Progressive transfer learning for advancing machine learning-based reduced-order modeling.

To maximize knowledge transfer and improve the data requirement for data-driven machine learning (ML) modeling, a progressive transfer learning for reduced-order modeling (p-ROM) framework is proposed. A key concept of p-ROM is to selectively transfer knowledge from previously trained ML models and effectively develop a new ML model(s) for unseen tasks by optimizing information gates in hidden layers. The p-ROM framework is designed to work with any type of data-driven ROMs. For demonstration purposes, we evaluate the p-ROM with specific Barlow Twins ROMs (p-BT-ROMs) to highlight how progress learning can apply to multiple topological and physical problems with an emphasis on a small training set regime. The proposed p-BT-ROM framework has been tested using multiple examples, including transport, flow, and solid mechanics, to illustrate the importance of progressive knowledge transfer and its impact on model accuracy with reduced training samples. In both similar and different topologies, p-BT-ROM achieves improved model accuracy with much less training data. For instance, p-BT-ROM with four-parent (i.e., pre-trained models) outperforms the no-parent counterpart trained on data nine times larger. The p-ROM framework is poised to significantly enhance the capabilities of ML-based ROM approaches for scientific and engineering applications by mitigating data scarcity through progressively transferring knowledge.

Development and Characterization of a Sustainable Bio-Polymer Concrete with a Low Carbon Footprint.

Polymer concrete (PC) has been used to replace cement concrete when harsh service conditions exist. Polymers have a high carbon footprint when considering their life cycle analysis, and with increased climate change concerns and the need to reduce greenhouse gas emission, bio-based polymers could be used as a sustainable alternative binder to produce PC. This paper examines the development and characterization of a novel bio-polymer concrete (BPC) using bio-based polyurethane used as the binder in lieu of cement, modified with benzoic acid and carboxyl-functionalized multi-walled carbon nanotubes (MWCNTs). The mechanical performance, durability, microstructure, and chemical properties of BPC are investigated. Moreover, the effect of the addition of benzoic acid and MWCNTs on the properties of BPC is studied. The new BPC shows relatively low density, appreciable compressive strength between 20-30 MPa, good tensile strength of 4 MPa, and excellent durability resistance against aggressive environments. The new BPC has a low carbon footprint, 50% lower than ordinary Portland cement concrete, and can provide a sustainable concrete alternative in infrastructural applications.

Enhancing high-fidelity nonlinear solver with reduced order model.

We propose the use of reduced order modeling (ROM) to reduce the computational cost and improve the convergence rate of nonlinear solvers of full order models (FOM) for solving partial differential equations. In this study, a novel ROM-assisted approach is developed to improve the computational efficiency of FOM nonlinear solvers by using ROM's prediction as an initial guess. We hypothesize that the nonlinear solver will take fewer steps to the converged solutions with an initial guess that is closer to the real solutions. To evaluate our approach, four physical problems with varying degrees of nonlinearity in flow and mechanics have been tested: Richards' equation of water flow in heterogeneous porous media, a contact problem in a hyperelastic material, two-phase flow in layered porous media, and fracture propagation in a homogeneous material. Overall, our approach maintains the FOM's accuracy while speeding up nonlinear solver by 18-73% (through suitable ROM-assisted FOMs). More importantly, the proximity of ROM's prediction to the solution space leads to the improved convergence of FOMs that would have otherwise diverged with default initial guesses. We demonstrate that the ROM's accuracy can impact the computational efficiency with more accurate ROM solutions, resulting in a better cost reduction. We also illustrate that this approach could be used in many FOM discretizations (e.g., finite volume, finite element, or a combination of those). Since our ROMs are data-driven and non-intrusive, the proposed procedure can easily lend itself to any nonlinear physics-based problem.

Reduced order modeling for flow and transport problems with Barlow Twins self-supervised learning.

