Strategies for Managing Risk due to Back Diffusion

Abstract

Back diffusion of contaminants from secondary sources may hamper site remediation if it is not properly addressed in the remedial design. A review of all reported technologies and strategies that have been or could be applied to address plume persistence due to back diffusion as published in the peer-reviewed literature is provided. We classify these into four major categories. The first category consists of those approaches that do not include active measures to specifically address contamination in the low permeable zones (LPZs) and can therefore be considered passive LPZ management approaches. A disadvantage of these approaches is the long duration that may be required to meet acceptable endpoints; however, this allows degradation to potentially play a significant part even at modest rates. The remaining three categories all use approaches to specifically address contaminants in the LPZ. The second category consists of strategies that promote contaminant destruction through the forward diffusion of amendments into the LPZ. A variety of laboratory tests indicate concentration or flux reductions range from no improvement, to reductions as high as four orders-of-magnitude depending on the evaluation metric. The third category consists of strategies that alter physical characteristics of the secondary source, and includes viscosity modification, fracturing, and soil mixing. Each of these offer unique advantages and are often used to deliver one or more amendments for contaminant treatment. The final category consists of thermal and electrokinetic remediation, both less susceptible to permeability contrast limitations. However, they are not routinely used for secondary-source treatment.

Introduction

In the context of groundwater contaminants, back diffusion refers to the contamination of a high permeability (k, [L²]) layer or zone (HPZ) by the diffusive transport of contaminants out of an adjacent low k layer or zone (low permeable zone, LPZ). Consequently, the term refers to a setting with regions dominated by two distinct types of transport: one dominated by advection, or advection and dispersion, and the other dominated by diffusion. Contaminants reside in the LPZs because at some earlier time a concentration gradient was directed from the HPZ into the LPZ, resulting in diffusive transport into the LPZ (i.e., forward diffusion). For example, the release of a nonaqueous phase liquid (NAPL), such as fuel or chlorinated solvent, into the subsurface results in a NAPL source zone. This serves as the primary contaminant source for dissolved contaminant transport through the HPZ initially, which in turn creates concentration gradients that drive contaminants into the LPZ through forward diffusion (Figure 1a). Once the primary contaminant source (e.g., NAPL) has been removed, isolated, or exhausted, the HPZ is subsequently swept clean, resulting in a reversal of the concentration gradients which now drives contaminant mass back out of the LPZ and into the HPZ (Figure 1b). Back diffusion can be a significant factor at some sites that sustains contaminant plumes after the primary contaminant source is eliminated, and as such, contaminant in LPZs is considered a secondary contaminant source.

Figure 1.

Conceptual illustrations of (a) primary source creating secondary source; (b) secondary source sustaining plume after primary source removed; (c) downgradient plume containment; (d) amendment injection into the HPZ, with amendment forward diffusion; (e) HPZ/LPZ interface treatment zone; and (f) fracturing of the LPZ.

Citations related to contaminant transport in HPZs with diffusional transport in adjacent LPZs are numerous. Examples of geologic settings in published works where diffusional transport in LPZs has been a significant feature include fractured chalk (Foster 1975; Lawrence et al. 1990), clay (McKay et al. 1993; Falta 2005; Huang and Goltz 2015), and rock (Neretnieks 1980; Parker et al. 1997; Bodin et al. 2003). Examples of unconsolidated geologic settings include an aquifer with LPZ inclusions or beds (Guswa and Freyberg 2000; Parker et al. 2008; McDade et al. 2013), or an adjacent aquitard (Sudicky et al. 1985; Ball et al. 1997; Parker et al. 2004). Modeling studies have ranged from analytical solutions for aqueous solute transport in a single fracture through consolidated media (Grisak and Pickens 1980; Tang et al. 1981; Houseworth et al. 2013), to three-dimensional (3D) numerical solutions for multiphase flow in a fractured network (Slough et al. 1999; Reynolds and Kueper 2004). Investigations have ranged in scale from bench-top studies using layered unconsolidated media (Sudicky et al. 1985; Sale et al. 2008) to field studies at contaminated sites (Foster 1975; Ball et al. 1997).

Early works demonstrated that forward diffusion into LPZs can be a significant mechanism for retardation and attenuation of dissolved-phased contaminant transport in HPZs (Foster 1975; Grisak and Pickens 1980; Neretnieks 1980; Tang et al. 1981; McKay et al. 1993; Zuber and Motyka 1994). Plume attenuation and retardation due to diffusion in fractured media has been noted in later work as well (Reynolds and Kueper 2001; Lipson et al. 2005; Chapman et al. 2013). Another common feature that has been well documented is tailing in solute breakthrough curves due to diffusional transport in LPZs (Sudicky et al. 1985; Goltz and Roberts 1986; Carrera et al. 1998; Hantush and Mariño 1998). Parker et al. (1994) explicitly consider NAPL in fractured media along with diffusion into LPZs, and demonstrated that for typical fractured rock characteristics, the storage capacity in the matrix for dissolved and sorbed contaminant phases exceeded the storage capacity for NAPL in factures. They also showed that transfer of the contaminant from the NAPL in the fractures to the matrix through forward diffusion may occur in a relatively short time. Others have likewise noted the relatively high contaminant storage capacity of the LPZ in fractured rock and other geologic settings (Sudicky and Frind 1982; Parker et al. 1997; Chapman and Parker 2005; Falta 2005; Sale et al. 2008; Brown et al. 2012).

Contaminant mass in the LPZ, if not degraded through biotic or abiotic mechanisms, will return to the HPZ through back diffusion once the concentrations in the HPZ decrease below those in the LPZ. Back diffusion and its ability to sustain contaminant plumes after primary source remediation has been a more recent research focus (Liu and Ball 2002; Chapman and Parker 2005; Parker et al. 2008, 2010; Sale et al. 2008; Rasa et al. 2011; Chapman et al. 2012; Dearden et al. 2013; Wilking et al. 2013; Yang et al. 2017). The duration required to remove the contaminant mass from the LPZ has been shown to be much longer than the time for forward diffusion (Parker et al. 1994, 1997; Chapman and Parker 2005; Falta 2005; Brown et al. 2012; Yang et al. 2016). Moreover, back diffusion may sustain contaminant plumes over durations longer than typically envisioned to meet regulatory standards through remediation activities of a few decades. In some cases, it has been noted that plumes may persist for centuries due to back diffusion (LaBolle and Fogg 2001; Polak et al. 2003; Chapman et al. 2013).

Given the potential for back diffusion to sustain contaminant groundwater plumes, it is important to understand remedial technologies and strategies that may be applied to address the risk posed. Remediation in environments that create back diffusion is challenging because the LPZs impede contaminant removal by flushing as well as the delivery of amendments to promote abiotic or biotic degradation (Parker et al. 1994; Falta 2005). Freeze and McWhorter (1997) present a framework for evaluating dense nonaqueous phase liquid (DNAPL) contaminant risk in low-k media, but noted a lack of information on remedial performance in LPZs as a limitation to properly evaluate risk reduction associated with contaminant mass removal in LPZs. Walden (1997) provided a review of seven technologies applicable to LPZs with a focus on petroleum contaminants. Technologies were divided into three groups, those for contaminant removal (soil vapor extraction, bioventing), mobility enhancement (thermal processes, surfactant flushing, in situ soil mixing), and k enhancement (hydraulic and pneumatic fracturing). While this review still offers valuable insight, it is more than two decades old and does not include some options currently available, such as electrokinetic (EK) treatment. Kuppusamy et al. (2016) provide a comprehensive review of all in situ remediation technologies but is not specific to the issue of back diffusion. Horst et al. (2019) provided a review of selected remediation technologies for low-k media, and categorized nine technologies into three groups: complete treatment (excavation; high-temperature thermal; and soil mixing, stabilization, and solidification), source reduction (EKs, low-temperature thermal, and hydraulic fracturing), and flux control (passive barrier systems, phytohydraulics, phytoremediation, and horizontal reactive media treatment wells). However, as noted the review was selective and relied in part on the authors’ own remediation experiences and did not include references to many citations with direct relevance to the topic. Most recently, You et al. (2020) provided an extensive review on the topic of diffusive transport between high and LPZs, but the review was not limited to remediation.

Consequently, provided herein is a more complete review of all reported technologies and strategies that could be applied to address plume persistence due to back diffusion as published in the peer-reviewed literature. A focus was placed on works published in peer-reviewed journals to limit the scope of the review. In many cases, technologies have been proposed or used to address contaminants in low k media without specifically addressing back diffusion. These references are also included given their potential utility to address back diffusion. This review is intended to provide a state-of-the-science resource to assist in evaluating treatment options at sites where back diffusion has been identified as a significant factor.

Overview

To aid with the review, remedial strategies and techniques have been organized into four major categories as illustrated in Figure 2. The cost and difficulty of implementation, as well as remedial performance, generally increases from left to right. The first major category consists of those approaches that do not include active measures to specifically address contamination in the LPZs, and can therefore be considered passive LPZ management approaches. Approaches in this category can be subdivided into monitored natural attenuation (MNA) for both HPZ and LPZ, or those that apply treatments that remove or degrade contamination in the HPZ and rely on diffusional transport as the mechanism to remove contamination from LPZs. Conventional pump-and-treat (PAT) or down gradient permeable reactive barriers (PRBs) would fall into the latter category.

Figure 2.

Summary of strategies and treatments to address plume persistence due to back diffusion. The cost and difficulty of implementation, as well as remedial performance generally increases from left to right.

The remaining three major categories all use approaches to specifically address contaminants in the LPZ, and can therefore be considered active LPZ management approaches. Of these, the first major active category includes those approaches that attempt to promote contaminant destruction through the forward diffusion of amendments into the LPZ, and is subdivided into two categories based on amendment type: amendments for in situ chemical oxidation (ISCO) or in situ chemical reduction, or amendments to promote bioremediation. The second major active category consists of those strategies that alter physical characteristics of the subsurface to minimize the influence of back diffusion. This category is subdivided into two major types, the first of which consists of approaches that minimize k contrasts using viscosity amendments to change hydraulic conductivity (K, [LT⁻¹]). The second subdivision includes those approaches that alter the geologic stratigraphy. The most dramatic example of this is excavation, but the category also includes hydraulic/pneumatic fracturing and soil mixing. The final category includes those approaches that use remedial techniques free from, or at least less susceptible to, limitations associated with k contrasts, and includes thermal and EK remediation techniques.

