framework, aiming to overcome challenges posed by complex canopy structures and weak SIF signals for improved physiological activity estimation.

Study Configuration

• Spatial Scale: Kiwifruit orchard in Matapihi region of Tauranga, New Zealand (37°41′46.0″S, 176°12′15.4″E). Canopy reconstruction used 1 meter by 1 meter grids. LiDAR point cloud data was analyzed within 1.5 meter × 1.5 meter spatial windows.
• Temporal Scale: Aerial surveys were conducted in December 2022 and February 2024 during midday under clear skies. Field campaigns for ground-truth measurements were conducted concurrently. Weather data was collected in December, with daytime temperatures up to 30.4 °C and 116 mm of rain.

Methodology and Data

• Models used:
  • Radiative Transfer Models (RTMs): PROSPECT-PRO, FLUSPECT, LESS (LargE-Scale remote sensing data and image Simulation Framework), MODTRAN (for atmospheric correction).
  • Functional-Structural Plant Model (FSPM): GroIMP (3D Growth Grammar-related Interactive modelling platform).
  • Machine Learning: Kernel Ridge Regression (KRR).
  • SIF Retrieval Methods for comparison: Standard Fraunhofer Line Depth (sFLD), Three-band Fraunhofer Line Depth (3FLD), Improved Fraunhofer Line Depth (iFLD), Spectral Fitting Method (SFM).
• Data sources:
  • Airborne Hyperspectral Imagery: AisaFENIX sensor (350 nm to 2500 nm, 448 bands, full width at half-maximum (FWHM) of 4.8 nm from 377 nm to 968 nm and 12 nm from 972 nm to 2500 nm).
  • Airborne LiDAR Data: Hovermap LiDAR mounted on an Unmanned Aerial Vehicle (UAV) for 3D canopy structure, leaf area index (LAI), leaf inclination angle, and leaf area density (LAD).
  • Ground-truth Measurements:
    • Light-adapted leaf-level fluorescence quantum yield (MultispeQ).
    • Solar irradiance (ASD FieldSpec4 spectroradiometer with cosine corrector).
    • Georeferenced LAI measurements (Delta-T Devices SunScan System).
    • Leaf biochemical traits (chlorophyll, nitrogen, phosphorus, sodium, potassium, calcium, chloride ions).
  • Simulated Data: A comprehensive look-up table (LUT) of 100,000 simulations generated using PROSPECT-PRO, FLUSPECT, and LESS RTMs, with input parameters sampled using Latin hypercube sampling.

Main Results

• The hybrid 3D RTM-KRR model achieved the highest accuracy for SIF retrieval (coefficient of determination R² = 0.72, normalized root mean square error nRMSE = 9.3%) when validated against ground-truth data, outperforming empirical methods.
• Among the Fraunhofer Line Discrimination (FLD) methods, 3FLD achieved the highest accuracy (R² = 0.63, nRMSE = 10.2%), while the Spectral Fitting Method (SFM) showed slightly higher accuracy (R² = 0.65, nRMSE = 9.8%).
• The 3D RTM-KRR model successfully generated a spatially continuous SIF map across the orchard (ranging from 0 to 8 mW·m⁻²·sr⁻¹·nm⁻¹) without artifacts, providing insights into physiological heterogeneity.
• SIF exhibited strong positive correlations with leaf chlorophyll content (Pearson correlation coefficient r > 0.77).
• Moderate positive correlations were observed between SIF and leaf concentrations of nitrogen, phosphorus, and sodium.
• Negative correlations were identified between SIF and concentrations of potassium, calcium, and chloride ions.

Contributions

• Developed and validated a novel hybrid LESS-KRR framework for accurate SIF mapping in heterogeneous kiwifruit canopies, integrating 3D radiative transfer modeling with machine learning.
• Demonstrated the superior performance of the 3D RTM-KRR approach over traditional empirical (FLD) and spectral fitting methods for SIF retrieval from airborne hyperspectral data.
• Provided a robust methodology for reconstructing realistic 3D canopy structures using LiDAR data and functional-structural plant modeling (GroIMP), enhancing the fidelity of RTM simulations.
• Generated high-resolution spatial SIF maps for kiwifruit orchards, offering valuable insights into physiological heterogeneity for precision agriculture.
• Quantified the relationships between SIF and key biochemical traits (chlorophyll, N, P, Na, K, Ca, Cl), advancing understanding of SIF as an integrative indicator of canopy physiological status in horticultural settings.

Funding

• MBIE, New Zealand, Smart Ideas programme (contract number SCONU2101).

Citation

@article{Pullanagari2025Mapping,
  author = {Pullanagari, Reddy R. and Dehghan-Shoar, Mohammad Hossain and Zhu, Junqi and Orsi, Alvaro A. and Yule, Ian J.},
  title = {Mapping of sun-induced fluorescence (SIF) in kiwifruit canopy using a 3D radiative transfer modeling and airborne hyperspectral imaging},
  journal = {Remote Sensing Applications Society and Environment},
  year = {2025},
  doi = {10.1016/j.rsase.2025.101840},
  url = {https://doi.org/10.1016/j.rsase.2025.101840}
}