Predicting remediation efficiency of LNAPLs using surrogate polynomial chaos expansion model and global sensitivity analysis

A multi-phase and multi-component numerical simulator assessed the removal efficiencies of light non-aqueous phase liquid (LNAPL) consisting of benzene, toluene, ethylbenzene, and xylene-p. Scenarios of the LNAPL-spilling, natural distribution, and remediation stages were designed in the full-physics numerical modeling. For LNAPLs remediation, a multi-phase extraction (MPE) and a steam injection technique were employed. The removal efficiencies of LNAPLs were computed by systematically varying 6 factors that determine the configuration of the remediation wells. Then, surrogate polynomial chaos expansion (PCE) models mathematically predicting the removal efficiencies were developed through 600 training datasets representing 4 scenario cases; different permeability and the location of SI well were considered in the scenario cases. The PCE models were utilized for Sobol global sensitivity analysis and stochastic Monte Carlo prediction. As a result, the depth of the MPE well was identified as the most significant factor in determining the removal efficiency of the LNAPLs. The removal efficiency was maximized when the MPE well was positioned 1.5 m below the groundwater table. Additionally, the contributions of influencing factors were significantly changed by the field permeability. This study proposed a general framework that efficiently predicts LNAPLs remediation efficiency and identifies key influencing factors by combining advanced numerical modeling, PCE-based surrogate modeling, and sensitivity analyses.