Table 1 summarizes and compares four flux equations relevant to remedial treatment in LPZs. Values of K for geologic media may range across many orders of magnitude, and heterogeneity in K at a given site leads to plume persistence due to back diffusion. When K of the LPZ is sufficiently small, q will likewise be small and advection will no longer be the dominant transport mechanism. It may not be appropriate to assume advection is zero in the LPZ, since it has been suggested that diffusion and slow advection may be important LPZ transport processes (Sale et al. 2013). Nonetheless, the most extreme cases of plume persistence occur when the dominant LPZ transport process is diffusion. Walden (1997) and Sorenson et al. (2019) suggested K< ~1 cm/day (10⁻⁵ cm/s) as a LPZ characteristic, while Horst et al. (2019) suggested K< ~10 cm/day (10⁻⁴ cm/s) as measured from bulk aquifer hydraulic tests, as one of six characteristics in a practical definition of LPZ material. The other characteristics were geology, direct-push tip stress measurements, hydraulic gradients, extraction rates, and injection rates. Sale et al. (2013) highlighted the importance of contaminant mass storage in the LPZ, and noted the following as key LPZ features: contaminant sorption characteristics, reducing conditions, thickness and areal extent, and the historical presence of DNAPL at HPZ/LPZ interfaces.

Table 1.

Summary of Transport Flux Equations as the Underlying Basis for Contaminant Remediation

Name and Type	Formula	Definitions	Approximate Parameter Range	Equation
Darcy’s law, groundwater flux	$q=-K\frac{\partial h}{\partial x}$	q = groundwater flux [LT−1], K = hydraulic conductivity [LT−1], h = hydraulic head [L], x = length [L]	10−5 ≤ K ≤ 106 cm/daya	(1)
Fick’s first law, diffusive chemical flux	$J=-\eta D_{eff}\frac{\partial C}{\partial x}$	J = diffusive flux [ML−2 T−1], η = porosity [–], Deff = effective diffusion coefficient [L2T−1], C = aqueous concentration [ML−3]	0.001 ≤ Deff ≤ 1 cm2/dayb,c	(2)
Fourier’s law, heat flux	$q_H=-\lambda \frac{\partial \theta }{\partial x}$	qH = heat flux [MT−3], λ = thermal conductivity [LMΘ−1 T−3], θ = temperature [Θ]	0.002 ≤ λ ≤ 0.08 W/cm/Kd	(3)
Helmholtz-Smoluchowski equation, chemical flux	$J_E=C\left(u-k_e\right)\frac{\partial E}{\partial x}$	JE = chemical flux [ML−2 T−1], u = effective ionic mobility [T2IM−1], ke = electroosmotic permeability [T2IM−1], E = electric potential [L2MT−3I−1]	0.04 ≤ u ≤ 40 cm2/day/Ve 0.01 ≤ ke ≤ 10 cm2/day/Vf	(4)

^aFreeze and Cherry (1979).

^bThe range for D_eff is based on τD_aq, where τ = tortuosity [–] and D_aq = aqueous diffusion coefficient [L²T⁻¹]; typical ranges of 0.01 ≤ τ ≤ 0.8 (Shackelford 1991) and 0.1 ≤ D_aq ≤ 1 cm²/day for most organic groundwater contaminants were assumed.

^cThe assumption is often made that D_eff≠f(C), which is appropriate for dilute solutions. In cases where solutions of remedial amendments have high C, the question of whether D_eff = constant appears to be unexplored.

^dBased on data from Santa et al. (2020): values range from ∼0.5 to ∼8 W/m/K depending on porosity and water saturation for rock and ∼ 0.2 to ∼3 W/m/K based on media type and water saturation for unconsolidated media.

^eBased on the range shown for D_eff and the Nernst-Einstein-Townsend equation, assuming a valency of 1 and a temperature of 10°C. The range is also consistent with values reported by Gill et al. (2014).

^fBased on information from Acar et al. (1995) and Gill et al. (2014).

The extent to which contaminant plumes in the HPZ are sustained by diffusive contaminant transport in the LPZs depends not only on the absolute value of k in the LPZ, but also on the k contrast between the HPZ and LPZ. Large contrasts in k between the LPZ and HPZ promote regions with distinct transport mechanisms (i.e., advective/dispersive vs. diffusive dominated transport), while small contrasts in k suggest minimum differences in transport mechanisms. While some tailing may result from small contrasts, significant plume persistence due to back diffusion stems from distinct differences in HPZ and LPZ transport mechanisms due to order-of-magnitude differences in k. To assist with comparisons between published works, we use κ to characterize the contrast in K between the HPZ and LPZ:

$$\kappa =\log \left(K_{HPZ}/K_{LPZ}\right),$$ (5)

where K_HPZ and K_LPZ are hydraulic conductivity in the HPZ and LPZ, respectively. In this manner, κ represents order-of-magnitude differences in hydraulic conductivity between the HPZ and LPZ.

Literature Review

Passive LPZ Management

Whether intentional or not, the oldest form of management for back diffusion is treatment or containment of the contaminants in the HPZs without specific activities to address contaminants in the LPZs. As such, LPZ management can be considered a passive activity. Back diffusion and natural LPZ abiotic/biotic degradation are the sole mechanisms for contaminant removal from the LPZ. While any number of remediation treatments could be used to address HPZ contamination, two strategies most often used are containment, including physical containment, hydraulic control using PAT systems (Figure 1c), or downgradient permeable reactive barriers; or conventional amendment injection (Figure 1d). In the context of this paper, conventional amendment injection refers to aqueous phase amendment injection without specific intent to overcome limitations stemming from k contrasts, as discussed below. Amendments include any chemical that promotes abiotic or biotic degradation, or that serves as a contaminant sorbent (e.g., injection of powder activated carbon).

Pump and Treat Systems

Limitations on PAT systems due to diffusional transport were first noted roughly three decades ago (Keely 1989; Mackay and Cherry 1989; Mercer et al. 1990), and publications on the topic have continued (Mackay et al. 2000; LaBolle and Fogg 2001; Ishimori et al. 2006; McDade et al. 2013; Seyedabbasi et al. 2013; Guo et al. 2019). Several publications have indicated that PAT in settings with back diffusion may require timeframes on the order of one or more centuries for restoration (LaBolle and Fogg 2001; Lemming et al. 2012; McDade et al. 2013; Guo et al. 2019). For example, LaBolle and Fogg (2001) simulated PAT remediation using the alluvial aquifer characteristics found at the Lawrence Livermore National Laboratory Superfund Site in Livermore, California. The model used four hydrofacies to represent the subsurface geology, and κ = 5 using the highest and lowest K estimates. Results of their simulations indicated that mass removal due to PAT from the leading edge of the plume was more rapid than mass removal from the trailing edge of the plume near the source area because the latter experienced a longer duration of forward diffusion than the former. For a nondegrading contaminant, their results suggested that the residence time in the LPZ for the system modeled would be on the order of centuries to millennia, and that mass recovery from the LPZ due to PAT would be insignificant on the time scale of decades. They concluded that “…effective molecular diffusivity, and abundance and architecture of low-K materials are key factors governing time to clean-up in many remediation projects.”

Several strategies have been proposed to improve PAT performance in heterogeneous environments. These include chaotic advection (Trefry et al. 2012; Cho et al. 2019), directed groundwater recirculation (Suthersan et al. 2015), enhanced flushing (Sale et al. 2013), and vertical circulation wells (Tatti et al. 2019). Case studies illustrating how well these strategies may address plume persistence due to back diffusion at actual contaminated sites were not found in the literature. Cho et al. (2019) reported results from a field trial with chaotic advection to promote mixing through transient switching of pressure in a series of wells, but it was a proof-of-concept study conducted in a mildly heterogeneous aquifer (κ ~ 1). In large part, the success of these enhanced PAT strategies to address plume persistence due to back diffusion may depend on the architecture of the LPZs. While it seems reasonable to suspect beneficial outcomes in some settings (e.g., sites with mild κ and thin, discontinuous LPZs) or when addressing issues related to slow advection, further study appears needed to demonstrate their effectiveness as the complexity of the LPZ architecture increases (e.g., sites with large κ, and thick, continuous LPZs).

Amendment Injection for HPZ Treatment

Amendment injection into the HPZ to promote abiotic or biotic degradation is another approach that can be used to address the risk associated with contaminant diffusion from the LPZ. A distinction is made herein as to whether the primary intent of the injection is to address contaminants in the HPZs with a passive approach for contaminants in the LPZ, vs. injection in the HPZ with the specific goal of amendment delivery to the LPZ through forward diffusion, or EKs as discussed in subsequent sections. The process of forward diffusion will of course occur whether it is intended or not, but for organization purposes addressing LPZ contaminants through forward amendment diffusion is discussed in the next section. The intent of amendment injection could be to promote contaminant degradation throughout the HPZ, or a focus could be placed on the HPZ/LPZ interface (Marble et al. 2014; Puigserver et al. 2016) (Figure 1e). Another advantage to reducing contaminant concentrations in the HPZ through amendment injection (or PAT) is that concentration gradients can be increased to accelerate contaminant removal from the LPZs.

As an example, Révész et al. (2014), Shapiro et al. (2018), and Tiedeman et al. (2018) describe a bioremediation experiment in a chlorinated solvent contaminated fractured rock setting that included both biostimulation and bioaugmentation. The experiment was conducted in the West Area source zone at the Naval Air Warfare Center in West Trenton, New Jersey, and Shapiro et al. (2018) noted that 1000 kg was an order-of-magnitude estimate of the trichloroethylene (TCE) mass in the study area. They also estimated that the pretreatment diffusive TCE mass discharge from the rock matrix was ~7 kg/year and that several hundreds of years would therefore be needed to remove the contaminant mass from the rock matrix through diffusion. After bioremediation treatment, however, they estimated that the diffusive TCE mass discharge from the rock matrix had increased to ~45 kg/year, or higher using less conservative estimation methods. If that rate could be sustained, then the remediation timescale would be reduced to decades compared to centuries using prebioremediation estimates. The authors noted however that multiple injection events would most likely be needed to sustain elevated TCE diffusive mass discharge, and even then, diffusive flux is expected to decrease with time as mass is removed from the rock matrix.

As noted in the case above, a potential disadvantage to a focus on the HPZ is that periodic treatment may be required over the duration that contaminants enter the HPZ from the LPZ. Indeed, back diffusion may be a large contributing factor in many of the cases where contaminant rebound has been observed after amendment delivery (Mundle et al. 2007; Manoli et al. 2012; Tressler and Uchrin 2014), or if PAT or some other treatment is terminated prematurely for that matter (Gomezlahoz et al. 1994; de Barros et al. 2013). Nonetheless, periodic, batch injection of amendments is one strategy that can be used (Huling et al. 2017) to address contaminant back diffusion, as well as constant-head injection designs for slow but steady amendment delivery (Pac et al. 2014). Another strategy that has emerged is the use of controlled release materials designed to slowly release amendments and therefore sustain active treatment over prolonged periods (Christenson et al. 2012). O’Connor et al. (2018) provide a review of contaminant treatment using controlled release materials. Christenson et al. (2016) provide a 5-year review of a field application to treat dissolved chlorinated solvent using these materials, and the most recent results indicate an 89% reduction in TCE concentration.

