Representative Elementary Volume Estimation and Neural Network-Based Prediction of Change Rates of Dense Non-Aqueous Phase Liquid Saturation and Dense Non-Aqueous Phase Liquid–Water Interfacial Area in Porous Media

Abstract

Investigation of the change rate for contaminant parameters is important to characterize dense non-aqueous phase liquid (DNAPL) transport and distribution in groundwater systems. In this study, four experiments of perchloroethylene (PCE) migration are conducted in two-dimensional (2D) sandboxes to characterize change rates of PCE saturation (So) and PCE–water interfacial area (AOW) under different conditions of salinity, surface active agent, and heterogeneity. Associated representative elementary volume (REV) of the change rate of So (So rate) and change rate of AOW (AOW rate) is derived over the long-term transport process through light transmission techniques. REV of So rate (SR-REV) and REV of AOW rate (AR-REV) are estimated based on the relative gradient error (εgi). Regression analysis is applied to investigate the regularity, and a model based on a back-propagation (BP) neural network is built to simulate and predict the frequencies of SR-REV and AR-REV. Experimental results indicated the salinity, surface active agent, and heterogeneity are important factors that affect the So rate, AOW rate, SR-REV, and AR-REV of the PCE plume in porous media. The first moment of the PCE plume along the vertical direction is decreased under conditions of high salinity, surface active agent, and heterogeneity, while these factors have different effects on the second moment of the PCE plume. Compared with the salinity and surface active agent, heterogeneity has the greatest effect on the GTP, the distributions of the So rate and AOW rate along the depth, and dM, dI. For SR-REV, the standard deviation is increased by the salinity, surface active agent, and heterogeneity. Simultaneously, the salinity and heterogeneity lead to lower values of the mean value of SR-REV, while the surface active agent increases the mean value of SR-REV. However, the mean and standard deviation of AR-REV have no apparent difference under different experimental conditions. These findings reveal the complexity of PCE transport and scale effect in the groundwater system, which have important significance in improving our understanding of DNAPL transport regularity and promoting associated prediction.