Quantitative evaluation of hydrogeologic properties controlling field-scale DNAPL contamination across porous and fractured media

Abstract

This study presents a novel framework that combines high-resolution 3D numerical simulations with surrogate model-assisted parametric analyses to quantitatively evaluate the effects of field properties on DNAPL contamination and the uncertainty of its entry risk into vulnerable fractured rock. To simulate DNAPL distribution and retention across porous media (PM) and fractured rock (FR), a site-characterized model integrating both media was developed. The target simulation outputs included DNAPL mass in PM and FR, and residually trapped mass in PM. Then, the influence of four hydrogeologic (permeability anisotropy, water table depth, mean porosity of FR, fracture aperture size) and three multiphase-fluid parameters (residual saturation of DNAPLs, DNAPL-water capillary pressures in PM and fracture) on these outputs was assessed through the surrogate model-assisted parametric analyses. Two surrogate models, polynomial chaos expansion (PCE) and random forest (RF), representing distinct theoretical foundations, were employed. Despite their different prediction patterns, both models accurately estimated the simulation outputs (training R2: PCE = 0.943–0.984; RF = 0.974–0.991). Global sensitivity analysis using both models consistently identified permeability anisotropy, water table depth, and residual saturation of DNAPLs as the most influential factors affecting subsurface DNAPL distribution and its risk potential. Finally, risk uncertainty analysis indicated the potential for substantial misprediction due to poorly measurable field properties (i.e., permeability anisotropy, residual saturation), and the aggravated risk under groundwater drawdown at a fractured aquifer site.