Modeling the impact of ethanol on the persistence of benzene in gasoline-contaminated groundwater

Transcription

1 WATER RESOURCES RESEARCH, VOL. 38, NO. 1, 1003, /2001WR000589, 2002 Modeling the impact of ethanol on the persistence of benzene in gasoline-contaminated groundwater J. W. Molson, J. F. Barker, and E. O. Frind Department of Earth Sciences, University of Waterloo, Waterloo, Ontario, Canada M. Schirmer Interdisciplinary Department of Industrial and Mining Landscapes, Umweltforschungszentrum, Centre for Environmental Research Leipzig-Halle, Leipzig, Germany Received 4 April 2001; revised 23 July 2001; accepted 27 July 2001; published 22 January [1] The effect of ethanol on the persistence of benzene in gasoline-contaminated aquifers is simulated using a multicomponent reactive transport model. The conceptual model includes a residual gasoline source which is dissolving at the water table into an aquifer containing a limited amount of dissolved oxygen. The coupled processes include nonaqueous phase liquid (NAPL) source dissolution, transport of the dissolved components, and competitive aerobic biodegradation. Comparisons are made between dissolved benzene plumes from a gasoline spill and those from an otherwise equivalent spill containing 10% ethanol (gasohol). Simulations have shown that under some conditions a 10% ethanol component in gasoline can extend the travel distance of a benzene plume by up to 150% relative to that from an equivalent ethanol-free gasoline spill. The increase occurs because ethanol preferentially consumes oxygen, which reduces the biodegradation rate of benzene. The impact is limited, however, because sufficient oxygen disperses behind the ethanol plume into the slightly retarded benzene plume. A sensitivity analysis for two common spill scenarios showed that background oxygen concentrations and benzene retardation have the most significant influence on ethanol-induced benzene persistence. The results are highly relevant in light of the increasing use of ethanol-enhanced fuels throughout the world and the forthcoming ban of methyl tertiary-butyl-ether (MTBE) in California and its probable replacement by ethanol by the end of INDEX TERMS: 1831 Hydrology: Groundwater quality, 1832 Hydrology: Groundwater transport; KEYWORDS: Groundwater contamination, numerical modeling, ethanol, BTEX, biodegradation 1. Introduction [2] The use of oxygenated gasoline has become a standard method of decreasing vehicular carbon monoxide and ozone precursor emissions into the atmosphere. The most commonly used gasoline oxygenate is methyl-tertiary-butyl-ether (MTBE). Although MTBE has shown low toxicity, its low biodegradability in groundwater and strong taste and odor at micrograms per liter levels have resulted in its ban in California by the end of Ethanol is the second most common oxygenate in gasoline and seems to be emerging as the preferred choice to replace MTBE. [3] Alcohols such as ethanol and methanol are easily degradable in groundwater environments and, owing to their high solubility, are expected to readily dissolve from a spill or source zone [Stockholm et al., 1998; Reichhardt, 1999; de Oliveira, 1997]. One potential problem in the spill of an ethanol-gasoline mixture, however, is the preferential use of this alcohol as a substrate by the microbial community. This preferential use depletes the available electron acceptors and may increase the persistence of the remaining benzene, toluene, ethylbenzene, and xylene (BTEX) components [Hunt et al., 1997; Corseuil et al., 1998]. At sufficiently high concentrations, ethanol may also act as a cosolvent, increasing the concentration and mobility of the remaining BTEX components [Powers et al., 2001]. The BTEX components of gasoline, and of Copyright 2002 by the American Geophysical Union /02/2001WR benzene in particular, are of primary concern because of their relatively high solubility and toxicity at low concentrations. [4] In this study, we provide further insight into the influence of ethanol on BTEX persistence in gasoline-contaminated groundwater by applying a numerical model to simulate BTEX migration from two conceptual models of gasoline spills. The first conceptual model addresses the impact of a local spill of 10% ethanol gasoline into a pristine aquifer. This type of spill is expected, for example, from leaking underground storage tanks [Rice et al., 1999]. In the second conceptual model we simulate a pure ethanol spill within an existing residual gasoline source and dissolved plume. This model represents a spill from an ethanol storage tank into a previous gasoline spill at a hydrocarbon fuel processing plant [Rice et al., 1999]. In each conceptual model, comparisons are made between the ethanol-impacted benzene plume and the equivalent ethanolfree benzene plume, and various system parameters are adjusted to evaluate sensitivity and uncertainty. 2. Review of Existing Modeling Approaches [5] Applications of models to predict the behavior of benzene in ethanol-impacted groundwater are scarce and very limited in scope. Most references to ethanol in groundwater are with respect to its use as a cosolvent for enhanced remediation of nonaqueous phase liquid (NAPL) impacted groundwater [e.g., Lunn and Kueper, 1997]. Interest in the effect of ethanol on BTEX migration in groundwater has only recently increased because of the potential use of ethanol as 4-1

