Subsurface Source Zone Characterization and Uncertainty Quantification Using Discriminative Random Fields

Abstract

A novel statistical approach is developed and implemented for the stochastic reconstruction of nonaqueous phase liquid (NAPL) source zone realizations and the quantification of source zone metrics and associated uncertainty. The approach employs discriminative random field (DRF) models, to simulate the spatial distributions and relationships among source zone properties (i.e., permeability, NAPL saturation, and aqueous concentration distributions) consistent with commonly collected field data. Application of DRF models requires a limited number of full‐scale simulations to train the model parameters. Monte Carlo sampling methods based on these trained models then provide an efficient method to generate contaminant mass realizations conditioned on measured boreholes, bypassing the need to run computationally intensive, partial differential equation‐based simulations of physical flow and transport. Postprocessing of these realizations yields approximations of uncertainty to inform further sampling for characterization and remediation. The reconstructed contaminant mass realizations provide sufficient information for calculating averaged characterization metrics, such as total contaminant mass and pool fraction, used to predict source zone longevity, mass recovery behavior, and remedial performance. The model performance is evaluated through comparisons of these predicted source zone metrics with those obtained from the “true” mass distributions generated with validated flow and transport models. These comparisons clearly demonstrate that stochastic application of a DRF model can reconstruct realistic saturation and concentration fields, conditioned to borehole data at different times. The present study should be viewed as the first step in generating a three‐dimensional characterization tool that can be applied over a wide range of conditions observed at contaminated sites.

Hydrogeophysical Characterization of Nonstationary DNAPL Source Zones by Integrating a Convolutional Variational Autoencoder and Ensemble Smoother

“…Nevertheless, the large number of realizations required by rejection sampling remains a computational bottleneck. To alleviate the computational efforts, Arshadi et al (2020) coupled the discriminative random field model (DRF, Kumar & Hebert, 2006) and the Monte Carlo (MC) sampling method to generate physically realistic SZA realizations conditioning on borehole measurements. While the results indicated a promising performance of their machine-learning interpolation method (DRF-MC), they only accounted for direct, invasive observations (e.g., permeabilities, NAPL saturations).…”

Integration of Deep Learning‐Based Inversion and Upscaled Mass‐Transfer Model for DNAPL Mass‐Discharge Estimation and Uncertainty Assessment

“…A nonstationary model is one in which the field of interest (e.g., S N ) cannot be represented as the superposition of sinusoidal waves; thus, the mean is not constant and the covariance is not a function of only the separation (Kitanidis, 1997; Li et al., 2007). Because of the significant deviation of the stationary prior model from physical reality, conventional geostatistical methods, especially with sparse data, tend to produce oversmoothed SZA estimates and miss features of importance for DNAPL mass discharge, like dispersed ganglia or lateral and vertical pools (Abriola et al., 2012; Arshadi et al., 2020). To overcome this problem, the alternative methods are needed that can better represent the nonstationary S N field.…”

Numerical modeling and data-worth analysis for characterizing the architecture and dissolution rates of a multicomponent DNAPL source

“…Although scale-dependent mass transfer rate coefficients may simplify grid discretization requirements, the parameterization of NAPL source zones for inverse numerical modeling and uncertainty quantification with spatially-correlated random parameter fields is not straightforward (Arshadi et al, 2020;Kang et al, , 2021bKock & Nowak, 2015 Given that NAPL source zones have complex spatial morphologies with sharp saturation transitions at fine scales, traditional interpolation and geostatistical methods used in groundwater flow modeling may be not suitable for parameterizing NAPL source zones (Arshadi et al, 2020;. Alternative methods proposed for parameterizing NAPL source zones include deep learning algorithms trained with images of saturation distributions generated with multiphase flow simulations on highly-resolved permeability fields (Arshadi et al, 2020;Kang et al, , 2021b, posing additional data requirements and uncertainties on porous media characteristics and model parameters (Abriola, 1989;Agaoglu et al, 2015;. Moreover, these parameterization methods have been tested with synthetically-generated source zones to estimate categories of NAPL saturations through inverse modeling conditioned by borehole data (Arshadi et al, 2020), or by aqueous-phase concentrations under LEA (Kang et al, , 2021b.…”

“…Alternative methods proposed for parameterizing NAPL source zones include deep learning algorithms trained with images of saturation distributions generated with multiphase flow simulations on highly-resolved permeability fields (Arshadi et al, 2020;Kang et al, , 2021b, posing additional data requirements and uncertainties on porous media characteristics and model parameters (Abriola, 1989;Agaoglu et al, 2015;. Moreover, these parameterization methods have been tested with synthetically-generated source zones to estimate categories of NAPL saturations through inverse modeling conditioned by borehole data (Arshadi et al, 2020), or by aqueous-phase concentrations under LEA (Kang et al, , 2021b. Although these methods can generate physically-based, spatially-correlated categorical parameters, they are computationally expensive and require further validation and verification with field data.…”