Qi et al. (2025) Retrieving fine-scale leaf and soil spectral properties from canopy reflectance with differentiable 3D radiative transfer

Identification

• Journal: Remote Sensing of Environment
• Year: 2025
• Date: 2025-11-04
• Authors: Jianbo Qi, Jifan Wei, Meng Xia, Donghui Xie
• DOI: 10.1016/j.rse.2025.115115

Research Groups

• Faculty of Geographical Science, Beijing Normal University, China
• Yale University, USA

Short Summary

This study introduces 3D-Diff, a novel differentiable three-dimensional (3D) radiative transfer approach, to accurately retrieve fine-scale leaf and soil spectral properties from canopy reflectance using remote sensing imagery and known 3D canopy structures, demonstrating robust performance in both virtual and real-world validations.

Objective

• To develop and validate an innovative differentiable 3D radiative transfer approach (3D-Diff) for accurately estimating leaf and soil spectral properties or biochemical parameters from canopy-level reflectance observations, addressing the challenges posed by complex canopy structures.

Study Configuration

• Spatial Scale: Fine-scale, with validation experiments showing accuracy differences at 0.05 meters and 2.5 meters spatial resolutions.
• Temporal Scale: Not explicitly defined, but the method is applicable to retrieving properties for vegetation mapping and monitoring applications.

Methodology and Data

• Models used: 3D-Diff (differentiable three-dimensional radiative transfer approach) implemented within the LESS (LargE-Scale remote sensing data and image Simulation) model. It employs computer graphics-derived differentiable modeling and a gradient-based optimization algorithm.
• Data sources: Virtual experiments and real measurements.

Main Results

• 3D-Diff accurately estimated the reflectance/transmittance of all scene parameters, including non-directly-observed elements.
• For virtual experiments, the maximum Root Mean Square Error (RMSE) was 0.015.
• For real measurements, the maximum RMSE was 0.096.
• Spatial resolution significantly impacted accuracy: relative RMSE values were 0.07 % for a 0.05 meter resolution and 1.4 % for a 2.5 meter resolution in a dense virtual forest scene.
• The approach, while computationally slower than analytic models, establishes differentiable radiative transfer as a promising framework for fine-scale vegetation mapping, enabling simultaneous multi-parameter retrieval with physical interpretability.

Contributions

• Introduction of 3D-Diff, an innovative differentiable 3D radiative transfer approach, for retrieving fine-scale leaf and soil spectral properties from canopy reflectance.
• Demonstration of robust performance in retrieving multiple scene parameters, including non-directly-observed elements, through both virtual experiments and real measurements.
• Establishment of differentiable radiative transfer as a promising framework for fine-scale vegetation mapping, offering simultaneous multi-parameter retrieval while maintaining physical interpretability.

Funding

• Not specified in the provided text.

Citation

@article{Qi2025Retrieving,
  author = {Qi, Jianbo and Wei, Jifan and Xia, Meng and Xie, Donghui},
  title = {Retrieving fine-scale leaf and soil spectral properties from canopy reflectance with differentiable 3D radiative transfer},
  journal = {Remote Sensing of Environment},
  year = {2025},
  doi = {10.1016/j.rse.2025.115115},
  url = {https://doi.org/10.1016/j.rse.2025.115115}
}