Uranyl Adsorption onto HydrousFerric OxidesA Re-Evaluation forthe Diffuse Layer Model DatabaseJ O H N J . M A H O N E Y , * , † S O N Y A A . C A D L E , ‡

A N D R Y A N T . J A K U B O W S K I §

Mahoney Geochemical Consulting LLC, Lakewood, Colorado80226, GeoTrans, Inc., Louisville, Colorado 80027, and MWHAmericas, Inc. Steamboat Springs, Colorado 80487

Received May 30, 2009. Revised manuscript receivedOctober 28, 2009. Accepted October 31, 2009.

The diffuse layer model (DLM) database of Dzombak andMorel was developed to quantify the adsorption of dissolvedspeciesonto thehydrousferricoxide(HFO)surface,andcontainednumerous surface complexation reactions, including surfacecomplexationreactionsforuranyl (UO2

+2)consistingofHfo_sOUO2+

and Hfo_wOUO2+. However, these constants were not based

upon experimentally obtained data, but rather were derived fromlinear free energy relationships (LFER) using log KMOH values.Whencomparedtoexperimentaldata, theLFER-derivedconstantsfor uranyl were shown to overestimate adsorption by afactor of 10 in some cases. At least 14 uranyl HFO data setshave been previously published and were used to re-estimateconstants by coupling the geochemical computer codePHREEQCwithUCODE_2005,anautomatedparameteroptimizationprogram. Five uranyl-bearing surface complexation reactionswere initially evaluated; the constants were optimized by allowingUCODE to incrementally vary selected log Kx

int values untilthe best fit to the experimental data was obtained. Assumptionsconsistent with the original DLM were retained. Changes tothe K1

int and K2int constants, and addition of uranyl monocarbonate

and uranyl dicarbonate surface complexes, will update andcorrect the uranyl sorption reactions in this widely used database.

IntroductionPredicting the fate and mobility of uranium in groundwateris important in the licensing and decommissioning ofuranium mills and associated tailings management facilities.Under oxidizing conditions, uranyl [U(VI)] is the dominantform of uranium, and most uranyl-bearing minerals (suchas schoepite) are relatively soluble (1). Consequently, modelsused to describe the behavior of uranyl in groundwater relypredominantly upon adsorption processes. Thus, accuratesorption constants for the modeling of uranium are criticalto uranium processing facility environmental assessments.

A significant amount of experimental work has beenperformed to quantify and describe the behavior of uranylin these systems (2-9). During the past 25 years, numeroussurface ionization and complexation models, and resultantdatabases, have been developed that describe trace metaladsorption onto surfaces such as hydrous ferric oxide (HFO)

or ferrihydrite, as well as goethite, quartz, and clays. Onedatabase that has achieved a large degree of usage wasdeveloped by Dzombak and Morel (10) (D&M). In theirseminal work, previously published adsorption data sets wereevaluated and surface complexation reactions were definedassuming that a diffuse layer model (DLM) described thecharge potential relationships on the HFO surface. The modeldeveloped from their effort was based upon a series of well-considered parameters, such as surface site densities andsurface areas. The utility of this database is evident in itsinclusion in numerous modeling packages such as PHREEQC,MINTEQA2, and Geochemist’s Workbench (11-13).

The original constants provided with the D&M databasethat quantified the adsorption of the uranyl ion (UO2

+2) ontoHFO were not based upon fitting experimental results, butrather used correlations between the constant for theformation of various metal hydroxides (MeOH+), and surfacecomplexation constants for cation complexation on strong(Hfo_sOH) and weak (Hfo_wOH) surface sites. Using theirlinear free energy relationship (LFER), and a log KMOH of 8.2for the UO2(OH)+ complex, D&M calculated log K values forHfo_sOUO2

+ (log K1int of 5.2) and Hfo_wOUO2

+ (log K2int of

2.8).

Hfo_sOH + UO2+2 ) Hfo_sOUO2

+ + H+ K1int

Hfo_wOH + UO2+2 ) Hfo_wOUO2

+ + H+ K2int

As part of a solute transport model developed for licensinga uranium mill tailings facility, the LFER-derived parameterswere initially compared to two different sets of experimentaldata. Unfortunately, the LFER-derived constants show pooragreement with these two data sets (Figures 1 and 2). TheD&M model parameters tended to overpredict the amountof adsorption under low-pH conditions and under-predictthe amount of adsorption under high-pH conditions. Thefailure of the uranyl surface complexation constants devel-oped by D&M to agree with experimental data may be relatedto the fact that the ions used to develop the LFER correlationwere all simple mononuclear cations whereas uranyl is anoxygen-bearing polynuclear cation. Additionally, the log KMOH

value of 8.2 for the UO2(OH)+ complex, which was obtained

* Corresponding author phone: (303) 986-7643; e-mail:jmahoney@mahoneygeochem.com.

