I

Mo

DCa

b

a

ARRAA

KACSPST

1

olaChhlserz

BT

h1

ARTICLE IN PRESSG ModelJGGC-1491; No. of Pages 10

International Journal of Greenhouse Gas Control xxx (2015) xxx–xxx

Contents lists available at ScienceDirect

International Journal of Greenhouse Gas Control

journa l homepage: www.e lsev ier .com/ locate / i jggc

odeling the impact of carbon dioxide leakage into an unconfined,xidizing carbonate aquifer

iana H. Bacon a,∗, Nikolla P. Qafoku a, Zhenxue Dai b, Elizabeth H. Keating b,hristopher F. Brown a

Pacific Northwest National Laboratory, P.O. Box 999, Richland, WA 99352, USALos Alamos National Laboratory, P.O. Box 1663, Los Alamos, NM 87545, USA

r t i c l e i n f o

rticle history:eceived 15 September 2014eceived in revised form 28 March 2015ccepted 14 April 2015vailable online xxx

eywords:quiferO2

imulationHREEQcTOMP

a b s t r a c t

Multiphase, reactive transport modeling was used to identify the mechanisms controlling trace metalrelease under elevated CO2 conditions from a well-characterized carbonate aquifer. Modeling was con-ducted for both batch and column experiments. The column experiments resulted in higher trace metalconcentrations because the rock to water ratio was higher. A kinetic desorption model fits the overalltrends in release for seven trace metals observed in batch and column experiments exposing EdwardsAquifer material to elevated concentrations of CO2. Observed and predicted trace metal concentrationsare compared to groundwater concentrations from this aquifer to determine the potential for leaking CO2

to adversely impact drinking water quality. Finally, a three-dimensional multiphase flow and reactive-transport simulation of CO2 leakage from an abandoned wellbore into a generalized model of the shallow,unconfined portion of the aquifer is used to determine potential impacts on groundwater quality. As ameasure of adverse impacts on groundwater quality, both the EPA’s regulatory limits and the maximum

race metals trace metal concentration observed in the aquifer were used as threshold values. Results of the field-scale simulations indicate that CO2 leakage into a carbonate aquifer is likely to cause decreases in pHand increases in TDS beyond observed ranges in the aquifer and beyond regulatory limits. However,trace metal concentrations are not predicted to exceed either the observed maximums or the regulatorylimits.

© 2015 Elsevier Ltd. All rights reserved.

. Introduction

Although storage reservoirs are evaluated and selected basedn their ability to safely and securely store emplaced fluids,

eakage of CO2 from storage reservoirs is a primary risk factornd potential barrier to the widespread acceptance of geologicO2 sequestration. Harvey et al. (2013) conducted a compre-ensive review of the recently published literature regardingow elevated CO2 levels may affect geochemical processes under

ow-temperature, low-pressure conditions characteristic of near-urface environments. Emphasis was placed on CO2-induced

Please cite this article in press as: Bacon, D.H., et al., Modeling the

carbonate aquifer. Int. J. Greenhouse Gas Control (2015), http://dx.doi

ffects on dissolution/precipitation and adsorption/desorptioneactions, and consequences for the geochemistry of the vadoseone and potable aquifers. The review concluded that a significant

∗ Corresponding author at: Pacific Northwest National Laboratory, 902 Battelleoulevard, P.O. Box 999, MSIN: K9-33, Richland, WA 99352, USA.el.: +1 509 372 6132.

E-mail address: diana.bacon@pnnl.gov (D.H. Bacon).

ttp://dx.doi.org/10.1016/j.ijggc.2015.04.008750-5836/© 2015 Elsevier Ltd. All rights reserved.

amount of new scientific evidence suggests that CO2 intrusion intopotable aquifers or the vadose zone may have both beneficial anddeleterious outcomes.

A number of modeling studies have focused on the potential formobilization of toxic metals in groundwater due to acidificationof the aquifers; many of these studies have focused on sandstoneaquifers. For instance, Vong et al. (2011) modeled a glauconiticsandstone aquifer exposed to CO2 over 10 years and predictedelevated Cd, Pb, and Zn concentrations due to acidification anddissolution of greenockite, galena, and sphalerite. Jacquemet et al.(2011) developed a similar model with SOx and NOx as impuri-ties in the CO2 gas stream and found increased aqueous Fe and Mnmineral dissolution. Wang and Jaffe (2004) modeled buffered andunbuffered aquifers exposed to CO2 for 8 years and found increasedPb concentrations from acidic dissolution of galena. Zheng et al.(2009) modeled an Eastern Coastal Plain aquifer exposed to CO2

impact of carbon dioxide leakage into an unconfined, oxidizing.org/10.1016/j.ijggc.2015.04.008

for 100 years and predicted increased aqueous Pb and As due togalena and arsenian pyrite dissolution. A modeling study of thevadose zone exposed to CO2 for up to 40 days by Altevogt andJaffe (2005) predicted the potential mobilization of toxic metals

ING ModelI

2 f Gree

d(wBtiweiiNaatC

ewcambhimsciAlniAmg(mimEtgtcneac

a

TSo

ARTICLEJGGC-1491; No. of Pages 10

D.H. Bacon et al. / International Journal o

ue to acidification of the vadose zone pore fluids. Zheng et al.2012) modeled CO2 injection for 30 days into the shallow ground-ater at the Zero Emission Research and Technology field site in

ozeman, Montana, and predicted increased dissolution of reac-ive Fe-bearing minerals and that calcite dissolution would resultn increased solution concentrations of calcium and carbonate that

ould drive ion-exchange reactions in the sediment. Viswanathant al. (2012) developed an inverse reactive-transport model tonterpret As mobilization observed in a batch experiment with sed-mentary samples collected from a shallow sandstone aquifer inew Mexico where CO2 is actively upwelling and has been studieds a natural analog. Because these studies focused on sandstonequifers, the effects of acidification are likely to be more severehan in a carbonate aquifer which can buffer pH changes due toO2 leakage.

One experimental study has included carbonate aquifers. Yangt al. (2014) performed inverse reactive-transport modeling ofater–rock–CO2 batch experiments for potable aquifers, three

arbonate-poor and three carbonate-rich, from the Gulf Coast areand found that release of Ba, Mn, and Sr were likely regulated byineral dissolution (Type I), and that As solubility was controlled

y adsorption/desorption from clay (Type II). Also, a modeling studyas focused on trace metal release from pure minerals, includ-

ng calcite. Navarre-Sitchler et al. (2013) used reactive-transportodeling to predict the release of Pb from either galena or calcite

olid solution and found that regardless of the Pb source, the Pboncentrations in solution remained below the maximum contam-nant level (MCL) for Pb set by the U.S. Environmental Protectiongency (15 �g/L). However, the formation of aqueous complexes

ike PbCO3(aq) that would form under reducing conditions wasot considered, which would significantly increase the solubil-

ty of galena and increase the total aqueous concentration of Pb.lso, the EPA’s Class VI Rule requires groundwater geochemistryonitoring above the confining zone to detect changes in aqueous

eochemistry resulting from fluid leakage out of the injection zone40CFR 146.90(d)) where the results of groundwater monitoring

ay be compared against baseline geochemical data collected dur-ng site characterization to obtain evidence of fluid movement that

ay impact underground sources of drinking water (USDWs) (U.S.nvironmental Protection Agency, 2012). Therefore the MCL is nothe correct threshold for determining the impact of CO2 leaks onroundwater. It is important to establish the baseline groundwa-er chemistry that captures the natural variability of trace metaloncentrations in USDWs, and to look for changes greater than theatural variability with space and time (Last et al., 2013). Bacont al. (2014) modeled trace metal release from a shallow carbon-

Please cite this article in press as: Bacon, D.H., et al., Modeling the

carbonate aquifer. Int. J. Greenhouse Gas Control (2015), http://dx.doi

te aquifer in response to CO2 leakage using equilibrium surfaceomplexation reactions taken from the literature.

Experimental measurements are needed to identify the mech-nisms controlling trace metal release from a particular aquifer

able 1ummary statistics and EPA primary drinking water standards for pH, TDS, and MCLs for trf the Edwards Aquifer (Musgrove et al., 2010).

