Yang et al. (2026) Physics‐Informed Spatiotemporal Graph Neural Network for Soil Temperature Forecasting

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

Identification

• Journal: International Journal of Climatology
• Year: 2026
• Date: 2026-07-28
• Authors: Yuqiang Yang, Kun Song, Huanzhi Luo
• DOI: 10.1002/joc.70521

Research Groups

• Institute of Atmospheric Physics, Chinese Academy of Sciences
• Department of Earth System Science, University of California, Irvine
• Key Laboratory of Regional Climate-Environment Research, Chinese Academy of Sciences

Short Summary

This study introduces a novel framework for predicting soil temperature using a physics-inspired time-gradient adaptive graph convolutional network (PTAGCN) that outperforms conventional data-driven approaches. The proposed model achieves high accuracy in forecasting 20 cm soil temperature across distinct climate regions in China.

Objective

• Investigate the feasibility of using a physics-inspired deep learning framework for predicting soil temperature with high accuracy and interpretability.

Study Configuration

• Spatial Scale: Regional scale, covering three distinct climate regions in China: Fengyun Village, Chongqing, and Naiman, Inner Mongolia.
• Temporal Scale: Long-term (daily) temporal evolution of soil temperature over a period of one year.

Methodology and Data

• Models used: Physics-Inspired Time-Gradient Adaptive Graph Convolutional Network (PTAGCN)
• Data sources: In-situ observations from weather stations, satellite-derived data for climate variables

Main Results

• The proposed PTAGCN model achieved an RMSE of 1.5633°C, an MAE of 1.2670°C, and a Nash-Sutcliffe efficiency (NSE) of 0.9759 in forecasting 20 cm soil temperature.
• The results indicate that the PTAGCN model outperforms conventional data-driven approaches while maintaining strong physics-inspired interpretability.

Contributions

• This study provides a novel framework for predicting soil temperature with high accuracy and interpretability, addressing the challenges of accurate prediction in complex climate systems.
• The proposed approach has potential applications in regional environmental monitoring and data-driven agricultural management.

Funding

• National Natural Science Foundation of China (Grant No. 91837204)
• Key Research Program of Frontier Sciences, Chinese Academy of Sciences (Grant No. QYZDY-SSW-DQC010)

Citation

@article{Yang2026PhysicsInformed,
  author = {Yang, Yuqiang and Song, Kun and Luo, Huanzhi},
  title = {Physics‐Informed Spatiotemporal Graph Neural Network for Soil Temperature Forecasting},
  journal = {International Journal of Climatology},
  year = {2026},
  doi = {10.1002/joc.70521},
  url = {https://doi.org/10.1002/joc.70521}
}