2 4-2 MOLSON ET AL.: MODELING THE IMPACT OF ETHANOL a replacement for MTBE. Furthermore, there is a lack of well-defined field spill sites on which models can be tested. However, model applications to gasoline contaminants [Poulsen et al., 1992; Munoz and Irarrazaval, 1998; Lu et al., 1999], or to MTBE-impacted contamination sites [Schirmer and Barker, 1998], have been well documented, and much insight can be gained from these past efforts. [6] Ethanol behavior in groundwater systems has been studied with several existing models, including those used by Heermann and Powers [1998], Stockholm et al. [1998], Ulrich [1999], and McNab et al. [1999]. Rice et al. [1999] provide a detailed summary of the processes involved and review the current state of the art in terms of modeling ethanol impacts in groundwater. [7] Although adding some insight, all previous approaches have been limited to some degree. Stockholm et al. [1998], for example, used a two-dimensional (2-D) analytical model and first-order degradation rates to compare benzene plume lengths under ethanol and ethanol-free conditions. The approach did not consider dynamic effects of cotransport and only activated benzene degradation once the ethanol concentrations decreased below a given threshold. They predicted benzene plume length increases on the order of 17 34%. [8] Ulrich [1999] applied MODFLOW/MT3D to simulate ethanol impacts on benzene. Although a 3-D approach was adopted, several simplifying assumptions limited the insights gained. First, the source was an existing dissolved plume of benzene; dissolution from a residual source was not considered. Second, since only a single component was transported for each simulation, dynamic interactions between benzene and ethanol were not considered. Instead, it was assumed that no degradation of BTEX occurred where ethanol was present from a separate simulation. Where degradation was active, first-order rates were assumed. Application to a Borden-like flow system induced increases of 10% in the migration distance of the ethanol-impacted benzene plume. Including acetic acid as a more persistent by-product of ethanol degradation increased the length by a factor of 1.8. [9] McNab et al. [1999] used a Monte Carlo approach to evaluate ethanol impacts using an analytical transport solution from Baetsle [1969], modified to include a finite source release area. By correlating benzene degradation rates with a variety of parameters and by using superposition to quantify biological oxygen demand (BOD) from ethanol, they obtained probability distributions of benzene concentrations downgradient of the source area. The influence of retardation factors and competitive substrate utilization was neglected, and first-order degradation rates were assumed. Their results suggested that ethanol would increase benzene plume lengths by less than a factor of 2. [10] Heermann and Powers [1996] simulated the transport of m- xylene and ethanol from a gasohol pool into a 2-D aquifer system. The influence of ethanol was restricted to cosolvency alone and sorption and biodegradation were not considered. They found maximum ethanol-induced increases in benzene plume travel distances on the order of 10%. There was essentially no effect when lower interphase mass transfer coefficients were assumed. A relatively high transverse vertical dispersivity of 0.1 m may have caused excess plume dispersion and contributed to the low predicted impact. 3. Theoretical Development 3.1. Reactive Mass Transport [11] In this study, we use the model BIONAPL [Frind et al., 1999; Molson, 2001] to simulate the multicomponent reactive system. The coupled processes include groundwater flow, multicomponent dissolution from a residual gasoline or gasohol source, advective-dispersive transport, oxygen-limited competitive biodegradation, and microbial growth, all within a 3-D porous medium. [12] The governing equation for mass transport of component m can be i D m m m v i l m Cm w þ lm DIS Cm s Cw m i R w ; where C m w (kg m 3 ) is the aqueous (water) phase concentration of organic contaminant m, D m ij (m 2 d 1 ) is the hydrodynamic dispersion tensor [Molson et al., 1992], v i (m d 1 ) is the m groundwater flow velocity, l BIO (d 1 ) is the biodegradation rate, m l DIS (d 1 ) is the gasoline dissolution rate coefficient, R w is the retardation, t is time (days) and x i (m) represents the spatial dimensions (x i = x, y, z). [13] In(1),C m s is the effective solubility of the NAPL component defined, assuming validity of Raoult s law [Mackay et al., 1991], as ð1þ C m s ¼ X m C m o ; ð2þ where X m is the mole fraction and C m o (kg m 3 ) is the pure phase aqueous solubility of component m. Assuming linear equilibrium partitioning to the aquifer solids, the retardation term R w can be expressed by R w =1+r b K d q 1, where r b (kg m 3 ) is the bulk density, K d (m 3 kg 1 ) is the distribution coefficient, and q is the porosity. [14] The biodegradation rate l m BIO for organic component m in (1) can be expressed as l m BIO ¼ XNA n¼1 k m;n M n 1 K m;n C þ Cm w K m;n A þ A n ; ð3þ where M (kg microbes /m 3 ) is the microbial concentration, subscript w refers to water, k m,n (kg substrate kg microbes 1 day 1 ) is the maximum utilization rate, A n is the concentration of electron acceptor n, and K c m,n and K A m,n are the half-utilization rate concentrations for the organic and electron acceptor, respectively. Equation (3) represents the net organic biodegradation rate due to parallel degradation using N A electron acceptors. [15] The electron acceptor concentration (A n ) in (3) is assumed to be dissolved in the aqueous phase, and therefore transport is also governed by advection and dispersion D i;j n v j A n l n An ; ð4þ n where the electron acceptor rate term l BIO represents the net reaction rate due to biodegradation of all N m organic components according to l n BIO ¼ XNm k m;n M n c m;n m¼1 K m;n C Cw m þ Cm w K m;n A 1 þ A n ; ð5þ where c m,n is the stoichiometric mass ratio of electron acceptor to organic substrate.