Measurement Number of samples Number of detections Minimum

pH, standard units 90 90 6.28

TDS (mg/L) 90 90 201

Sb (�g/L) 87 3 <.05

As (�g/L) 87 54 <.9

Ba (�g/L) 87 87 22.9

Be (�g/L) 87 1 <.06

Cd (�g/L) 87 0 <.04

Cr (�g/L) 87 67 <.8

Cu (�g/L) 87 50 <.4

Pb (�g/L) 83 24 <.08

Se (�g/L) 87 56 <.08

Th (�g/L) 52 4 <.04

PRESSnhouse Gas Control xxx (2015) xxx–xxx

material. Experimental work has largely focused on batch stud-ies. Little and Jackson (2010) conducted a yearlong CO2-nanopurewater-sediment study with sediments from aquifers that overlaypotential geologic carbon storage sites in the United States andfound that aqueous concentrations of some species (Mn, Fe, Co, Ni,and Zn) increased, some (e.g., Mo) decreased, and others remainedunaffected. Lu et al. (2010) exposed sediments from carbonate-poor and carbonate-rich potable aquifers in the U.S. Gulf Coastregion to CO2 for two weeks and observed that “Type I” cationsthat were controlled by mineral dissolution (Ca, Mg, Si, K, Sr, Mn, B,Zn) rapidly increased and reached stable concentrations and that“Type II” cations that were controlled by adsorption/desorptionfrom clay (Fe, Al, Mo, U, V, As, Cr, Cs, Rb, Ni and Cu) increased butthen declined. Wei et al. (2011) exposed variably saturated soils toCO2 at 25 bar for 3 days and observed that aqueous concentrationsof some species increased (Mg, K, Al, V, Cr, Mn, Fe, Co, Cu, Rb, Sr, Ba,Pb, and U) and others (Zn and Cd) decreased. However, the mecha-nisms that control the release of elements into the aqueous phaseare not currently well understood.

The objective of this paper is to use modeling to identify themechanisms controlling trace metal release under elevated CO2conditions from a well-characterized carbonate aquifer, the shal-low/urban unconfined portion of the Edwards aquifer near SanAntonio, Texas, containing little clay, which is an USDW. Modelingwas conducted for two experimental scenarios: batch experimentsto simulate sudden, fast, and short-lived release of CO2 as wouldoccur in the case of well failure during injection, and column experi-ments to simulate more gradual leaks such as those occurring alongundetected faults, fractures, or well linings. Observed and predictedtrace metal concentrations are compared to groundwater from thisaquifer to determine the potential for leaking CO2 to adverselyimpact drinking water quality. Finally, a three-dimensional multi-phase flow and reactive-transport simulation of CO2 leakage froman abandoned wellbore into a generalized model of the shallow,unconfined portion of the Edwards Aquifer is used to determinepotential impacts on groundwater quality. As a measure of adverseimpacts on groundwater quality, we use both the EPA’s MCL lim-its and the maximum trace metal concentration observed in theaquifer.

2. Methods

2.1. Aquifer data

The National Risk Assessment Partnership (NRAP) consists offive U.S. Department of Energy national laboratories collaborating

impact of carbon dioxide leakage into an unconfined, oxidizing.org/10.1016/j.ijggc.2015.04.008

to develop a framework for predicting the risks associated withcarbon sequestration (https://edx.netl.doe.gov/nrap/). The NRAPGroundwater Protection Working Group chose Edwards Aquiferfor study because it is a well-characterized carbonate aquifer. As

ace metals in 90 groundwater samples from the shallow/urban unconfined portion

25th percentile Median 75th percentile Maximum MCL

6.8 6.87 6.92 7.29 6.5312 329 360 456 500

<.2 <.30 <1 .06 6<1 .27 .34 1.11 1032.1 36.9 42.3 69.9 2000

<.06 <.06 <1 .04 4<.04 <.04 <1 <1 5

.20 .44 2.09 5.57 100<.4 .33 1.04 57.3 1300<.08 <1 .05 .15 15

<1 .26 .35 1.4 50<.04 <.04 <.04 .05 2

IN PRESSG ModelI

f Greenhouse Gas Control xxx (2015) xxx–xxx 3

pUttd2siccacpctfifi2wa

2

di(iscetorfAtgAStg(wistC0CaEe

2

m

TM

ARTICLEJGGC-1491; No. of Pages 10

D.H. Bacon et al. / International Journal o

art of the National Water-Quality Assessment Program (2014), the.S. Geological Survey (USGS) collected and analyzed groundwa-

er samples from 1996 to 2006 from the San Antonio segment ofhe Edwards Aquifer in central Texas, a productive karst aquifereveloped in Cretaceous-age carbonate rocks (Musgrove et al.,010). The National Water-Quality Assessment (NAWQA) Programtudies provide an extensive data set of groundwater geochem-stry and water quality, consisting of 249 groundwater samplesollected from 136 sites (wells and springs), including (1) wellsompleted in the shallow, unconfined, and urbanized part of thequifer in the vicinity of San Antonio (shallow/urban unconfinedategory); (2) wells completed in the unconfined (outcrop area)art of the regional aquifer (unconfined category); and (3) wellsompleted in and springs discharging from the confined part ofhe regional aquifer (confined category). Table 1 summarizes theeld observations of 10 trace metals from the shallow/urban uncon-ned portion of the Edwards Aquifer collected between 1996 and006 (Musgrove et al., 2010). All data reported for these ground-ater samples were below the respective regulatory limits for the

nalytes of interest.

.2. Experimental Data

Qafoku et al. (2013) used two experimental methodologies toetermine the impact of CO2 release: batch and column exper-

ments. Readers are referred to the original report Qafoku et al.2013) and subsequent journal article (Wang et al., 2015) for exper-mental details, which are summarized here. The objective of thetudy was to investigate impacts on the potable aquifer pH andhemical composition caused by the intrusion of CO2 gas. Thexperiments were conducted at room temperature and were openo the atmosphere to assess the effects of the leaking CO2 gasn mineral dissolution/precipitation and/or sorption/desorptioneactions in the natural solid materials affected by CO2 migratingrom deep geologic storage. Representative rocks from the Edwardsquifer were used for testing. A synthetic groundwater (SGW) solu-

ion, which was similar in composition to the Edwards Aquiferroundwater, based on USGS groundwater data of the Edwardsquifer, was used in the batch and column studies. Samples of theGW used in each experiment underwent full chemical analysiso determine the final chemical composition of each batch. CO2as was injected continuously in the batch reactors. High-purity99.998%) or industrial-grade (99.9%) CO2 gas delivered by Oxarcas used in these experiments. The column experiments modeled

n this paper were initially leached with the appropriate SGW foreveral hours to achieve hydrological equilibrium (i.e., full satura-ion) and then they were leached with CO2 gas-saturated SGW. TheO2 gas was continuously purged in the influent bottle at a rate of.5 mL/min. Flow through the column was stopped twice after theO2 gas-saturated SGW was used as the influent solution in order tossess the effect of kinetics on the system. Changes in effluent pH,h, and chemical concentration with time were followed in thesexperiments.

Please cite this article in press as: Bacon, D.H., et al., Modeling the

carbonate aquifer. Int. J. Greenhouse Gas Control (2015), http://dx.doi

.3. Modeling

For the modeling reported herein, we consider only traceetals that are listed in the EPA primary drinking water regula-

able 2ineral reactions considered in simulations of batch experiments, column experiments a

Kinetic reaction Equilibrium coefficient at 25 ◦C, log

Calcite + H+ = HCO3− + Ca+2 1.847

Dolomite + 2H+ = 2HCO3− + Ca+2 + Mg+2 3.533

Strontianite = CO3− + Sr+2 −9.271

Fig. 1. Energy dispersive spectrum (EDS) showing elemental analysis of pointlabeled spectrum 1 on rock surface shown in background SEM photograph.

tions and were detected during the batch or column experiments(Qafoku et al., 2013). To place the experimental results in context,we will compare the experimental observations with historicalgroundwater data for these trace metals. Cd was not detected in anygroundwater samples nor in the batch and column experiments.Be and Th were detected in very few groundwater samples, and Beand Th were not measured in the experiments. Although Sb wasdetected in only three groundwater samples, it was detected in theexperiments. As, Ba, Cr, Cu, Pb, and Se were detected in a signifi-cant number of groundwater samples, and were also measured inthe batch and column experiments.

