Machine learning estimation of unsaturated soil hydraulic conductivity functions using literature-derived pedotransfer functions and an interactive graphical user interface

Measurement of unsaturated soil hydraulic conductivity K ( h ), remains one of the most expensive and time-consuming procedures in geotechnical and hydrological engineering. This study presents a machine learning (ML) framework coupled with an interactive graphical user interface (GUI) for estimating unsaturated soil hydraulic properties from basic soil characterization data. The framework employs Random Forest and XGBoost ensemble algorithms to predict van Genuchten parameters ( θ r , θ s , α , n , K sat ) using readily available soil properties, including texture, bulk density, organic carbon, and particle size distribution. The models were trained and validated on a literature-derived synthetic dataset of 1500 soil samples generated using established empirical relationships from the ROSETTA, UNSODA, and Cosby databases. A 75/25 train-test split was used, with five-fold cross-validation performed on the training subset for hyperparameter optimization. Test-set performance yielded R 2 . Five-fold cross-validation yielded R 2 values of 0.766 for θ r , 0.937 for θ s , 0.995 for α , 0.984 for n , and 0.151 for log( K sat ). The predicted parameters were subsequently used to generate full Soil Water Retention Curves (SWRC) and unsaturated hydraulic conductivity curves K ( h ) via the Mualem-van Genuchten model. Comparison with classical empirical pedotransfer functions and multiple linear regression demonstrated comparable performance for K sat prediction while offering superior capabilities for full-curve generation. It is explicitly acknowledged that the reported accuracy reflects the models’ ability to reproduce synthetic reference targets rather than independent experimental observations. The accompanying interactive GUI enables real-time prediction, batch processing, and export of results for both research and practical applications in agriculture and environmental engineering.