Abdelhedi et al. (2026) Machine learning prediction of effective porosity and water content in unsaturated zones: application to the Merguellil Basin in the arid Mediterranean region of central Tunisia

Identification

• Journal: Euro-Mediterranean Journal for Environmental Integration
• Year: 2026
• Date: 2026-01-08
• Authors: Mohamed Abdelhedi, Anis Ammari, Dhouha Ben Othman, Habib Abida, Hakim Gabtni, Chedly Abbes
• DOI: 10.1007/s41207-025-01048-x

Research Groups

• Research Laboratory GEOMODELE (LR16ES17), Department of Earth Sciences, Faculty of Sciences, University of Sfax, Sfax, Tunisia
• Water Research and Technology Center, Georessources Laboratory, CERTE, Soliman, Tunisia

Short Summary

This study developed an innovative methodology combining ultrasonic waves and machine learning to accurately predict effective porosity and water content in the unsaturated zone of the Merguellil Basin, central Tunisia, providing crucial insights for groundwater recharge assessment in arid regions.

Objective

• To create a methodology for efficiently and accurately estimating effective porosity and water content of soil in the unsaturated zone using ultrasonic waves and machine learning algorithms.

Study Configuration

• Spatial Scale: Merguellil Basin, downstream of the El Houareb Dam, central Tunisia. Soil samples were collected from eight transverse sections along the Merguellil Wadi.
• Temporal Scale: Soil sampling campaigns were conducted on July 11, 2021, and December 12, 2021.

Methodology and Data

• Models used: Multilayer Perceptron (MLP) Regressor, Extreme Gradient Boosting (XGB) Regressor, Random Forest Regressor.
• Data sources:
  • 32 soil samples collected from 8 monitoring stations.
  • Laboratory measurements of effective porosity and water content using conventional techniques.
  • Field measurements of ultrasonic wave velocities (direct, semi-direct, and surface transmission methods) using a 25 kHz transducer and PUNDIT LAB+ instrument.
  • Density measurements of soil samples.
  • Geographic coordinates (latitude, longitude, altitude) for each sampling point.
  • Input parameters for ML models: density, altitude, and ultrasonic velocity (from semi-direct and direct methods).

Main Results

• The MLP Regressor model demonstrated the highest accuracy for both parameters.
• For water content prediction, the MLP Regressor achieved a coefficient of determination (R²) of 0.56, with a cross-validation R² of 0.83. The XGB Regressor and Random Forest Regressor achieved R² values of 0.45 and 0.48, respectively.
• For effective porosity prediction, the MLP Regressor achieved an R² of 0.76, with a cross-validation R² of 0.80. The XGB Regressor and Random Forest Regressor achieved R² values of 0.58 and 0.42, respectively.
• Strong positive correlations were observed between ultrasonic velocity and effective porosity, and an inverse correlation between ultrasonic velocity and water content.
• The models' reliability was further validated through rigorous cross-validation, with mean absolute errors below 0.05 for critical parameters.

Contributions

• Introduces a novel and rapid approach combining ultrasonic waves and machine learning for efficient and accurate estimation of effective porosity and water content in unsaturated zones.
• Offers a significant improvement over traditional, often time-consuming and data-intensive methods, by being non-destructive and computationally efficient.
• Provides valuable insights for forecasting infiltration and groundwater recharge, particularly relevant for sustainable water resource management in arid and semi-arid regions like the Mediterranean basins.
• Highlights the capability of machine learning techniques to accurately predict soil hydrodynamic properties in geophysical studies, advancing understanding and offering practical tools for hydrologists and geophysicists.

Funding

• Ministry of Higher Education and Scientific Research of the Republic of Tunisia.

Citation

@article{Abdelhedi2026Machine,
  author = {Abdelhedi, Mohamed and Ammari, Anis and Othman, Dhouha Ben and Abida, Habib and Gabtni, Hakim and Abbes, Chedly},
  title = {Machine learning prediction of effective porosity and water content in unsaturated zones: application to the Merguellil Basin in the arid Mediterranean region of central Tunisia},
  journal = {Euro-Mediterranean Journal for Environmental Integration},
  year = {2026},
  doi = {10.1007/s41207-025-01048-x},
  url = {https://doi.org/10.1007/s41207-025-01048-x}
}