Results from X-ray diffraction (XRD) and scanning electronmicroscope (SEM) analyses (Fig. 1) indicate that the limestoneconsists of 100% calcite (Qafoku et al., 2013; Wang et al., 2015).Previous publications indicate that the predominant mineral in thislimestone is calcite with a small amount of dolomite (Maclay andSmall, 1994), and a minor amount of the mineral strontianite wasincluded in solid solution with calcite (Tesoriero and Pankow, 1996)(Table 2).

Adsorption/desorption reactions for As, Ba, Cu, Cr, Cd, Pb, Se,and Sb on calcite were also included in the reaction network forthe Edwards Aquifer material. A surface complexation model forcalcite was first presented by van Cappellen et al. (1993) and fur-ther developed for divalent cations (Pokrovsky and Schott, 2002),and then extended to arsenate (Sø et al., 2008) and phosphate (Søet al., 2011). The model had two types of surface sites, >Ca+ and>CO3

−. Cations were assumed to sorb to the >CO3− sites, and anions

to the >Ca+ sites. However, these equilibrium sorption models didnot fit the column experiment results. The trace metal release wasclearly rate-controlled in our experiments, so it was modeled using

impact of carbon dioxide leakage into an unconfined, oxidizing.org/10.1016/j.ijggc.2015.04.008

a kinetic adsorption isotherm (Tebes-Stevens et al., 1998):

∂C∗

∂t= kf

(C − C∗

Kd

)(1)

nd groundwater aquifer.

Forward rate at 25 ◦C, mol/m2/s Activation energy, kJ/mol

1.5 × 10−6 23.52.9 × 10−8 52.2Same as calcite Same as calcite

ARTICLE IN PRESSG ModelIJGGC-1491; No. of Pages 10

4 D.H. Bacon et al. / International Journal of Greenhouse Gas Control xxx (2015) xxx–xxx

Table 3Aqueous complexation reactions considered in simulations.

Equilibrium Reaction Equilibrium Coefficient at 25 ◦C, log

HCO3− + H+ = CO2 + H2O 6.3447

HCO3− + Ca+2 = CaHCO3

+ 1.0467HPO4

−2 + Ca+2 = CaHPO4 2.7400HPO4

−2 + Ca+2 = CaPO4− + H+ −5.8618

SO4−2 + Ca+2 = CaSO4 2.1111

Cl− + Cd+2 = CdCl+ 2.7059HCO3

− + Cd+2 = CdHCO3+ 1.5000

HPO4−2 + H+ = H2PO4

− 7.2054Mg+2 + HCO3

− = MgHCO3+ 1.0357

Mg+2 + HPO4−2 = MgHPO4 2.9100

SO4−2 + Mg+2 = MgSO4 2.4117

H2O = OH− + H+ −13.9951Na+ + HCO3

− = NaHCO3 0.1541Pb+2 + HCO3

− = PbCO3 + H+ −3.7488Pb+2 + H2O = PbOH+ + H+ −7.6951Sr+2 + SO4

−2 = SrSO4 2.30002H2O + Cr+3 = Cr(OH)2

+ + 2H+ −9.70003H2O + Cr+3 = Cr(OH)3 + 3H+ −18.0000H2O + Cr+2 = CrOH+2 + H+ −4.0000HCO − + Cu+2 = CuCO + H+ −3.3735

wktaip

PderEurmt

dtbmc

fl

Table 4Ranges of statistical hydraulic parameters assumed for the Edwards Aquifer.

Parameter Minimum Maximum Unit

Permeability variance 0.017 0.79 km2

Horizontal correlation length 1 3.95 kmAnisotropy 1.1 49.1 –

3 3

H2O + Cu+2 = CuOH+ + H+ −7.2875SeO3

−2 + H+ = HSeO3− 7.2861

here C* = sorbed concentration, C = aqueous concentration,f = forward rate constant, kd = distribution coefficient (L/g). Allrace metals were modeled using a forward rate of 1 mol/daynd an equilibrium coefficient of 1.0 × 10−3 L/g, but with differentnitial adsorbed concentrations. Fitted values for the initial C* areresented in Section 3.

Modeling of the batch experiments was conducted usingHREEQC (Parkhurst and Appelo, 1999) and the thermo.com.V8.R6atabase (Wolery and Jarek, 2003). Modeling of the columnxperiments was conducted using the multiphase flow andeactive-transport solver STOMP-CO2-R (White et al., 2012) withCKEChem (White and McGrail, 2005). The STOMP simulationssed a smaller subset of equilibrium aqueous complexationeactions extracted from PHREEQC simulations of the batch experi-

ents and groundwater samples (Musgrove et al., 2010) taken fromhe unconfined/urban portion of the Edwards Aquifer (Table 3).

Modeling of CO2 leakage into the Edwards Aquifer was also con-ucted using the same reaction network. To account for the fact thathe aquifer rock in situ is not fresh ground material, as used in theatch and column experiments, the initial amount of adsorbed trace

Please cite this article in press as: Bacon, D.H., et al., Modeling the

carbonate aquifer. Int. J. Greenhouse Gas Control (2015), http://dx.doi

etals was assumed to be in equilibrium with median groundwateroncentrations (Table 1).

A three-dimensional, heterogeneous model of groundwaterow and reactive transport in the Edwards Aquifer was developed

Fig. 2. Three-dimensional model grid (a) and heterogeneous permeability fie

Mean horizontal permeability −13.5 −10.6 Log10 m2

Mean porosity 0.05 0.34 –

to determine the impact of a hypothetical CO2 leak on groundwaterquality. The model is a generalization of the shallow, unconfinedportion of the aquifer near San Antonio, Texas. The aquifer isassumed to be 150 m thick, and we focus on an 8 km × 5 km area(Fig. 2). The grid is refined at the assumed leak point at X = 6500 m,Y = 7000 m. There are 38 grid cells in the X direction, 35 cells in theY direction, and 10 cells in the Z direction.

Aquifer heterogeneity in the Edwards Aquifer is controlledby large and unpredictable variations in karst features (Lindgrenet al., 2005). The geostatistical model for the Edwards Aquifer wasdeveloped by Dai et al. (2014a). They assumed random Gaussianvariations in permeability, using mean, variance, and horizontalcorrelation length determined for this aquifer by Painter et al.(2007) and Lindgren et al. (2005). Since there is not enough datato define the vertical correlation length, we assume it is 100 timesless than the horizontal one.

Stochastic fields of heterogeneous permeability were generatedusing the pilot point method and random Gaussian interpolation(Deutsch and Journel, 1998; Harp et al., 2008). All nodes wereassumed to have anisotropic intrinsic permeability. The anisotropyfactor (Kxy/Kz) varies between 1.1 and 49.1. Porosity was allowedto vary spatially along with permeability:

k = a�b (2)

where, k [m2] is permeability in horizontal direction, � is poros-ity, and a and b are coefficients (a = 4.84 × 10−10 and b = 3) (Bernabeet al., 2003; Deng et al., 2012). By using both of the permeability andporosity data obtained from Painter et al. (2007) and Lindgren et al.(2005), we tested the empirical relationship in Eq. (2) and the com-puted porosity (from known permeability) is within the lower andupper bounds as listed in Table 4. Permeability in x and y directionswas assumed to be the same, while a vertical anisotropy factor wasassigned for the vertical permeability at each node. Once the hor-

impact of carbon dioxide leakage into an unconfined, oxidizing.org/10.1016/j.ijggc.2015.04.008

izontal permeability was estimated, the vertical permeability wascalculated based on the anisotropy factor (Dai et al., 2014b; Yanget al., 2014). Uncertainty ranges for the statistical parameters arelisted in Table 4.

ld (b) of the shallow/unconfined urban portion of the Edwards Aquifer.

ARTICLE IN PRESSG ModelIJGGC-1491; No. of Pages 10

D.H. Bacon et al. / International Journal of Greenhouse Gas Control xxx (2015) xxx–xxx 5

Table 5Initial aqueous species concentrations.