LPZ Natural Degradation

The length of time over which plumes may persist due to back diffusion can be significantly less if contaminants in LPZs are degraded through naturally occurring abiotic or biotic processes. West and Kueper (2010) demonstrate through a modeling study of fractured rock that degradation in the matrix may be the most significant mechanism for the attenuation of the plume in the fracture. Abiotic degradation has been demonstrated experimentally in crushed sandstone (Darlington et al. 2008, 2013; Yu et al. 2018a) and sandstone cores (Yu et al. 2018b). Schaefer et al. (2013) note that naturally occurring reactive metals, such as ferrous iron exposed within the rock pore space, could substantially mitigate the impact of TCE back diffusion from rock matrices, and subsequent work by Schaefer et al. (2015) confirmed abiotic degradation in rock samples with long-term TCE exposure. Based on rates of abiotic degradation from the rock samples used, they estimated that abiotic degradation would produce an order-of-magnitude decrease in contaminant mass over a decade compared to the case without degradation. Moreover, degradation in fractured rock settings is not limited to abiotic processes. Biotic degradation has also been demonstrated in rock matrices (Darlington et al. 2008; Yu et al. 2018a, 2018b). Lima et al. (2012) reported the presence of dechlorinating microorganisms in a contaminated fractured sandstone-dolostone sequence, and the authors concluded that the potential for biotic degradation in fractured rock systems should not be ignored.

The long-term significance of abiotic or biotic degradation in aquitards or other unconsolidated LPZ media relative to back diffusion has also been demonstrated experimentally (Wanner et al. 2016; Schaefer et al. 2018) and in modeling simulations (Chambon et al. 2010; Wanner et al. 2018). Chambon et al. (2010) assessed remediation of a chlorinated solvents contaminated fractured clay aquitard with enhanced anaerobic dechlorination. Model simulations explored four scenarios: no biodegradation, biodegradation in fractures only, biodegradation in the fractures plus a reaction zone extending 5 cm from the fractures, and degradation in both the fractures and clay matrix. For the hypothetical site modeled, remedial time frames needed to meet TCE drinking water standards at an assumed downgradient monitoring well were 140 years and 25 years for the third and fourth scenarios, respectively, compared to 619 years for the first two scenarios.

Limitations in biodegradation due to bacterial exclusion from LPZ small pores, or restrictions in the spatial extent of biodegradation to narrow zones near HPZs have been noted in the literature (Lima and Sleep 2007) and suggested in the results of others (Manoli et al. 2012; Sale et al. 2013). Results from other works suggest that such restrictions may not occur at all sites (Takeuchi et al. 2011; Lu et al. 2014). Wanner et al. (2016) evaluated naturally occurring degradation of chlorinated volatile organic contaminants (CVOCs) in a clayey aquitard resulting from a controlled release experiment at the Borden research site in Ontario, Canada. Vertical concentration profiles and compound-specific isotope analysis of groundwater and soil samples were used to demonstrate contaminant degradation in the aquitard. Comparisons between field data and modeling simulations indicated that the degradation varied with depth in the aquitard, which most likely stemmed from limitations associated with the diffusion of nutrient into the aquitard from the aquifer.

Treatment through Forward Diffusion

Forward diffusion of amendments (Figure 1d) to address contamination in LPZs may be the most common, nonpassive remedial approach used historically, with or without explicit recognition of back diffusion as a limitation to site remediation. The rate at which the amendment moves into the LPZ through forward diffusion and the resulting spatial distribution in the LPZ will depend on the amendment concentration in the HPZ, the diffusion coefficient of the amendment, the LPZ tortuosity factor, retardation, reaction with contaminants, and reaction with other, nontarget materials.

Laboratory Studies

One-dimensional (1D) column studies on amendment forward diffusion have been reported by Struse et al. (2002), Huang et al. (2014), and Yu et al. (2018b). Struse et al. (2002) used permanganate $\left({\text{MnO}}_4^{-}\right)$ and a silty clay with a reported hydraulic conductivity of less than 0.86 cm/day. The experimental system consisted of 3.81 cm diameter soil cores 2.54 cm in length with constant concentration boundary conditions. In one set of experiments, replicate columns were injected with 2.9 mg of TCE DNAPL, and the columns were equilibrated for 3 days prior to the diffusion experiments with ${\text{MnO}}_4^{-}$. The initial soil concentration was 64 mg/kg assuming the mass of TCE injected into the columns was distributed uniformly. A ${\text{MnO}}_4^{-}$ concentration difference of 5000 mg/l was established across the column length for 35 days; the columns were then disassembled, and extracts of the entire column were collected. Analytical results were below reported method detection limits of 1.75 mg/kg, which indicated the treatment efficiency was 97% or greater. Similar experiments using sedimentary rocks were reported by Huang et al. (2014), but with very different results relative to ${\text{MnO}}_4^{-}$ transport. A concentration difference of 2000 mg/l was maintained across 1-cm thick rock cores for durations ranging from 49 to 56 days. Scanning electron microscopy was used to measure permanganate penetration, which was reported to vary from approximately 0.005 to 0.04 cm (50 to 400 μm). These short distances were attributed to two factors. First, the porosity of the consolidated samples was low, ranging from 0.05 to 0.074. This suggests tortuosity and the effective diffusion coefficient was likewise low (factors which will fortunately also impact contaminant forward diffusion). Second, the reaction between ${\text{MnO}}_4^{-}$ with total organic carbon in the rock limited ${\text{MnO}}_4^{-}$ transport. Based on the results obtained, the authors noted that it was unlikely that oxidant treatment would address contamination in the rock matrix beyond a few hundred microns. Yu et al. (2018b) evaluated biostimulation using lactate in sandstone cores. They noted that the lactate significantly increased TCE reduction to cis-1,2-dichloroethylene (cDCE), which resulted in a higher rate of CVOC removal, possibly due to three mechanisms: higher concentration gradients, lower cDCE sorption, and a higher cDCE diffusion coefficient.

A variety of experiments have been conducted in two-dimensional (2D), laboratory-scale physical aquifer models related to amendment forward diffusion (Hønning et al. 2007; Marble et al. 2010; Sale et al. 2013; Cavanagh et al. 2014; Clifton et al. 2014; Cavanagh et al. 2017; Chowdhury et al. 2017b). The physical models contained one or more LPZs and one or more HPZs. Contaminant was loaded into the LPZs using a variety of methods, ranging from contaminant forward diffusion (Sale et al. 2013; Chowdhury et al. 2017b) to mixing contaminant NAPL in with the LPZ during packing (Marble et al. 2010; Cavanagh et al. 2017). Remedial amendments were then flushed through the HPZs, often then followed by water flushing. Remedial amendments used in these experiments were one of two types: amendments for ISCO (${\text{MnO}}_4^{-}$, persulfate $\left[{\text{S}}_2{\text{O}}_8^{2-}\right]$) or amendments for biostimulation and bioaugmentation (e.g., lactate, KB-1 proprietary bioaugmentation culture, etc.). Reductions in effluent contaminant concentration or contaminant flux from back diffusion were common metrics to evaluate treatment performance. Figure 3a summarizes results from studies that used amendments for ISCO, while Figure 3b primarily summarizes experiments that used amendments for biostimulation and bioaugmentation. However, two additional types of experiments (enhanced hydraulic flushing and ${\text{MnO}}_4^{-}$ flushing) are also included in Figure 3b to show the entire set of tests completed by Sale et al. (2013).

Figure 3.

Summary of concentration reductions in forward diffusion experiments from laboratory physical aquifer models. Panel (a) summarizes studies that used amendments for in situ chemical oxidation and panel (b) summarizes studies that used amendments for biostimulation and bioaugmentation. Panel (b) also includes two additional types of experiments to show the entire set of tests completed by Sale et al. (2013). The legend in panel (b) pertains to both panels. Letters represent citations: (A) Sale et al. 2013; (B) Chowdhury et al. 2017b; (C) Marble et al. 2010; (D) Hønning et al. 2007; (E) Cavanagh et al. 2014; (F) Cavanagh et al. 2017; and (G) Clifton et al. 2014. The notes of (N) indicate the contaminant was introduced as a NAPL, (U) indicates the amendment was not activated, and DL indicates results were less than detection limit, and the detection limit was used as the post treatment result.

Reductions in effluent concentration and flux are shown as order-of-magnitude estimates and percent reductions in Figure 3. Reductions in concentration or flux immediately following treatment, relative to concentrations or flux immediately prior to treatment are shown by the blue bars, and range from 0.1 to 4 orders-of-magnitude for ISCO treatment, and from 0.2 to 1 for biostimulation or augmentation amendments. This order-of-magnitude range represents percent reductions ranging from 22% to 99.99%. The median reduction in concentration or flux for all results based on this comparison is 1.1 orders-of-magnitude or 92%. The orange bars represent reductions after water flushing following amendment injection, and orange bars less than blue bars represent contaminant rebound. The overall range of this data is 0.1 to 3 orders-of-magnitude, and the median reduction is 1.0 order-of-magnitude or 90%. Contaminant rebound occurred in most cases involving ISCO amendments and enhanced flushing, but in only half the cases involving biostimulation/bioaugmentation amendments. Contaminant rebound to the molar sum of CVOC compounds was not observed in the experiments reported by Sale et al. (2013), although rebound in the parent TCE compound was observed (data not shown).

Another approach used to evaluate remedial performance in 2D physical models is to compare results after amendment flushing to those based on hydraulic flushing without any amendment. On one hand, such comparisons illustrate the benefits obtained relative to hydraulic flushing only. On the other hand, such comparisons should be made bearing in mind that hydraulic flushing achieves concentration reductions by flushing contaminant mass downstream, presumably to an extraction (e.g., PAT) or treatment system (e.g., PRB), as opposed to chemical degradation through abiotic and biotic reactions. The gray bars in Figure 3 represent reductions in concentration or flux in the water flush after amendment injection, relative to the concentration or flux due to water flushing only for a similar length of time, and these values range from −0.1 to 1.5 (which correspond to percent reductions of −26 to 97%). The median reduction in concentration for all results based on this comparison was 0.5 orders-of-magnitude or 68%. Marble et al. (2010) reported a higher TCE concentration after flushing with ${\text{MnO}}_4^{-}$ compared to hydraulic flushing only (e.g., percent reduction of −26%). While they noted that the difference in concentration was small relative to experimental variability (290 vs. 230 mg/l), it was suggested that the higher concentration following ${\text{MnO}}_4^{-}$ treatment may reflect improved flushing of the LPZs due to the formation of manganese-dioxide precipitates in the neighboring HPZ near the LPZ. The extent to which enhanced flushing of the LPZ occurred however may be a function of the initial k contrast, and the mild contrast (0.8 for one LPZ layer and 1.8 in the other) may have been a factor.

