Reactive Transport Modeling of Column Experiments for the

Apr 23, 2004 - The variable x is the distance along the column length. ..... M.Sc. Thesis, University of Waterloo, Waterloo, ON, Canada, 1995. There i...
0 downloads 0 Views 124KB Size
Download PDF
Environ. Sci. Technol. 2004, 38, 3131-3138

Reactive Transport Modeling of Column Experiments for the Remediation of Acid Mine Drainage R I C H A R D T . A M O S , * ,† K. ULRICH MAYER,† DAVID W. BLOWES,‡ AND C A R O L J . P T A C E K ‡,§ Department of Earth and Ocean Sciences, University of British Columbia, Vancouver, British Columbia, V6T 1Z4 Canada, Department of Earth Sciences, University of Waterloo, Waterloo, Ontario, N2L 3G1 Canada, and National Water Research Institute, Environment Canada, Burlington, Ontario, L7R 4A6 Canada

Reactive transport modeling was used to evaluate the performance of two similar column experiments. The experiments were designed to simulate the treatment of acid mine drainage through microbially mediated sulfate reduction and subsequent sulfide mineral precipitation by means of an organic carbon permeable reactive barrier. Principal reactions considered in the simulations include microbially mediated reduction of sulfate by organic matter, mineral dissolution/precipitation reactions, and aqueous complexation/hydrolysis reactions. Simulations of column 1, which contained composted leaf mulch, wood chips, sawdust, and sewage sludge as an organic carbon source, accurately predicted sulfate concentrations in the column effluent throughout the duration of the experiment using a single fixed rate constant for sulfate reduction of 6.9 × 10-9 mol L-1 s-1. Using the same reduction rate for column 2, which contained only composted leaf mulch and sawdust as an organic carbon source, sulfate concentrations at the column outlet were overpredicted at late times, suggesting that sulfate reduction rates increased over the duration of the column experiment and that microbial growth kinetics may have played an important role. These modeling results suggest that the reactivity of the organic carbon treatment material with respect to sulfate reduction does not significantly decrease over the duration of the 14-month experiments. The ability of the columns to remove ferrous iron appears to be strongly influenced by the precipitation of siderite, which is enhanced by the dissolution of calcite. The simulations indicate that while calcite was available in the column, up to 0.02 mol L-1 of ferrous iron was removed from solution as siderite and mackinawite. Later in the experiments after ∼300 d, when calcite was depleted from the columns, mackinawite became the predominant iron sink. The ability of the column to remove ferrous iron as mackinawite was estimated to be ∼0.005 mol L-1 for column 1. As the precipitation of mackinawite is sulfide limited at later times, the amount of iron removed will ultimately depend * Corresponding author phone: (604)827-5607; fax: (604)822-6088; e-mail: [email protected]. † University of British Columbia. ‡ University of Waterloo. § Environment Canada. 10.1021/es0349608 CCC: $27.50 Published on Web 04/23/2004

 2004 American Chemical Society

on the reactivity of the organic mixture and the amount of sulfate reduced.

Introduction Groundwater contamination from mine tailings impoundments is often characterized by high concentrations of SO4, Fe(II), and other metals and a near-neutral pH (1, 2). Left untreated, the ferrous iron in contaminated groundwater will be oxidized to ferric iron upon discharge to oxygenated surface water, and the subsequent precipitation of ferric oxyhydroxides would produce acidic conditions with detrimental effects to the ecosystem. One currently employed treatment technique is the use of an organic carbon permeable reactive barrier to stimulate sulfate reduction and subsequent metal sulfide precipitation (3, 4). The reduction of sulfate by organic carbon can be represented by

2CH2O + SO42- f 2HCO3- + H2S

(1)

where CH2O represents short-chain organic carbon molecules that are capable of being oxidized by sulfate-reducing bacteria (SRB). The sulfides produced from the above reaction form sparingly soluble sulfide minerals:

Me2+ + H2S f MeS(solid) + 2H+

(2)

where Me2+ represents divalent metal cations such as Fe2+, Zn2+, and Ni2+. Column experiments were previously conducted to simulate the treatment of acidic mine drainage contaminated groundwater and to assess the rates of sulfate reduction and metal removal by selected organic carbon mixtures (5, 6). Two columns were constructed consisting of a reactive mixture containing creek sediment as a bacterial source, limestone as a neutralizing agent, silica sand as a nonreactive porous medium, and an organic carbon source that consisted of composted leaf mulch, wood chips, sawdust, and sewage sludge. The column input solution was simulated acid mine drainage comparable in characteristics to the Nickel Rim plume (2). At the Nickel Rim mine site near Sudbury, ON, a full-scale reactive barrier was installed to treat groundwater emanating from the tailings impoundment. The barrier, which was installed in 1995, was shown to reduce sulfate and decrease downgradient metal loading (3). The reactive transport code MIN3P (7) was used to model physical heterogeneities and temporal variations in temperature that impact the performance of the Nickel Rim barrier (8, 9). Here, we use MIN3P (7) to conduct quantitative simulations of the remediation process for the more controlled conditions of the column experiments (5, 6). The focus of the present study is to test the existing conceptual model using reactive transport modeling, based on the detailed data set obtained from the column influent and effluent chemistry, and to investigate if and how changes in organic matter reactivity and the mineralogical assemblage within the treatment material affect the long-term metal removal capability of the reactive mixture. An additional goal of this study is to demonstrate the usefulness of multicomponent reactive transport modeling as an analysis tool, which can help to identify nonintuitive interactions between controlling processes in the column experiments. VOL. 38, NO. 11, 2004 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