Species Concentration Unit

pH 6.87HCO3

− 3.29 × 102 mg/LCa+2 1.01 × 102 mg/LCl− 1.22 × 101 mg/LHPO4

−2 1.40 × 10−2 mg/LMg+2 8.10 × 100 mg/LNa+ 6.76 × 100 mg/LSO4

−2 1.15 × 101 mg/LSr+2 1.37 × 10−1 mg/LH2AsO4

− 2.70 × 10−1 �g/LBa+2 3.69 × 101 �g/LCd+2 0.00 × 100 �g/LCr(OH)2

+ 4.40 × 10−1 �g/LCu+2 3.30 × 10−1 �g/L

+2 −2

mngP

(swbwwaaiWabtcsbtfpuwfogR

ss1

3

3

scstvd

100

101

102

103

104

105

106

107

0 50 100 150 200 250 300 350 400

Aqu

eous

Con

cent

ratio

n,

g/µL

Time, h r

Ca (sample 1)Ca (sample 2)Ca (sample 3)

Mg (sample 1)Mg (sample 2)Mg (sample 3)

Sr (sample 1)Sr (sample 3)Ca (modeled)

Mg (modeled)Sr (modeled)

Fig. 3. Major ion model results for batch experiments (rock samples 1–3).

101

102

103

104

0 50 100 150 200 250 300 350 400

Ba

Con

cent

ratio

n,

Time, hr

Sample 1Sample 2Sample 3

Modeled (Samples 1,2)Modeled (Sample 3)

Aquifer MedianAquifer Maximum

MCL

g/µL

adsorbed concentration of 9.0 × 10−9 mol/g.

-1

100

101

102

103

Con

cent

ratio

n,

Cr (sample 1)Modeled (Sample 1)

Aquifer MedianAquifer Maximum

MCLg/µL

Pb 6.40 × 10 �g/LHSeO3

− 2.60 × 10−1 �g/LSb(OH)3 1.22 × 10−2 �g/L

Volume averaging was used to upscale porosity in the geologicodel to the reactive-transport model grid. The principal compo-

ents of the upscaled permeability tensor were computed as theeometric mean of the Cardwell and Parsons bounds (Cardwell andarsons, 1945; Li et al., 2001).

Groundwater flows into the domain at the north boundaryY = 8 km) at a rate that varies with permeability. A horizontal pres-ure gradient of 8.5 × 10−6 MPa/m in the Y direction was used, andas based on observed heads (Hutchison and Hill, 2011). The south

oundary (Y = 0) is an outflow boundary. Since the regional ground-ater flow direction is from north to south, we assumed that thereas no flow across the east (X = 5 km) and west (X = 0) boundaries

nd they were assigned a no-flow boundary condition. The perme-bility in the bedrock is about 3 orders of magnitude less than thatn the Edwards Aquifer (Painter et al., 2007; Lindgren et al., 2005).

e assumed it was an impermeable bedrock and a no-flow bound-ry was assigned to the bottom boundary, except for the CO2 andrine leaks at X = 6.5 km, Y = 7 km. The top boundary was the waterable, which was a no-flow boundary for the liquid phase, but aonstant-pressure boundary of 0.101325 MPa (atmospheric pres-ure) for the gas phase. The species concentrations on the inflowoundary were the same as the initial concentrations (Table 5). Ini-ial conditions were based on the median aqueous concentrationsor 90 groundwater samples from the shallow/urban unconfinedortion of the Edwards aquifer (Musgrove et al., 2010). Median val-es for Cd, Pb and Sb were below detection limit, so average valuesere calculated by Last et al. (2013) based on an interwell approach

or determining background groundwater concentrations as rec-mmended in the U.S. Environmental Protection Agency’s unifieduidance for statistical analysis of groundwater monitoring data atCRA Facilities (U.S. Environmental Protection Agency, 2009).

A mass source of CO2 at the bottom of the aquifer was used toimulate a leak from an abandoned wellbore into the aquifer. Sevenimulations were run with leak rates ranging from 1 × 10−7 kg/s to

× 10−1 kg/s. Aquifer simulations were run for 50 years.

. Results

.1. Batch experiments

The batch experiments were modeled by including kinetic dis-olution of calcite, dolomite, and strontianite in solid solution withalcite, and kinetic desorption of Ba and Cr. Assuming an ideal solid

Please cite this article in press as: Bacon, D.H., et al., Modeling the

carbonate aquifer. Int. J. Greenhouse Gas Control (2015), http://dx.doi

olution, the best fit to the data is obtained assuming volume frac-ions of 0.01 for dolomite and 0.001 for strontianite (Fig. 3). Theseolume fractions correspond to a mole fraction of 7.46 × 10−3 forolomite and 1.26 × 10−3 for strontianite. These mole fractions are

Fig. 4. Ba model results for batch experiments (rock samples 1–3).

too low to measure via X-ray diffraction (XRD), and this explainswhy dolomite and strontianite were not detected in the samples.Ba was the only trace metal of regulatory concern detected in allthree batch experiments. Some variability is observed in the Ba con-centrations (Fig. 4). The Ba sorption was modeled using a forwardrate of 1 mol/day and equilibrium coefficients of 1.0 × 10−3 L/g, butwith different initial adsorbed concentrations. The initial amountof sorbed Ba2+ was 1.3 × 10−6 mol/g for samples 1 and 2, and4.0 × 10−6 mol/g for sample 3. Cr was detected in sample 1 only(Fig. 5). Cr sorption was modeled using a forward rate of 1 mol/dayand equilibrium coefficient of 1.0 × 10−3 L/g, and with an initial

impact of carbon dioxide leakage into an unconfined, oxidizing.org/10.1016/j.ijggc.2015.04.008

10 0 50 100 150 200 250 300 350 400

Time, hr

Fig. 5. Cr ion model results for batch experiments (rock sample 1).

ARTICLE IN PRESSG ModelIJGGC-1491; No. of Pages 10

6 D.H. Bacon et al. / International Journal of Greenhouse Gas Control xxx (2015) xxx–xxx

101

102

103

104

105

106

0 100 200 300 400 500 600 6

7

8

9

10

11

Con

cent

ratio

n, u

g/L

pH

Modeled pHModeled CO2

Modeled CaModeled Mg

Modeled SrpH (Sample 1)Ca (Sample 1)

Mg (Sample 1)Sr (Sample 1)

pH (Sample 3)Ca (Sample 3)

Mg (Sample 3)Sr (Sample 3)

3

wistafooMtdicawe

owc1BmmfTtrdmr

c1cm

TI

10-2

10-1

100

101

102

0 100 200 300 400 500 600

As

Con

cent

ratio

n,

Time, h r

ModeledSample 1Sample 3

Aquifer MedianAquifer Maximum

MCL

g/µL

Fig. 7. As modeling results compared to column experiment results and aquiferconcentrations.

100

101

102

103

104

0 100 200 300 400 500 600

Ba

Con

cent

ratio

n,

Time, hr

ModeledSample 1Sample 3

Aquifer MedianAquifer Maximum

MCLg/µL

below the aquifer median during the flow periods (Fig. 8). Cr con-centrations spike above the aquifer maximum during the stop-flowperiods, but fall below the aquifer median during the flow periods(Fig. 9). Cu concentrations remain between the aquifer median and

103ModeledSample 1

Time, h r

Fig. 6. Major ion modeling results compared to column experiment results.