We propose a unified data-driven reduced order model (ROM) that bridges the performance gap between linear and nonlinear manifold approaches. Deep learning ROM (DL-ROM) using deep-convolutional autoencoders (DC-AE) has been shown to capture nonlinear solution manifolds but fails to perform adequately when linear subspace approaches such as proper orthogonal decomposition (POD) would be optimal. Besides, most DL-ROM models rely on convolutional layers, which might limit its application to only a structured mesh. The proposed framework in this study relies on the combination of an autoencoder (AE) and Barlow Twins (BT) self-supervised learning, where BT maximizes the information content of the embedding with the latent space through a joint embedding architecture. Through a series of benchmark problems of natural convection in porous media, BT-AE performs better than the previous DL-ROM framework by providing comparable results to POD-based approaches for problems where the solution lies within a linear subspace as well as DL-ROM autoencoder-based techniques where the solution lies on a nonlinear manifold; consequently, bridges the gap between linear and nonlinear reduced manifolds. We illustrate that a proficient construction of the latent space is key to achieving these results, enabling us to map these latent spaces using regression models. The proposed framework achieves a relative error of 2% on average and 12% in the worst-case scenario (i.e., the training data is small, but the parameter space is large.). We also show that our framework provides a speed-up of [Formula: see text] times, in the best case, and [Formula: see text] times on average compared to a finite element solver. Furthermore, this BT-AE framework can operate on unstructured meshes, which provides flexibility in its application to standard numerical solvers, on-site measurements, experimental data, or a combination of these sources.

Permeability-controlled migration of induced seismicity to deeper depths near Venus in North Texas.

Migration of seismic events to deeper depths along basement faults over time has been observed in the wastewater injection sites, which can be correlated spatially and temporally to the propagation or retardation of pressure fronts and corresponding poroelastic response to given operation history. The seismicity rate model has been suggested as a physical indicator for the potential of earthquake nucleation along faults by quantifying poroelastic response to multiple well operations. Our field-scale model indicates that migrating patterns of 2015-2018 seismicity observed near Venus, TX are likely attributed to spatio-temporal evolution of Coulomb stressing rate constrained by the fault permeability. Even after reducing injection volumes since 2015, pore pressure continues to diffuse and steady transfer of elastic energy to the deep fault zone increases stressing rate consistently that can induce more frequent earthquakes at large distance scales. Sensitivity tests with variation in fault permeability show that (1) slow diffusion along a low-permeability fault limits earthquake nucleation near the injection interval or (2) rapid relaxation of pressure buildup within a high-permeability fault, caused by reducing injection volumes, may mitigate the seismic potential promptly.

Fast and scalable earth texture synthesis using spatially assembled generative adversarial neural networks.

The earth texture with complex morphological geometry and compositions such as shale and carbonate rocks, is typically characterized with sparse field samples because of an expensive and time-consuming characterization process. Accordingly, generating arbitrary large size of the geological texture with similar topological structures at a low computation cost has become one of the key tasks for realistic geomaterial reconstruction and subsequent hydro-mechanical evaluation for science and engineering applications. Recently, generative adversarial neural networks (GANs) have demonstrated a potential of synthesizing input textural images and creating equiprobable geomaterial images for stochastic analysis of hydrogeological properties, for example, the feasibility of CO2 storage sites and exploration of unconventional resources. However, the texture synthesis with the GANs framework is often limited by the computational cost and scalability of the output texture size. In this study, we proposed a spatially assembled GANs (SAGANs) that can generate output images of an arbitrary large size regardless of the size of training images with computational efficiency. The performance of the SAGANs was evaluated with two and three dimensional (2D and 3D) rock image samples widely used in geostatistical reconstruction of the earth texture and Lattice-Boltzmann (LB) simulations were performed to compare pore-scale flow patterns and upscaled permeabilities of training and generated geomaterial images. We demonstrate SAGANs can generate the arbitrary large size of statistical realizations with connectivity and structural properties and flow characteristics similar to training images, and also can generate a variety of realizations even on a single training image. In addition, the computational time was significantly improved compared to standard GANs frameworks.

Connectivity-informed drainage network generation using deep convolution generative adversarial networks.

Stochastic network modeling is often limited by high computational costs to generate a large number of networks enough for meaningful statistical evaluation. In this study, Deep Convolutional Generative Adversarial Networks (DCGANs) were applied to quickly reproduce drainage networks from the already generated network samples without repetitive long modeling of the stochastic network model, Gibb's model. In particular, we developed a novel connectivity-informed method that converts the drainage network images to the directional information of flow on each node of the drainage network, and then transforms it into multiple binary layers where the connectivity constraints between nodes in the drainage network are stored. DCGANs trained with three different types of training samples were compared; (1) original drainage network images, (2) their corresponding directional information only, and (3) the connectivity-informed directional information. A comparison of generated images demonstrated that the novel connectivity-informed method outperformed the other two methods by training DCGANs more effectively and better reproducing accurate drainage networks due to its compact representation of the network complexity and connectivity. This work highlights that DCGANs can be applicable for high contrast images common in earth and material sciences where the network, fractures, and other high contrast features are important.

