Dong et al. (2026) Unraveling meteorological contributions and lagged effects on multi-depth soil moisture prediction using Transformer-SHAP framework

Identification

• Journal: Journal of Hydrology
• Year: 2026
• Date: 2026-09-03
• Authors: Jiankang Dong, Wenxue Xia, Huiqing Hao, Yi Lian, Furong Yu, Yonghong Hao, Tian‐Chyi Jim Yeh
• DOI: 10.1016/j.jhydrol.2026.136349

Research Groups

Faculty of Geography, Tianjin Normal University, China Tianjin Key Laboratory of Water Resources and Environment, Tianjin Normal University, China College of Water Resources, North China University of Water Resources and Electric Power, China Department of Hydrology and Atmospheric Sciences, The University of Arizona, USA

Short Summary

This study employs the Transformer-SHAP framework to predict soil moisture across multiple depths and investigates the meteorological contributions and lagged effects on these predictions.

Objective

• Investigate the role of meteorological factors in predicting soil moisture at different depths using the Transformer-SHAP framework.
• Identify the optimal historical input window for each soil depth.

Study Configuration

• Spatial Scale: Soil profile scale, with separate models developed for 5, 20, 45, and 80 cm soil layers.
• Temporal Scale: Daily to weekly time scales, with a focus on understanding lagged effects of meteorological variables on soil moisture predictions.

Methodology and Data

• Models used: Transformer-based models for each soil depth, incorporating SHAP analysis for interpretable results.
• Data sources: Not specified in the provided text.

Main Results

• The Transformer-SHAP framework achieved NSE values of 0.82, 0.78, 0.70, and 0.90 for the 5, 20, 45, and 80 cm soil layers, respectively.
• Optimal historical input windows increased from 20 days at 5 cm to 21, 29, and 40 days at 20, 45, and 80 cm, indicating longer dependence on antecedent meteorological information with increasing depth.
• SHAP analysis revealed depth-related changes in meteorological contributions, with a shift from rapid shallow responses to near-surface energy conditions and precipitation toward more delayed deep responses associated with antecedent wetness and profile-scale moisture redistribution.

Contributions

• This study provides reliable predictions of soil moisture across multiple depths using the Transformer-SHAP framework.
• The results enable interpretable comparisons across soil depths, shedding light on the complex relationships between meteorological variables and soil moisture dynamics.

Funding

• Not specified in the provided text.

Citation

@article{Dong2026Unraveling,
  author = {Dong, Jiankang and Xia, Wenxue and Hao, Huiqing and Lian, Yi and Yu, Furong and Hao, Yonghong and Yeh, Tian‐Chyi Jim},
  title = {Unraveling meteorological contributions and lagged effects on multi-depth soil moisture prediction using Transformer-SHAP framework},
  journal = {Journal of Hydrology},
  year = {2026},
  doi = {10.1016/j.jhydrol.2026.136349},
  url = {https://doi.org/10.1016/j.jhydrol.2026.136349}
}