3 MOLSON ET AL.: MODELING THE IMPACT OF ETHANOL 4-3 [16] Equations (1) and (4) are discretized using a Galerkin finite element approach with brick elements. The groundwater velocity field (v i ) is coupled within the model using the same discretization scheme. Nonlinear terms are handled using Picard iteration with convergence traps for oxygen and each organic component. Microbe concentrations are updated at each iteration, and the matrix equations are solved with a conjugate gradient solver. Further details are described by Frind et al. [1999, 1989] Gasoline Source Dissolution [17] The source in this study is assumed to be dissolving at equilibrium, which assumes ideal component behavior and a large contact area between the residual NAPL and the groundwater. Cline et al. [1991] conclude that Raoult s law is valid for gasoline mixtures and there is evidence that residual NAPLs can be dissolved in the subsurface under conditions that have reached equilibrium [Frind et al., 1999]. Source heterogeneities and nonideal component behavior can, however, lead to apparent or real nonequilibrium dissolution at field contamination sites [Soerens et al., 1998]. Effluent concentrations (C m s ) from the dissolving source (equation (2)) are computed using a mixing-cell model [Molson, 2001] and are applied in (1) using a mass-flux boundary condition expressed as q 0 C m s q ¼ vc m w D m ; where q 0 is the Darcy flux and n is the direction normal to the boundary. Mass balance is maintained between the dissolving source and the underlying aquifer Cosolvency [18] When the aqueous volume fraction of ethanol becomes greater than 0.1, the effective solubilities of the remaining gasoline components can increase significantly as they preferentially dissolve into the water-ethanol solution [Heermann and Powers, 1998]. This cosolvency effect can be expressed as a dissolution enhancement factor E [Ji and Brusseau, 1998] defined for each organic component as E ¼ S m Sw ¼ 10 sfc ; ð6þ ð7þ are not considered. In particular, the buoyancy effect of a highconcentration ethanol plume is neglected. 3. The residual gasoline source is immobile, which is considered valid for low pressure gradients and gasoline saturations <0.1. Interfacial tension changes due to ethanol are assumed not significant. 4. BTEX sorption is assumed linear and reversible and is not affected by ethanol. 5. Dissolution of the residual gasoline occurs at equilibrium (or cosolvent-enhanced equilibrium) according to Raoult s law. 6. A single electron acceptor (oxygen) and a single (immobile) microbial population are present. Although additional electron acceptors may be present, oxygen is often the most significant for BTEX components. Anaerobic ethanol degradation is assumed minimal; instead, we assume a fermentation pathway in which degradation by-products will be more persistent oxygen depleters. 7. Substrate inhibition is neglected; the preferential degradation of ethanol is included by assuming k ethanol >k benzene and by transporting all components simultaneously where they compete for the available oxygen. 8. Only the dissolved (aqueous) phase organics can biodegrade, not the sorbed phase organics. 9. The reaction stoichiometries assume complete organic degradation to CO 2 and H 2 O. 4. Conceptual Model 4.1. Flow System [21] The conceptual flow model for this study is provided in Figure 1. The hydrogeological system is loosely based on the shallow, unconfined Borden aquifer, perhaps the best characterized sand aquifer in the world. Numerous natural gradient reactive tracer tests [Sudicky, 1986; Hubbard et al., 1994] and modeling studies [Frind and Hokkanen, 1987; Molson et al., 1992] provide an excellent database of flow system parameters. The flow system properties assumed for this study are provided in Table 1. [22] The aquifer system is 400 m long by 10 m thick by 12 m wide. We take advantage of a vertical symmetry plane and only simulate a 6-m-wide section passing through the source center. A flux of 2.5 m yr 1 is assigned along the left inflow boundary, and 0.20 m yr 1 is assigned as recharge along the top water table boundary. The right exit boundary is assigned a head of 10.0 m, and where S m represents the solubility of the gasoline component in the water-cosolvent mixture, S w is the solubility of the gasoline component in water, s is the cosolvency factor, and f c is the volume fraction of ethanol in the aqueous phase. [19] The cosolvency power can be expressed as s = log(s c S w 1 ), where S c is the solubility of the component in pure cosolvent. Linear/log linear functions have also been employed to predict cosolvency based on observed laboratory data [Heermann and Powers, 1998] Assumptions [20] Several assumptions were made in applying the BION- APL model to the gasoline/gasohol system. They are consistent with the conceptual model, and most can be modified in the model for site-specific cases (with the exception of a mobile source and changes to interfacial tension). The assumptions include the following: 1. The aquifer flow system is saturated, isotropic, homogeneous, and at steady state. 2. Contaminant concentrations are dilute, and density effects Figure 1. Conceptual model of the steady state flow system for the gasoline/gasohol simulations. Inset shows the location of the source with effluent concentrations equal to the effective solubility of each component (C m s ).