.2. Column experiments

The column experiments utilizing unweathered rock samplesere modeled by including calcite, and dolomite and strontian-

te in solid solution with calcite. The proportions of dolomite andtrontianite were the same as calibrated during modeling ofhe batch experiments. Trace metal concentrations were fit bydjusting the forward reaction rate and equilibrium coefficientor the adsorption reactions. For sample 1, the concentrationf dissolved CO2 in the inlet water was adjusted to match thebserved pH (Fig 6). Good agreement between observed Ca andg concentrations with time is also seen (Fig 6), indicating that

hese cations are controlled by kinetic dissolution of calcite andolomite. Strontium increases to a relatively stable concentration

ndicating kinetic mineral dissolution, whereas the trace metaloncentrations decrease steadily with time, indicating kineticdsorption/desorption. The mole fraction of Sr in solid solutionith calcite is estimated to be 9.68 × 10−4, somewhat lower than

stimated for the batch experiments.The trace metals were all modeled using a forward rate (kf)

f 1 mol day−1 and distribution coefficient (Kd) of 1 × 10−3 L/g, butith different initial adsorbed concentrations. The initial adsorbed

oncentrations, C*, of As, Ba, Cu, Cr, Pb, Se, and Sb ranged from × 10-8 to 1 × 10−6 mol/g (Table 6). The initial adsorbed value fora, 1 × 10−6 mol/g, agrees well with the value fit to the batch experi-ents: 1.3 × 10−6–4.0 × 10−6 mol/g. The other trace metals all haveuch lower initial adsorbed concentrations ranging from 1 × 10−8

or As and Pb to 6 × 10−8 for Sb and 1 × 10−7 for Cu, Cr, and Se.his is consistent with the other trace metals not being detected inhe batch experiments. It should be noted that the solid-to-solutionatio is considerably higher in the column experiments, and that theifferences in concentrations between batch and column experi-ents are more pronounced for the elements that undergo sorption

eactions.Figs. 7–13 show the model predictions for trace metal con-

Please cite this article in press as: Bacon, D.H., et al., Modeling the

carbonate aquifer. Int. J. Greenhouse Gas Control (2015), http://dx.doi

entrations compared to column experimental results for samples and 3, and are compared to aquifer median and maximum con-entrations, as well as the EPA MCL values. The release of the traceetals with time is very different in shape than that of the major

able 6nitial adsorbed trace metal concentrations, C*, fit to column experiments.

Trace metal Initial adsorbed concentration, mol/g

As 1 × 10−8

Ba 1 × 10−6

Cu 1 × 10−7

Cr 1 × 10−7

Pb 1 × 10−8

Se 1 × 10−7

Sb 6 × 10−8

Fig. 8. Ba modeling results compared to experimental results and aquifer concen-trations.

ions. Whereas the major ions increase to a stable value with timeand show small spikes during the stop-flow events, the trace metalconcentrations generally decrease with time and show relativelylarge spikes during the stop-flow events. As concentrations duringthe three flow periods decrease with time, spike above the MCLduring the first stop-flow period, and reach the MCL at the startof the second stop-flow period (Fig. 7). Ba concentrations spikeabove the aquifer maximum during the stop-flow periods, but fall

impact of carbon dioxide leakage into an unconfined, oxidizing.org/10.1016/j.ijggc.2015.04.008

10-1

100

101

102

0 100 200 300 400 500 600

Cr

Con

cent

ratio

n,

Time, hr

Sample 3Aquifer Median

Aquifer MaximumMCL

g/µL

Fig. 9. Cr modeling results compared to experimental results and aquifer concen-trations.

ARTICLE IN PRESSG ModelIJGGC-1491; No. of Pages 10

D.H. Bacon et al. / International Journal of Greenhouse Gas Control xxx (2015) xxx–xxx 7

10-3

10-2

10-1

100

101

102

103

104

0 100 200 300 400 500 600

Cu

Con

cent

ratio

n,

Time, hr

ModeledSample 1Sample 3

Aquifer MedianAquifer Maximum

MCL

g/µL

Fig. 10. Cu modeling results compared to experimental results and aquifer concen-trations.

10-2

10-1

100

101

102

103

0 100 200 300 400 500 600

Pb C

once

ntra

tion,

Time, hr

ModeledSample 1Sample 3

Aquifer MaximumMCLg/µ

L

Ft

mtstta(tf

mC

Ft

10-3

10-2

10-1

100

101

102

103

0 100 200 300 400 500 600

Se C

once

ntra

tion,

Time, h r

ModeledSample 1Sample 3

Aquifer MedianAquifer Maximum

MCL

g/µL

Fig. 13. Se modeling results compared to experimental results and aquifer concen-trations.

Table 7Summary of column experiment trace metal concentrations at the end of eachexperimental stage, relative to aquifer maximum concentration and MCL regulatorylimits.

Experiment stage As Ba Cr Cu Pb Sb Se

Flow 1 1.00 19.51 0.59 1.52 1.32 0.69 3.01Stop flow 1 44.68 202.70 2.03 58.15 817.10 17.78 149.90Flow 2 0.46 10.78 0.58 1.80 0.18 0.17 1.07Stop flow 2 10.08 88.24 18.11 10.89 44.97 8.64 34.53Flow 3 0.75 5.66 0.46 0.66 0.15 0.21 2.56

Threshold As Ba Cr Cu Pb Sb Se

ig. 11. Pb modeling results compared to experimental results and aquifer concen-rations. Aquifer median concentrations are below detection.

aximum for the duration of the experiment (Fig. 10). Pb concen-rations spike above the MCL threshold initially and during bothtop-flow events, finally falling beneath the aquifer maximum nearhe end of the experiment (Fig. 11). Sb concentrations spike abovehe MCL thresholds during the two stop-flow events, and remainbove the aquifer maximum for the duration of the experimentFig. 12). Se concentrations spike above the MCL threshold duringhe two stop-flow events and average close to the aquifer maximumor the duration of the experiment (Fig. 13).

Please cite this article in press as: Bacon, D.H., et al., Modeling the

carbonate aquifer. Int. J. Greenhouse Gas Control (2015), http://dx.doi

The outcomes of the column experiments relative to the aquiferaximum concentrations and MCLs are summarized in Table 7.

oncentrations are generally higher during the stop-flow periods.

10-2

10-1

100

101

102

0 100 200 300 400 500 600

Sb C

once

ntra

tion,

Time, hr

ModeledSample 1Sample 3

Aquifer MaximumMCLg/µ

L

ig. 12. Sb modeling results compared to experimental results and aquifer concen-rations. Aquifer median concentrations are below detection.

Aquifer max mg/L 1.11 69.90 5.57 57.30 0.15 0.06 1.40MCL (mg/L) 10 2000 100 1300 15 6 50

At the end of the first stop-flow period, all of the trace metals exceptfor Cr exceed the maximum aquifer-observed values, and As, Pb, Sb,and Se exceed the MCL limits. However, concentrations of the tracemetals generally decrease during the flow periods, and at the end ofthe second stop-flow period Ba and Cr exceed the observed aquifermaximums and only As, Pb and Sb exceed the MCL. Concentrationsare generally lower during the flow periods. By the last flow period,only Sb and Se exceed the aquifer maximum threshold and no tracemetals exceed their respective MCLs.

3.3. Aquifer simulations

A three-dimensional multiphase flow and reactive-transportsimulation of CO2 leakage from an abandoned wellbore into a gen-eralized model of the shallow, unconfined portion of the EdwardsAquifer is used to determine potential impacts on groundwaterquality. As a measure of adverse impacts on groundwater quality,we use both the EPA’s regulatory limits and the maximum tracemetal concentration observed in the aquifer. CO2 leakage froman abandoned wellbore was modeled as a point source. A rangeof leakage rates from 1 × 10−7 kg/s to 1 × 10−1 kg/s were imposedfor a 50-year period. The volume of aquifer with concentrationsbeyond the MCL or aquifer maximum (or minimum in the caseof pH) were calculated for pH, TDS, and each of the trace metals(Fig. 14). As CO2 leaks into the aquifer, the pH decreases to 4.86due to dissolution and dissociation of CO2 in the groundwater. ThepH does not fall below the aquifer minimum for the slowest leakrate of 1 × 10−7 kg/s. Plume sizes for the MCL and aquifer minimumthreshold are relatively similar. CO2 is a gas at the aquifer temper-

impact of carbon dioxide leakage into an unconfined, oxidizing.org/10.1016/j.ijggc.2015.04.008

atures and pressures, so it rises buoyantly toward the water tablerather than spreading horizontally (Fig. 15). The TDS plumes aresimilar in size to those for pH. Increases in TDS are due to calcitedissolution. In none of the cases do the trace metals exceed the

ARTICLE ING ModelIJGGC-1491; No. of Pages 10

8 D.H. Bacon et al. / International Journal of Gree

102

103

104

105

106

107

0 10 20 30 40 50Vol

ume

pH <

MC

L o

r A

quif

er M

inim

um, m

3

Time, yr

CO2 Rate, kg/s

Min 1.e-1Min 1.e-2Min 1.e-3Min 1.e-4Min 1.e-5Min 1.e-6Min 1.e-7

MCL 1.e-1MCL 1.e-2MCL 1.e-3MCL 1.e-4MCL 1.e-5MCL 1.e-6MCL 1.e-7

Fig. 14. Volume of aquifer where pH is less than MCL or observed minimum valuein aquifer, as a function of CO2 leak rate.

