Quantitative characterization of LNAPL contaminant concentrations in the vadose zone by integrating environmental factors and interpretable support vector machine models

Accurate monitoring of light non-aqueous phase liquids (LNAPLs) contaminant in the vadose zone is essential for groundwater protection. This study employed diesel as a representative LNAPL contaminant and systematically analyzed the response relationships between key environmental factors (volumetric water content (θ), electrical conductivity (EC), pH, and oxidation-reduction potential (ORP)) and contaminant transport using sand tank experiments. Based on these results, a novel method for quantifying contaminant concentration was developed using multi-environmental factors. Utilizing high-precision experimental data, this research identified three transport stages of LNAPL in fine-sand vadose zones. Within this framework, an interpretable concentration calculation model was constructed by integrating the support vector machine (SVM) algorithm and SHapley Additive exPlanations (SHAP) method coupled with multi-environmental factors. Results demonstrate that the model exhibits excellent generalization performance (R² > 0.90), enabling precise quantification of LNAPL contamination plumes and accurate prediction of leakage dynamics. SHAP analysis clarified the contribution of each input variable, confirming θ as the most influential factor in predicting the spatial variation of total petroleum hydrocarbon (TPH) concentration. Furthermore, it quantitatively revealed significant spatial heterogeneity in the key environmental factors dominating TPH concentration and their relative contributions. In conclusion, this study provides theoretical and modeling support for developing real-time, environmental factor-based risk assessment methods for soil LNAPL contamination.