4 4-4 MOLSON ET AL.: MODELING THE IMPACT OF ETHANOL Table 1. Flow System Properties for Gasoline/Gasohol Transport Simulations Parameter Value Hydraulic conductivity 10 4 ms l Porosity 0.35 Average hydraulic gradient Ground water velocity m d l (variable along the flowpath) Aquifer recharge 0.20 m yr l the lower boundary is considered impermeable. The flow system is assumed to be at steady state for all simulations. The model domain is resolved using a grid of (= 25,440) elements in the longitudinal, transverse horizontal, and vertical directions, respectively, satisfying the grid Peclet and Courant constraints Transport System [23] This study considers two conceptual models representing different spill scenarios. In the first conceptual model we consider a four-component gasoline mixture composed of benzene, toluene, ethylbenzene, and the xylenes (TEX), ethanol, and a relatively more inert remaining fraction which is treated as a single equivalent compound (this would include, for example, alkanes and cycloalkanes). Asingle electron acceptor (oxygen) and a dynamic microbial population are also included. Anaerobic degradation is not considered because benzene biodegradation is not always observed under anaerobic conditions [e.g., Barker and Wilson, 1997]. All simulation scenarios include a simple gasoline source, as well as a gasoline source mixed with a 10% volume fraction of ethanol to represent gasohol. [24] In the second conceptual model the basic source geometry and chemistry are the same; however, we consider a spill of pure ethanol into an existing BTEX source and plume generated from the first conceptual model. This spill scenario is one identified by Rice et al. [1999] as being of particular concern given the large number of hydrocarbon-handling sites with existing groundwater contamination. [25] The residual source in both models occupies a volume 10 m long, 1 m wide, and 0.50 m thick. The source is located at the water table, starting 10 m from the left boundary, and is assumed to be dissolving at equilibrium into groundwater moving through the residual gasoline (Figure 1). Properties of the gasoline components and residual source are provided in Tables 2 and 3, Table 3. Assumed Characteristics of Residual Gasoline Source Component Source Mass, kg Gasoline Source Mole fraction a Gasohol Source Benzene TEX Other components Ethanol 3.5 na a PS-6 gasoline standard from Brookman et al. [1985] and Poulsen et al. [1991]; na, not applicable. respectively. The gasoline source is based on a PS-6 gasoline standard as defined by Brookman et al. [1985], which is often used as a source fuel for laboratory experiments [see, e.g., Poulsen et al., 1991; Donaldson et al., 1994]. The initial condition for conceptual model 1 is a clean aquifer with a background oxygen concentration of 3.8 mg L 1 and an initial microbial concentration of 3.0 mg L 1. Oxygen is allowed to enter the domain at background levels along the left inflow and top water table boundaries using a mass-flux boundary condition. [26] Dispersivities of 1.0, 0.02, and 0.01 m in the longitudinal, transverse horizontal, and vertical directions, respectively, are consistent with estimates based on statistical analyses of Borden aquifer properties [Sudicky, 1986]. These dispersivities will tend to minimize plume spreading and to minimize mixing of substrates and the electron acceptor. Field evidence tends to support this concept of low dispersion as long and thin BTEX and MTBE plumes have been well documented in the literature [Davis et al., 1999; Einarson et al., 1999]. A time step of 1 day was used to ensure convergence of the reactive transport system Biodegradation [27] A variety of anaerobic electron acceptors (e.g., SO 4,Fe 3+, CO 2 ) can be used in ethanol biodegradation, producing by-products such as organic acids, methane, etc. The by-products still represent a significant, but reduced, biological oxygen demand. These electron acceptors may be quite limited relative to the high loading of ethanol from a gasohol spill. In such cases, fermentation of ethanol to CO 2 and a variety of organic acids, H 2, and CH 4 [Powers et al., 2001] likely dominates anaerobic degradation. These fermentation by-products have the same BOD as the original ethanol. In this study, the fermentation pathway is assumed, and ethanol represents the complete oxygen demand of the alcohol itself plus the reaction products. This approach is meant to Table 2. Physical Properties of Gasoline Components Assumed in the Model Compound MW, a kg mol 1 Solubility, b kg m 3 Density, c kg m 3 R d m 2 s 1 Pure Phase Difficulty Coefficient, a Benzene TEX Other components Ethanol e a Wiedemeier et al. [1999, Appendix B], assuming a tortuosity of 0.7. b Johnson et al. [1990], as referenced by Wiedemeier et al. [1999, p. 88]. c Weast [1980]. d Retardation factors from Wiedemeier et al. [1999] and Hubbard et al. [1994]. e The maximum possible solubility, which allows all of the ethanol to dissolve in the initial source pore volume of groundwater. Note that Heermann and Powers [1998] concluded that >99% of the ethanol in a gasohol source will partition into the aqueous phase. Similarly, Poulsen et al. [1992] found >99% of methanol dissolved into the initial pore volume of water.

5 MOLSON ET AL.: MODELING THE IMPACT OF ETHANOL 4-5 Table 4. Assumed Biodegradation Parameters for Gasoline Components, Base Case Scenario a Organic Component Maximum Rate Organic Half Constant k, mg org mg 1 mic d 1 K C,mgL 1 O 2 Half Constant K A,mgL 1 Microbial Yield Y, mg mic mg org 1 Benzene TEX Other components Ethanol a From Bekins et al. [1998], Schirmer et al. [2000], and Frind et al. [1989]. represent the worst-case scenario having the greatest oxygen demand. The stoichiometry of the overall reaction for the case where ethanol biodegradation occurs via fermentation to CH 4, with subsequent aerobic degradation of the CH 4,is Anaerobic fermentation of ethanol 2CH 3 CH 2 OH! 3CH 4 þ CO 2 Aerobic biodegradation of methane 3CH 4 þ 6O 2! 3CO 2 þ 6H 2 O Overall stoichiometry CH 3 CH 2 OH þ 3O 2! 2CO 2 þ 3H 2 O [28] The stoichiometry of 3 mol of oxygen required for each mole of either ethanol or methane is therefore maintained. Here we assume a constant BOD for ethanol and its by-products. [29] Kinetic biodegradation parameters for the base case are provided in Table 4. Published parameters based on local site conditions vary considerably, and defining a representative parameter set can be difficult [see, e.g., Bekins et al., 1998; Alvarez et al., 1991]; a sensitivity analysis is used to address this parameter uncertainty. The oxygen:substrate mass ratios in Table 5 assume complete organic degradation to CO 2 and H 2 O. While the model tracks biomass change, it does not take the required carbon from the substrate. Since this biomass carbon is a minor portion of the carbon mineralized, this approach should not introduce significant bias for our purposes. [30] Cosolvency is not considered in the first conceptual model because at and below the ethanol concentration in the source area of 2000 mg L 1, laboratory data suggest that cosolvency effects are not significant [Poulsen et al., 1992; Heermann and Powers, 1998]. Similarly, no cosolvency modification of BTEX retardation was considered. 