Shang et al. (2025) High-Resolution Mapping and Spatiotemporal Dynamics of Cropland Soil Temperature in the Huang-Huai-Hai Plain, China (2003–2020)

Identification

• Journal: Remote Sensing
• Year: 2025
• Date: 2025-11-19
• Authors: Guofei Shang, Yiran Tian, Xiangyang Liu, Xia Zhang, Zhe Li, Shunqing An
• DOI: 10.3390/rs17223765

Research Groups

• Hebei International Joint Research Center for Remote Sensing of Agricultural Drought Monitoring, Hebei GEO University, Shijiazhuang, China
• State Key Laboratory of Efficient Utilization of Arable Land in China, Institute of Agricultural Resources and Regional Planning, Chinese Academy of Agricultural Sciences, Beijing, China
• Hebei Utilization and Planning Institute of Natural Resources, Shijiazhuang, China

Short Summary

This study developed a high-resolution (1 km) monthly cropland soil temperature dataset for the Huang-Huai-Hai Plain (2003–2020) at 0–5 cm and 5–15 cm depths using a Random Forest model with recursive feature elimination and cross-validation, revealing complex spatiotemporal dynamics and a shift from cooling to rapid warming after 2012.

Objective

• To develop a high-accuracy monthly soil temperature prediction framework for cropland at 0–5 cm and 5–15 cm depths in the Huang-Huai-Hai Plain using a Random Forest algorithm integrated with recursive feature elimination and cross-validation, utilizing Land Surface Temperature (LST) as a primary predictor.
• To generate 1 km monthly cropland soil temperature maps for the Huang-Huai-Hai Plain from 2003 to 2020 and analyze their long-term trends, seasonal characteristics, and depth differences to understand spatiotemporal variation patterns.

Study Configuration

• Spatial Scale: Huang-Huai-Hai Plain, China (approximately 469,500 square kilometers, 32–40°N latitude, 114–121°E longitude), with output maps at 1 km spatial resolution.
• Temporal Scale: 2003–2020, with monthly temporal resolution for predictions and analysis.

Methodology and Data

• Models used: Random Forest (RF) algorithm for prediction, Recursive Feature Elimination with Cross-Validation (RFE-CV) for optimal feature selection.
• Data sources:
  • In situ observations: Approximately 3000 soil temperature measurements (2003–2020) from Lembrechts et al. (2022) dataset, filtered for 30–50°N latitude.
  • Predictor variables (19 geo-environmental covariates):
    • Environmental: Land Surface Temperature (LST) (Zenodo), Evapotranspiration (ET) (Google Earth Engine), Normalized Difference Vegetation Index (NDVI) (Google Earth Engine), Shortwave Radiation (Reflectance_1 to _7) (Google Earth Engine).
    • Soil properties: Soil Moisture (SM) (National Tibetan Plateau Data Center), Bulk density, Sand content, Soil Organic Carbon (SOC), Soil pH (SoilGrids database).
    • Topographic factors: Elevation, Slope, Surface Roughness, Terrain Ruggedness Index (TRI) (EarthEnv Topography).
  • Auxiliary variables: 30 m annual land cover dataset (Jie Yang, 2024) resampled to 1 km for cropland mask; 1 km soil type data (1:1,000,000 Soil Map of the People’s Republic of China, 1995) aggregated to Soil Order level.

Main Results

• Model Performance: High accuracy achieved for both depths. For 0–5 cm: MeanR² = 0.8–1.0, MeanRMSE = 1.0–1.9 °C, MeanMAE = 0.7–1.0 °C, MeanNSE = 0.8–1.0, MeanBias = 0.0–0.3 °C. For 5–15 cm: MeanR² = 0.9–1.0, MeanRMSE = 1.1–1.6 °C, MeanMAE = 0.8–1.1 °C, MeanNSE = 0.9–1.0, MeanBias = 0.0 °C.
• Feature Importance: Environmental variables were generally most important, exhibiting a "decrease-then-increase" (U-shaped) seasonal pattern. Soil properties gained influence in spring–summer, while topographic factors became more important in autumn–winter. Deeper soil layers showed reduced dependence on environmental variables and increased reliance on soil properties and topography.
• Long-Term Trends (2003–2020): Cropland soil temperature showed an overall warming trend, but with two distinct phases: a cooling trend from 2003–2012 (−0.60 °C/decade at 0–5 cm; −0.52 °C/decade at 5–15 cm) followed by a more rapid warming trend from 2012–2020 (1.04 °C/decade at 0–5 cm; 0.84 °C/decade at 5–15 cm). Shallow soil warmed faster than deeper soil.
• Spatial Heterogeneity: Soil temperature decreased with increasing latitude (0.52 °C per 1° for 0–5 cm; 0.44 °C per 1° for 5–15 cm) and elevation (0.46 °C per 100 m for 0–5 cm; 0.36 °C per 100 m for 5–15 cm). Alisol, low-latitude Fluvisol, and low-latitude–high-altitude Luvisol soil types exhibited stronger warming rates and greater temperature variability.
• Seasonal Characteristics: Temperature anomalies showed a consistent warming pattern across seasons, with the largest amplitudes in spring–summer (e.g., 0–5 cm spring anomaly from −0.5 °C to 0.5 °C). Seasonality (annual max-min temperature difference) was greater in the 0–5 cm layer (24.27–27.62 °C) than in the 5–15 cm layer (21.22–23.96 °C), with an increasing trend (0.46 °C/decade for shallow, 0.13 °C/decade for deep).
• Depth Differences: A very strong correlation (0.9205–0.9990) and no significant lag effect (peak correlation at 0 months, 0.9954) were observed between the two soil layers, indicating synchronous temperature changes. The increasing trend in temperature differences between layers suggests a reduction in soil thermal inertia, particularly in specific soil types.

Contributions

• Generation of a novel, high-resolution (1 km) monthly cropland soil temperature dataset for the Huang-Huai-Hai Plain (2003–2020) at 0–5 cm and 5–15 cm depths, crucial for precision agriculture and land management.
• Development of an enhanced Random Forest modeling framework incorporating 19 comprehensive geo-environmental variables and utilizing Land Surface Temperature (LST) as a primary predictor, improving upon traditional air temperature-based approaches.
• Implementation of a transparent and systematic variable selection strategy (RFE-CV) that reveals dynamic, month-to-month shifts in predictor importance and their variation with soil depth, providing insights for future model design and observational experiments.
• Comprehensive spatiotemporal analysis of soil temperature dynamics, including long-term trends, seasonal characteristics, and depth differences, highlighting the influence of latitude, elevation, and soil type on thermal regimes in a major agricultural region.

Funding

• National Key R&D Program of China (2022YFB3903005, 2022YFB3903005-4)
• Central Public-interest Scientific Institution Basal Research Fund (No. Y2024QC17)
• National Natural Science Foundation of China (Grant No. 42101371, 42401484)

Citation

@article{Shang2025HighResolution,
  author = {Shang, Guofei and Tian, Yiran and Liu, Xiangyang and Zhang, Xia and Li, Zhe and An, Shunqing},
  title = {High-Resolution Mapping and Spatiotemporal Dynamics of Cropland Soil Temperature in the Huang-Huai-Hai Plain, China (2003–2020)},
  journal = {Remote Sensing},
  year = {2025},
  doi = {10.3390/rs17223765},
  url = {https://doi.org/10.3390/rs17223765}
}