9

3131

TABLE 1. Reaction Stoichiometries for Sulfate Reduction and Mineral Precipitation/Dissolution Reactionsa

(1)

CH2O + 0.5SO42- f CO32- + 0.5HS- + 1.5H+

sulfate reduction

keff (mol L-1 bulk s-1)

Ks (mol L-1)

6.9 × 10-9

1.62 × 10-3

Mineral Precipitation/Dissolution Reactions

(2) (3) (4) (5) (6) (7)

Fe2+ + HS- f FeS + H+ Fe2+ + CO32- f FeCO3 Ca2+ + CO32- f CaCO3 Ca2+ + SO42- + 2H2O f CaSO4‚2H2O Ni2+ + HS- f NiS + H+ Zn2+ + HS- f ZnS + H+

mackinawite siderite calcite gypsum millerite sphalerite

keff (mol L-1 bulk s-1)

log K

1.2 × 10-11 4.0 × 10-8 4.0 × 10-8 4.0 × 10-9 1.2 × 10-12 1.2 × 10-12

4.6480 10.4500 8.4750 4.5800 8.0420 11.6180

a All reactions are written in terms of the components used in the model. All constants apply to both column simulations except where noted in the text.

Conceptual Model The conceptual model is based on the column experiments described in Waybrant (5) and Waybrant et al. (6) and on field studies of the full-scale permeable barrier at the Nickel Rim site by Benner et al. (3, 10-12) and Herbert et al. (13). In the reactive medium sulfate is reduced by organic carbon to produce sulfide and carbonate according to reaction 1 in Table 1. Microbiological studies and sulfur isotope data confirm that this is a bacterially mediated reaction (10), although the model does not currently consider microbial growth and decay, uptake of nutrients, or byproducts of bacterial organic decay such as phosphate. Furthermore, the reactivity of the organic matter is considered to be constant. The model assumes that sulfate is exclusively reduced to sulfide, which either forms an aqueous sulfide species or precipitates in the form of metal sulfides. However, mineralogical studies suggest that native sulfur and/or pyrite may also form in the reactive medium (13). A possible reaction pathway involves the reduction of sulfate to sulfide followed by oxidation of the sulfide to form sulfur species of intermediate redox states (14). This is supported by microbiological studies at the Nickel Rim barrier that show elevated numbers of sulfide-oxidizing bacteria within the reactive media (11). It is unclear what oxidant is available in the barrier to facilitate this reaction. The rate of sulfate reduction by organic matter is dependent on the rate of organic carbon decomposition and on the sulfate concentrations (15-17). The approach described by Westrich and Berner (15) and Tarutis and Unz (16) treats the pool of degradable organic carbon as consisting of various groups of compounds each with their own intrinsic reactivity. The rate of the organic matter degradation as a whole is then the sum of the reactivity of each of the individual fractions:

∑k G

ROC ) -

i

(3)

i

where ki is the rate constant for decay of fraction i and Gi is the concentration of the fraction. Boudreau and Westrich (17) show that the rate of sulfate reduction by organic matter in marine sediments is dependent on the concentration of sulfate only when sulfate concentrations are low. The rate expression for sulfate reduction is then

[

]

[SO4] 1 RSO4 ) ROC 2 Ks + [SO4]

(4)

where the factor 1/2 is the ratio of moles of sulfate reduced to moles of organic carbon oxidized (eq 1) and Ks is the halfsaturation constant, which is assumed to be equal to 1.62 3132

9

ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 38, NO. 11, 2004

mmol L-1, as determined by Boudreau and Westrich (17). For SO4 concentrations significantly above 1.62 mmol L-1, RSO4 is essentially constant, and the rate of sulfate reduction is independent of sulfate concentrations. The Ks value of 1.62 mmol L-1 was determined for high-sulfate marine sediments and likely represents a maximum value. Given that the sediment used in the columns was acquired from a freshwater setting, the indigenous sulfate-reducing bacteria likely have a much higher sulfate affinity and therefore a much lower Ks (18). However, because the sulfate concentrations are generally much higher than Ks (i.e., influent sulfate concentration are >10 mmol L-1, and effluent sulfate concentrations only approach zero in the first 100 d of the experiment (5, 6)) the simulation results are not particularly sensitive to Ks. Column data (5, 6) and model simulations, which will be discussed below, suggest that the rate of organic matter decomposition does not change appreciably over the duration of the experiments. Therefore, it becomes possible to describe organic matter decomposition and sulfate reduction as an overall reaction, and a simplified version of eqs 3 and 4 can be used to quantify the rate of overall reaction progress:

RCH2O-SO4 ) -kCH2O-SO4

[

[SO4]

]

Ks + [SO4]