5. Simulation Results 5.1. Conceptual Model 1: A 10% Ethanol Gasoline (Gasohol) Spill Into a Pristine Aquifer Scenario 1: Base case. [31] A base case simulation is first completed with both an ethanol-free gasoline source and a gasohol source containing 10% ethanol by volume. The parameters are later varied in a sensitivity analysis, in which all simulations are compared to the base case scenario. [32] As the four gasoline components dissolve from the source and enter the aquifer, their concentrations will vary over time according to their effective solubilities (Figure 2). In the ethanolfree gasoline source the initial concentrations for benzene, TEX, and the remaining components are 31, 46, and 16 mg L 1, respectively. Under the assumed conditions, benzene concentrations began decreasing after 2500 days and reached a concentration of 10 mg L 1 by 3600 days (not shown). The TEX and the remaining fraction, which are relatively less soluble and have higher mole fractions, persist longer in the source. [33] The concentration of ethanol in the gasohol source reaches a maximum of 2000 mg L 1, which is sufficiently low to justify neglecting cosolubility effects. Because of ethanol s high solubility and limited mass in the gasohol source, all the ethanol is dissolved instantaneously and is completely flushed from the source within one pore volume, or 320 days (Figure 2). The addition of ethanol to the gasoline causes a small reduction in the mole fraction for BTEX and the remaining components at early time. This reduces their effective solubilities and source concentrations but does not significantly impact plume concentrations because ethanol is rapidly flushed from the source. The effective solubilities of the remaining components then return to the levels of the ethanol-free case. [34] The one pore volume flush time for ethanol is a consequence of its high solubility, its limited mass, and the volume of water within the available saturated pore space in the source area. This approach neglects kinetic limitations or substrate interactions, which may inhibit equilibrium dissolution. Donaldson et al. [1994] provide evidence supporting this behavior for methanol and conclude that methanol can dissolve quite rapidly, within a few pore volumes. Ethanol is expected to behave similarly. [35] The organic concentration profiles for the base case (ethanol-free gasoline) simulation at 10 and 20 years are shown in Figure 3 (for clarity, we only show the domain up to 200 m). The organic plumes are long and thin, with high vertical concentration gradients resulting from a low transverse dispersivity and low numerical dispersion (note the vertical exaggeration). Table 5. Reaction Stoichiometries for Simulated Gasoline Components Gasoline Component Reaction Stoichiometry O 2 :Substrate Mass Ratio X Benzene C 6 H /2 O 2 ) 6CO 2 +3H 2 O 3.08 TEX C 7 H 8 +9O 2 ) 7CO 2 +4H 2 O 3.13 Others C 8 H /2 O 2 ) 8CO 2 +9H 2 O 3.5 Ethanol C H 3 CH 2 OH+3O 2 ) 2CO 2 +3H 2 O 2.09 Figure 2. Simulated aqueous source concentrations from the dissolving residual gasoline and gasohol sources for conceptual model 1 and base case scenario. Concentrations are computed from equation (2) and are applied in a mass-flux boundary condition across the water table.

6 4-6 MOLSON ET AL.: MODELING THE IMPACT OF ETHANOL Figure 3. Simulated plumes of benzene, toluene, ethylbenzene, and the xylenes (TEX) and remaining components from the base case (ethanol-free) gasoline source at (top) 10 years and (bottom) 20 years. Note vertical exaggeration of 7 times. [36] The benzene plume, as defined by the 10 mg L 1 contour at 10 years, is 100 m long and 3 m thick and has almost completely dissolved from the source. The plume length is similar to many BTEX plumes observed at the field scale [see, e.g., Davis et al., 1999; Thierrin et al., 1995; Wiedemeier et al., 1999, Table A2]. After 20 years, biodegradation has removed most of the benzene mass, leaving only a small remnant of the plume from 100 to 140 m. In this ethanol-free benzene plume, degradation is quite rapid. We shall later compare this plume with those codegrading with ethanol. [37] Because of their relatively higher mass fraction and lower solubility, TEX and the remaining components continue to dissolve beyond 20 years, and their plumes are therefore still connected to the source. The TEX plume is more extensive than the benzene plume because of its longer source persistence, higher retardation (which reduces the degradation rate), and higher source concentrations (see Figure 2). While the source concentration for the remaining components is about half that of benzene, the maximum degradation rate k is 10 times less. Therefore the remaining component plume also becomes more extensive. [38] The corresponding oxygen depletion plumes are provided in Figure 4 at 10 and 20 years. They can be correlated with the organic plumes, although they are somewhat longer and wider because of a higher diffusion coefficient and unretarded oxygen transport. Oxygen concentrations vary from the initial background level of 3.8 mg L 1 at the far left and top right inflow boundaries to 0 mg L 1 where biodegradation has consumed all the available oxygen. [39] As benzene dissolves from the source and enters the aquifer, the aqueous plume mass increases until benzene becomes significantly depleted from the source at 2800 days (Figure 5). The mass lost to biodegradation continues to increase but increases at a slower rate once the source is depleted and benzene concentrations drop significantly below the half-utilization constant (K C ). A relatively minor fraction is sorbed to the solids. The sum of these mass components over the 25-year simulation time remains within 1.5% of the initial source mass of 0.50 kg. [40] The influence of ethanol in the base case (scenario 1) simulation is evident from Figure 6, which shows the concentration profiles for benzene originating from the gasohol source at 10, 15, and 20 years. The addition of ethanol has extended the benzene plume migration distance by 135% at 20 years, as defined by the 10 mgl 1 contour, relative to the travel distance of the ethanol-free benzene plume. The difference increases to 150% at 21.5 years. Because benzene is now competing with ethanol for the available oxygen and because ethanol reacts at a faster rate (k ethanol > k benzene ), the rate of benzene mass loss is reduced, and the plume migrates farther relative to the ethanol-free (gasoline source) benzene plume. The presence of ethanol has effectively stretched and split the benzene plume into two lobes. The trailing lobe is relatively unaffected by the presence of ethanol since ethanol had long since dissolved from the source before this part of the benzene plume entered the aquifer. The advanced lobe, however, persists within the advancing ethanol plume since ethanol is preferentially degraded. The nonretarded ethanol plume is continually advancing relative to the benzene plume and will eventually completely degrade. Figure 4. Simulated oxygen depletion plumes at 10 and 20 years for the base case scenario with no ethanol. Figure 5. Simulated benzene mass distribution showing source mass depletion, mass lost to biodegradation and sorption, and aqueous plume mass for the base case scenario with no ethanol.