Operational and geological controls of coupled poroelastic stressing and pore-pressure accumulation along faults: Induced earthquakes in Pohang, South Korea.

Coupled poroelastic stressing and pore-pressure accumulation along pre-existing faults in deep basement contribute to recent occurrence of seismic events at subsurface energy exploration sites. Our coupled fluid-flow and geomechanical model describes the physical processes inducing seismicity corresponding to the sequential stimulation operations in Pohang, South Korea. Simulation results show that prolonged accumulation of poroelastic energy and pore pressure along a fault can nucleate seismic events larger than Mw3 even after terminating well operations. In particular the possibility of large seismic events can be increased by multiple-well operations with alternate injection and extraction that can enhance the degree of pore-pressure diffusion and subsequent stress transfer through a rigid and low-permeability rock to the fault. This study demonstrates that the proper mechanistic model and optimal well operations need to be accounted for to mitigate unexpected seismic hazards in the presence of the site-specific uncertainty such as hidden/undetected faults and stress regime.

Mineral Fabric as a Hidden Variable in Fracture Formation in Layered Media.

Two longstanding goals in subsurface science are to induce fractures with a desired geometry and to adaptively control the interstitial geometry of existing fractures in response to changing subsurface conditions. Here, we demonstrate that microscopic mineral fabric and structure interact with macroscopic strain fields to generate emergent meso-scale geometries of induced fractures. These geometries define preferential directions of flow. Using additively manufactured rock, we demonstrate that highly conductive flow paths can be formed in tensile fractures by creating corrugated surfaces. Generation, suppression and enhancement of corrugations depend on the relative orientation between mineral fabric and layering. These insights into the role of micro-scale structure on macro-scale flow provide a new method for designing subsurface strategies to maximize potential production or to inhibit flow.

Pore-Scale Analysis of Calcium Carbonate Precipitation and Dissolution Kinetics in a Microfluidic Device.

In this work, we have characterized the calcium carbonate (CaCO3) precipitates over time caused by reaction-driven precipitation and dissolution in a micromodel. Reactive solutions were continuously injected through two separate inlets, resulting in transverse-mixing induced precipitation during the precipitation phase. Subsequently, a dissolution phase was conducted by injecting clean water (pH = 4). The evolution of precipitates was imaged in two and three dimensions (2-, 3-D) at selected times using optical and confocal microscopy. With estimated reactive surface area, effective precipitation and dissolution rates can be quantitatively compared to results in the previous works. Our comparison indicates that we can evaluate the spatial and temporal variations of effective reactive areas more mechanistically in the microfluidic system only with the knowledge of local hydrodynamics, polymorphs, and comprehensive image analysis. Our analysis clearly highlights the feedback mechanisms between reactions and hydrodynamics. Pore-scale modeling results during the dissolution phase were used to account for experimental observations of dissolved CaCO3 plumes with dissolution of the unstable phase of CaCO3. Mineral precipitation and dissolution induce complex dynamic pore structures, thereby impacting pore-scale fluid dynamics. Pore-scale analysis of the evolution of precipitates can reveal the significance of chemical and pore structural controls on reaction and fluid migration.

Metabolism-Induced CaCO3 Biomineralization During Reactive Transport in a Micromodel: Implications for Porosity Alteration.

The ability of Pseudomonas stutzeri strain DCP-Ps1 to drive CaCO3 biomineralization has been investigated in a microfluidic flowcell (i.e., micromodel) that simulates subsurface porous media. Results indicate that CaCO3 precipitation occurs during NO3(-) reduction with a maximum saturation index (SIcalcite) of ∼1.56, but not when NO3(-) was removed, inactive biomass remained, and pH and alkalinity were adjusted to SIcalcite ∼ 1.56. CaCO3 precipitation was promoted by metabolically active cultures of strain DCP-Ps1, which at similar values of SIcalcite, have a more negative surface charge than inactive strain DCP-Ps1. A two-stage NO3(-) reduction (NO3(-) → NO2(-) → N2) pore-scale reactive transport model was used to evaluate denitrification kinetics, which was observed in the micromodel as upper (NO3(-) reduction) and lower (NO2(-) reduction) horizontal zones of biomass growth with CaCO3 precipitation exclusively in the lower zone. Model results are consistent with two biomass growth regions and indicate that precipitation occurred in the lower zone because the largest increase in pH and alkalinity is associated with NO2(-) reduction. CaCO3 precipitates typically occupied the entire vertical depth of pores and impacted porosity, permeability, and flow. This study provides a framework for incorporating microbial activity in biogeochemistry models, which often base biomineralization only on SI (caused by biotic or abiotic reactions) and, thereby, underpredict the extent of this complex process. These results have wide-ranging implications for understanding reactive transport in relevance to groundwater remediation, CO2 sequestration, and enhanced oil recovery.

