Wang et al. (2026) Enhanced saturated hydraulic conductivity estimation in fine-grained soils: a voting regressor ensemble framework

Identification

• Journal: Journal of Hydrology
• Year: 2026
• Date: 2026-01-30
• Authors: Yan Wang, You Gao, De’an Sun
• DOI: 10.1016/j.jhydrol.2026.135041

Research Groups

• School of Civil & Environmental Engineering and Geography Science, Ningbo University, Ningbo, China
• Department of Civil Engineering, Shanghai University, Shanghai, China

Short Summary

This study developed a robust machine learning-based Voting Regressor (VR) ensemble framework to accurately estimate saturated hydraulic conductivity (ks) in fine-grained soils. The proposed VR model, integrating SVM, MLP, and GEP, demonstrated superior predictive performance (R² = 0.992, RMSLE = 0.00049 m/s) compared to traditional methods, using only basic soil physical parameters.

Objective

• To develop an accurate, robust, and computationally efficient method for estimating saturated hydraulic conductivity (ks) in fine-grained soils, addressing the limitations of costly, time-consuming laboratory methods and complex calibration requirements of theoretical models.

Study Configuration

• Spatial Scale: Soil samples/types (focus on fine-grained soils and diverse soil types); computational study.
• Temporal Scale: Instantaneous prediction of a static soil property.

Methodology and Data

• Models used: eXtreme Gradient Boosting (XGBoost), Categorical Boosting (CatBoost), Random Forest (RF), Gene Expression Programming (GEP), Support Vector Machine (SVM), Multilayer Perceptron (MLP), and a proposed Voting Regressor (VR) ensemble framework.
• Data sources: Augmented dataset using the synthetic minority over-sampling technique, and measured data from the literature for validation.

Main Results

• XGBoost, RF, and CatBoost exhibited significant overfitting and poor extrapolation capabilities.
• SVM, MLP, and GEP showed superior generalization capabilities, effectively capturing physical patterns.
• The proposed Voting Regressor (VR) ensemble model, integrating SVM, MLP, and GEP, significantly enhanced prediction accuracy in both interpolation and extrapolation scenarios.
• The VR model accurately simulates ks across diverse soil types using only basic soil physical parameters.
• It achieved superior performance with a coefficient of determination (R²) of 0.992 and a Root Mean Squared Logarithmic Error (RMSLE) of 0.00049 m/s, outperforming traditional theoretical models when validated against measured data.

Contributions

• Proposes a novel hybrid ensemble method based on the Voting Regressor framework for enhanced saturated hydraulic conductivity estimation in fine-grained soils.
• Identifies specific machine learning algorithms (SVM, MLP, GEP) with robust generalization capabilities for predicting soil hydraulic properties.
• Demonstrates superior accuracy and robustness of the proposed VR model compared to individual machine learning algorithms and traditional theoretical models.
• Provides a reliable computational tool with practical applications in precision irrigation management in agriculture and groundwater pollution risk assessment.

Funding

• Not specified in the provided text.

Citation

@article{Wang2026Enhanced,
  author = {Wang, Yan and Gao, You and Sun, De’an},
  title = {Enhanced saturated hydraulic conductivity estimation in fine-grained soils: a voting regressor ensemble framework},
  journal = {Journal of Hydrology},
  year = {2026},
  doi = {10.1016/j.jhydrol.2026.135041},
  url = {https://doi.org/10.1016/j.jhydrol.2026.135041}
}