† Mahoney Geochemical Consulting LLC.‡ GeoTrans, Inc.§ MWH Americas, Inc.

FIGURE 1. Comparison of Hsi’s (2) experimentally derivedmeasurements of uranyl uptake onto HFO to DLM calculationsusing LFER-derived constants of D&M uranyl surfacecomplexation reactions (model fits shown by dashed lines);based upon 10-5 m U(VI) sorbed onto 0.15 (triangles) and 1 g/L(lined X’s) of ferrihydrite in a 0.1 NaNO3 solution with nocarbonate.

Environ. Sci. Technol. 43, 9260–9266

9260 9 ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 43, NO. 24, 2009 10.1021/es901586w 2009 American Chemical SocietyPublished on Web 11/20/2009

from Baes and Mesmer (14), is no longer the generallyaccepted value for this reaction. The revised value, followingthe convention used by D&M, is now 8.8 (1, 15). Using theupdated constant would increase the log K1

int to 5.9 and logK2

int to 3.5. These values would further increase the extentof uranyl sorption and move the adsorption edge to evenlower pH values. This suggests that for these adsorptionexperiments, the polynuclear uranyl ion achieves a greaterrelative stability in solution when compared to the simplermononuclear cations. Reduced adsorption relative to themononuclear cations may also be caused by a steric effect.Finally, the original DLM database did not include surfacecomplexation constants for uranyl carbonate surface com-plexes. Much work has demonstrated the importance of theseconstants when uranyl adsorption is evaluated in carbonate-bearing waters (1-9).

Although other models have been developed to describeadsorption of uranyl onto HFO, their assumptions sum-marized in the Supporting Information (SI) are not consistentwith assumptions in the original D&M model. For example,Hsi and Langmuir (2, 3) used a triple layer model (TLM), butbecause the underlying assumptions between various modelsare so different the TLM parameters cannot be easilyconverted to the DLM (16). Some models claim to fit withinthe DLM framework, but do not provide a fully defensibleconnection with the rest of the DLM database. Workperformed primarily by Waite and Payne (4-6) and Morrison,Spangler, and Tripathi (8) used different site densities thanthose cited in the original DLM. Morrison, Spangler, andTripathi also eliminated the electrostatic terms.

While the importance of individual attempts to produceupdated constants should not be discounted, none of themreadily fit into the internally consistent D&M database thathas proven to be a valuable resource in surface complexationand solute transport modeling. The primary purpose of thiswork is to provide corrections to the DLM database for uranyladsorption, while maintaining the basic tenets such as surfacesite densities with those presented by D&M in 1990 (10).Also, we have maintained the original conventions relatedto strong and weak site definitions, mainly that high affinity(strong sites) model only cation sorption. Work using

extended X-ray adsorption fine structure (EXAFS) (6) suggeststhat at least some uranyl surface complexes are bidendate,forming bonds with two oxygens within an iron centeredoctahedron. Several geochemical modeling codes, includingPHREEQC (11), can accommodate bidendate complexeswithin the DLM framework, but it requires some exceptionsto the charge and elemental balance checking routines tofunction properly; specifically the no_check option must beincluded to define bidendate type surface complexes inPHREEQC. Because this effort was primarily designed to fixissues with the DLM database we opted to keep the definitionsof surface complexes as simple as possible and to keep thenumber of surface complexes to a minimum. For consistencywith the original database compilation we assumed that thesurface complexes are monodendate. The uranyl surfacecomplexes defined in this work are similar to the three surfacecomplexes defined by Wazne et al. (7), which were selectedbased upon a combination of zeta potential and Fouriertransform infrared spectroscopic measurements.

Materials and MethodsAn evaluation of the published uranyl complexation reactionsin solution was undertaken as the first step in this study.Various database compilations were examined and it wasdecided to use the WATEQ4F database as our primary sourceof thermodynamic data for this work. The evaluationindicated that the WATEQ4F.dat thermodynamic database,included in the PHREEQC distribution package, was con-sistent with the generally accepted dominant aqueouscomplexes and their stability constants listed in the NuclearEnergy Agency (NEA) database (15); despite the generalagreement, some minor differences in the stability constantswere noted (see SI). The primary reason the WATEQ4Fthermodynamic database was selected was because it isgenerally available and is likely to be used in any initial modelof uranyl sorption. However, to evaluate the sensitivity ofthe surface complexation coefficients to other databasecompilations additional fits were obtained using the NEAdatabase (15).

