Quantifying temperature and salinity effects on real-time moisture dynamics of clay-sand liners via explainable...

Area of Science:

• Geotechnical Engineering
• Environmental Science
• Data Science

Background:

• Effective water management in subsurface irrigation is crucial, especially in arid regions.
• Clay-sand liner performance is significantly impacted by coupled thermal and salinity stresses.
• Accurate prediction of moisture content is a prerequisite for optimizing irrigation systems.

Purpose of the Study:

• To develop a real-time prediction framework for volumetric moisture content (ClayMoist) in clay-sand liners.
• To utilize an extreme gradient boosting approach for enhanced predictive accuracy.
• To identify key predictors influencing moisture content in liners.

Main Methods:

• A dataset of 6,226 observations from a field-instrumented clay-sand liner was used.
• An extreme gradient boosting framework was trained and validated, with strict data partitioning to prevent leakage.
• Hyperparameter tuning was performed using systematic grid-search, and SHAP analysis was employed for interpretability.

Main Results:

• The model achieved high predictive accuracy, with R² values of 0.9993 (training) and 0.9966 (test), validated by 10-fold cross-validation (mean R² = 0.994).
• Electrical conductivity (EC) was identified as the dominant predictor, followed by clay temperature (ClayTemp).
• Partial dependence plots revealed specific thresholds and non-monotonic relationships for EC and ClayTemp influencing ClayMoist.

Conclusions:

• The data-driven framework provides a physically interpretable basis for sensor-guided moisture monitoring in clay-sand liners.
• The model's predictive scope is limited to the specific mineralogical and plasticity characteristics of the tested material.
• Future research should focus on multi-site validation and diverse soil compositions to assess transferability and generalizability.