(5)

where kCH2O-SO4 is an effective rate constant. In response to the changes in geochemistry of the pore water due to sulfate reduction, a number of mineral precipitation and dissolution reactions occur. In addition to the precipitation of iron sulfides, the formation of iron carbonates occurs due to the increase in dissolved carbonate. The solubility of mackinawite (FeS) is much lower than that of siderite (FeCO3) and ultimately controls the concentration of ferrous iron in solution provided that sufficient sulfide is present. The nickel and zinc sulfide minerals (millerite and sphalerite, respectively) will also become oversaturated within the reactive medium and precipitate. The coprecipitation of nickel and zinc with iron sulfides is not considered in the model. The concentrations of these metals in solution is much lower than iron so that these minerals are less important in terms of controlling the overall aqueous chemistry in the pore water. Experimental data show a significant increase in calcium concentrations in the column, indicating a source of calcium within the reactive mixture (5, 6). The reactive mixture contains limestone, which is potentially dissolving in response to geochemical changes in the pore water, including changes in pH and decreases in carbonate concentration from siderite precipitation. Gypsum may also be present in

TABLE 2. Composition of Reactive Mixtures Used in Column Experiments (from ref 5)

composted leaf mulch wood chips sawdust sewage sludge creek sediment agricultural limestone silica sand

column 1 (wt %)

column 2 (wt %)

19.5 8 10.5 10 43 2 7

23 0 22 0 44 3 8

TABLE 3. Physical Parametersa Used in the Model porosity hydraulic conductivity (m s-1) diffusion coefficient (m s-1) dispersivity (m) flow rate (0-65 d) (m s-1) flow rate (>65 d) (m s-1) a

column 1

column 2

0.53 1.2 × 10-5 1.0 × 10-9 0.019 1.1 × 10-7 1.6 × 10-7

0.61 1.2 × 10-5 1.0 × 10-9 0.013 1.1 × 10-7 1.88 × 10-7

All parameters are values determined from column experiments

(5).

the column as a result of the column preparation techniques. In the current simulation, gypsum was not added to the reactive mixture but was allowed as a secondary phase. The pH in the system is controlled by the complex interaction between the sulfate reduction reactions, the mineral dissolution-precipitation reactions noted above, and speciation reactions. Column effluent pH initially starts at approximately 7.0 and decreases to 6.5 over the length of the experiment. This is an increase over influent pH which varies from 6.0 to 6.5 (5, 6).

Model Parameters The experiments were run in 40 cm long columns, 5 cm in diameter with a 5 cm silica sand layer at the influent end and a 1.5 cm silica sand layer at the effluent end. The silica layer at the influent end also contained crushed pyrite to ensure that the influent water was oxygen-free. Because the influent water was previously deoxygenated (6) and because of the low solubility of oxygen, it is unlikely that pyrite oxidation was sufficient to affect the geochemical composition of the influent water in any appreciable way. Therefore, the presence of crushed pyrite was neglected in the simulations. The remainder of the columns contained reactive mixtures as outlined in Table 2 (5, 6). The columns were modeled with a 40 cm long one-dimensional solution domain discretized into 40, 1 cm control volumes. The volume fractions of calcite in the reactive mixtures were set to 0.0031 for column 1 and 0.0036 for column 2, which is representative of the measured volume fraction in the columns but was adjusted to fit model data as described below. For organic carbon, the initial volume fraction was arbitrarily set to 0.1 to represent the amount of the readily degradable organic carbon in the mixture. In this case, the volume fraction of the organic carbon was sufficiently high so that the reaction progress was not limited by the availability of organic material. The remainder of the solid phase in the reactive mixture is considered to be nonreactive, as are the silica sand layers. Physical properties used in the model are listed in Table 3 and are assumed to be homogeneous throughout the entire solution domain. All properties are those measured from the column experiments (5, 6). The simulations were run for a 425 d period, which approximately corresponds to the length of time the column experiments continued. The column temperature was assumed to be 25 °C.

The following 15 components are used to describe the geochemical system: Ar(aq), Br, Ca, CO3, Cl, Fe(II), H+, H2S, K, Mg, Mn, Na, Ni, SO4, and Zn. On the basis of the WATEQ4F (19) and MINTEQA2 (20) databases, 68 aqueous species were identified and included in the model to accurately determine mineral solubilities. Equilibrium constants for speciation reactions are from the WATEQ4F (19) and MINTEQA2 (20) databases (see Supporting Information). The sulfate reduction reaction (reaction 1, Table 1) is assumed to be a kinetically controlled irreversible reaction, limited by the availability of sulfate only, and is modeled with a Monod-type rate expression of the form shown in eq 5. The rate constant given in Table 1 was calibrated to match the observed sulfate concentrations from column 1 data. Mineral dissolution and precipitation reactions are modeled as kinetically controlled reversible reactions with an empirical rate expression of the following form:

(

R ) -keff 1 -

IAP K

)

(6)

where keff is the effective rate constant for the dissolution of the mineral phase, IAP is the ion activity product, and K is the equilibrium constant. This rate expression allows calibrating rate constants to match the observed reaction progress and simultaneously honors the thermodynamic constraints of the reactions (21, 22). Effective rate constants are calibrated to match the observed data and are shown in Table 1. Equilibrium constants are from the WATEQ4F (19) and MINTEQA2 (20) databases (Table 1). The column input water chemistry shown in Table 4 (5, 6) was used for input boundary conditions. Note that from 0 to 96 d the input solution contained only high sulfate concentrations (10.51 mmol L-1) and the simulated acid mine drainage was not used as input solution until after 96 d. During feed 1, bromide was added to the input solution for the purposes of a tracer test. The initial water chemistry is the same as feed 0.