Adaptation of Delftia acidovorans for degradation of 2,4-dichlorophenoxyacetate in a microfluidic porous medium.

Delftia acidovorans MC1071 can productively degrade R-2-(2,4-dichlorophenoxy)propionate (R-2,4-DP) but not 2,4-dichlorophenoxyacetate (2,4-D) herbicides. This work demonstrates adaptation of MC1071 to degrade 2,4-D in a model two-dimensional porous medium (referred to here as a micromodel). Adaptation for 2,4-D degradation in the 2 cm-long micromodel occurred within 35 days of exposure to 2,4-D, as documented by substrate removal. The amount of 2,4-D degradation in the adapted cultures in two replicate micromodels (~10 and 20 % over 142 days) was higher than a theoretical maximum (4 %) predicted using published numerical simulation methods, assuming instantaneous biodegradation and a transverse dispersion coefficient obtained for the same pore structure without biomass present. This suggests that the presence of biomass enhances substrate mixing. Additional evidence for adaptation was provided by operation without R-2,4-DP, where degradation of 2,4-D slowly decreased over 20 days, but was restored almost immediately when R-2,4-DP was again provided. Compared to suspended growth systems, the micromodel system retained the ability to degrade 2,4-D longer in the absence of R-2,4-DP, suggesting slower responses and greater resilience to fluctuations in substrates might be expected in the soil environment than in a chemostat.

Environmental risk analysis of hazardous material rail transportation.

An important aspect of railroad environmental risk management involves tank car transportation of hazardous materials. This paper describes a quantitative, environmental risk analysis of rail transportation of a group of light, non-aqueous-phase liquid (LNAPL) chemicals commonly transported by rail in North America. The Hazardous Materials Transportation Environmental Consequence Model (HMTECM) was used in conjunction with a geographic information system (GIS) analysis of environmental characteristics to develop probabilistic estimates of exposure to different spill scenarios along the North American rail network. The risk analysis incorporated the estimated clean-up cost developed using the HMTECM, route-specific probability distributions of soil type and depth to groundwater, annual traffic volume, railcar accident rate, and tank car safety features, to estimate the nationwide annual risk of transporting each product. The annual risk per car-mile (car-km) and per ton-mile (ton-km) was also calculated to enable comparison between chemicals and to provide information on the risk cost associated with shipments of these products. The analysis and the methodology provide a quantitative approach that will enable more effective management of the environmental risk of transporting hazardous materials.

NAPL source zone depletion model and its application to railroad-tank-car spills.

We developed a new semi-analytical source zone depletion model (SZDM) for multicomponent light nonaqueous phase liquids (LNAPLs) and incorporated this into an existing screening model for estimating cleanup times for chemical spills from railroad tank cars that previously considered only single-component LNAPLs. Results from the SZDM compare favorably to those from a three-dimensional numerical model, and from another semi-analytical model that does not consider source zone depletion. The model was used to evaluate groundwater contamination and cleanup times for four complex mixtures of concern in the railroad industry. Among the petroleum hydrocarbon mixtures considered, the cleanup time of diesel fuel was much longer than E95, gasoline, and crude oil. This is mainly due to the high fraction of low solubility components in diesel fuel. The results demonstrate that the updated screening model with the newly developed SZDM is computationally efficient, and provides valuable comparisons of cleanup times that can be used in assessing the health and financial risk associated with chemical mixture spills from railroad-tank-car accidents.

Estimation of interfacial tension between organic liquid mixtures and water.