Fig. 15. Extent of pH below the observed aquifer minimum value.

Foe

aii

4

aueoe

ig. 16. Pb concentrations in the aquifer after 50 years of CO2 leakage. Isosurfacesf 3.8 × 10−5, 6.4 × 10−5 and 8.9 × 10−5 mg/L are delineated. Concentrations do notxceed observed maximum values of 1.5 × 10−4 mg/L.

quifer maximum or MCL values. For instance, Pb values increasen the vicinity of the leak, but do not exceed the natural variationsn Pb values (Fig. 16).

. Discussion

Modeling of batch and column experiments was conducted inn effort to identify the mechanisms controlling trace metal release

Please cite this article in press as: Bacon, D.H., et al., Modeling the

carbonate aquifer. Int. J. Greenhouse Gas Control (2015), http://dx.doi

nder elevated CO2 conditions from carbonate aquifer rock. Mod-ling of the batch experiments was not as conclusive as modelingf the column experiments. The only trace metals seen in the batchxperiments are Ba and Cr. The water/rock ratio was higher for

PRESSnhouse Gas Control xxx (2015) xxx–xxx

the batch experiments (5 mL/g) than for the column experiments(0.123 mL/g) (Qafoku et al., 2013). The solid-to-solution ratio inbatch tests is an important variable because some elements occurat low concentrations that can only be detected when the tests areconducted at a low water/rock ratio (Qafoku et al., 2013).

Our use of the stop-flow column technique to address the issueof leaking CO2 on groundwater quality allows identification ofthe mechanisms controlling trace metal release from these lime-stone samples. A kinetic desorption model fits the overall trends inrelease for seven trace metals observed in batch and column experi-ments exposing Edwards Aquifer rock to elevated concentrations ofCO2. In the column experiments, the concentrations of As, Ba, Cu, Cr,Pb, Se, and Sb decrease with time after exposure to CO2. This indi-cates that release is controlled by kinetic desorption, rather thanmineral dissolution. In the column experiments, concentrations ofCa, Mg, and Sr reach equilibrium with time, indicating release iscontrolled by solid solution dissolution of calcite. The trace met-als show large spikes during the stop-flow events, likely becausethey are released by faster kinetic desorption. The stop-flow eventsincrease the residence time and allow concentrations to increase.The major cations are released by relatively slower dissolution ofcalcite and so show less-significant increases in concentration.

As presented in Section 1 of this paper, there is very little pre-vious work on the effect of CO2 leaks on carbonate rocks; mostprevious experimental and modeling studies are related to sand-stones and so the results are not directly comparable. Yang et al.(2014) found that Sr was controlled by solid solution dissolution ofdolomite for carbonate-rich formations. However, they predictedthat Ba was also controlled by solid solution release, whereas ourresults indicate kinetic desorption. Further, they concluded thatAs release was controlled by desorption from illite and kaolinite,rather than from calcite as was found in our study.

Adsorption site densities estimated in other studies for arse-nate on clays such as illite, 1.16 × 10−6 mol/g, and kaolinite,2.44 × 10−6 mol/g (Manning and Goldberg, 1996), are comparableto adsorption site densities estimated in other studies for arsenateon calcite, 7.04 × 10−7–2.11 × 10−6 mol/g (Sø et al., 2008), indicat-ing that adsorption of arsenate on calcite can be significant even ifclays are present. The adsorption site densities calculated by (Søet al., 2008) are higher than the initial adsorbed concentrationsestimated for our column experiments, 1 × 10−8 mol/g. This maybe due to competitive sorption with other oxyanion-forming met-als, Se, Sb, and Cr, because the total initial adsorbed concentrationsof As, Se, Sb, and Cr are higher, 2.7 × 10−7 mol/g (Table 6).

Although the model fits the overall trend of trace metal release,the peak concentrations of some species, particularly Pb, are notfit particularly well. This is the best fit that can be achieved witha single-rate model, indicating that perhaps a multi-rate surfacecomplexation model, such as that developed for uranium release atthe Hanford Site (Liu et al., 2013, 2008; Qafoku et al., 2005) mightprovide a better fit. More work is needed to extend existing equi-librium surface complexation models for trace metals on calcite toa kinetic multi-rate model.

Using the models fit to experimental data, a three-dimensionalmultiphase flow and reactive-transport simulation of CO2 leak-age from an abandoned wellbore into a generalized model of theshallow, unconfined portion of the Edwards Aquifer is used todetermine the risk of potential impacts on groundwater quality. Asa measure of adverse impacts on groundwater quality, we use boththe EPA’s regulatory limits and the maximum trace metal concen-tration observed in the aquifer to delineate the volume of aquiferabove (or in the case of pH, below) the defined limit. Results of

impact of carbon dioxide leakage into an unconfined, oxidizing.org/10.1016/j.ijggc.2015.04.008

these field-scale simulations indicate that CO2 leakage into a car-bonate aquifer is likely to cause decreases in pH and increases inTDS beyond observed ranges in the aquifer and beyond regulatorylimits. However, trace metal concentrations are not predicted to

ING ModelI

f Gree

eif

litubwt

aafimcwwbhalbeta

A

hPoiueMfRw

R

U

A

B

B

C

C

D

D

D

ARTICLEJGGC-1491; No. of Pages 10

D.H. Bacon et al. / International Journal o

xceed either the observed maximums or the MCL regulatory lim-ts. This is consistent with reactive-transport modeling of Pb releaserom pure calcite presented by Navarre-Sitchler et al. (2013).

This study does not consider the impact of trace metals mobi-ized by brine or CO2 from the formation where CO2 is initiallynjected (Karamalidis et al., 2013). Carroll et al. (2014) consideredhe impact of leaking CO2 and brine on groundwater aquifers, butsed only simple geochemical models that considered pH bufferingy calcite and treated trace metals as tracers. More experimentalork is needed to determine the capacity of aquifer rock to adsorb

hese additional trace metals.This study enhances understanding of the impact of CO2 leak-

ge on a carbonate aquifer by modeling batch and column testsnd using the calibrated model for trace metal release to develop aeld scale model. Our approach is innovative in that the large scaleodel is build upon the metal releasing mechanism revealed by the

olumn test, which make the model concept more reliable. Finally,e use the field-scale results to predict the volume of aquifer thatill be adversely impacted based on a statistical assessment of

ackground water quality and regulatory thresholds. These resultsave important implications for the safety of geologic carbon stor-ge. Coupled with realistic assessments of the risk of carbon dioxideeakage from the carbon storage reservoir from abandoned well-ores (Pan et al., 2009; Viswanathan et al., 2008; William Careyt al., 2010) these results indicate that the risk of trace metal con-amination due to carbon dioxide leakage alone into a carbonatequifer may be small.

cknowledgements

The U.S. Department of Energy’s (DOE’s) Office of Fossil Energyas established the National Risk Assessment Partnership (NRAP)roject. The research presented in this report was completed as partf the groundwater protection task of the NRAP Project. NRAP fund-

ng was provided to Pacific Northwest National Laboratory (PNNL)nder DOE contract number DE-AC05-76RL01830. A portion of thexperimental research was performed using the Environmentalolecular Sciences Laboratory (EMSL), a national scientific user

acility sponsored by DOE’s Office of Biological and Environmentalesearch and located at PNNL. A portion of the modeling researchas performed using PNNL Institutional Computing facilities.

eferences

SGS National Water-Quality Assessment (NAWQA) Program, 2014. U.S.Geological Survey.