7 MOLSON ET AL.: MODELING THE IMPACT OF ETHANOL 4-7 Figure 6. Simulated benzene plumes from the gasohol source at 10, 15, and 20 years. Also shown are the ethanol plume and dissolved oxygen depletion plume at 20 years. [41] In Figure 6 an oxygen depletion zone surrounds the ethanol plume (including its anaerobic degradation products) at 20 years, and, somewhat overlapping it to the rear, a second oxygen depletion zone surrounds the plumes of benzene, TEX, and the remaining components. There appears to be only a limited, long-term residual oxygen shadow caused by the ethanol since oxygen eventually disperses back into the groundwater following the ethanol plume. This is only partially developed in Figure 6 since the oxygen-consuming hydrocarbon plumes are close behind. The trailing benzene plume, lagging behind the ethanol plume because of its higher retardation, is eventually exposed to background oxygen concentrations and can degrade at a rate similar to that of the ethanol-free benzene plume. Furthermore, most of the benzene still remains in the source after the ethanol has been completely flushed. Hydrodynamic dispersion provides some mixing near the source at later times; however, a significant fraction of the benzene plume mass never comes in contact with the ethanol plume. The plumes of TEX and remaining components remain relatively unaffected by the presence of ethanol (not shown). Because of their even slower dissolution rate and higher retardation relative to benzene (R = 2.0 and 2.8, respectively) these components lag even further behind the ethanol plume and are exposed to the same oxygen levels as in the ethanol-free case. [42] The mass of the aqueous benzene plume also increases in the presence of ethanol because less benzene is biodegraded (Figure 7). Although the small benzene plume in the leading lobe (Figure 6) does not contain much mass, it remains significant because the concentrations still exceed typical drinking water standards (>10 mg L 1 ) Dimensionality: [43] Field-scale reactive mass transport is an inherently three-dimensional process. In field systems involving aerobic biodegradation a 3-D model representation is essential because biodegradation occurs at a significant rate only in the fringe areas of the plume where the organics can mix with oxygen. A 2-D approximation would be invalid because the surface of the plume would be incorrectly represented. [44] A 3-D perspective of the 10-ppb benzene plume from the gasohol source (at 15 years) illustrates the extensive surface area available for dispersive mixing and biodegradation (Figure 8). In the transverse horizontal direction the persistence of the benzene plume due to ethanol is particularly striking. A 1-D or 2-D approach would not capture this behavior correctly. An equivalent 2-D simulation, for example, produces a significantly larger benzene plume with peak concentrations over 20 times higher than in the 3-D simulation (Figure 9). While the 3-D plume has almost completely degraded by 20 years, the 2-D plume takes over 50 years to degrade and travels over 1000 m. Frind et al. [1989] and Engesgaard et al. [1994] present further examples showing dimensionality effects in reactive transport systems. [45] The computational effort of the 3-D approach is quite manageable. A 25-year simulation, for example, with 6 components (4 gasohol components + O 2 + microbes) on a 30,000-node Figure 7. Comparison showing the effect of ethanol on the aqueous and biodegraded mass of benzene. Figure 8. Three-dimensional (3-D) perspective of the benzene plume from the gasohol source at 15 years. Outer isosurface corresponds to a benzene concentration of 10 ppb.

8 4-8 MOLSON ET AL.: MODELING THE IMPACT OF ETHANOL Figure 9. Effect of dimensionality on the simulated 20-year benzene plume from the gasohol source. (top) A 3-D simulation (within the vertical symmetry plane) and (bottom) an otherwise equivalent 2-D simulation. mesh, required about 65 Mb core memory and 7 hours on a Pentium III/400 MHz machine Sensitivity analysis: [46] The base case scenario represents only a single set of system parameters, many of which show considerable variation and uncertainty in published laboratory and field data [see, e.g., Bekins et al., 1998]. Molson and Frind [1990], for example, show that parameter uncertainty can lead to nonunique predictive simulations of biodegrading organics. A sensitivity analysis was therefore performed on the base case scenario to determine the influence of several of these parameters on system behavior. To isolate the effects of ethanol alone, each new simulation was completed with and without ethanol, and results were expressed as a percent change in benzene travel distance relative to the ethanol-free benzene plume. A more complete analysis was performed in a 2-D system by Molson et al. [2000]. Those scenarios showing the greatest effects were repeated here in 3-D and presented in Table Scenario 2: Higher benzene retardation. [47] In this scenario the retardation factor for benzene is increased from the base case value of R = 1.1 to R = 1.5. The retardation factors for TEX and the remaining components are not changed since sorption of these components would have little impact on the behavior of benzene. Ethanol is again assumed nonretarded. The corresponding evolution of the benzene plume from both the gasoline source and gasohol source is provided in Figure 10. [48] In a simple reactive system, increasing the retardation factor slows the plume advance and decreases the plume extent. In a bioreactive system, however, a higher retardation factor also reduces the effective biodegradation rate since the sorbed phase is assumed nonbiodegradable. These processes will have opposite Figure 10. Simulated evolution of the benzene plume from (left) a gasoline source and (right) a gasohol source, assuming a relatively higher benzene retardation of 1.5. effects on the size of the benzene plume. In the ethanol-free case at 20 years, for example, the higher retardation factor has almost doubled the plume length (compare the 20-year gasoline source benzene plumes in Figure 3, R ben = 1.1 and in Figure 10, R ben = 1.5). [49] The effect of adding ethanol to the source in this highretardation scenario is again an increase in the benzene