Machine Learning-Based Estimation of Soil’s True Air-Entry Value from GSD Curves

The application of machine learning (ML) methods has proven to be promising in dealing with a wide range of geotechnical engineering problems in recent years. ML methods have already been used for the prediction of soil water retention curves (SWRC) and estimation of air-entry values (AEV). However, the reported works in the literature are generally based on limited data and conventional, less accurate approaches for AEV estimation. In this paper, a large database, known as UNsaturated SOil hydraulic DAtabase (UNSODA), is studied and the conventional and true AEVs of 790 soil samples are estimated based on determination methods reported in the literature. A ML approach is then employed for the development of a predictive model for the estimation of true AEV from water content-based SWRCs of a wide range of soil types taking into account the impact of bulk density and grain size distribution parameters. The obtained results reveal an enhanced accuracy in AEV determination, featuring R2 values of 0.964, 0.901 and 0.851 for training, validation, and testing data, respectively, which confirm the marked performance of the developed ML model. Based on the results of a sensitivity analysis, the particle sizes of 50 and 250 µm are found to have the highest impact on the AEV estimation.