Evaluation of Stratified Prediction Methods for Spatial Distribution of Groundwater Contaminants (Benzene, Total Petroleum Hydrocarbons, and MTBE) at Abandoned Petrochemical Sites

This study evaluates the accuracy of various Geographic Information System interpolation methods in predicting the stratified spatial distribution of organic pollutants (Benzene, Total Petroleum Hydrocarbons [TPH], and Methyl Tert-butyl Ether [MTBE]) in groundwater at a petrochemical-contaminated site. Given the limitations of traditional monitoring methods in predicting spatial distribution, this study focuses on the spatial computational prediction of volatile organic compound concentrations at a former petrochemical industrial site. Three interpolation methods—Inverse Distance Weighting (IDW), Radial Basis Function (RBF), and Ordinary Kriging (OK)—were applied and evaluated. Prediction accuracy was assessed using leave-one-out cross-validation, with performance quantified through key metrics: Root Mean Square Error, Coefficient of Determination, and Spearman’s Rank Correlation Coefficient. Results demonstrate significant variations in optimal prediction methods depending on pollutant type and depth stratum. For pollutants predominantly enriched in shallow and middle layers (Benzene, TPH), OK yielded the highest accuracy and stability. Conversely, for predictions of pollutants primarily concentrated in deeper layers, RBF achieved superior performance. IDW consistently underperformed across all strata and pollutants. All interpolation methods generally exhibited systematic overestimation of pollutant concentrations (mean cross-validation error > 0). Through a hierarchical evaluation of the accuracy and interpolation effectiveness of these methods, this study develops a more accurate modeling framework to describe the composite groundwater contamination patterns at petrochemical sites. This study systematically evaluates the spatial prediction accuracy of various non-aqueous phase liquid species under differing groundwater-table depths, identifies the most robust interpolation method, and thereby provides a benchmark for enhancing predictive fidelity in subsurface contaminant mapping.