plume travel distance relative to the ethanol-free benzene plume (Figure 10). The increase, however, is not as pronounced as in the base case (R = 1.1) because the ethanol and benzene plumes are now traveling at significantly different rates and the plume overlap time is reduced. [50] The effect of retardation and ethanol on the travel distance of the 10-ppb benzene concentration is shown in Figure 11. In the base case gasoline source scenario (R = 1.1) the benzene plume steadily advances over the first 15 years, reaches a steady state limit, and is completely degraded at 21.5 years. In the highretardation (R = 1.5) gasoline source scenario the apparent advance of the plume is retarded, but only slightly, because the slower travel velocity is offset by the lower effective biodegradation rate since the sorbed phase is not degrading. The benzene plume in this case persists until 27 years. [51] The benzene plumes from the gasohol source (Figure 11) advance more rapidly relative to those from the gasoline plume Table 6. Overview of Simulation Scenarios and Influence of Ethanol on the Extent of the Benzene Plume, Conceptual Model 1 Scenario Parameters Change, a % 1, base case See Tables 2, 3, and (at 20 years) 2 R benzene increased from 1.1 to (at 17 years) 3 O 2 decreased from 3.8 to 2.0 mg/l +100 (at 25 years) b 4 Multiple spill, 4 kg/yr for 10 years +120 (at 40 years) b a Maximum percent change in the distance traveled from the source center to the downgradient edge of the benzene plume at 10 mg L 1. b Maximum percent change not reached; change may grow in time. Figure 11. Simulated advance of the 10-ppb benzene contour from the gasoline and gasohol sources, comparing the effect of benzene retardation for the base case (R = 1.1) and high-retardation (R = 1.5) scenarios.

9 MOLSON ET AL.: MODELING THE IMPACT OF ETHANOL 4-9 Figure 12. Simulated benzene plumes assuming a relatively lower background oxygen concentration of 2.0 mg L 1. because the leading lobe is cotransported within the oxygenconsuming ethanol plume. The difference in travel distance for the base case scenario (R = 1.1), for example, reaches a maximum of 150% at 21.5 years, and the plume is completely degraded by 24 years. With a higher retardation the plume advances at a slower rate until 17.5 years when the advanced lobe is completely biodegraded. Because more benzene is sorbed with higher retardation, the effective biodegradation rate is less, and the trailing lobe persists until about 27 years, similar to the ethanol-free benzene plume. [52] Where the mobility of benzene and ethanol is significantly different (i.e., with higher benzene retardation), the potential for extension of the benzene plume front in a gasohol spill is therefore reduced relative to that from a lower-retardation scenario. Aquifers rich in organic matter are therefore more likely to be protected with respect to the influence of ethanol on benzene plume travel distances. [53] These findings contradict the assumptions made by McNab et al. [1999], who suggest that effects of sorption could be neglected for assessing ethanol impacts. This assumption was necessary for their solution approach to be valid. Furthermore, Stockholm et al. [1998] predicted increasing benzene plume lengths with higher retardation, assuming continuous source functions for both ethanol and benzene. These previous studies, however, were based upon relatively simplified conceptual models which did not consider limited source persistence and fully coupled competitive biodegradation as described in the reaction terms of (3) and (5) Scenario 3: Low background O 2. [54] In this scenario the background oxygen concentration is reduced from 3.8 to 2.0 mg L 1, which is well within the observed range for unconfined aquifers (oxygen saturation at 5 C is 12.8 mg L 1, [Appelo and Postma, 1993]). Oxygen is still allowed to enter from the upper water table boundary through recharge. [55] Reducing the initial background oxygen concentration from 3.8 to 2.0 mg L 1 reduces the degree of benzene degradation and produces a larger ethanol-free benzene plume relative to the base case (Figure 12). In the gasohol plume the aerobic degradation rate of ethanol (and its products) is also reduced, which increases the duration of overlap between the ethanol and benzene plumes. The result is an increase in the benzene plume length of 64% at 20 years, relative to the gasoline source benzene plume. The difference increases further at later times (not shown), reaching 100% at 25 years. Proximity of the model boundary prevented estimates beyond 25 years, although the ethanol plume will eventually separate completely from the benzene plume and the difference will vanish. [56] Aquifers relatively low in dissolved oxygen are therefore potentially more susceptible to benzene persistence owing to the presence of ethanol in gasoline. Since higher oxygen concentrations promote rapid biodegradation and therefore a shorter plume overlap time, aquifers rich in oxygen should be relatively safe from increased benzene persistence [Molson et al., 2000] Scenario 4: Multiple spill. [57] Scenario 4 considered a multiple spill of gasoline and gasohol distributed once per year over a 10-year period. The total mass spilled remained the same as in the base case. The benzene plumes (not shown) showed similar behavior to the base case, although the relative extension due to ethanol was somewhat less. In this scenario the ethanol plume is repeatedly flushed through the benzene plume, but the lower mass flux of ethanol into the aquifer consumes less oxygen, leaving more available to benzene. The impact is therefore lessened since benzene can degrade at a higher rate. The impact is summarized in Table Conceptual Model 2: A Pure Ethanol Spill Into an Existing Gasoline Plume [58] In this conceptual model, pure ethanol is spilled into the remaining source mass and the existing 5-year ethanol-free gasoline plume from the base case scenario of conceptual model 1. Targeting an initial aqueous phase source concen- Figure 13. Simulated source concentrations for conceptual model 2 for a pure ethanol spill at 5 years into an existing gasoline spill. Concentrations are obtained from equation (2) and applied in a mass-flux boundary condition across the water table.