Results and Discussion Tracer Tests. Column effluent breakthrough curves for bromide from the model simulations and the column experiments, shown in Figure 1, show good agreement. This indicates that the model accurately simulates flow conditions using measured values of physical parameters from the column experiments (Table 3). Sulfate Reduction and Iron Removal: Column 1. Breakthrough curves of modeled effluent sulfate concentrations compared to column influent and effluent sulfate concentrations for column 1 are shown in Figure 2. Both the column effluent and modeled effluent sulfate concentrations show a decrease over influent sulfate concentrations as is expected due to sulfate reduction. The rate constant for sulfate reduction in the model was calibrated to provide a best fit for these data and was kept constant over the entire simulation period. The modeled data show a good fit to the column data throughout the 425 d period, indicating that the reactivity of the organic matter does not decrease significantly over time. Simulated sulfide concentrations show a good fit to the column data where experimental data are available (Figure 2). Sulfide concentrations are controlled by the amount of sulfide produced from sulfate reduction (eq 1, Table 1) minus the amount of sulfide removed through metal sulfide precipitation (eqs 2, 6, and 7, Table 1). The decreasing trend in sulfide is therefore due to a decrease in sulfate reduction or an increase in metal sulfide precipitation. However, the rate constant for sulfate reduction is constant, and through most of the simulation the concentration of sulfate remains well above the half-saturation constant (with the exception VOL. 38, NO. 11, 2004 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

9

3133

For feed 0: pH, alkalinity, and Ca calculated based on water containing 10.51 mmol b a Reproduced with permission from Environ. Sci. Technol. 2002, 36, 1349-1356. Copyright 2002 American Chemical Society). L-1 CaSO4 and in equilibrium with calcite and atmospheric CO2 as per Waybrant (5). c na, not available.

na 0.4003 0.136 0.146 0.217 0.026 0.4353 0.4951 0.4071 na 9.95 13.2 13.0 11.8 4.98 3.4 3.2 na na 0.1775 0.2343 0.2321 0.3604 0.2570 0.1631 0.1569 na na 0.016 0.024 0.022 0.21 0.040 0.027 0.057 na 0 95.9 135 163.6 205.6 254.6 298.1 344.6 395.6 feed 0b feed 1 feed 2 feed 3 feed 4 feed 5 feed 6 feed 7 feed 8

7.8 6.40 5.85 6.41 6.46 6.60 6.16 6.61 6.43

0.37 2.94 1.10 0.90 0.68 0.52 0.22 0.56 0.48

10.51 33.40 40.52 35.95 38.09 20.81 16.89 14.08 14.57

nac 12.43 21.15 15.68 17.12 7.944 6.793 5.977 5.663

na 0.011 0.015 0.0089 0.014 0.019 0.013 0.0087 na

na 2.04 2.17 2.12 2.12 0.861 0.613 0.565 na

10.6 12.6 2.5 2.3 2.1 5.1 5.3 5.5 na

na 0.691 0.811 0.826 0.747 0.302 0.20 0.18 na

0 0.33 0 0 na na na na na

Cl (mmol L-1) Br (mmol L-1) Mg (mmol L-1) K (mmol L-1) Ca (mmol L-1) Mn (mmol L-1) Ni (mmol L-1) start of feed (d)

pH

alkalinity (mequiv/L)

SO4 (mmol L-1)

Fe (mmol L-1)

Zn (mmol L-1)

Na (mmol L-1) 9

input solution

TABLE 4. Input Solution Chemistry from Column Studies and Used as Boundary Conditions for Modela 3134

FIGURE 1. Comparison of simulated (s) and observed effluent (eff) breakthrough for conservative Br tracer. Column data from ref 5.

ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 38, NO. 11, 2004

FIGURE 2. Breakthrough curves of simulated (s) effluent aqueous component concentrations for SO4, H2S, and Fe(II) compared to observed influent (in) and effluent (eff) data for column 1. The influent data have been shifted by 1 pore vol for comparison purposes. Column data from ref 5. of the first 100 d). Therefore, the rate of sulfate reduction and the amount of sulfide produced is constant throughout most of the simulation. Furthermore, the concentrations of Ni and Zn are very low, and the amount of millerite and sphalerite precipitation will not significantly affect aqueous sulfide concentrations. The decreasing trend in sulfide concentrations must therefore be predominantly due to an increase in the amount of iron sulfide precipitation over the course of the experiment. The mechanism by which mackinawite precipitation increases over time will be discussed in detail below. Another possible mechanism of sulfide removal, which is not considered in the model, is the oxidation of sulfide resulting in the formation of native sulfur or pyrite. Modeled ferrous iron concentrations show a good fit to column data. Both modeled and column effluent data show a significant decrease in the ferrous iron concentrations over influent data due to mackinawite and siderite precipitation. However, initially (up to ∼250 d) all ferrous iron is removed from solution, equivalent to ∼0.015 mol L-1, while at later times (after ∼250 d) ferrous iron remains in the effluent and the amount removed is only ∼0.005 mol L-1. Therefore, the rate of iron removal decreases substantially as the column experiments progress. The decrease in sulfide concentration and the increase in ferrous iron concentrations in the column effluent over the duration of the experiments can be examined through the use of spatial profiles of total aqueous component concentrations for Fe(II), SO4, and H2S at 150 d (Figure 3A) and

FIGURE 4. Simulated (s) effluent aqueous component concentrations compared to influent (in) and effluent (eff) data for (A) Ca and (B) pH for column 1. The influent data have been shifted by 1 pore vol for comparison purposes. Column data from ref 5.

