Liu et al. (2025) Transformer-based soil moisture simulation for understanding future drying trend globally

Identification

• Journal: Journal of Hydrology
• Year: 2025
• Date: 2025-11-30
• Authors: Yangxiaoyue Liu, Yuan Tian, Ying Xin, Shenghai Yuan, Jiangyuan Zeng, Min Feng, Chunqiao Song
• DOI: 10.1016/j.jhydrol.2025.134709

Research Groups

• Institute of Geographic Sciences and Natural Resources Research, Chinese Academy of Sciences, Beijing, China.
• South China Sea Institute of Planning and Environment Research, State Oceanic Administration, Guangzhou, China.
• Xi’an University of Finance and Economics, Xi’an, China.
• Aerospace Information Research Institute, Chinese Academy of Sciences, Beijing, China.
• Institute of Tibetan Plateau Research, Chinese Academy of Sciences, Beijing, China.
• Nanjing Institute of Geography and Limnology, Chinese Academy of Sciences, Nanjing, China.

Short Summary

This study introduces TSMSNet, a Transformer-based deep learning model designed to simulate global soil moisture (SM) from 2016 to 2099 under various climate scenarios. The research identifies a significant global drying trend that intensifies with higher greenhouse gas emission pathways, particularly affecting habitable regions and agricultural lands.

Objective

• To develop a high-reliability global soil moisture simulation framework (TSMSNet) that overcomes the trend-blurring limitations of traditional multi-model ensemble averages.
• To analyze future global soil moisture variations and drying trends under SSP1-2.6, SSP2-4.5, and SSP5-8.5 scenarios.

Study Configuration

• Spatial Scale: Global terrestrial surface.
• Temporal Scale: 2016–2099 (Future projections); model training and validation utilize historical periods.

Methodology and Data

• Models used: TSMSNet (Transformer SM Simulation Net), with CSMSNet (Convolutional Long Short Term Memory) used as a performance benchmark.
• Data sources:
  • Inputs: Nine CMIP6 future SM datasets, spatial distribution of error parameters, and geographic auxiliary data.
  • Learning Target: A merged dataset combining Soil Moisture Active Passive (SMAP) satellite observations and European Centre for Medium-Range Weather Forecasts Reanalysis v5-Land (ERA5-Land) SM.
  • Scenarios: Shared Socioeconomic Pathways SSP1-2.6 (sustainable), SSP2-4.5 (middle of the road), and SSP5-8.5 (fossil-fueled development).

Main Results

• Model Performance: TSMSNet achieved superior accuracy (R = 0.68, ubRMSE = 0.045 m³/m³) compared to CSMSNet (R = 0.65, ubRMSE = 0.047 m³/m³).
• Drying Trends: The model predicts an overwhelming global drying trend. The magnitude of soil moisture decline increases as the scenario shifts from SSP1-2.6 to SSP5-8.5.
• Land Cover Impact: The most pronounced drying trends were observed in forest and cropland ecosystems.
• Socio-Geographic Impact: Soil moisture shows a faster rate of descent in habitable areas compared to uninhabitable areas, posing risks to human settlement and food security.

Contributions

• Methodological Innovation: Demonstrates the first application of a Transformer-based architecture for long-term global soil moisture simulation, proving more effective at capturing long-term trends than LSTM-based models or simple ensemble averaging.
• Dataset Development: Provides a reliable, high-resolution future soil moisture dataset that integrates the precision of satellite-reanalysis fusion with the predictive power of CMIP6 models.
• Environmental Insights: Quantifies the acceleration of soil desiccation in relation to different socioeconomic development pathways, highlighting specific vulnerabilities in habitable zones.

Funding

• Not explicitly detailed in the provided text snippet.

Citation

@article{Liu2025Transformerbased,
  author = {Liu, Yangxiaoyue and Tian, Yuan and Xin, Ying and Yuan, Shenghai and Zeng, Jiangyuan and Feng, Min and Song, Chunqiao},
  title = {Transformer-based soil moisture simulation for understanding future drying trend globally},
  journal = {Journal of Hydrology},
  year = {2025},
  doi = {10.1016/j.jhydrol.2025.134709},
  url = {https://doi.org/10.1016/j.jhydrol.2025.134709}
}

Generated by BiblioAssistant using gemini-3-flash-preview (Google API)