ltevogt, A.S., Jaffe, P.R., 2005. Modeling the effects of gas phase CO2 intrusion onthe biogeochemistry of variably saturated soils. Water Resour. Res. 41,W09426, http://dx.doi.org/10.1029/2004wr003819

acon, D.H., Dai, Z., Zheng, L., 2014. Geochemical impacts of carbon dioxide, brine,trace metal and organic leakage into an unconfined, oxidizing limestoneaquifer. Energy Procedia 63, 4684–4707 http://dx.doi.org/10.1016/j.egypro.2014.11.502

ernabe, Y., Mok, U., Evans, B., 2003. Permeability-porosity relationships in rockssubjected to various evolution processes. Pure Appl. Geophys. 160, 937–960,http://dx.doi.org/10.1007/Pl00012574

ardwell, W.T., Parsons, R.L., 1945. Average permeabilities of heterogeneous oilsands. T. Am. I. Min. Met. Eng. 160, 34–42.

arroll, S.A., Keating, E., Mansoor, K., Dai, Z., Sun, Y., Trainor-Guitton, W., Brown, C.,Bacon, D., 2014. Key factors for determining groundwater impacts due toleakage from geologic carbon sequestration reservoirs. Int. J. Greenhouse GasControl 29, 153–168, http://dx.doi.org/10.1016/j.ijggc.2014.07.007

ai, Z., Keating, E., Bacon, D., Viswanathan, H., Stauffer, P., Jordan, A., Pawar, R.,2014a. Probabilistic evaluation of shallow groundwater resources at ahypothetical carbon sequestration site. Sci. Rep. 4, 4006, http://dx.doi.org/10.1038/srep04006

ai, Z., Middleton, R., Viswanathan, H., Fessenden-Rahn, J., Bauman, J., Pawar, R.,

Please cite this article in press as: Bacon, D.H., et al., Modeling the

carbonate aquifer. Int. J. Greenhouse Gas Control (2015), http://dx.doi

Lee, S.-Y., McPherson, B., 2014b. An integrated framework for optimizing CO2

sequestration and enhanced oil recovery. Environ. Sci. Technol. Lett. 1, 49–54,http://dx.doi.org/10.1021/ez4001033

eng, H., Stauffer, P.H., Dai, Z., Jiao, Z., Surdam, R.C., 2012. Simulation ofindustrial-scale CO2 storage: multi-scale heterogeneity and its impacts on

PRESSnhouse Gas Control xxx (2015) xxx–xxx 9

storage capacity, injectivity and leakage. Int. J. Greenhouse Gas Control 10,397–418, http://dx.doi.org/10.1016/j.ijggc.2012.07.003

Deutsch, C.V., Journel, A.G., 1998. Gslib Geostatistical Software Library and User’sGuide, 2nd ed. Oxford University Press, New York.

Harp, D.R., Dai, Z.X., Wolfsberg, A.V., Vrugt, J.A., Robinson, B.A., Vesselinov, V.V.,2008. Aquifer structure identification using stochastic inversion. Geophys. Res.Lett. 35, L08404, http://dx.doi.org/10.1029/2008gl033585

Harvey, O.R., Qafoku, N.P., Cantrell, K.J., Lee, G., Amonette, J.E., Brown, C.F., 2013.Geochemical implications of gas leakage associated with geologic CO2

storage-a qualitative review. Environ. Sci. Technol. 47, 23–36, http://dx.doi.org/10.1021/Es3029457

Hutchison, W.R., Hill, M.E., 2011. Recalibration of the Edwards (Balcones FaultZone) Aquifer – Barton Springs Segment – Groundwater Flow Model, Austin,Texas. <http://www.twdb.state.tx.us/groundwater/models/alt/ebfz b/EBFZ BModel Recalibration Report.pdf>.

Jacquemet, N., Picot-Colbeaux, G., Vong, C.Q., Lions, J., Bouc, O., Jeremy, R., 2011.Intrusion of CO2 and impurities in a freshwater aquifer – impact evaluation byreactive transport modelling. Energy Procedia 4, 3202–3209, http://dx.doi.org/10.1016/JEgypro.2011.02.236

Karamalidis, A.K., Torres, S.G., Hakala, J.A., Shao, H.B., Cantrell, K.J., Carroll, S., 2013.Trace metal source terms in carbon sequestration environments. Environ. Sci.Technol. 47, 322–329, http://dx.doi.org/10.1021/Es304832m

Last, G.V., Murray, C.J., Brown, C.F., Jordan, P.D., Sharma, M., 2013. No-ImpactThreshold Values for NRAP’s Reduced Order Models, PNNL-22077, Richland,Washington, p.72. <http://www.pnnl.gov/main/publications/external/technical reports/PNNL-22077pdf>.

Li, D., Beckner, B., Kumar, A., 2001. A new efficient averaging technique for scaleupof multimillion-cell geologic models. SPE Reservoir Eval. Eng. 4,297–307.

Lindgren, R.J., Dutton, A.R., Hovorka, S.D., Worthington, S.R.H., Painter, S., 2005.Conceptualization and simulation of the Edwards Aquifer, San Antonio region,Texas, SIR 2004-5277. Survey, U.S.G., Reston, Virginia. <http://pubs.water.usgs.gov/sir20045277/>.

Little, M.G., Jackson, R.B., 2010. Potential impacts of leakage from deep CO2

geosequestration on overlying freshwater aquifers. Environ. Sci. Technol. 44,9225–9232, http://dx.doi.org/10.1021/Es102235w

Liu, C., Shang, J., Kerisit, S., Zachara, J.M., Zhu, W., 2013. Scale-dependent rates ofuranyl surface complexation reaction in sediments. Geochim. Cosmochim.Acta 105, 326–341, http://dx.doi.org/10.1016/j.gca.2012.12.003

Liu, C.X., Zachara, J.M., Qafoku, N.P., Wang, Z.M., 2008. Scale-dependent desorptionof uranium from contaminated subsurface sediments. Water Resour. Res. 44,W08413, http://dx.doi.org/10.1029/2007wr006478

Lu, J.M., Partin, J.W., Hovorka, S.D., Wong, C., 2010. Potential risks to freshwaterresources as a result of leakage from CO2 geological storage: a batch-reactionexperiment. Environ. Earth Sci. 60, 335–348, http://dx.doi.org/10.1007/S12665-009-0382-0

Maclay, R.W., Small, T.A., 1994. Carbonate geology and hydrology of the EdwardsAquifer in the San Antonio area, Texas, OFR 83-537. <http://pubs.er.usgs.gov/thumbnails/usgs thumb.jpg>.

Manning, B.A., Goldberg, S., 1996. Modeling arsenate competitive adsorption onkaolinite, montmorillonite and illite. Clay Clay Miner. 44, 609–623, http://dx.doi.org/10.1346/Ccmn.1996.0440504

Musgrove, M., Fahlquist, L., Houston, N.A., Lindgren, R.J., Ging, P.B., 2010.Geochemical evolution processes and water-quality observations based onresults of the National Water-Quality Assessment Program in the San Antoniosegment of the Edwards aquifer, 1996–2006, Scientific Investigations Report2010-5129, Survey, U.S.G., Reston, Virginia. <http://pubs.usgs.gov/sir/2010/5129/>.

Navarre-Sitchler, A.K., Maxwell, R.M., Siirila, E.R., Hammond, G.E., Lichtner, P.C.,2013. Elucidating geochemical response of shallow heterogeneous aquifers toCO2 leakage using high-performance computing: implications for monitoringof CO2 sequestration. Adv. Water Resour. 53, 45–55, http://dx.doi.org/10.1016/j.advwatres.2012.10.005

Painter, S.L., Woodbury, A.D., Jiang, Y.F., 2007. Transmissivity estimation for highlyheterogeneous aquifers: comparison of three methods applied to the EdwardsAquifer, Texas, USA. Hydrogeol. J. 15, 315–331, http://dx.doi.org/10.1007/s10040-006-0071-y

Pan, L.H., Oldenburg, C.M., Wu, Y.S., Pruess, K., 2009. Wellbore flow model forcarbon dioxide and brine. Energy Procedia 1 (71–78), 4984 http://dx.doi.org/http://dx.doi.org/10.1016/j.egypro.2009.01.012

Parkhurst, D.L., Appelo, C.A.J., 1999. User’s Guide to PHREEQC (Version 2): AComputer Program for Speciation, Batch-Reaction, One-DimensionalTransport, and Inverse Geochemical Calculations. Water ResourcesInvestigations, Washington, D.C, Report 99-4259 http://wwwbrr.cr.usgs.gov/projects/GWC coupled/phreeqc/html/final.html

Pokrovsky, O.S., Schott, J., 2002. Surface chemistry and dissolution kinetics ofdivalent metal carbonates. Environ. Sci. Technol. 36, 426–432, http://dx.doi.org/10.1021/Es010925u

Qafoku, N.P., Lawter, A.R., Brown, C.F., Harvey, O.R., Wang, G., Bowden, M.E.,Sullivan, E.C., 2013. Geochemical Impacts of Leaking CO2 from SubsurfaceStorage Reservoirs to Unconfined and Confined Aquifers, PNNL-22420,

impact of carbon dioxide leakage into an unconfined, oxidizing.org/10.1016/j.ijggc.2015.04.008

Richland, Washington. <http://www.pnnl.gov/main/publications/external/technical reports/PNNL-22420pdf>.

