Zhao et al. (2025) A 10-m resolution dataset of plastic-mulched farmland distributions on the Chinese Loess Plateau (2019–2021)

Identification

• Journal: Scientific Data
• Year: 2025
• Date: 2025-12-11
• Authors: Cheng Zhao, Yadong Luo, Xiangyu Chen, Lingmei Jiang, Zhao Wang, Hao Feng, Qiang Yu, Jianqiang He
• DOI: 10.1038/s41597-025-06304-x

Research Groups

• Key Laboratory for Agricultural Soil and Water Engineering in Arid Area of Ministry of Education, Northwest A&F University, Yangling, China
• Institute of Water-Saving Agriculture in Arid Areas of China, Northwest A&F University, Yangling, Shaanxi, China
• Key Laboratory of Eco-Environment and Meteorology for the Qinling Mountains and Loess Plateau, Shaanxi Provincial Meteorology Bureau, Xi’an, Shaanxi, China
• State Key Laboratory of Soil Erosion and Dryland Farming on the Loess Plateau, Institute of Water and Soil Conservation, Northwest A&F University, Yangling, Shaanxi, China

Short Summary

This study generated the first 10-meter resolution plastic-mulched farmland (PMF) distribution maps for the Chinese Loess Plateau (2019–2021) by developing a novel framework that couples automatic training sample generation and classifier transfer methods. The resulting maps achieved satisfactory accuracies (F1-scores 0.80–0.86) and demonstrated good agreement with agricultural census data (R² ≥ 0.87).

Objective

• To generate the first 10-meter resolution plastic-mulched farmland (PMF) distribution maps for the Chinese Loess Plateau (PMF-LP) spanning 2019–2021.
• To address the scarcity of training samples that hinders large-scale PMF mapping by developing a novel framework.
• To assess the spatiotemporal transferability of pre-trained classifiers for PMF mapping.

Study Configuration

• Spatial Scale: Chinese Loess Plateau (approximately 6.4 × 10⁵ km²), with a spatial resolution of 10 meters.
• Temporal Scale: Annual maps for the years 2019, 2020, and 2021.

Methodology and Data

• Models used:
  • Random Forest (RF) classifier (pixel-based)
  • Harmonic regression for time-series curve fitting
  • Two novel plastic-mulched farmland (PMF) indices: Max Blue Band-based Plastic-mulched Farmland Index (MBPMFI) and Blue Band-based Plastic-mulched Farmland Index (BPMFI)
• Data sources:
  • Sentinel-2 surface reflectance (SR) data (March to October, 2019–2021) from Google Earth Engine (GEE).
  • Cropland layers: GLC_FCS30D, CACD, and CLCD (resampled to 10-meter resolution).
  • Google Earth high-resolution images (GE-HRIs) for validation sample collection and optimal threshold determination.
  • Municipal-level agricultural census data (2019–2021) for PMF area validation.

Main Results

• The generated PMF-LP maps exhibited satisfactory accuracies, with F1-scores ranging from 0.80 to 0.86 and overall accuracies (OA) from 0.82 to 0.87 across the three years.
• The estimated PMF areas derived from the PMF-LP dataset showed strong consistency with municipal-level agricultural census data, with coefficients of determination (R²) of 0.92 (2019), 0.93 (2020), and 0.87 (2021).
• Temporal classifier transfer (Scenario–T) proved more suitable for multi-year PMF mapping, with an average F1-score change of -2.92% for 2019 and -6.07% for 2021, outperforming spatial-temporal transfer (Scenario–ST).
• The proposed framework demonstrated strong cross-regional generalization capabilities, achieving F1-scores exceeding 0.92 in other diverse agro-climatic regions across China.
• PMF was widely distributed across the Loess Plateau, showing extensive dispersion and localized clustering, with the highest intensity in the Hetao Irrigation District (18%), Northern Shanxi (17%), and Eastern Gansu-Southern Ningxia (16%).

Contributions

• Generated the first 10-meter resolution, time-continuous (2019–2021) plastic-mulched farmland distribution dataset for the Chinese Loess Plateau.
• Developed a novel mapping framework that integrates automatic training sample generation using two new blue band-based PMF indices (MBPMFI and BPMFI) with machine learning classifiers.
• Demonstrated the feasibility and effectiveness of temporal classifier transfer for multi-year PMF mapping, reducing the need for year-specific training samples.
• The hybrid approach addresses key limitations of existing methods, such as cloud contamination in index-based approaches and training sample scarcity in machine learning, enabling seamless PMF map production.
• The proposed framework is adaptable for PMF mapping using Landsat data due to consistent band usage.

Funding

• National Key Research and Development Program of China (No. 2021YFD1900700)
• National Natural Science Foundation of China (52579046)
• “111 Project” of China (No. B12007)

Citation

@article{Zhao202510m,
  author = {Zhao, Cheng and Luo, Yadong and Chen, Xiangyu and Jiang, Lingmei and Wang, Zhao and Feng, Hao and Yu, Qiang and He, Jianqiang},
  title = {A 10-m resolution dataset of plastic-mulched farmland distributions on the Chinese Loess Plateau (2019–2021)},
  journal = {Scientific Data},
  year = {2025},
  doi = {10.1038/s41597-025-06304-x},
  url = {https://doi.org/10.1038/s41597-025-06304-x}
}