FIGURE 3. Simulated spatial profiles through column 1 at 150 (panels A and B), 200 (panels C and D), and 300 d (panels E and F). A positive rate indicates precipitation while a negative rate indicates dissolution. The variable x is the distance along the column length. Dissolution/precipitation rates in mol L-1 bulk day-1. SO4(in) is the corresponding column influent concentration assuming only advective flow. dissolution-precipitation rates for various minerals at the same time (Figure 3B). Sulfate concentrations show a steady decrease throughout the profile due to sulfate reduction. Concentrations remain well above the half-saturation constant throughout the profile, indicating that the sulfate reduction rate, and therefore sulfide production, will remain nearly constant throughout the column. A linear trend would be expected for a constant sulfate reduction rate; however, changes in input concentrations (Figure 3A) and dissolution and precipitation of gypsum (Figure 3B) produce deviations from the linear trend. Sulfide concentrations remain low initially in the column due to mackinawite precipitation and begin to increase at 0.2 m, once mackinawite precipitation rates decline (Figure 3A,B). Mackinawite is initially oversaturated in the column due to high influent concentration of ferrous iron and the production of sulfide. However, due to the complete removal of Fe(II) at approximately 0.2 m, mackinawite is no longer oversaturated and precipitation ceases. Ferrous iron levels are controlled by precipitation of both mackinawite and siderite. The rate of precipitation for siderite is much greater than that of mackinawite; therefore, siderite is the predominant sink for ferrous iron at this point in time (Figure 3B). Siderite solubility is controlled by carbonate and ferrous iron concentration. The precipitation of siderite consumes carbonate leading to undersaturation with respect to calcite, which causes calcite to dissolve and liberates more carbonate. The net result is the mutual enhancement of the two processes resulting in increased siderite precipitation, enhanced iron removal, increased calcite dissolution, and increased dissolved calcium concentrations. As is seen in Figure 3A,B, the precipitation of siderite and the dissolution of calcite cease at 0.2 m as the ferrous iron concentrations approach zero. Likewise, these processes begin at 0.1 m,

which corresponds to the point in the column where calcite exists. At 150 d, calcite has been depleted in the column from 0 to 0.1 m due to dissolution. Figure 3C,D shows spatial profiles through the column at 200 d. The patterns observed are very similar to those seen at 150 d (Figure 3A,B). However, in this case siderite precipitation and calcite dissolution occur much closer to the outflow end of the column, ∼0.2-0.3 m. The delay in the commencement of these processes is due to the depletion of calcite in the column up to ∼0.2 m. This causes ferrous iron concentrations to remain high through most of the column, allowing mackinawite precipitation to continue through a longer portion of the column. Mackinawite precipitation is still iron limited in this case; however, more sulfide is removed, resulting in a lower sulfide concentration in the column effluent (see also Figure 2). The spatial profiles at 300 d are shown in Figure 3E,F. The most notable difference compared to earlier profiles is that siderite precipitation does not proceed to any significant degree. This is because calcite is now completely dissolved from the column. The result in this case is that iron levels remain relatively high throughout the column and that mackinawite precipitation is no longer limited by ferrous iron. Mackinawite precipitation then continues throughout the column, and sulfide concentrations approach zero, as is observed in the column effluent (Figure 2). As was noted above, the amount of solid-phase calcite included in the simulations was adjusted to fit the data. As no analysis of solid-phase calcite was done on the columns, the availability of calcite at various times cannot be verified. However, other evidence suggest that calcite dissolution is occurring in the column and that calcite is exhausted from the column at ∼300 d. Breakthrough curves of modeled effluent calcium concentrations compared to column influent and effluent calcium concentrations for column 1 are shown in Figure 4A. Both modeled and column effluent calcium concentrations show a significant increase over influent concentration in the period from ∼100 to 300 d, which corresponds to the time period when model results indicate that calcite is dissolving and iron is removed by siderite precipitation. Although the concentrations of calcium in the modeled and column data do not agree exactly, the general VOL. 38, NO. 11, 2004 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