Knowledge of IFT values for chemical mixtures helps guide the design and analysis of various processes, including NAPL remediation with surfactants or alcohol flushing, enhanced oil recovery, and chemical separation technologies, yet available literature values are sparse. A comprehensive comparison of thermodynamic and empirical models for estimating interfacial tension (IFT) of organic chemical mixtures with water is conducted, mainly focusing on chlorinated organic compounds for 14 ternary, three quaternary, and one quinary systems. Emphasis is placed on novel results for systems with three and four organic chemical compounds, and for systems with composite organic compounds like lard oil and mineral oil. Seven models are evaluated: the ideal and nonideal monolayer models (MLID and MLNID), the ideal and nonideal mutual solubility models (MSID and MSNID), an empirical model for ternary systems (EM), a linear mixing model based on mole fractions (LMMM), and a newly developed linear mixing model based on volume fractions of organic mixtures (LMMV) for higher order systems. The two ideal models (MLID and MSID) fit ternary systems of chlorinated organic compounds without surface active compounds relatively well. However, both ideal models did not perform well for the mixtures containing a surface active compound. However, for these systems, both the MLNID and MSNID models matched the IFT data well. It is shown that the MLNID model with a surface coverage value (0.00341 mmol/m2) obtained in this study can practically be used for chlorinated organic compounds. The LMMM results in poorer estimates of the IFT as the difference in IFT values of individual organic compounds in a mixture increases. The EM, with two fitting parameters, provided accurate results for all 14 ternarysystems including composite organic compounds. The new LMMV method for quaternary and higher component systems was successfully tested. This study shows that the LMMV may be able to be used for higher component systems and it can be easily incorporated into compositional multiphase flow models using only parameters from ternary systems.

Numerical and experimental investigation of DNAPL removal mechanisms in a layered porous medium by means of soil vapor extraction.

The purpose of this work is to identify the mechanisms that govern the removal of carbon tetrachloride (CT) during soil vapor extraction (SVE) by comparing numerical and analytical model simulations with a detailed data set from a well-defined intermediate-scale flow cell experiment. The flow cell was packed with a fine-grained sand layer embedded in a coarse-grained sand matrix. A total of 499 mL CT was injected at the top of the flow cell and allowed to redistribute in the variably saturated system. A dual-energy gamma radiation system was used to determine the initial NAPL saturation profile in the fine-grained sand layer. Gas concentrations at the outlet of the flow cell and 15 sampling ports inside the flow cell were measured during subsequent CT removal using SVE. Results show that CT mass was removed quickly in coarse-grained sand, followed by a slow removal from the fine-grained sand layer. Consequently, effluent gas concentrations decreased quickly at first, and then started to decrease gradually, resulting in long-term tailing. The long-term tailing was mainly due to diffusion from the fine-grained sand layer to the coarse-grained sand zone. An analytical solution for a one-dimensional advection and a first-order mass transfer model matched the tailing well with two fitting parameters. Given detailed knowledge of the permeability field and initial CT distribution, we were also able to predict the effluent concentration tailing and gas concentration profiles at sampling ports using a numerical simulator assuming equilibrium CT evaporation. The numerical model predictions were accurate within the uncertainty of independently measured or literature derived parameters. This study demonstrates that proper numerical modeling of CT removal through SVE can be achieved using equilibrium evaporation of NAPL if detailed fine-scale knowledge of the CT distribution and physical heterogeneity is incorporated into the model. However, CT removal could also be fit by a first-order mass transfer analytical model, potentially leading to an erroneous conclusion that the long-term tailing in the experiment was kinetically controlled due to rate-limited NAPL evaporation.

An environmental screening model to assess the consequences to soil and groundwater from railroad-tank-car spills of light non-aqueous phase liquids.

North American railroads transport a wide variety of chemicals, chemical mixtures and solutions in railroad tank cars. In the event of an accident, these materials may be spilled and impact the environment. Among the chemicals commonly transported are a number of light non-aqueous phase liquids (LNAPLs). If these are spilled they can contaminate soil and groundwater and result in costly cleanups. Railroads need a means of objectively assessing the relative risk to the environment due to spills of these different materials. Environmental models are often used to determine the extent of contamination, and the associated environmental risks. For LNAPL spills, these models must account for NAPL infiltration and redistribution, NAPL dissolution and volatilization, and remediation systems such as pump and treat. This study presents the development and application of an environmental screening model to assess NAPL infiltration and redistribution in soils and groundwater, and to assess groundwater cleanup time using a pumping system. Model simulations use parameters and conditions representing LNAPL releases from railroad tank cars. To take into account unique features of railroad-tank-car spill sites, the hydrocarbon spill screening model (HSSM), which assumes a circular surface spill area and a circular NAPL lens, was modified to account for a rectangular spill area and corresponding lens shape at the groundwater table, as well as the effects of excavation and NAPL evaporation to the atmosphere. The modified HSSM was first used to simulate NAPL infiltration and redistribution. A NAPL dissolution and groundwater transport module, and a pumping system module were then implemented and used to simulate the effects of chemical properties, excavation, and free NAPL removal on NAPL redistribution and cleanup time. The amount of NAPL that reached the groundwater table was greater in coarse sand with high permeability than in fine sand or silt with lower permeabilities. Excavation can reduce the amount of NAPL that reaches the groundwater more effectively in lower permeability soils. The effect of chemical properties including vapor pressure and the ratio of density to viscosity become more important in fine sand and silt soil due to slow NAPL movement in the vadose zone. As expected, a pumping system was effective for high solubility chemicals, but it was not effective for low solubility chemicals due to rate-limited mass transfer by transverse dispersion and flow bypassing. Free NAPL removal can improve the removal efficiency for moderately low solubility chemicals like benzene, but cleanup times even after free NAPL removal can be prolonged for very low solubility chemicals like cyclohexane and styrene.