Factors that influenced variability between experiments included the amount of amendment injected relative to the amount of contaminant in the models, reaction rates between amendment and contaminant, and noncontaminant consumption of the amendments. For example, results from Sale et al. (2013) indicated a percent reduction in TCE effluent concentration relative to the saturated TCE solution as 99.9% (a reduction from 1300 mg/l to approximately 1.3 mg/l), while results from Chowdhury et al. (2017b) indicated a percent reduction in TCE effluent concentration relative to the saturated TCE solution as 99.98% (a reduction from 1000 mg/l to approximately 0.2 mg/l). Both Sale et al. (2013) and Chowdhury et al. (2017b) introduced the contaminant into their respective models by flushing the HPZ with high TCE concentrations, thereby inducing contaminant forward diffusion into the LPZs. Even so, the difference in final TCE effluent concentration between Sale et al. (2013) and Chowdhury et al. (2017b) is approximately an order of magnitude. One factor which presumably contributed significantly to this is the order-of-magnitude higher ${\text{MnO}}_4^{-}$ concentration used by Chowdhury et al. (2017b) relative to Sale et al. (2013). Another factor which most likely played a significant part was the permanganate injection duration relative to the contaminant injection duration, which was ~4 times larger in the Chowdhury et al. (2017b) experiment. The ratio of the MnO₄ injection duration to the TCE injection duration was ~0.5 for Sale et al. (2013) and ~17 for Chowdhury et al. (2017b), indicating a relatively larger amount of amendment was used by Chowdhury et al. (2017b).

Modeling and Field Studies

Mundle et al. (2007) used a numerical finite difference model to investigate the sensitivity of hydrogeological and engineering parameters on tetrachloroethylene (PCE) and TCE concentration rebound following the application of ISCO in fractured clay. The conceptual model assumed a clay matrix 15 m long by 20 m wide by 5 m high, containing parallel fractures 15 m long by 20 m wide with a base case aperture of 7.5 × 10⁻⁵ m (75 μm). Contamination of the domain resulted from a 5-m long DNAPL source zone, initially placed within the fracture in the center of the domain with a saturation of 0.34. An initial 2-year period of natural attenuation was followed by a 5-year period (base case) of permanganate treatment at a concentration of 5000 mg/l, ending with a 12-year period of post-remedial monitoring. For the base case used in the study, 98% of the DNAPL had completely dissolved in the first 44 days of the natural attenuation period, and by the time it had completely dissolved, 98% of the contaminant mass was located in the clay matrix due to forward diffusion. At the end of the simulation, the relative mass destroyed by oxidation was 15.3% and the percent reduction in concentration was 94.1%. For the other cases investigated, the percent mass destroyed ranged from roughly 1% to 30% and the percent reductions in concentration ranged from roughly 30% to 97%. An exception was one case where the aperture height was reduced by half, resulting in no TCE concentration reductions because the fracture hydraulic conductivity was so low permanganate was not flushed through the complete fracture length. Moreover, it was noted that if the fracture velocity was too large the majority of the permanganate would be flushed from the fracture before it had the chance to diffuse into the matrix and react with the contaminants. Finally, results suggested that ISCO would be more effective if applied early after contaminant release to prevent forward diffusion of contaminants into the matrix. While this observation may have limited applicability at legacy sites, it is an important observation for sites with recent contaminant spills.

Arshadi and Rajaram (2015) investigated the behavior of spatially integrated reaction rates associated with bimolecular diffusion and reaction. Their analysis suggests these systems can broadly be divided into two domains, early- and late-time, based on the relative dominance of transport vs. chemical reaction. Permanganate diffusion from a fracture into a contaminated rock matrix for contaminant treatment was used as an example application in their study. Early in the process (i.e., early-time domain), the rate of transport relative to the rate of chemical reaction is large, such that the oxidant advances into the matrix and spatially overlaps with the contaminants. In the late-time domain, the rate of reaction is large relative to the rate of diffusional transport, such that a sharp, moving reaction front is created. The reaction front divides the matrix into a contaminant-free region behind the front, and an oxidant-free region ahead of the front. Example applications were given for permanganate reactions with TCE, PCE, methyl tert-butyl ether (MTBE), and cyclotrimethylenetrinitramine (RDX). For the cases noted, estimates of the transition time from early- to late-time regimes ranged from a few seconds for TCE, to more than a year for select cases with RDX. Extension of their analysis to address reactions with natural organic matter may be useful to help evaluate treatment efficiency relative to natural oxidant demand.

Both Manoli et al. (2012) and Wanner et al. (2018) conducted modeling investigations of enhanced reductive dechlorination (ERD) with assumed diffusional transport limitations. Using soil cores from a field site as the basis for modeling conditions, Manoli et al. (2012) used a 1D model to explore ERD of CVOCs in a 5-m section of fractured clay till with 3-mm wide sand stringers spaced ~1 m apart. Dechlorination of TCE was assumed to be limited to 5-cm wide bioactive zones around the sand stringers (2.5 cm on either side). The biostimulation amendment was assumed to be initially distributed in a 3-cm wide zone centered on the sand stringers (1.5 cm on either side) based on the premise of forward amendment diffusion. For the conditions modeled, the injected substrate was depleted in 5 years, and the TCE concentration reduction was ~90% (from an initial concentration of 35 mg/l to ~3 mg/l). The concentration increased however to 20 mg/l at the end of the simulation 5 years after substrate depletion, giving a final concentration reduction of ~43%. It was also predicted that without donor re-injection, TCE mass removal would stall at 18% after 5 years.

Wanner et al. (2018) conducted modeling studies of a hypothetical aquifer underlain by a clayey aquitard that included spatially distributed rates of first-order degradation (TCE to cDCE without further degradation). Degradation rates were largest at the aquifer/aquitard interface and decreased nonlinearly with depth into the aquitard, assuming nutrient limitations due to diffusional transport as suggested by Wanner et al. (2016). Without degradation, concentrations of TCE remained higher than the maximum contaminant level (MCL) for more than 100 years in a simulated downgradient well for the conditions used in the study. Assuming spatially uniform aquitard degradation with a half-life of 30 days resulted in the same well reaching MCL concentrations for TCE as soon as the primary source was removed, but cDCE persisted above its MCL for more than 100 years. Various results in between these bounding cases for TCE were obtained as degradation rates varied with depth in the aquitard, but cDCE persisted above its MCL for more than 100 years in all cases.

Despite the fact that there is most likely a large number of sites where forward amendment diffusion into LPZs is a significant feature for site remediation, detailed field studies that focus on this aspect in the peer-reviewed literature are lacking. An exception is literature associated with fracturing which is discussed in the next section. Case studies of ISCO at fractured rock sites were reported by Werner and Helmke (2003) and Goldstein et al. (2004). Performance in both cases was based on aqueous samples from monitoring wells, and detailed analyses of amendment or contaminant distributions in the rock matrix after treatment were not made. Werner and Helmke (2003) explicitly note the need to address contamination in both fractures and matrix, the latter to prevent recontamination of the former through back diffusion. They describe an unsuccessful initial attempt at full-scale remediation using Fenton’s reagent, followed by a subsequent, re-designed attempt using sodium permanganate. The site was primarily contaminated with PCE at concentrations of 8.7 and 25 mg/l for an average and high, respectively, prior to remediation. The last sampling event referenced in the paper was ~30 days after the final treatment, and the percent reduction in average groundwater concentration relative to pretreatment was ~89%. Goldstein et al. (2004) present results from a chemical oxidation pilot test using potassium permanganate in a fractured shale environment. Prior to the pilot test, a detailed characterization effort identified the contaminants, primarily consisting of PCE, in both the fractures and the rock matrix, and the importance of selecting a remediation technology that could address contaminant mass in both locations was noted. Results from the pilot test showed an average concentration reduction of >96% at monitoring locations, but reductions were short lived. Concentrations had returned to pretreatment levels within 0.5 years due to upgradient contamination and from back diffusion of contaminants from the rock matrix. Nonetheless, the pilot test was considered successful and full-scale remedial design was planned.

As noted in the last case study, contaminant rebound may result from processes other than back diffusion. Others have also noted that contaminant rebound may stem from several causes (Krembs et al. 2010; Huling et al. 2017). In general, contaminant rebound may result from any rate-limited, mass-transfer process, and may be caused by incomplete primary source treatment (e.g., dissolution from NAPL remaining after partial source treatment) or because of contaminant mass outside of the primary source zone. Our focus in this review is back diffusion, with contamination in LPZs acting as secondary sources. While such activity is not the focus of this review, it is nonetheless important to note that site-specific characterization information, as for example obtained from soil cores or other appropriate characterization tests, should first identify the cause of plume persistence before remedial strategies are implemented to address it.

The distinction between a remedial design that uses conventional amendment injection vs. a remedial design that includes amendment forward diffusion to address contaminants in the LPZs is an explicit recognition of the contaminant mass in the LPZ and the intention to promote contaminant degradation in the LPZ with a prescribed amount of amendment. As highlighted in the discussion of the published laboratory experiments, the key components of the remedial design relevant to amendment LPZ loading are the amendment concentration delivered to the HPZ/LPZ interface and the duration of amendment injection. Strategies used in conventional amendment injection, such as batch injections (Huling et al. 2017) or constant head injection schemes (Pac et al. 2014), are equally applicable for this purpose. Once amendment injection is terminated and amendment concentrations at the HPZ/LPZ interface decrease, back diffusion of any remaining amendment that has not yet reacted with target contaminants or nontarget material is expected in an analogous manner to contaminant back diffusion. Amendment back diffusion and its potential to continue downstream groundwater remediation after amendment injection in the HPZs has terminated has been identified as one process in the concept described by Adamson et al. (2011) as sustained remediation.