9

3135

trends are very similar. The discrepancy in the calcium concentration is likely due to the difficulties in simulating pH and its effect on carbonate speciation, as described below. Furthermore, at later times (>300 d) calcium influent and effluent concentrations are equal, indicating that the source of calcium no longer exists. This corresponds to the period of time when the model results suggest that calcite is depleted from the column. An additional line of evidence supporting siderite precipitation is simple mass balance calculations, which indicate that the amount of sulfide produced cannot account for the large decrease in ferrous iron in the column through mackinawite precipitation alone. For example, at 150 d ∼0.01 mol L-1 of sulfate is removed from solution (Figure 2). Using the stoichiometry of eqs 1 and 2 in Table 1, 1 mol of sulfate produces 1 mol of sulfide, which will precipitate with 1 mol of ferrous iron to produce 1 mol of mackinawite. At this time ∼0.005 mol L-1 sulfide remains in the column effluent, indicating that only ∼0.005 mol L-1 sulfide is used for mackinawite precipitation. However, ∼0.02 mol L-1 ferrous iron is removed from solution, which suggests that ∼0.015 mol L-1 ferrous iron is removed via an alternate sink. The pH data from the model simulations and the column data (5, 6) are shown in Figure 4B. The model underpredicts pH at all times and more significantly at later times, suggesting some inconsistency in the conceptual model, which may arise from the simplified approach to the sulfate reduction reactions. These reactions involve the decomposition of complex organic matter that contains other compounds such as ammonia and phosphate. Column effluent was sampled for ammonium and phosphate and did not show any increase in the amount of ammonium but did show phosphate concentrations up to 10 mg L-1 (5). Studies at the Nickel Rim site have found sulfide-oxidizing bacteria in the barrier (11), suggesting that intermediate sulfur species may be formed, which could not be accounted for in the simulations. A further unknown in these experiments is the composition of the sediment that was used as the bacterial source. This sediment contained uncharacterized minerals that may have affected the geochemical evolution of the system. All of these processes may have contributed to the discrepancies between observed and simulated pH of the effluent water. Nickel, Zinc, and Other Cations. The removal Ni and Zn is simulated by the precipitation of millerite and sphalerite, although, it is likely that coprecipitation with iron sulfides occurred in the column experiment instead of, or in addition to, these precipitation reactions. Millerite and sphalerite are very insoluble minerals and precipitate quickly with small amounts of sulfide present. Ni and Zn are only minor constituents of the simulated acid mine drainage and do not have a significant effect on the geochemical evolution of the system. Experimental data (5) show that manganese is removed from the column up to ∼250 d, corresponding to the time period when calcite is dissolving in the column (Figure 4) and suggesting that the precipitation of rhodochrosite (MnCO3) is a potential sink for manganese. The simulations do not indicate that precipitation of rhodochrosite is occurring but do show that rhodochrosite approaches saturation near the column effluent. The discrepancy in the simulated pH (i.e., lower simulated pH than observed) would have the effect of increasing the solubility of rhodochrosite and could account for this inconsistency in the model. Manganese concentrations are low in the column experiments and will not significantly affect the overall geochemical evolution of the system. Other cations such as sodium, magnesium, and potassium are nonreactive in the simulations, which is consistent with the column data (5, 6). Column 2. Breakthrough curves of simulated effluent sulfate concentrations compared to observed column influent and effluent sulfate concentrations for column 2 are shown 3136

9

ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 38, NO. 11, 2004

FIGURE 5. Breakthrough curves of simulated (s) effluent aqueous component concentrations for SO4, H2S, and Fe(II) compared to observed influent (in) and effluent (eff) data for column 2 with a sulfate reduction rate of (A) 6.9 × 10-9 and (B) 1.5 × 10-8 mol L-1 bulk s-1. The influent data have been shifted by 1 pore vol for comparison purposes. Column data from ref 5. in Figure 5A using the same rate constant for sulfate reduction, 6.9 × 10-9 mol L-1 bulk s-1, as was calibrated with column 1. The simulated data show a reasonably good fit to the observed data initially but overpredicts sulfate concentrations at later times. This may suggest that the sulfate reduction rate in the column experiment is increasing with time. A possible explanation is that microbial population growth needs to be considered. Microbial communities are potentially growing during the initial stages of the experiment and do not reach their maximum population until beyond approximately 150 d. Currently the model does not consider microbial growth. Figure 5B shows the results of a simulation run with a sulfate reduction rate constant of 1.5 × 10-8 mol L-1 bulk s-1. This rate constant produces a better fit to the later sulfate data, suggesting a maximum sulfate reduction rate constant, but also overpredicts sulfide at all times. Although the simulations for column 2 were unable to exactly reproduce the observed trends, the general agreement, specifically the decline in sulfide concentration, the increase in the ferrous iron concentration, and the increase in the calcium concentration (data not shown) would tend to indicate that the conceptual model used is a reasonable representation of the geochemical system in the column. Model Calibration, Parameter Sensitivity, and Model Uniqueness. The most uncertain model parameters are the calibrated rate constants for sulfate reduction and the mineral dissolution-precipitation reactions (Table 1). The main difficulty in the calibration process is the inherent nonlinearity of the system, implying that multiple major sources and sinks exist for the relevant aqueous components (SO4: sulfate reduction and gypsum dissolution-precipitation; H2S: sulfate reduction and mackinawite precipitation; Fe: mackinawite and siderite; CO3: siderite and calcite; Ca:

