Scholarlink

<p>Accurate prediction of moisture content in clay-sand liners is a prerequisite for effective water management in subsurface irrigation systems, particularly in arid and semi-arid regions where coupled thermal and salinity stresses critically govern liner performance. This study presents an extreme gradient boosting framework grounded on 6,226 observations for real-time prediction of volumetric moisture content (ClayMoist) in a field-instrumented clay-sand liner comprising 20% highly plastic clay and 80% sand, monitored continuously at hourly intervals over a March-November 2014 campaign in the Eastern Province of Saudi Arabia. The model was trained and validated on a strictly partitioned dataset, with all preprocessing parameters derived exclusively from the training subset to preclude data leakage, and optimized through systematic grid-search hyperparameter tuning. Six input features were considered: clay temperature (ClayTemp), ambient temperature, electrical conductivity (EC), watering state, date, and time. The framework achieved exceptional predictive accuracy, with R<sup>2</sup> values of 0.9993 and 0.9966 for training and test subsets, respectively, and validated by 10-fold cross-validation (mean R<sup>2</sup> = 0.994 ± 0.003). SHAP-based interpretability analysis identified EC as the dominant predictor, followed by ClayTemp, with both watering state and temporal variables contributing marginally; partial dependence plots further revealed that the EC-ClayMoist relationship exhibits a saturation-like gradient change beyond approximately 2.1 mS/cm, while the ClayTemp-ClayMoist relationship is non-monotonic and partly confounded by seasonal collinearity. These findings provide a physically interpretable, data-driven basis for sensor-guided moisture monitoring in clay-sand liner systems, though the model’s predictive scope is explicitly limited by the mineralogical and plasticity characteristics of the training material, and extrapolation to soils of markedly different composition requires site-specific recalibration. Future work should extend validation to multi-site, multi-composition datasets and higher temperature regimes to assess the transferability and operational generalizability of the proposed framework.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

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

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年7月9日

查看全文
  • Muawia Dafalla,
  • Yassir M. Abbas

摘要

Accurate prediction of moisture content in clay-sand liners is a prerequisite for effective water management in subsurface irrigation systems, particularly in arid and semi-arid regions where coupled thermal and salinity stresses critically govern liner performance. This study presents an extreme gradient boosting framework grounded on 6,226 observations for real-time prediction of volumetric moisture content (ClayMoist) in a field-instrumented clay-sand liner comprising 20% highly plastic clay and 80% sand, monitored continuously at hourly intervals over a March-November 2014 campaign in the Eastern Province of Saudi Arabia. The model was trained and validated on a strictly partitioned dataset, with all preprocessing parameters derived exclusively from the training subset to preclude data leakage, and optimized through systematic grid-search hyperparameter tuning. Six input features were considered: clay temperature (ClayTemp), ambient temperature, electrical conductivity (EC), watering state, date, and time. The framework achieved exceptional predictive accuracy, with R2 values of 0.9993 and 0.9966 for training and test subsets, respectively, and validated by 10-fold cross-validation (mean R2 = 0.994 ± 0.003). SHAP-based interpretability analysis identified EC as the dominant predictor, followed by ClayTemp, with both watering state and temporal variables contributing marginally; partial dependence plots further revealed that the EC-ClayMoist relationship exhibits a saturation-like gradient change beyond approximately 2.1 mS/cm, while the ClayTemp-ClayMoist relationship is non-monotonic and partly confounded by seasonal collinearity. These findings provide a physically interpretable, data-driven basis for sensor-guided moisture monitoring in clay-sand liner systems, though the model’s predictive scope is explicitly limited by the mineralogical and plasticity characteristics of the training material, and extrapolation to soils of markedly different composition requires site-specific recalibration. Future work should extend validation to multi-site, multi-composition datasets and higher temperature regimes to assess the transferability and operational generalizability of the proposed framework.