Alteration of LPZ Architecture

Fluid Viscosity Amendments

The injection of a more viscous fluid into the HPZ decreases preferential flow and improves sweep efficiency. For this reason, viscosity modification coupled with amendment injection promote a more uniform amendment distribution in heterogeneous environments and amendment residence time increases due to reduced fluid mobility in the HPZ, resulting in improved reactions with contaminants. It may also promote greater mass transfer into the LPZ through forward diffusion since elevated concentrations in the HPZ are maintained for longer durations. The concept of viscosity modification is mathematically reflected in the relationship between K and dynamic viscosity (μ, [ML⁻¹ T⁻¹]):

$$K=\frac{k\rho g}{\mu },$$ (6)

where ρ is fluid density [ML⁻³] and g is gravitational acceleration [LT⁻²]. Assuming variation in density is negligible, then substitution of Equation (6) into Equation (5) yields

$$\kappa =\log \left[\left(\frac{k_{HPZ}}{k_{LPZ}}\right)\left(\frac{{\mu }_{LPZ}}{{\mu }_{HPZ}}\right)\right].$$ (7)

Consequently, limitations due to contrasts in k can be offset by appropriate contrasts in μ between the fluids in the HPZ and LPZ media.

A number of laboratory studies using columns and 2D tanks have been reported in the literature to investigate the effects of viscosity modification on LPZ sweep efficiency (Martel et al. 1998; Darwish et al. 2003; Zhong et al. 2008, 2011; Silva et al. 2012, 2013; Chokejaroenrat et al. 2013, 2014; Kananizadeh et al. 2015). Xanthan was used in all cases, except for Darwish et al. (2003) who used polyacrylamides. Porous media configurations in 2D flow cells consisted of two to five heterogeneous layers, or one to two LPZ lens inside higher-k media. The high-k media was typically 20/30 silica sand or a comparable media, and κ was generally one or less. A summary of reported results is shown in Figure 4, which presents reported sweep efficiency improvement using xanthan in the concentration range of 200 to 800 mg/l relative to the flushing solution without viscosity amendment. Results are shown as reported and the figure serves as a general summary rather than a formal comparison accounting for all experimental variations. For a homogeneous media, a single injected pore volume (PV) results in a sweep efficiency of 1, while more than one PV is needed under heterogeneous conditions. Figure 4 illustrates that for the conditions tested a single PV without amendment produced a sweep efficiency as low as ~0.2, or that as many as 6 PVs were needed to achieve a sweep efficiency of 1. With xanthan, however, results were generally much closer to the ideal conditions of 100% sweep efficiency in a single PV.

Figure 4.

Summary of laboratory experiments using xanthan to modify viscosity in heterogeneous 2D flow cells. High permeability media was typically 20/30 sand or similar material, and the permeability contrast with the LPZ media is represented using the size of the data symbols according to the legend. Letters represent citations: (A) Chokejaroenrat et al. 2013; (B) Chokejaroenrat et al. 2014; (C) Martel et al. 1998; (D) Robert et al. 2006; (E) Silva et al. 2012; (F) Silva et al. 2013; (G) Zhong et al. 2008; and (H) Zhong et al. 2011.

Model studies have also been used to evaluate improved sweep efficiency through viscosity modification (Zhong et al. 2008; Silva et al. 2012, 2013; Chokejaroenrat et al. 2013; Kananizadeh et al. 2015). Results similar to those summarized in Figure 4 have been reported. For example, Silva et al. (2012) conducted simulations of a three-layered domain with permeabilities of 10⁻⁸ cm² (1 Darcy), 10⁻⁷ cm² (10 Darcy), and 10⁻⁶ cm² (100 Darcy) (κ = 2 using the minimum and maximum k), and the number of PVs for 100% sweep efficiency was 7.7 ± 0.0 in the cases without xanthan, and 3.4 ± 0.4 in the cases with a xanthan concentration of 500 mg/l. Kananizadeh et al. (2015) reported 100% sweep efficiency at PV = {1.0, 2.0, 4.0, 4.5} for an amendment solution consisting of 500 mg/l xanthan and 10,000 mg/l ${\text{MnO}}_4^{-}$ for heterogeneous media with κ = {0.3, 0.9, 1.2, 1.9} respectively.

Silva et al. (2017) and Truex et al. (2015) investigated viscosity modification to improve sweep efficiency in field tests at the Marine Corps Base Camp Lejeune, North Carolina, and Joint Base Lewis McChord, Washington, respectively. The hydrogeological characteristics of the target zone at the Marine Corps Base consisted of a layer of interbedded silty sands and silt underlain by a layer of fine to medium sand. Estimates of K for this zone ranged from 30 to 340 cm/day (κ = 1.1), and hydraulic testing within the demonstration plot indicated K ranged from 30 to 520 cm/day (κ = 1.2). The test interval at Joint Base Lewis McChord consisted of mixed till (50 cm/day ≤ K ≤ 600 cm/day) and outwash (1000 cm/day ≤ K ≤ 5000+ cm/day). Based on these estimates of K, ~0.2 ≤ κ ≤ ~2+. Both sites were contaminated with CVOCs, and the predominant contaminant found at Marine Corps Base and Joint Base Lewis McChord was PCE and TCE, respectively. Truex et al. (2015) specifically noted higher concentrations of CVOCs in LPZs, which was suspected to act as a secondary source to the down-gradient plume. Silva et al. (2017) used a xanthan-amended ${\text{MnO}}_4^{-}$ solution (500 mg/l xanthan, 5000 mg/l sodium permanganate, and 3000 mg/l sodium hexametaphosphate) and reported a sweep efficiency of 67% for the xanthan-amended test, compared to 33% for the xanthan-free test after one injected PV. Truex et al. (2015) used a viscosity modified amendment solution consisting of 800 mg/l xanthan, 1000 mg/l ethyl lactate, and 480 mg/l potassium chloride and reported a 69% sweep efficiency for the xanthan-amended test, compared to 49% for the xanthan-free test for two injected PVs.

Investigations of fluid viscosity modification to improve remedial performance have predominantly been conducted in settings with relatively mild k contrasts (0.1≤ κ ≤ 2). Silva et al. (2012) note that for k contrasts >10 (κ > 1), increasingly larger volumes of amended solutions will be needed to treat LPZs, and a recirculation-well strategy may be needed in these cases. This may reflect theoretical and practical limitations to the extent that viscosity modifications may offset k contrasts. For example, xanthan viscosity ranges from a maximum associated with a shear rate of zero (${\mu }_p^0$) to a minimum that approaches the viscosity of the unamended solution (e.g., water, μ_w). The maximum possible offset is therefore ${\mu }_w/{\mu }_p^0$ and data from Silva et al. (2012), for example, indicates that ${\mu }_w/{\mu }_p^0$ is approximately {0.14, 0.025, 0.01, 0.005} for xanthan concentrations of {250, 500, 750, 1000} mg/l, respectively, in a 400 mg/l calcium chloride solution. Consequently, the maximum possible offset is roughly two orders-of-magnitude. In theory, higher offsets are possible, for example, Chokejaroenrat et al. (2013) reported ${\mu }_p^0$ was ~5600 cP $\left({\mu }_w/{\mu }_p^0=0.0002\right)$ for a solution consisting of 1250 mg/l xanthan and 12,500 mg/l ${\text{MnO}}_4^{-}$; however, viscosities were reported to significantly decline after 0.5 days due to the reaction between xanthan and ${\text{MnO}}_4^{-}$.

Physical Alteration

In some environments, it may be possible to physically alter the stratigraphy of the subsurface environment to minimize or eliminate the risk of plume persistence due to back diffusion. This can be accomplished by fracturing the LPZ material (Figure 1f), or using soil mixing to convert a heterogeneous environment into a homogenous one.

Fracturing

Pneumatic fracturing refers to the injection of pressured air into consolidated (United States Environmental Protection Agency [U.S. EPA], 1995) and unconsolidated (Walden 1997) formations. In the case of unconsolidated media, the fractures will close with time because there is no injected rigid media to keep the fractures open. In contrast, hydraulic fracturing typically involves the injection of water, sand, and a gel. Once injected, the sand serves to maintain the fracture opening. A technique related to hydraulic fracturing is direct-push injection, but sand and gel are not used as part of the injected solution (Christiansen et al. 2010). Purposes for LPZ fracturing include enlarging k for flushing or injecting aqueous-based amendments, emplacing solid reactive amendments (U.S. EPA 1994; Horst et al. 2019), and decreasing the diffusional transport distance in the LPZ which therefore accelerates remediation (Walden 1997). Hydraulic fracturing has been reported to increase k by an order of magnitude or more (Horst et al. 2019; Sorenson 2019). Sorenson et al. (2019) provide a guidance document for the design, implementation, and monitoring of fracturing for subsurface remediation. They also provide a detailed review of case studies as reported in the literature.

The characteristics of emplaced fractures and the distribution of the fractures relative to the contaminant distribution are key features of this approach (Swift et al. 2012). For example, Chambon et al. (2010) concluded that 0.0025-cm thick fractures spaced 1 m apart would not be sufficient to ensure site remediation in clayey till through ERD in a reasonable time frame, while Scheutz et al. (2010) suggest that 1- to 2-cm thick fractures spaced 0.1 to 0.4 m apart may be sufficient to ensure remediation through ERD in a reasonable time frame. Christiansen et al. (2010) suggest that fracture spacing may need to be on the order of 0.1 to 0.25 m to ensure remediation of a clayey till is completed in a 10-year time frame.

A variety of pilot- or full-scale fracturing tests have been reported in the literature, and these works can loosely be placed into one of two categories: investigations of fracturing techniques and resulting fracture characteristics (Murdoch et al. 2006; Christiansen et al. 2008, 2010, 2012; Sorenson 2019), and tests of contaminant treatment with fracturing techniques. The latter category can be divided based on remediation strategy: fractures to promote abiotic degradation through oxidation/reduction reactions (Siegrist et al. 1999; Swift et al. 2012; Sorenson 2019), fractures to promote biodegradation (Scheutz et al. 2010; Swift et al. 2012; Horst et al. 2019; Sorenson 2019), fractures for steam flushing (Nilsson et al. 2011), or fractures to promote EK treatment (Murdoch and Chen 1997; Roulier et al. 2000).

Siegrist et al. (1999) conducted a comparative study in which hydraulic fracturing was used to emplace amendment into a TCE contaminated clay and silt formation. The tests were conducted in the vadose zone, but with high water saturation (90%) and nonetheless provide useful information for LPZ treatment in the saturated zone. Two side by side test cells were used in the study, and zero-valent iron (ZVI) was emplaced within one cell while permanganate was emplaced in the other. An important distinction between the remedial approaches was noted: remediation using ZVI is accomplished through a reaction of TCE on the metal surface, while remediation using permanganate occurs through an aqueous-phase reaction. In the former case, contaminant must diffuse to the metal surface, while in the latter case, both contaminant and amendment may diffuse to reach one another. Evidence of this was observed in the study. Batch tests using ZVI media collected from fractures through coring showed that the reactivity was equal to or better than batch tests using fresh ZVI. In contrast, there was little evidence of geochemical changes in the media 1 cm away from the ZVI filled fracture after 0.83 years. Batch tests were conducted using LPZ material collected from an area next to the fracture, but this material was not capable of degrading TCE. In comparison, a 15-cm thick permanganate reaction zone was established on either side of the permanganate filled-fracture after 0.83 years, and the symmetry of the distribution about the fracture was taken as evidence for diffusional transport. Core samples were again collected, and subsequent batch tests with the reactive material emplaced in the fracture, as well as the LPZ material adjacent to the fracture, showed that both were capable of degrading TCE.