calcite and gypsum). Initially, the rate of sulfate reduction was calibrated based on the assumption that gypsum dissolution-precipitation is an insignificant sink for SO4. This was accomplished by adjusting the rate constant for sulfate reduction such that the modeled effluent sulfate concentration matched the observed values as shown in Figure 2. With the calibrated rate and reaction stoichiometry (eq 1), the amount of H2S produced is determined. This approach allows the estimation of the rate of mackinawite precipitation in an attempt to match H2S in the effluent. Subsequently, the rates for siderite and calcite dissolution-precipitation were estimated to match Fe and Ca. Setting a rate for gypsum dissolutionprecipitation closes the system and affects the simulated SO4 concentrations, providing feedback on the initial assumption. This cycle has been repeated in an iterative fashion until the results presented here were obtained. The rates of millerite and sphalerite dissolutionprecipitation are inconsequential to the overall geochemical evolution of the system, because the concentrations of Ni and Zn in the influent water are minor as compared to the concentrations of the major cations Fe and Ca (Table 4). These rate constants are therefore only constrained by the fact that Ni and Zn are removed almost completely from solution before the pore water exits the column. It is instructive to compare the calibrated sulfate reduction rate to the original interpretation of the column experiments and results from related field studies (12). For column 1, the calibrated rate constant, equivalent to194 mmol of SO4 (L of H2O)-1 yr-1, compares favorably with the average value determined from the column experiments, 186 mmol of SO4 (L of H2O)-1 yr-1 (5, 6), which is not surprising considering the simulations were calibrated to these data. However, the calibrated rate constant is higher than that determined for the full-scale barrier at Nickel Rim, 47 mmol of SO4 (L of H2O)-1 yr-1 (12). This discrepancy can be in part be explained by the effect of temperature on the reaction rate, given that the average annual temperature of the field-scale barrier is 9 °C (12) and the column experiments were run at room temperature (5, 6). The Arrhenius equation expressed as

log

[

k1 Ea 1 1 ) k2 2.303R T2 T1

]

(7)

(where k1 and k2 are the rate constants at temperatures T1 and T2 (K), respectively; Ea (kJ mol-1) is the activation energy, and R (8.314 × 10-3 kJ mol-1 K-1) is the gas constant) can be used to relate the change in temperature on the rate of the reaction. Here, Ea is defined as an apparent activation energy, which measures the response of the bacterial community to a change in temperature and is dependent on the age and reactivity of the organic substrate (23). We use a value of 40 kJ mol-1 for Ea based on the Nickel Rim reactive barrier, which contained similar organic material (12). This value must be considered an estimate because Ea values can be highly variable, even within similar substrates (23). Previous values determined for Ea range from 23 to 132 kJ mol-1 (23, 24). Given that k1 is the calibrated rate constant, 194 mmol of SO4 (L of H2O)-1 yr-1, T1 equal to 298 K, and T2 equal to 282 K (12), the estimated rate constant at 9 °C for sulfate reduction with the organic material in the column is 78 mmol of SO4 (L of H2O)-1 yr-1. This value is still higher than the sulfate reduction rate determined from the full-scale barrier (12), but considering uncertainties regarding the composition and age of the different materials used in the laboratory and field studies, some discrepancies in the temperaturenormalized sulfate reduction rates would be expected. It is difficult to state with certainty that the model parameters and conceptual model presented are unique. In fact, the discrepancies noted in the simulated and observed

effluent pH and calcium concentrations indicate that the model as presented is incomplete, which likely has some effect on the calibrated model parameters. However, given the constraints noted above, we are confident that the principal results are adequately supported. Implications for Long-Term Performance. In this modeling study, a fixed rate constant was used for sulfate reduction throughout the simulation period. For column 1, this fixed rate was adequate for reproducing the sulfate concentrations throughout the experiment, suggesting that the sulfate reduction rate and the reactivity of the organic matter was not significantly decreasing. The modeling results for column 2 indicate that the sulfate reduction rate increased during the experiment, suggesting that microbial population growth may be occurring, but again indicating no significant decrease in sulfate reduction rate or organic carbon reactivity. These observations are contrary to the models described by Westrich and Berner (15) and Tarutis and Unz (16), which suggest an initially high organic matter degradation rate that declines quickly as the labile fraction of organic matter is used up. However, this apparent contradiction is not surprising considering the relatively short duration of the experiments and the relatively large organic carbon pool. On the basis of the measured compositions of the organic matter (5, 6), both columns contained 5300 mmol of carbon. Given average sulfate reduction rates of 186 and 277 mmol (L of H2O)-1 yr-1 and pore volumes of 0.420 and 0.480 L for columns 1 and 2, respectively, the total amount of sulfate reduced over a 425 d period for each of the columns is 91 and 155 mmol, respectively. This is equivalent to 182 mmol of C or 3.4% of the available carbon for column 1 and 310 mmol of C or 5.8% for column 2. Although a large portion of the available carbon is likely of a more recalcitrant nature, considering the small percentages of the organic carbon pool used, it is possible that the most labile portion of the pool was not exhausted during the experiments. The ability of the reactive mixture to remove ferrous iron is of primary importance to this remediation strategy. Furthermore, the form in which iron is immobilized may also have significant implications. The solubility of mackinawite is much lower than that of siderite. Therefore, mackinawite is the preferred iron sink in terms of providing a stable solid phase with minimal risk of dissolution at a later date. Data and simulations indicate that early in the experiments the predominant iron sink is siderite, which inhibits mackinawite precipitation. The precipitation of siderite does have the ability to remove larger amounts of iron, although this situation may be considered only temporary, as it will only continue as long as calcite dissolution occurs in the system. Furthermore, due to its relatively high solubility, siderite has the potential to redissolve as conditions in the inflowing pore water change; this includes changes in pH and/or the concentration of Fe2+ and total carbonate. Later in the column life, once all the calcite had been exhausted, siderite no longer precipitates to any significant extent and mackinawite becomes the predominant iron sink. At this point less iron is removed from solution. It is therefore apparent that Fe removal at later times give a better indication of the longterm ability of the column to remediate acid mine drainage.