Qafoku, N.P., Zachara, J.M., Liu, C.X., Gassman, P.L., Qafoku, O.S., Smith, S.C., 2005.Kinetic desorption and sorption of U(VI) during reactive transport in a

ING ModelI

1 f Gree

S

S

T

T

U

U

v

V

V

V

Zheng, L.G., Apps, J.A., Spycher, N., Birkholzer, J.T., Kharaka, Y.K., Thordsen, J., Beers,S.R., Herkelrath, W.N., Kakouros, E., Trautz, R.C., 2012. Geochemical modelingof changes in shallow groundwater chemistry observed during the MSU-ZERTCO2 injection experiment. Int. J. Greenhouse Gas Control 7, 202–217, http://dx.

ARTICLEJGGC-1491; No. of Pages 10

0 D.H. Bacon et al. / International Journal o

contaminated Hanford sediment. Environ. Sci. Technol. 39, 3157–3165, http://dx.doi.org/10.1021/Es048462q

ø, H.U., Postma, D., Jakobsen, R., Larsen, F., 2008. Sorption and desorption ofarsenate and arsenite on calcite. Geochim. Cosmochim. Acta, 5871–5884,http://dx.doi.org/10.1016/j.gca.2008.09.023

ø, H.U., Postma, D., Jakobsen, R., Larsen, F., 2011. Sorption of phosphate ontocalcite: results from batch experiments and surface complexation modeling.Geochim. Cosmochim. Acta 75, 2911–2923, http://dx.doi.org/10.1016/j.gca.2011.02.031

ebes-Stevens, C., Valocchi, A.J., VanBriesen, J.M., Rittmann, B.E., 1998.Multicomponent transport with coupled geochemical and microbiologicalreactions: model description and example simulations. J. Hydrol. 209, 8–26,http://dx.doi.org/10.1016/S0022-1694(98)104-8

esoriero, A.J., Pankow, J.F., 1996. Solid solution partitioning of Sr2+ , Ba2+ , and Cd2+

to calcite. Geochim. Cosmochim. Acta 60, 1053–1063, http://dx.doi.org/10.1016/0016-7037(95)449-1

.S. Environmental Protection Agency, 2012. Draft Underground Injection Control(UIC) Program Class VI Well Testing and Monitoring Guidance, EPA816-D-10-009. <http://water.epa.gov/type/groundwater/uic/class6/upload/GSTestingMonitoring Guidance DRAFT 01262012 508pdf>.

.S. Environmental Protection Agency, 2009. Statistical Analysis of GroundwaterMonitoring Data at RCRA Facilities, Unified Guidance, EPA 530/R-09-007,Washington, D.C. <http://www.epa.gov/epawaste/hazard/correctiveaction/resources/guidance/sitechar/gwstats/index.htm>.

an Cappellen, P., Charlet, L., Stumm, W., Wersin, P., 1993. A surface complexationmodel of the carbonate mineral–aqueous solution interface. Geochim.Cosmochim. Acta 57, 3505–3518, http://dx.doi.org/10.1016/0016-7037(93)90135-J

iswanathan, H., Dai, Z., Lopano, C., Keating, E., Hakala, J.A., Scheckel, K.G., Zheng,L., Guthrie, G.D., Pawar, R., 2012. Developing a robust geochemical and reactivetransport model to evaluate possible sources of arsenic at the CO2

sequestration natural analog site in Chimayo, New Mexico. Int. J. GreenhouseGas Control 10, 199–214, http://dx.doi.org/10.1016/j.ijggc.2012.06.007

iswanathan, H.S., Pawar, R.J., Stauffer, P.H., Kaszuba, J.P., Carey, J.W., Olsen, S.C.,Keating, G.N., Kavetski, D., Guthrie, G.D., 2008. Development of a hybridprocess and system model for the assessment of wellbore leakage at a geologicCO(2) sequestration site. Environ. Sci. Technol. 42, 7280–7286, http://dx.doi.org/10.1021/Es800417x

Please cite this article in press as: Bacon, D.H., et al., Modeling the

carbonate aquifer. Int. J. Greenhouse Gas Control (2015), http://dx.doi

ong, C.Q., Jacquemet, N., Picot-Colbeaux, G., Lions, J., Rohmer, J., Bouc, O., 2011.Reactive transport modeling for impact assessment of a CO2 intrusion on traceelements mobility within fresh groundwater and its natural attenuation forpotential remediation. Energy Procedia 4, 3171–3178, http://dx.doi.org/10.1016/JEgypro.2011.02.232

PRESSnhouse Gas Control xxx (2015) xxx–xxx

Wang, G., Qafoku, N.P., Lawter, A.R., Bowden, M.E., Harvey, C.F., Sullivan, E.C.,Brown, C.F., 2015. Submitted geochemical impacts of leaking CO2 fromsubsurface storage reservoirs to an unconfined aquifer. Int. J. Greenhouse GasControl, in press.

Wang, S., Jaffe, P.R., 2004. Dissolution of a mineral phase in potable aquifers due toCO2 releases from deep formations: effect of dissolution kinetics. EnergyConvers. Manage. 45, 2833–2848, http://dx.doi.org/10.1016/JEnconman.2004.01.002

Wei, Y., Maroto-Valer, M., Steven, M.D., 2011. Environmental consequences ofpotential leaks of CO2 in soil. Energy Procedia 4, 3224–3230, http://dx.doi.org/10.1016/JEgypro.2011.02.239

White, M.D., Bacon, D.H., McGrail, B.P., Watson, D.J., White, S.K., Zhang, Z.F., 2012.STOMP Subsurface Transport Over Multiple Phases: STOMP-CO2 andSTOMP-CO2e Guide: Version 1.0, PNNL-21268. Richland, WA. <http://www.pnnl.gov/main/publications/external/technical reports/PNNL-21268pdf>.

White, M.D., McGrail, B.P., 2005. STOMP, Subsurface Transport Over MultiplePhases, Version 1.0, Addendum: ECKEChem, Equilibrium-Conservation-KineticEquation Chemistry and Reactive Transport, PNN-15482, Richland,Washington. <http://www.pnl.gov/main/publications/external/technicalreports/PNNL-15482pdf>.

William Carey, J., Svec, R., Grigg, R., Zhang, J., Crow, W., 2010. Experimentalinvestigation of wellbore integrity and CO2–brine flow along thecasing–cement microannulus. Int. J. Greenhouse Gas Control 4, 272–282,http://dx.doi.org/10.1016/j.ijggc.2009.09.018

Wolery, T.W., Jarek, R.L., 2003. Software User’s Manual, EQ3/6, Version 8.0,Albuquerque, New Mexico.

Yang, C., Dai, Z., Romanak, K.D., Hovorka, S.D., Trevino, R.H., 2014. Inversemodeling of Water–Rock–CO2 batch experiments: potential impacts ongroundwater resources at carbon sequestration sites. Environ. Sci. Technol. 48,2798–2806, http://dx.doi.org/10.1021/es4041368

Zheng, L., Apps, J.A., Zhang, Y., Xu, T., Birkholzer, J.T., 2009. On mobilization of leadand arsenic in groundwater in response to CO2 leakage from deep geologicalstorage. Chem. Geol. 268, 281–297, http://dx.doi.org/10.1016/j.chemgeo.2009.09.007

impact of carbon dioxide leakage into an unconfined, oxidizing.org/10.1016/j.ijggc.2015.04.008

doi.org/10.1016/j.ijggc.2011.10.003.