10 4-10 MOLSON ET AL.: MODELING THE IMPACT OF ETHANOL Figure 14. Benzene plumes from conceptual model 2 for a pure ethanol spill into a 5-year gasoline plume, including enhanced solubility of benzene in ethanol (consolvency effect). tration of 500,000 mg L 1, kg of ethanol is spilled into a 2.5 m 3 source volume, producing an initial aqueous volume fraction for ethanol of At these high-ethanol concentrations, cosolvency effects become important and must therefore be considered by the model. [59] Cosolvency is here represented as an increase in solubility using the enhancement factor approach given in (7). In the base case for conceptual model 2, we assign a cosolvency factor s = 2.0, which produces a maximum enhancement factor at early time of approximately E = 18. The objective was to produce a maximum enhancement factor for benzene that is consistent with observed laboratory data. Heermann and Powers [1998], for example, suggest maximum enhancement factors for the BTEX components in an ethanol-gasoline-water mix with an aqueous volume fraction fc = 0.6, on the order of The aqueous volume fraction of ethanol ( fc) is updated during the simulation as ethanol disperses and migrates away from the source area. [60] Here we assume identical cosolvency factors for all gasoline components, although they are likely somewhat higher for those less soluble. Because the less soluble components would remain persistent regardless of a brief cosolvency effect, any differences in their cosolvency factors are expected to have minimal effect on the simulated outcome. The effect of cosolvency on the retardation of the gasoline components is not considered. [61] The simulated source concentrations of benzene, TEX, the remaining gasoline components, and ethanol for conceptual model 2 are shown in Figure 13. From 0 to 1860 days (5 years) the source concentrations follow those from the base case (ethanolfree) scenario of conceptual model 1. After 5 years, ethanol is introduced into the source, which causes a sharp increase in the effective solubilities and source concentrations of the gasoline components. As ethanol disperses from the source area and concentrations decline, the cosolvency effect is reduced, and the effective solubilities for TEX and the remaining components return to pre-ethanol spill levels. Benzene source concentrations continue to drop following the addition of ethanol because the source becomes depleted of benzene and the mole fraction declines. [62] The simulated benzene plumes for conceptual model 2 are shown in Figure 14 at 10, 15, and 20 years. The ethanol spill begins at 5 years when the benzene plume is 64 m long and about 37% of the original benzene mass remains in the source. Ethanol is assumed to immediately partition into the aqueous (water) phase and begins to disperse within the oxygen depletion shadow of the BTEX components. Because of its low retardation (R = 1) and higher diffusion coefficient, ethanol rapidly disperses and begins consuming oxygen ahead of the BTEX plumes. The benzene plume is more extended in the vertical direction in this scenario because of the more rapid source dissolution and because of significantly lower biodegradation within the larger ethanol plume. [63] In this cosolvency scenario the ethanol spill has increased the benzene plume travel distance by about 112% at 20 years (Figure 14), relative to the ethanol-free benzene plume from conceptual model 1 (Figure 3). The increase is a result of two counteracting processes affecting benzene plume concentrations. First, the 5-year delay in the ethanol release has allowed the benzene plume to degrade at relatively higher rates at early time, thereby reducing benzene concentrations at the leading edge. The 5-year delay, however, also increases the degree of overlap between the ethanol and benzene plumes, which tends to keep benzene concentrations higher because ethanol is preferentially consumed. The cosolvency effect further increases the extent of overlap by inducing a high-concentration pulse of dissolved benzene into the aquifer at the same time as ethanol is released. Because of its short duration, however, the cosolvency effect appears relatively less important. [64] As the ethanol plume disperses in this scenario, the benzene travel distance increases to at least 150% at 40 years (not shown), relative to the base case gasoline spill of conceptual model 1. After 40 years the plumes extend beyond the right outflow boundary, and the full influence of the ethanol cannot be determined. A further increase on the order of 10 20% is possible. The maximum influence of ethanol is delayed because of the 5-year spill delay but, more importantly, because the larger ethanol plume overlaps the benzene plume for a longer time. These results suggest that a late accidental release of pure ethanol, combined with the enhanced solubility of benzene, can lead to even longer benzene travel distances compared to gasohol spills in which the ethanol is released concurrently from a gasohol source. 6. Conclusions [65] The simulations suggest that using gasohol in place of gasoline has the potential to increase benzene plume travel distances by at least 50% relative to ethanol-free benzene plumes (as defined by the 10 mg L 1 contour). The increase is most pronounced with low background oxygen, with low retardation, and with a multiple gasohol spill. Maximum increases on the order of 150% were simulated for both conceptual model 1 (gasohol spill) and conceptual model 2 (pure ethanol spill). The influence is likely even higher in some cases since the plumes reached the model boundary before the full effect could be realized. The predicted impacts are somewhat higher than previous estimates which predicted at most a doubling of plume lengths. [66] The influence of ethanol on benzene persistence depends primarily on the degree of overlap between the benzene and ethanol plumes. Under most scenarios the rapid dissolution of ethanol combined with slow dissolution and higher retardation of benzene created limited overlap between the two plumes. Oxygen has time to replenish the aquifer behind the ethanol plume and becomes available in sufficient quantity to the trailing organics. Lower benzene retardation and low background oxygen concentrations increase plume overlap, which allows ethanol (and its degradation products) to preferentially deplete oxygen over a longer time. The rate of benzene biodegradation is therefore reduced, and the plume travels farther. Field situations that would