Acknowledgments Funding for this research was provided by a Canadian Water Network grant awarded to D.W.B. and Rejean Samson.

Supporting Information Available Aqueous complexes and hydrolysis products, stoichiometric coefficients, and equilibrium constants used in simulations of columns 1 and 2. This material is available free of charge via the Internet at http://pubs.acs.org. VOL. 38, NO. 11, 2004 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

9

3137

Literature Cited (1) Dubrovsky, N. M.; Morin, K. A.; Cherry, J. A.; Smyth, D. J. A. Water Pollut. Res. J. Can. 1984, 19, 55-89. (2) Johnson, R. H.; Blowes, D. W.; Robertson, W. D.; Jambor, J. L. J. Contam. Hydrol. 2000, 41, 49-80. (3) Benner, S. G.; Blowes, D. W.; Ptacek, C. J. Ground Water Monit. Rem. 1997, 17, 99-107. (4) Ludwig, R. D.; McGregor, R. G.; Blowes, D. W.; Benner, S. G.; Mountjoy, K. Ground Water 2002, 40, 59-66. (5) Waybrant, K. R. The prevention of acid mine drainage using in situ porous reactive walls: A laboratory study. M.Sc. Thesis, University of Waterloo, Waterloo, ON, Canada, 1995. (6) Waybrant, K. R.; Ptacek, C. J.; Blowes, D. W. Environ. Sci. Technol. 2002, 36, 1349-1356. (7) Mayer, K. U.; Frind, E. O.; Blowes, D. W. Water Resour. Res. 2002, 38, 1174-1194. (8) Benner, S. G.; Mayer, K. U.; Blowes, D. W. Abstract H22F-05; AGU Fall Meeting, San Francisco, CA, December 13-17, 1999. (9) Mayer, K. U.; Benner, S. G.; Blowes, D. W.; Frind, E. O. In Sudbury ‘99; Conference on Mining and the Environment, Sudbury, Ontario, 1999; Vol. 1, pp 145-154. (10) Benner, S. G.; Blowes, D. W.; Gould, W. D.; Herbert, R. B., Jr.; Ptacek, C. J. Environ. Sci. Technol. 1999, 33, 2793-2799. (11) Benner, S. G.; Gould, W. D.; Blowes, D. W. Chem. Geol. 2000, 169, 435-448. (12) Benner, S. G.; Blowes, D. W.; Ptacek, C. J.; Mayer, K. U. Appl. Geochem. 2002, 17, 301-320. (13) Herbert, R. B., Jr.; Benner, S. G.; Blowes, D. W. Appl. Geochem. 2000, 15, 1331-1343. (14) Luther, G. W.; Church, T. M. In Sulfur Cycling on the Continents; Howarth, R. D., Stewart, J. W. B., Ivanov, M. V., Eds.; John Wiley & Sons: New York, 1992,

3138

9

ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 38, NO. 11, 2004

(15) Westrich, J. T.; Berner, R. A. Limnol. Oceanogr. 1984, 29, 236249. (16) Tarutis, W. J., Jr.; Unz, R. F. Water Sci. Technol. 1994, 29, 219226. (17) Boudreau, B. P.; Westrich, J. T. Geochim. Cosmochim. Acta 1984, 48, 2503-2516. (18) Roden, E. E.; Tuttle, J. H. Biogeochemistry 1993, 22, 81-105. (19) Ball, J. W.; Nordstrom, D. K. User’s Manual for WATEQ4F, with revised thermodynamic database and test cases for calculating speciation of major, trace and redox elements in natural waters; USGS Open-File Report 91-183; U.S. Geological Survey: Reston, VA, 1991; 189 pp. plus diskette. (20) Allison, J. D.; Brown, D. S.; Novo-Gradac, K. J. MINTEQA2/ PRODEFA2, A geochemical assessment model for environmental systems: Version 3.0 Users’s Manual; U.S Environmental Protection Agency: Washington, DC, 1991; EPA/600/3-91/021. (21) Steefel, C. I.; van Cappellen, P. Geochim, Cosmochim, Acta 1990, 54, 2657-2677. (22) Steefel, C. I.; MacQuarrie, K. T. M. In Reactive transport in porous media; Lichtner, P. C., Steefel, C. I., Oelkers, E. H., Eds.; Reviews of Mineralogy, Vol. 34; Mineralogical Society of America: Warrendale, PA, 1996; Chapter 2 (23) Westrich, J. T.; Berner, R. A. Geomicrobiol. J. 1988, 6, 99-117. (24) Sagemann, J.; Jorgensen, B. B.; Greeff, O. Geomicrobiol. J. 1998, 15, 85-100.

Received for review September 2, 2003. Revised manuscript received March 4, 2004. Accepted March 15, 2004. ES0349608