Yin et al. (2026) Unraveling soil moisture dynamics with dual‐scale interpretable machine learning: Cover cropping and irrigation insights in semi‐arid agriculture

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

Identification

• Journal: Vadose Zone Journal
• Year: 2026
• Date: 2026-01-01
• Authors: Huichao Yin, Prakriti Bista, Rajan Ghimire, Hui Yang, Kenneth C. Carroll
• DOI: 10.1002/vzj2.70077

Research Groups

Not specified in the provided abstract.

Short Summary

This paper developed a novel dual-resolution framework integrating machine learning and deep learning with explainable AI to predict soil moisture under various cover crop treatments, demonstrating improved accuracy and interpretability for data-driven irrigation management.

Objective

• To develop and evaluate a novel, dual-resolution framework integrating machine learning and deep learning with explainable artificial intelligence to predict and interpret soil moisture dynamics under different cover crop treatments in arid and semi-arid agroecosystems.

Study Configuration

• Spatial Scale: Field-scale experiment conducted in Clovis, New Mexico, across five distinct cover crop treatments (fallow, pea, oat, pea–oat mixture, six-species mixture).
• Temporal Scale: Two-year experimental duration; data collected continuously at 5-minute intervals and aggregated to daily resolution.

Methodology and Data

• Models used: Machine Learning (Random Forest, Light Gradient Boosting Machine [LGBM], eXtreme Gradient Boosting), Deep Learning (Long Short-Term Memory [LSTM], Transformer models), Explainable AI (SHapley Additive exPlanations [SHAP]).
• Data sources: In-situ continuous monitoring of soil moisture, soil temperature, and meteorological variables (e.g., rainfall, solar radiation) from a 2-year field experiment.

Main Results

• Features from the three most recent days (72-hour lookback) were most influential for daily soil moisture predictions.
• Light Gradient Boosting Machine (LGBM) achieved the highest predictive accuracy among the tested models.
• Transformer models effectively reduced temporal lag in high-frequency (5-minute) soil moisture predictions.
• SHAP analysis revealed treatment-specific sensitivities and scale-dependent relationships between environmental variables and soil moisture.
• Historical soil moisture and total water input were the dominant predictors for daily soil moisture.
• Solar radiation, temperature, and recent water events were more influential at finer temporal scales (5-minute).

Contributions

• Introduces a novel, dual-resolution framework combining machine learning, deep learning, and explainable artificial intelligence for soil moisture prediction and interpretation.
• Enhances both the predictive accuracy and interpretability of soil water dynamics in complex agroecosystems.
• Provides a data-driven approach to optimize irrigation scheduling and improve water-use efficiency in arid and semi-arid regions.
• Identifies scale-dependent feature importance for soil moisture prediction, offering insights for model design and management strategies.

Funding

Not specified in the provided abstract.

Citation

@article{Yin2026Unraveling,
  author = {Yin, Huichao and Bista, Prakriti and Ghimire, Rajan and Yang, Hui and Carroll, Kenneth C.},
  title = {Unraveling soil moisture dynamics with dual‐scale interpretable machine learning: Cover cropping and irrigation insights in semi‐arid agriculture},
  journal = {Vadose Zone Journal},
  year = {2026},
  doi = {10.1002/vzj2.70077},
  url = {https://doi.org/10.1002/vzj2.70077}
}