Cai et al. (2025) A Normalization-Calibration Model for Multi-Source Ground-Based FPAR Observations in Mountainous Forests

Identification

• Journal: Remote Sensing
• Year: 2025
• Date: 2025-11-22
• Authors: Yi Cai, Ainong Li, Jinhu Bian, Zhengjian Zhang, Limin Chen, Xiaohan Lin, Yi Deng, Xi Nan, Guangbin Lei, Amin Naboureh
• DOI: 10.3390/rs17233797

Research Groups

• Center for Digital Mountain and Remote Sensing Application, Institute of Mountain Hazards and Environment, Chinese Academy of Sciences, Chengdu 610213, China
• College of Resources and Environment, University of Chinese Academy of Sciences, Beijing 101408, China
• Wanglang Mountain Remote Sensing Field Observation and Research Station of Sichuan Province, Mianyang 621000, China

Short Summary

This study developed and validated a normalization-calibration model for multi-source ground-based FPAR observations in mountainous forests, using FPARnet as a reference, to overcome systematic biases and enhance spatial representativeness for remote sensing product validation. The model significantly improved consistency among FPAR data from various instruments, particularly for LAI-NOS, LAI-2200, and DHP.

Objective

• To develop a PAIe-LAI-FPAR conversion model using the Beer–Lambert law for five ground-based observation techniques (FPARnet, LAI-NOS, LAINet, LAI-2200, and Digital Hemispherical Photography (DHP)) at Wanglang Station.
• To comparatively analyze the correlation and consistency of FPAR derived from these different observation techniques.
• To develop and validate a normalization-calibration model based on regression, using FPAR FPARnet as the reference, to reduce systematic biases and improve consistency among multi-source FPAR observations in mountainous forests.

Study Configuration

• Spatial Scale: Wanglang Mountain Remote Sensing Field Observation and Research Station of Sichuan Province, Pingwu County, Mianyang City, Sichuan Province, southwestern China (104°00′~104°04′E, 32°58′N~33°02′N). Elevation ranges from approximately 2300 to 4980 meters. The study area includes Deciduous Broadleaf Forests (DBF), Evergreen Needleleaf Forest (ENF), and Deciduous Shrubland (DSH).
• Temporal Scale: FPARnet and LAI-NOS data were collected daily from April 2024 to October 2024. LAINet data were collected daily from April 2024 to July 2024. LAI-2200 data were collected monthly in May, July, August, September, October 2023, and August, October 2024. DHP data were collected monthly in August, September, October 2024.

Methodology and Data

• Models used:
  • PAIe-LAI-FPAR conversion model (based on Beer–Lambert law)
  • Four-Component PAR dynamic decomposition method (for FPARnet)
  • Beer–Lambert law (for LAI to FPAR conversion)
  • Sobol global sensitivity analysis (for model parameters k, Ωe, α, γe)
  • Linear regression models (for normalization-calibration)
  • Statistical metrics: R (correlation coefficient), RMSEobs (Root Mean Square Error of observation), Bias, Limits of Agreement (LoA), R² (coefficient of determination), RMSEpred (Root Mean Square Error of prediction), R²CV, RMSECV (from Leave-One-Out Cross-Validation), t-test.
• Data sources:
  • Ground-based observations from five instruments:
    • FPARnet (fully automated Photosynthetically Active Radiation (PAR) instrument, quantum sensors)
    • LAI-NOS (automated Leaf Area Index (LAI) observation system, digital hemispherical photography)
    • LAINet (automated LAI instrument, quantum sensors)
    • LAI-2200 Plant Canopy Analyzer (fisheye optical sensor, manual measurements)
    • Digital Hemispherical Photography (DHP) (fisheye camera, manual measurements)
  • Data collected at three observation towers (DSH, DBF, ENF) and surrounding fixed plots.

Main Results

• FPARnet accurately captured the seasonal dynamics of DSH, DBF, and ENF, demonstrating high daily measurement consistency with errors mostly concentrated around zero.
• Parameter sensitivity analysis for the PAIe-LAI-FPAR conversion model revealed that the light extinction coefficient (k) was the most sensitive parameter, followed by the element clumping index (Ωe), while the woody-to-total area ratio (α) and the ratio of leaf area to shoot area (γe) had low sensitivity.
• FPAR LAI-NOS showed a strong positive correlation with FPAR FPARnet (R = 0.9, RMSEobs = 0.05), with good consistency (mean difference = 0.0025, 7.6% of points outside the Limits of Agreement).
• FPAR LAI-2200 and FPAR DHP exhibited very strong correlations with FPAR LAI-NOS (R = 0.95 for both, RMSEobs = 0.08 and 0.07 respectively), also demonstrating good consistency (2.2% and 2.3% of points outside the Limits of Agreement respectively).
• FPAR LAINet showed a weak correlation with FPAR FPARnet (R = 0.12).
• After applying the normalization-calibration model, the consistency among multi-source FPAR observations significantly improved. The correlation coefficient (R) remained unchanged, while the average RMSEobs decreased by approximately 7.8%, and sample points aligned more closely along the 1:1 line.
• The established calibration models were robust, with R² values ranging from 0.81 to 0.91 and RMSEpred values from 0.05 to 0.06, and similar performance in cross-validation.

Contributions

• Established a practical normalization-calibration framework for multi-source ground-based FPAR observations in mountainous forests, using FPARnet as a reliable reference.
• Overcame the spatial limitations and limited representativeness of single observation techniques in complex mountainous terrain, significantly enhancing spatial sampling and expanding the available ground-based dataset by 30 new points.
• Provided a high-quality, standardized ground FPAR reference dataset that meets CEOS LPV requirements, thereby improving the reliability of validation for large-scale remote sensing products.
• Offered practical guidance for field experiments, including optimal sensor selection (e.g., LAI-NOS, LAI-2200, DHP showed better performance than LAINet in dense evergreen needleleaf forests), deployment optimization, and measurement protocol design for diverse forest ecosystems.
• Demonstrated the transferability of the normalization-calibration method, requiring only re-estimation of specific model parameters (α, Ωe, γe, k) for application in new study areas.

Funding

• National Key Research and Development Program of China, grant number 2020YFA0608702
• National Natural Science Foundation project of China, grant number U23A2019

Citation

@article{Cai2025NormalizationCalibration,
  author = {Cai, Yi and Li, Ainong and Bian, Jinhu and Zhang, Zhengjian and Chen, Limin and Lin, Xiaohan and Deng, Yi and Nan, Xi and Lei, Guangbin and Naboureh, Amin},
  title = {A Normalization-Calibration Model for Multi-Source Ground-Based FPAR Observations in Mountainous Forests},
  journal = {Remote Sensing},
  year = {2025},
  doi = {10.3390/rs17233797},
  url = {https://doi.org/10.3390/rs17233797}
}