Liu et al. (2025) Physics-Driven Machine-Learning Retrieval and Uncertainty Quantification of Crop Leaf Area Index

⚠️ Warning: This summary was generated from the abstract only, as the full text was not available.

Identification

• Journal: Remote Sensing
• Year: 2025
• Date: 2025-12-04
• Authors: Wei Liu, Xiao‐Hua Zhu, S. Yang, Zhihai Gao
• DOI: 10.3390/rs17233924

Research Groups

Not explicitly mentioned in the provided text.

Short Summary

A physics-driven machine-learning framework, coupling the PROSAIL radiative transfer model with a genetic-algorithm-optimised multilayer perceptron, is developed for operational Leaf Area Index (LAI) retrieval and end-to-end uncertainty quantification, demonstrating improved accuracy and generalisation across different crop sites.

Objective

• To develop and validate a physics-driven machine-learning framework for operational Leaf Area Index (LAI) retrieval from Sentinel-2B data, coupled with end-to-end uncertainty quantification.

Study Configuration

• Spatial Scale: Minqin (entirely maize) and Zhangye (predominantly maize, with one sunflower plot) areas. LAI retrieval at 10 meter resolution.
• Temporal Scale: Not explicitly mentioned, but implies continuous monitoring for operational use.

Methodology and Data

• Models used: PROSAIL radiative transfer model, Genetic-Algorithm-optimised Multilayer Perceptron (NN–GA).
• Data sources: Synthetic training library (30,000 samples) generated by PROSAIL convolved with Sentinel-2B spectral response functions. Validation data from Sentinel-2B satellite imagery.

Main Results

• The framework achieved validation results of RMSE = 0.44 and R² = 0.73 in Minqin, and RMSE = 0.40 and R² = 0.56 in Zhangye, indicating reasonable cross-site generalisation.
• Total retrieval uncertainty was decomposed into a simulation component (PROSAIL parameter variability) and a training component (variability across repeated NN–GA trainings).
• Physical-driven contributions to uncertainty were 11.42% (Minqin) and 11.48% (Zhangye).
• Machine-learning contributions to uncertainty were 18.06% (Minqin) and 12.96% (Zhangye).
• The framework improves 10 meter LAI retrieval accuracy.

Contributions

• Development of a novel physics-driven machine-learning framework (PROSAIL coupled with NN–GA) for operational LAI retrieval.
• Implementation of an end-to-end uncertainty quantification method, decomposing uncertainty into physical-driven and machine-learning components.
• Demonstration of improved 10 meter LAI retrieval accuracy and reasonable cross-site generalisation.
• Provision of a reproducible, per-pixel uncertainty budget to guide product use and refinement.

Funding

Not explicitly mentioned in the provided text.

Citation

@article{Liu2025PhysicsDriven,
  author = {Liu, Wei and Zhu, Xiao‐Hua and Yang, S. and Gao, Zhihai},
  title = {Physics-Driven Machine-Learning Retrieval and Uncertainty Quantification of Crop Leaf Area Index},
  journal = {Remote Sensing},
  year = {2025},
  doi = {10.3390/rs17233924},
  url = {https://doi.org/10.3390/rs17233924}
}