Scheutz et al. (2010) investigated biological ERD of cDCE and vinyl chloride (VC) in a clayey till using hydraulic fracturing, biostimulation, and bioaugmentation. The primary objective of the pilot-scale study was to evaluate the performance of ERD as a means of improving contaminant mass transfer from clayey till into an established bioactive zone for dechlorination. A sand fracture within the clayey till at approximately 7.5 m below ground surface (bgs) was created by injecting a fracturing fluid composed of green colored sand, guar, enzymes (to degrade the guar once injected), and potable water. Core samples were used to characterize the fracture and results showed that it extended nonuniformly from the injection well at radial distances ranging from less than 1 m to at least 2 m, and ranged in thickness from a few mm to a few cm. After fracture placement, a 208-l batch mixture was injected, consisting of site groundwater, emulsified soybean oil, a lithium bromide tracer, and a proprietary bioaugmentation culture containing Dehaloccoides bacteria (KB-1). Prior to injection of the treatment mixture, the authors reported moderately anaerobic groundwater conditions. However, analysis of core samples showed that a dechlorinating bioactive zone was established in the fracture within ~30 days of injection. By 118 days, 98% of cDCE had been transformed to ethene or VC, and by 348 days, most of the ethene had been reduced to ethane. By 489 days, however, trace concentrations of cDCE were detected in the fracture, suggesting partial contaminant rebound once the electron donor supply was exhausted. It was estimated that complete dechlorination of cDCE and VC was stimulated in 540 days (1.5 years) within a zone of more than 5 cm and possibly as much as 34 cm into the clayey till on either side of the fracture.

Soil Mixing

Soil-mixing remediation techniques overcome limitations in remediation due to heterogeneity by using large augers to mechanically homogenize the subsurface. Amendments can also be added during the process, thus improving amendment delivery and contaminant degradation. The amendment can include clay to reduce the k of the resulting homogeneous mixture, and thereby decrease contaminant flux from the treated area and increase the residence time of the amendment in the treated area (Olson and Sale 2015). Several cases have been reported by researchers exploring the use of soil-mixing techniques (Olson et al. 2012; Fjordbøge et al. 2012a, 2012b; Kakarla et al. 2017). While these case studies did not explicitly address back diffusion from secondary sources, they nonetheless serve as examples of how the technology may be used to do so.

Olson et al. (2012) reported 1-year postremediation results for TCE and 1,1,2,2-tetrachloroethane at Site 89, Marine Corps Base Camp Lejeune, North Carolina. The site geology before mixing consisted of interbedded sand and silts, with a depth of 2 m to groundwater. In addition to the parent contaminants, primary degradation products present at the site before treatment included cDCE, trans1,2-dichlorethylene (tDCE), and VC. Total contaminant mass in the treatment zones was estimated to be 28,000 kg, distributed over an area of 3010 m² and to a depth of 8.5 m bgs. Based on measured soil and groundwater concentrations, the authors presumed DNAPL was likely present prior to remediation.

Contaminated soil was mixed with 2% ZVI and 3% sodium bentonite clay, on a dry soil weight basis, to degrade contaminants and decrease soil hydraulic conductivity. Soil mixing was completed in 80 days using a 3.0 m diameter auger, and mixed columns overlapped by approximately 18% to promote complete mixing across the treated volume. Initial soil concentrations were based on grab samples collected immediately after mixing, and a soil ZVI content greater than 2% was measured for all grab samples, indicating a uniform distribution of the ZVI amendment. Groundwater and soil analyses within 1-year following treatment indicated an average site-wide reduction of 81 and 97% respectively, in total CVOCs concentrations. The addition of bentonite reduced the pretreatment soil hydraulic conductivity by ~2.5 orders of magnitude, from 150 cm/day (1.7 × 10⁻⁵ m/s) to 0.45 cm/day (5.2 × 10⁻⁸ m/s). This reduced the potential for downgradient contaminant mass discharge, as well as increased the amendment residence time and therefore, increased the potential for contaminant degradation. Estimates of mass discharge to support this were not calculated due to the uncertainty in the hydraulic gradient in the first year following treatment.

Another case study of soil mixing to a depth of ~8 m bgs was reported by Fjordbøge et al. (2012a, 2012b). They used ZVI-Clay mixing to address PCE DNAPL contamination at a site in Skuldelev, Denmark. The geology of the site consists of quaternary deposits of alternating layers of sand and clay till, underlain by limestone. Fjordbøge et al. (2012a) reported >99% PCE source mass reduction within a year following ZVI-Clay mixing, and Fjordbøge et al. (2012b) reported a down-gradient PCE mass discharge reduction of 76% over a monitoring period of 1.6 years. Kakarla et al. (2017) reported that clean-up goals were met after 1 day of soil mixing at the Kearsarge Metallurgical Corporation (KMC) Superfund Site in Conway, New Hampshire. The soils at the site ranged from coarse sand to clay and silt, and the soil was mixed to depth intervals ranging from 2 to 5 m bgs over an area of approximately 929 m². Modified Fenton’s reagent was applied with an oxidant dose of 7 g hydrogen peroxide per kg of soil. Concentrations of 1,1,1-trichloroethane and 1,1-dichloroethylene were reduced to nondetect levels, or below clean-up goals of 0.15 mg/kg for 1,1,1-trichloroethane and 0.06 mg/kg for 1,1-dichloroethylene. Results from soil and groundwater sampling completed 0.5 and 1 year, respectively, after soil-mixing were consistent with the 1-day observations.

Soil mixing of course is impractical for those geological settings which prohibit mechanical mixing (e.g., fractured consolidated media), or maybe undesirable at sites where significant alteration of the existing geologic structure would produce negative consequences. Contamination depth may also be a limitation due to increasing treatment cost with depth. The Naval Facilities Engineering Command (NAVFAC) reported that although mixing equipment can reach 30 m or more, treatment depths are typically limited to the upper 18 m of soil due to the high cost associated with treatment at greater depths (NAVFAC 2019). Other potential limitations include the spatial scale of contamination and the presence of aboveground and subsurface structures (U.S. EPA 2019; NAVFAC 2019). According to NAVFAC (2019), buried objects over 0.5 m may limit access to portions of treatment zones, resulting in less than 100% treatment of the contaminated sites.

Treatments Free of Permeability Contrast Limitations

Thermal Treatment

Thermal treatment has evolved to cover a range of potential remedial strategies, from low-temperature (<80°C) applications designed to enhance biodegradation or thermally activate oxidants (Horst et al. 2018), to high-temperature (>1000°C) applications intended to melt contaminants and soil alike into a glass solid (i.e., in situ vitrification). Most applications however have target temperatures in the range of 80 to 110°C and employ electrical resistive heating or thermal conductive heating to introduce heat into the subsurface (Triplett Kingston et al. 2010). These applications are primarily intended to volatilize organic contaminants, which are subsequently captured using multiphase extraction wells or soil vapor extraction wells. Other techniques used to introduce heat into the subsurface include steam injection, hot water injection, radio-frequency heating, microwave-frequency heating, and smoldering. Reviews of thermal remediation have been provided by Davis (1997), Triplett Kingston et al. (2010), Kuppusamy et al. (2016), and Vidonish et al. (2016).

Most applications as described in the peer-reviewed literature have focused on treatment of primary source zones (e.g., NAPL contamination) rather than secondary source zones associated with back diffusion (Beyke and Fleming 2005; Davis et al. 2005; Heron et al. 2005, 2009, 2013, 2015; Smart 2005; Powell et al. 2007; Truex et al. 2009; Beyke et al. 2014). Assessments of thermal treatment as a DNAPL primary source zone remediation technology based on performance at multiple sites can be found in several papers (McGuire et al. 2006; Triplett Kingston et al. 2010, 2012; Baker et al. 2016). McGuire et al. (2006) report a median reduction in CVOC concentrations in the source of ~98%, while Triplett Kingston et al. (2010) reported concentration reductions ranging from one to three orders-of-magnitude for a variety of organic contaminants. Contaminant mass discharge reductions have also been used to evaluate performance (Triplett Kingston et al. 2010, 2012), and reductions ranged from less than one order-of-magnitude to more than three orders-of-magnitude. Triplett Kingston et al. (2012) suggested that the reason some sites did not perform as well as others may be due to improper source zone delineation and incomplete treatment.

Baker et al. (2016) summarized results from thermal conductive heating remediation of 10 source areas at 5 sites, all with similar glacial and glaciofluvial geology. A focus was placed on the response of groundwater concentrations in monitoring wells downgradient from the source treatment areas, in part, as evidence that back diffusion from secondary sources may not always negate downgradient benefits of primary source treatment. In some locations, drinking water standards had been sustained over durations ranging from a few years to 10 years after primary source zone remediation. Heron et al. (2016) present a case study involving full-scale thermal treatment of a PCE source zone using thermal conductive heating at a site with a sand and gravel aquifer underlain by a silt aquitard. At the time of thermal treatment in 2009, the contaminant plume extended 360 m downgradient where it was intercepted by a PAT system. The PCE plume was located within a larger petroleum plume due to contamination upgradient from the site, which helped to create anaerobic conditions in the PCE plume and therefore promote PCE biodegradation. A source-zone volume of ~7500 m³ was treated for 200 days, resulting in a PCE mass reduction of 99.97% on average. Likewise, reduction in mass discharge from the source zone was estimated to be 99.9%. Strong biodegradation conditions in the plume offset any potential effects of contaminant back diffusion from the underlying aquitard, and 5 years after source-zone thermal treatment, plume concentrations had been sufficiently reduced to discontinue the use of the downgradient PAT system used for hydraulic control.

