Wang et al. (2026) Spatially adaptive estimation of multi-layer soil temperature at a daily time-step across China during 2010–2020

Identification

• Journal: Earth system science data
• Year: 2026
• Date: 2026-01-06
• Authors: Xuetong Wang, Liang He, P. F. Li, Jiageng Ma, Yu Shi, Qi Tian, Gang Zhao, Jianqiang He, Hao Feng, Hao Shi, Qiang Yu
• DOI: 10.5194/essd-18-97-2026

Research Groups

• College of Natural Resources and Environment, Northwest A&F University, Yangling, China
• State Key Laboratory of Soil and Water Conservation and Desertification Control, Northwest A&F University, Yangling, China
• National Meteorological Center, Beijing, China
• Key Laboratory of Ecosystem Network Observation and Modeling, Institute of Geographic Sciences and Natural Resources Research, Chinese Academy of Sciences, Beijing, PR China
• Institute of Carbon Neutrality, Sino-French Institute for Earth System Science, College of Urban and Environmental Sciences, Peking University, Beijing, China
• College of Soil and Water Conservation Science and Engineering, Northwest A&F University, Yangling, Shaanxi, China
• Key Laboratory for Agricultural Soil and Water Engineering in Arid Area of Ministry of Education, Northwest A&F University, Yangling, China
• State Key Laboratory for Ecological Security of Regions and Cities, Research Center for Eco-Environmental Sciences, Chinese Academy of Sciences, Beijing, China
• College of Resources and Environment, University of Chinese Academy of Sciences, Beijing, China

Short Summary

This study developed a spatially adaptive machine learning framework to generate a high-resolution (1 km), daily, multi-layer soil temperature dataset for China from 2010 to 2020, integrating in-situ observations, satellite remote sensing, and reanalysis data. The resulting dataset provides accurate spatiotemporal soil temperature profiles, significantly improving upon existing products for applications in precision agriculture, ecosystem modeling, and climate-land feedback studies.

Objective

• To construct a spatially adaptive modeling system for soil temperature (Ts) estimation.
• To generate a multi-layer Ts dataset at a daily time-step and 1 kilometer resolution across China from 2010 to 2020.
• To evaluate the generated dataset through independent validation with flux tower observations and benchmarking against widely used Ts products.

Study Configuration

• Spatial Scale: China, at a 1 km spatial resolution.
• Temporal Scale: Daily time-step, covering the period from 2010 to 2020.

Methodology and Data

• Models used: Spatially adaptive layer-cascading Extreme Gradient Boosting (XGBoost) algorithm, incorporating a rotated-quadtree spatial partitioning strategy.
• Data sources:
  • In-situ observations: Daily mean soil temperature measurements at 0, 5, 10, 15, 20, and 40 cm depths from 2093 national weather stations operated by the China Meteorological Administration (CMA).
  • Satellite remote sensing:
    • MOD11A1 Land Surface Temperature (LST) product (daily, 1 km resolution, daytime and nighttime).
    • MODIS Surface Reflectance Product (MOD09GA) for Enhanced Vegetation Index (EVI) (daily, 500 m resolution).
  • Reanalysis data: ERA5-Land reanalysis dataset (daily mean 2 m air temperature, surface solar radiation, total precipitation) from ECMWF (0.1° spatial resolution, hourly temporal resolution).
  • Auxiliary data:
    • Shuttle Radar Topography Mission (SRTM) digital elevation model (Version 3) for elevation and slope (30 m spatial resolution).
    • China Soil Information Grid dataset for soil texture (clay, silt, sand proportions) at 0–5, 5–15, 15–30, and 30–60 cm depths (1 km spatial resolution).

Main Results

• The spatially adaptive XGBoost model achieved high accuracy, with median R² values ranging from 0.92 to 0.98 and median RMSE values from 1.6 to 2.4 K across different depths for both training and test sets.
• Independent validation using 18 ChinaFLUX tower sites showed robust performance (R² = 0.78–0.87; RMSE = 3.89–5.14 K), although a systematic positive bias of approximately +2 to +3 K was observed for annual mean Ts.
• Model performance varied with depth, being slightly weaker at 0 cm and 40 cm compared to intermediate depths (5–20 cm).
• Prediction accuracy varied across land cover types, with barren land showing the highest R² and cropland, forest, and grassland exhibiting slightly lower but still high performance.
• Seasonal variations in accuracy were observed, with higher R² in spring and autumn (0.48–0.98) and lower R² in summer and winter, particularly at deeper layers (20–40 cm). RMSE values were lower in spring and autumn (1.3–2.2 K) and higher in summer and winter.
• The generated 1 km resolution Ts dataset exhibited significantly finer spatial detail and higher site-level accuracy (R² = 0.94–0.97) compared to ERA5-Land (R² = 0.83–0.89) and GLDAS 2.1 (R² = 0.83–0.87) reanalysis products, especially in topographically complex regions.
• Spatial and temporal patterns showed that deeper soil layers were warmer than surface layers in winter, while the surface was warmer in summer, with deeper layers exhibiting greater thermal stability.

Contributions

• Developed a novel spatially adaptive machine learning framework using a rotated-quadtree strategy and XGBoost to address spatial heterogeneity and uneven data distribution in large-scale soil temperature estimation.
• Generated the first high-resolution (1 km), daily, multi-layer (0, 5, 10, 15, 20, and 40 cm) soil temperature dataset for China covering 2010–2020, filling a significant data gap.
• Demonstrated superior accuracy and spatial detail of the new dataset compared to existing global and regional reanalysis products (ERA5-Land, GLDAS 2.1).
• Provided a robust data foundation for high-resolution environmental modeling, precision agriculture, ecosystem monitoring, and climate impact assessments in China.

Funding

• National Key Research and Development Program of China (grant no. 2023YFF1303700)
• National Natural Science Foundation of China (grant no. 42375195)

Citation

@article{Wang2026Spatially,
  author = {Wang, Xuetong and He, Liang and Li, P. F. and Ma, Jiageng and Shi, Yu and Tian, Qi and Zhao, Gang and He, Jianqiang and Feng, Hao and Shi, Hao and Yu, Qiang},
  title = {Spatially adaptive estimation of multi-layer soil temperature at a daily time-step across China during 2010–2020},
  journal = {Earth system science data},
  year = {2026},
  doi = {10.5194/essd-18-97-2026},
  url = {https://doi.org/10.5194/essd-18-97-2026}
}