Impact of nonaqueous phase liquid (NAPL) source zone architecture on mass removal mechanisms in strongly layered heterogeneous porous media during soil vapor extraction.

An existing multiphase flow simulator was modified in order to determine the effects of four mechanisms on NAPL mass removal in a strongly layered heterogeneous vadose zone during soil vapor extraction (SVE): a) NAPL flow, b) diffusion and dispersion from low permeability zones, c) slow desorption from sediment grains, and d) rate-limited dissolution of trapped NAPL. The impacts of water and NAPL saturation distribution, NAPL-type (i.e., free, residual, or trapped) distribution, and spatial heterogeneity of the permeability field on these mechanisms were evaluated. Two different initial source zone architectures (one with and one without trapped NAPL) were considered and these architectures were used to evaluate seven different SVE scenarios. For all runs, slow diffusion from low permeability zones that gas flow bypassed was a dominant factor for diminished SVE effectiveness at later times. This effect was more significant at high water saturation due to the decrease of gas-phase relative permeability. Transverse dispersion contributed to fast NAPL mass removal from the low permeability layer in both source zone architectures, but longitudinal dispersion did not affect overall mass removal time. Both slow desorption from sediment grains and rate-limited mass transfer from trapped NAPL only marginally affected removal times. However, mass transfer from trapped NAPL did affect mass removal at later time, as well as the NAPL distribution. NAPL flow from low to high permeability zones contributed to faster mass removal from the low permeability layer, and this effect increased when water infiltration was eliminated. These simulations indicate that if trapped NAPL exists in heterogeneous porous media, mass transfer can be improved by delivering gas directly to zones with trapped NAPL and by lowering the water content, which increases the gas relative permeability and changes trapped NAPL to free NAPL.

Evaluation of simplified mass transfer models to simulate the impacts of source zone architecture on nonaqueous phase liquid dissolution in heterogeneous porous media.

Nonaqueous phase liquid (NAPL) dissolution was studied in three-dimensional (3D) heterogeneous experimental aquifers (25.5 cm x 9 cm x 8.5 cm) with two different longitudinal correlation lengths (2.1 cm and 1.1 cm) and initial spill volumes (22.5 ml and 10.5 ml). Spatial and temporal distributions of NAPL during dissolution were measured using magnetic resonance imaging (MRI). At high NAPL spill volume, average effluent concentrations initially increased during dissolution, as NAPL pools transitioned to NAPL ganglia, and then decreased as the total NAPL-water interfacial area decreased over time. Experimental results were used to test six dissolution models: (i and ii) a one-dimensional (1D) model using either specific NAPL-water interfacial area values estimated from MR images at each time step (i.e., 1D quasi-steady state model), or an empirical mass transfer (Sh') correlation (i.e., 1D transient model), (iii and iv) a multiple analytical source superposition technique (MASST) using either the NAPL distribution determined from MR images at each time step (i.e., MASST steady state model), or the NAPL distribution determined from mass balance calculations (i.e., MASST transient model), (v) an equilibrium streamtube model, and (vi) a 3D grid-scale pool dissolution model (PDM) with a dispersive mass flux term. The 1D quasi-steady state model and 3D PDM captured effluent concentration values most closely, including some concentration fluctuations due to changes in the extent of flow reduction. The 1D transient, MASST steady state and transient, and streamtube models all showed a monotonic decrease in effluent concentration values over time, and the streamtube model was the most computationally efficient. Changes during dissolution of the effective NAPL-water interfacial area estimated from imaging data are similar to changes in effluent concentration values. The 1D steady state model incorporates estimates of the effective NAPL-water interfacial area directly at each time point; the 3D PDM does so indirectly through mass balance and a relative permeability function, which causes reduced water flow through high saturation NAPL regions. Hence, when model accuracy is required, the results indicate that a surrogate of this effective interfacial area is required. Approaches to include this surrogate in the MASST and streamtube models are recommended.