Thermal technologies are well suited to address contaminants in heterogeneous environments because heat flux is proportional to the thermal conductivity of the porous media, and thermal conductivity of geologic media varies much less than hydraulic conductivity (Table 1). In some applications, heterogeneity considered disadvantageous to flushing technologies may offer a benefit in thermal treatment. Krol et al. (2014) conducted numerical modeling and Martin et al. (2016) conducted 2D physical laboratory experiments with a clay lens packed inside a more permeable sand media. Compared to a homogeneous sand pack, the heterogeneous cases with the clay lens achieved higher temperatures in a shorter amount of time using electrical resistive heating because the clay lens had a higher electrical conductivity than sand. In another application, contaminants in heterogeneous media were addressed by using different thermal technologies: electrical resistance heating for the LPZs was combined with steam-enhanced extraction for the HPZs (Heron et al. 2005).

Heterogeneity may nonetheless be a factor in hampering thermal treatment performance. Perhaps the biggest concern associated with thermal treatment in a heterogeneous environment is the convection of heat from the target zone by groundwater flowing through the HPZs. If the flow is too large, enough heat may be removed to negatively impact the efficiency with which target temperatures are reached, or worse, prevent some parts of the treatment zone from reaching target temperatures (Baston and Kueper 2009; Truex et al. 2009; Heron et al. 2013, 2015; Munholland et al. 2016). Once contaminants enter the gas phase, their removal depends on gas phase transport to the extraction wells and gas-phase transport may be hampered by heterogeneities. Martin and Kueper (2011) conducted laboratory experiments and observed that an overlying capillary barrier remained water saturated during ERH, which trapped gas and contaminants beneath it. Comparable results were observed in laboratory experiments by Munholland et al. (2016), who reported that approximately a third to a half of the original DNAPL that had been volatilized into the gas phase due to heating at its original location, condensed back into a DNAPL outside the treatment zone due to lateral gas transport induced by a capillary barrier.

Heterogeneity in temperatures may also have negative consequences. Gas production during thermal treatment and its potential to redistribute contaminants when the gas migrates into overlying colder regions, causing contaminants to condense, has been observed in 2D physical aquifer models and numerical models (Krol et al. 2011a; Hegele and Mumford 2014; Martin et al. 2017). Suggestions to offset the negative influence of colder regions above target treatment zones on gas transport include ensuring that regions between contaminants and extraction wells are heated to boiling temperatures (Hegele and Mumford 2014), or injecting a solution with dissolved gas to promote gas formation (Hegele and Mumford 2015; Molnar et al. 2019). Likewise, Krol et al. (2011b) and Krol et al. (2014) showed that contaminants may be redistributed due to buoyancy induced flow resulting from thermal treatment, which could result in contaminants being trapped below shallower LPZs.

While most published works related to thermal treatment have focused on the primary source zone treatment, experiments conducted by Chen et al. (2010) and Liu et al. (2014) considered contaminant mass in the LPZ associated with fractured consolidated media and fractured clay, respectively, which may be more representative of a secondary source zone. Both experiments used 1,2-dichloroethane as the contaminant at an initial concentration of ~200 to ~250 mg/l. Chen et al. (2010) conducted their experiment using a Berea sandstone core and reported 99% contaminant removal when ~38% of the PV had been removed as condensate. Liu et al. (2014) conducted experiments using Kaolin clay in test cells with rigid and flexible walls to explore impacts of test boundary conditions. Contaminant concentrations were reduced by two orders-of-magnitude once ~50% of the pore water was removed in both test cells.

Related modeling work was reported by Chen et al. (2012, 2015), who modeled the laboratory experiments dealing with fractured rock and field-scale thermal remediation in fractured rock. For the field-scale simulations, a 2D radially symmetric domain 20-m deep by 8-m radius was used with the groundwater table 4.5 m below ground surface. Chen et al. (2012) uniformly applied heat at a rate of 200 W/m³ to remediate 1,2-dichloroethane contamination at an initial concentration of 253 mg/l. After 28 days, temperatures reached 103°C and the heat was turned off, but extraction continued for an additional 7 days, at which time 28% of the pore water and >99% of the contaminant mass were removed. In the simulation reported by Chen et al. (2015), the contaminant was TCE at an initial aqueous concentration of 7.28 mg/l. Heat was uniformly applied to the saturated zone at a rate of 200 W/m³ for 30 days to induce boiling, and sensitivity of contaminant removal by a multiphase extraction well to matrix k, fracture spacing and aperture, and extraction well pressure was evaluated. For the range of conditions tested, ~99% of the contaminant mass was removed at the end of the 30-day test period. The rate at which contaminant was removed was most sensitive to matrix k: >99% of the TCE was removed in 20 days when k = 10⁻¹⁰ cm², while only 71% was removed at the end of 30 days when k = 10⁻¹³ cm². The reason noted for this was the impact of k on pressure gradients resulting from boiling. Higher pressures result in elevated boiling points, which inhibit treatment efficiency for a fixed energy input and duration.

EK Treatment

The transport of contaminants or amendments due to electrical gradients resulting from the application of direct current electrical energy is the basis for EK remediation. There are three general transport mechanisms associated with groundwater EK remediation: electromigration, electrophoresis, and electroosmosis. Electromigration refers to the movement of ions through a solution due to an electric field, and it is the primary mechanism for the EK remediation of ionic contaminants such as heavy metals. It can also be used to promote the distribution of ionic amendments such as ${\text{S}}_2{\text{O}}_8^{2-}$, ${\text{MnO}}_4^{-}$, or nitrate. Electrophoresis refers to the movement of chemicals sorbed to colloidal-sized charged particles in an electric field, and an example of its use for groundwater remediation is polymer-modified nanoscale ZVI (Jones et al. 2011). Electroosmosis is the movement of the bulk fluid due to the electric field. This movement stems from the attraction of counter ions in the double layer to the charged surface of the soil particles. Excess counter ions move in response to the electric field, which in turn creates bulk fluid movement in the pores due to viscous forces. Contaminants in the fluid, regardless of whether they have an ionic charge, are transported by this bulk motion.

Transport by EK is independent of hydraulic conductivity and is therefore effective in creating flow in LPZs. With respect to electromigration, the proportionality constant relating J_E to ∂E/∂x is u (Table 1), and u depends on both solute (ionic mobility at infinite dilution) and porous media properties (porosity and tortuosity). Likewise, for electroosmosis, the proportionality constant relating J_E to ∂E/∂x is k_e (Table 1), and k_e depends on both solution (viscosity, permittivity) and porous media (porosity) properties, and the combined effect of the two (zeta potential). The range of values for u and k_e is much less than the range for K (Table 1). However, the electrochemical interactions created by ionic transport in porous media under an electrical gradient may result in a more complex system to design and control compared to more conventional remediation systems. The electrical gradient may be impacted by ionic transport; transport by electromigration and electroosmosis may be in opposite directions, although electromigration is generally an order-of-magnitude higher than electroosmosis; and the direction of electroosmotic flux may change under certain conditions (Acar et al. 1995).

Multiple EK remediation reviews are available (Virkutyte et al. 2002; Reddy and Cameselle 2009; Yeung and Gu 2011; Lima et al. 2017), including reviews of EK remediation specific to organic contaminants (Saichek and Reddy 2005a; Gomes et al. 2012; Gill et al. 2014; Ottosen et al. 2019). Lima et al. (2017) note that while laboratory-scale EK experiments have been “highly successful,” the technology is not commonly used in practice. As an illustration of that point, it is noted that EK treatment was only selected twice out of 628 times as an in situ groundwater treatment technology at Superfund sites from 1982 to 2014 (U.S. EPA 2017). Given the volume of material on EK remediation and the number of available reviews, we defer to the latter for a summary of the literature that deals with homogeneous, albeit low-conductivity media. Our primary focus here is applications of EK remediation for organic contaminants in heterogeneous media since those works are most relevant to the topic of back diffusion, and there appears to be far fewer publications of this nature.

Laboratory experiments dealing with EK remediation in 2D heterogeneous physical aquifer models can be divided into two groups: experiments in uncontaminated media focusing of amendment distributions (Reynolds et al. 2008; Gill et al. 2015, 2016), and experiments in contaminated media focusing on treatment performance (Saichek and Reddy 2005b; Chowdhury et al. 2017b). In the first group, Reynolds et al. (2008) conducted EKs experiments evaluating ${\text{MnO}}_4^{-}$ distribution in two heterogeneous distributions. The first consisted of a single LPZ, and in a control experiment without an applied electric field, visual analysis demonstrated that the ${\text{MnO}}_4^{-}$ solution had mostly bypassed the LPZ as a result of flow due to a hydraulic gradient. After an elapsed time of 2.5 h, ~10% of the LPZ had been invaded by the ${\text{MnO}}_4^{-}$ solution due to forward diffusion, in contrast to ~53 to ~65% LPZ coverage by ${\text{MnO}}_4^{-}$ in 4 h with electromigration. The second heterogeneous pack consisted of 11 LPZ lenses consisting of kaolin clay within a glass bead pack. In a control experiment without EK treatment, the solution was hydraulically flushed through the model for 1.5 h. This was repeated for the test with EK treatment, and then an electric field was applied for 1.7 h. It was reported that the concentration of ${\text{MnO}}_4^{-}$ in the kaolin lenses was ~10% that of the concentration in the bulk solution, which was substantially more than that in the control experiment.

Groundwater remediation using EK biostimulation in layered media was studied by Gill et al. (2015) and Gill et al. (2016) using uncontaminated physical aquifer models. Gill et al. (2015) evaluated nitrate transport when the direction of the electrical field was normal to layered heterogeneity. They demonstrated that the effective ionic mobility decreased as hydraulic conductivity decreased, although decreases in the former were much less than decreases in the latter. As hydraulic conductivity decreased ~5 orders of magnitude, the effective ionic mobility only decreased by a factor of ~2 for the media investigated. However, differences in effective ionic mobility lead to electrical gradients which enhanced nitrate transport through the LPZ relative to an experimental system homogeneously packed with the LPZ material. This work was continued by Gill et al. (2016) with a focus on layered heterogeneity-oriented parallel to the applied voltage. Experimental observations along with supporting calculations demonstrated that transverse nitrate flux from the HPZ material into the LPZ material resulted in elevated nitrate concentrations in the LPZ, relative to a homogeneous case consisting of the LPZ material only. In the heterogeneous case, relative nitrate concentrations in the LPZ were ~20 to ~250% greater than those in experiments with homogeneously packed LPZ material. Gill et al. (2016) also conducted a conceptual modeling exercise estimating the amount of time needed to deliver nitrate to degrade BTEX located in a LPZ layer under assumed field conditions; 48 years would be needed to deliver enough nitrate based on diffusion alone, but only 0.17 years would be needed using electromigration.

