Wang et al. (2025) Enhancing Machine Learning-Based GPP Upscaling Error Correction: An Equidistant Sampling Method with Optimized Step Size and Intervals

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

Identification

• Journal: Remote Sensing
• Year: 2025
• Date: 2025-12-22
• Authors: Zegen Wang, Jiaqi Zuo, Zhiwei Yong, Xinyao Xie
• DOI: 10.3390/rs18010023

Research Groups

Not specified in the provided text.

Short Summary

This paper proposes an optimized equidistant sampling method to correct gross primary productivity (GPP) upscaling errors by precisely quantifying nonuniform density distributions of sub-pixel heterogeneity factors, demonstrating significant improvements in accuracy and transferability over existing methods and k-means clustering.

Objective

• To address limitations in current machine learning-based GPP upscaling error correction approaches by proposing an equidistant sampling method with optimized step size and intervals that precisely quantifies nonuniform density distributions and enhances correction precision.

Study Configuration

• Spatial Scale: GPP simulations at 480 m resolution, validated against 30 m resolution benchmarks.
• Temporal Scale: Annual (implied by GPP error units of gC m⁻² yr⁻¹).

Methodology and Data

• Models used: An eco-hydrological model (for GPP simulations). The proposed method is an optimized-step equidistant sampling method. K-means clustering was used for comparative analysis.
• Data sources: GPP simulations (from an eco-hydrological model), 30 m resolution GPP benchmarks. Heterogeneity factors include land cover, altitude, slope, Topographic Wetness Index (TWI), Topographic Nitrogen Index (TNI), Leaf Area Index (LAI), slope orientation, and Sky View Factor (SVF).

Main Results

• The proposed equidistant sampling method significantly improved GPP upscaling error correction, achieving a 0.27 increase in the determination coefficient (R²) and a 91.22 gC m⁻² yr⁻¹ reduction in root mean square error (RMSE) compared to previous elevation-based correction research (baseline R² = 0.48, RMSE = 285 gC m⁻² yr⁻¹).
• For comparative analysis, k-means clustering showed lesser improvements (ΔR² = 0.21, ΔRMSE = -63.54 gC m⁻² yr⁻¹).
• The optimized-step equidistant sampling method consistently surpassed k-means clustering in performance metrics when using identical statistical interval counts.
• Controlled variable experiments revealed that land cover, altitude, slope, and TNI are the most significant factors affecting GPP upscaling error correction, followed by LAI, while slope orientation, SVF, and TWI hold equal relevance.

Contributions

• Introduces an innovative equidistant sampling method with optimized step size and intervals that precisely quantifies nonuniform density distributions of sub-pixel heterogeneity factors, significantly enhancing GPP upscaling error correction accuracy.
• Overcomes critical limitations of previous machine learning-based approaches, specifically the failure to account for nonuniform density distributions and dependence on subjective classification thresholds.
• Provides an efficient solution that maintains high correction accuracy while minimizing computational costs.
• Demonstrates superior performance compared to k-means clustering for GPP upscaling error correction.

Funding

Not specified in the provided text.

Citation

@article{Wang2025Enhancing,
  author = {Wang, Zegen and Zuo, Jiaqi and Yong, Zhiwei and Xie, Xinyao},
  title = {Enhancing Machine Learning-Based GPP Upscaling Error Correction: An Equidistant Sampling Method with Optimized Step Size and Intervals},
  journal = {Remote Sensing},
  year = {2025},
  doi = {10.3390/rs18010023},
  url = {https://doi.org/10.3390/rs18010023}
}