In laboratory experiments with a contaminant, Saichek and Reddy (2005b) conducted two heterogeneous tests, one with a layer of Kaolin underlain by sand, and another with a 5-cm circular lens of Kaolin in an otherwise homogeneous sand media. The contaminant used in each test was phenanthrene, and it was mixed into the media prior to packing at an initial concentration of 500 mg/kg. The remedial mechanism was electroosmotic surfactant flushing using 5% nonionic octylphenol polyoxyethylene. The primary metric used to evaluate performance was the phenanthrene mass removed from the model. In the test with Kaolin underlain by sand, 63% of the contaminant mass was removed in 64 days using a combination of hydraulic flushing and EK treatment, while 120% was removed in the second test in 54 days (experimental uncertainty was reported as 30%). Post-test soil sampling was conducted in both tests and phenanthrene was detected in the Kaolin layer only.

Laboratory experiments related to the forward diffusion of ${\text{MnO}}_4^{-}$ to treat TCE contamination in LPZs as conducted by Chowdhury et al. (2017b) were discussed in Section 3.2. This work also included experimental replicates using EKs to enhance the penetration of the ${\text{MnO}}_4^{-}$ into the silt LPZ. The physical model consisted of two rectangular low k silt lenses within a high k sand. The replicate with EKs performed the best, with a 99.998% reduction in average TCE concentration at the end of the experiment (from ~1000 to 0.021 mg/l). The final concentration was an order-of-magnitude lower than the final concentration based on the forward diffusion of ${\text{MnO}}_4^{-}$ (~0.2 mg/l), and almost two orders of magnitude lower than the final concentration based on water flushing without amendment (~1.0 mg/l). Chowdhury et al. (2017a) described another concept involving a three-part treatment train of ISCO, EK, and thermal treatment. In this case, EK was used to transport ${\text{S}}_2{\text{O}}_8^{2-}$ into a physical aquifer model packed with silt and contaminated with PCE. After which, the same electrodes were used to thermally activate the ${\text{S}}_2{\text{O}}_8^{2-}$, and PCE concentrations were reduced to below detection limits (0.02 mg/l) within ~30 days.

Wu et al. (2012a) present a numerical model for simulating EK remediation, which was subsequently used by Wu et al. (2012b) and Wu et al. (2013) to investigate EK-ISCO using ${\text{MnO}}_4^{-}$ as the amendment. Wu et al. (2012b) describe a simulation in a domain 25 m long by 10 m high consisting of a correlated random k field with a mean k of 1.14 × 10⁻⁷ cm², and variance of ln transformed k of 7. The reported range in hydraulic conductivity within the domain was 1 × 10⁶ cm/day to 1 × 10⁻⁷ cm/day (κ = 13). The domain was contaminated by flushing with a PCE solution at the solubility limit until the LPZ concentrations were ~40% of the solubility limit. Wu et al. (2012b) reported ~16% PCE destruction in 100 days of treatment by ${\text{MnO}}_4^{-}$ injection without EK. Remedial performance improved to ~92% PCE destruction using the same ${\text{MnO}}_4^{-}$ injection scheme but combined with EK to enhance amendment transport. Using a pulsed ${\text{MnO}}_4^{-}$ injection scheme with the same electrode and injection well configuration, Wu et al. (2013) reported ~95% PCE destruction in ~70 days, with the remaining 5% flushed from the domain (i.e., complete PCE removal).

Documentation of field-scale remediation tests using EK treatment at organic contaminated sites are limited in the peer-reviewed literature, and no field-based studies where EK treatment was applied to specifically address back diffusion were found. A notable case study is remediation using the Lasagna technology at a site in Paducah, Kentucky (Ho et al. 1999a, 1999b; Athmer 2004). The Lasagna technology is a process that uses electroosmosis to transport organic contaminants to reactive zones installed in the contaminated area of the subsurface (Ho et al. 1995). The site was contaminated with TCE, and DNAPL was believed to be present based on sampling results with concentrations much greater than the solubility limit. Athmer and Ho (2009) noted the site soil consisted of clay loam with estimates of hydraulic conductivity ranging from 0.1 to 0.001 cm/day. Remedial goals were met in 2 years.

A recent pilot-scale field demonstration of enhanced bioremediation using EKs to deliver amendments to the intended treatment area was described by Cox et al. (2018). The site was contaminated with PCE, and the treatment area was a zone consisting predominantly of clay 12 m long by 12 m wide by 1.2 m deep. Amendments injected included potassium lactate as an electron donor, potassium carbonate for pH control, and KB-1 for microbial augmentation. At the end of the 1.2-year study, the monitoring well with the highest initial PCE concentration (7.64 mg/l) had a PCE reduction of 95%, with increases in dissolved ethene and biomarkers indicative of biodegradation. Concentrations of PCE in the clay were reduced on average 88% without significant increases in biodegradation intermediates. Energy requirements were reported to be low (1585 kW-h) and were comparable to energy requirements for two 100-W light bulbs for 1.2 years of operation.

Summary

Given the potential longevity of secondary contaminant sources and the risk they pose to groundwater resources, preventing contaminants from entering the LPZ in the first place should be viewed as highly beneficial. Aggressive action to address primary source zones at early stages of the contaminant spill when possible will be helpful for minimizing risk associated with secondary source zones later. Once contaminants enter LPZs, however, a variety of strategies and techniques are available to address contaminant diffusion from secondary sources (Figure 2). The first major category consists of those approaches that do not include active measures to address contamination in the LPZs, and as such rely on intrinsic contaminant degradation in the LPZ and diffusional transport of contaminants out of the LPZ. Included in this category are containment or treatment strategies for the HPZ, which have the advantage of providing a relatively rapid means to address the risk of contaminant transport over large areas. The disadvantage, however, of any LPZ passive treatment approach is the length of time it will take for back diffusion to remove contaminants from the LPZ. Diffusional transport can be increased by lowering contaminant concentrations in the HPZ through remedial action because that will increase the concentration gradient from the LPZ to the HPZ. Nonetheless, such strategies may still need to be in operation for a period much longer than typically considered acceptable in remedial designs (e.g., many decades if not centuries compared to several decades). In some cases, such options raise issues of long-term stewardship that may need to be more fully considered in the remedial selection and review process (NRC 2013). Prolonged treatment, however, does not necessarily mean that options that rely on back diffusion to remove contaminants from LPZs are the least favored remedial option. Containment options are sometimes viewed as the preferred option because of uncertainty in benefits or the expense associated with both primary and secondary source treatment (Lemming et al. 2012; Chapman et al. 2013).

The longevity of contaminants in LPZs however presents an opportunity for degradation, even at modest rates, to play a significant part in the management of plume persistence due to back diffusion. Abiotic and biotic degradation offers an inherent advantage as a cost-efficient, sustainable remedial treatment. Field-based studies have established that abiotic and biotic degradation can occur in LPZs, and that at some sites, it can mitigate secondary sources after primary source treatment. Modeling studies have also shown that degradation can significantly reduce the time to reach remedial endpoints. Moreover, improved rates of biodegradation through low-temperature thermal treatment may be an economical way to employ thermal treatment across distributed secondary sources. More research to better understand limiting factors in LPZ biomass migration and bioaugmentation is needed.

A variety of laboratory experiments in 2D physical models and numerical modeling studies offer insight on the forward diffusion of amendments into LPZs. Major factors that influence remedial performance include the amount of amendment injected relative to the amount of contaminant in the LPZs, reaction rates between amendment and contaminant, and noncontaminant consumption of the amendments. Reductions in concentration or flux in laboratory experiments immediately after treatment relative to concentrations or flux immediately prior to treatment ranged widely, from 0.1 to 4 orders-of-magnitude (median of 1.1 orders-of-magnitude). Reductions in concentration or flux immediately after treatment relative to the concentration or flux due to water flushing only for a similar length of time, however, ranged less, from −0.1 to 1.5 orders-of-magnitude (median of 0.5 orders-of-magnitude). Laboratory studies started with high-concentration primary sources (i.e., NAPLs or aqueous concentrations higher than 10% solubility limits), and studies addressing performance for primary sources with more mild concentration levels have apparently not been conducted. It would be helpful to evaluate secondary remedial performance under a wider range of forward loading conditions.

Strategies that alter the physical characteristics of the LPZ architecture include viscosity modification, fracturing, and soil mixing. To date, investigations of fluid viscosity modification to improve remedial performance have predominantly been conducted in mild conditions relative to those that promote plume persistence due to back diffusion. In some back-diffusion environments, like fractured consolidated media or fractured clay, viscosity modification to promote LPZ flushing will not be feasible because the conductivity of the LPZ is too small to accommodate any flow of practical significance for remediation. However, viscosity modification may be used as a strategy to lower the velocity in the HPZ, and thereby increase the duration over which injected amendment is in contact with the LPZ (Sale et al. 2013). This will in turn increase amendment mass transfer to the LPZ. Key factors in the success of fracturing are the spacing between fractures and their placement relative to the contaminant distribution. If the facture spacing is too large, then treatment durations may not be sufficiently reduced because diffusional transport remains a limitation. Except for excavation, soil mixing provides the largest modification to the LPZ architecture. It may also provide one of the better chances of achieving concentration-based remedial metrics throughout the secondary source. In cases where soil mixing includes an amendment to lower the hydraulic conductivity, care should be taken to ensure all upgradient sources of contaminants have been addressed, or that other amendments are also included at doses sufficient to degrade contaminants that may enter the mixed area. Otherwise the amended area may become contaminated through forward diffusion, and subsequently a secondary contaminant source.

Thermal and EK treatments offer distinct advantages in LPZ material and are much less likely to be hampered by diffusional transport limitations. Thermal treatment has been more widely documented in the peer-reviewed literature with respect to successful field-scale applications in primary source zones compared to EK treatment. A drawback to both treatments for secondary sources, however, is cost. In the case of thermal treatment, areal extents of ~4000 m² (1 acre) or less are common (Triplett Kingston et al. 2010). These treatments may not be viewed as acceptable for lower concentration secondary sources distributed throughout the plume. Advances in both technologies may help lower costs in future applications, and in particular, low-thermal applications combined with ISCO and biotreatment should be further investigated as cost-effective strategies to address secondary sources.

As noted, future work on several areas would be beneficial. A final recommendation for more field-based case studies is also made. While case studies have been reported where back-diffusion was found to be an issue after primary remedial treatment was implemented, field-based studies that report long-term results after treatment has been implemented to specifically address back diffusion were not found in the peer-review literature. Such studies are highly valuable in the evaluation of treatment strategies.

Article impact statement:

Back diffusion of contaminants from secondary sources may hamper site remediation if it is not properly addressed in the remedial design. A review of all reported technologies and strategies that have been or could be applied to address plume persistence due to back diffusion as published in the peer-reviewed literature was conducted and is intended to provide a state-of-the-science resource to assist in evaluating treatment options.