The D&M assumptions regarding site density [weak sites(Hfo_wOH) at 0.2 mol/mol Fe; strong sites (Hfo_sOH) at 0.005mol/mol Fe], and surface acidity constants (log Ka1

int of 7.29and log Ka2

int of -8.93), were the same for both strong andweak sites (10); a surface area of 600 m2/g and a formulaweight of 89 g/mol were used to complete the definition ofthe HFO surface (10). For the simulations that includedcarbonate, two additional surface complexation reactionswere included:

Hfo_wOH + CO3-2 + H+ ) Hfo_wOCO2

- +

H2O log KHfo_wOCO2-int ) 12.78 ( 0.48

Hfo_wOH + CO3-2 + 2H+ ) Hfo_wOCO2H +

H2O log KHfo_wOCO2Hint ) 20.37 ( 0.20

The log Kint values were reported by Appelo et al. (17) andare consistent with the initial assumption of the DLMdeveloped by D&M. The log Kint values for these two reactionswere not adjusted as part of this evaluation.

The inverse modeling program UCODE_2005 (18) wasused to optimize the fit of the surface complexation constantsto the experimental data. This program was originallydeveloped to optimize estimates of parameters used to modelgroundwater flow. However, UCODE_2005 is designed to bea universal inverse modeling code, so when coupled withPHREEQC, it can be used to model reactions in solution (19)or surface complexation reactions. The PHREEQC input fileconsisted of a series of models representing each individualdata point. Because the starting and final concentrations for

FIGURE 2. Comparison of Payne’s (5) experimentally derivedmeasurements of uranyl uptake onto HFO to DLM calculationsusing LFER-derived constants of D&M uranyl surfacecomplexation reactions; based upon 10-6 m U(VI) sorbed onto 1(filled triangles) and 20 (filled squares) mmole of HFO in a 0.1NaNO3 solution at a CO2 partial pressure of 10-3.5 atmospheres.Model fits include open triangles with lines for the 1 mmol/LHFO data set and open squares with lines for the 20 mmol/LHFO data set. Dashed vertical lines included to facilitatecomparisons between experimental data and modeledconcentrations.

VOL. 43, NO. 24, 2009 / ENVIRONMENTAL SCIENCE & TECHNOLOGY 9 9261

the various data sets cover several orders of magnitude,concentrations were converted to log values for purposes ofresidual calculation and optimization.

It should be noted that many of the previous attempts toestimate equilibrium constants used FITEQL (20). Theseprograms employ similar optimization methods. The maindifference between FITEQL and UCODE is that UCODE usesthe PHREEQC code to perform the solution reactionscalculations, while FITEQL is a self-contained programdesigned to calculate equilibrium constants. One of theprimary advantages of the approach incorporated as part ofthis study is the ability of PHREEQC to model a variety ofexperimental conditions in one input file including systemsthat are open to CO2 gas, and others that are closed systems(using total carbonate concentrations); additionally, changesin total uranium concentrations and sorbent amounts canalso be readily handled by PHREEQC. FITEQL is not able tomodel different partial pressures of CO2 gas simultaneously,let alone different total U concentrations or changes insorbent concentrations.

During parameter optimization, UCODE_2005 runs PHRE-EQC many times in succession, each time incrementallyvarying the input parameters so as to minimize the sum ofsquared, weighted residuals (SOSR). The maximum fractionalparameter change was 0.05. As part of the procedure,UCODE_2005 calculates the slope of the objective function(sensitivity) of each parameter. The residuals and sensitivitiesare used by UCODE_2005 to produce a better estimate of theparameter(s). The program minimizes the SOSR until con-vergence is reached. After each run, the residuals andparameters are compared to user-set convergence criterionto determine if convergence has been achieved. If thesimulation has not met the convergence criteria, newparameter estimates are calculated, and a new iteration isperformed. Convergence was attained when the estimatedparameters changed by less than one percent over twoiterations. Once convergence is reached, UCODE_2005calculates a series of statistical measurements of fit andproduces a summary of the iterations completed.

Table 1 provides a summary of the data used in theevaluation. In most cases, data were scanned and theindividual data points were digitized from the publicationfigures to obtain the pH and reported concentrations foreach data set. Data from Payne (5, 6) were taken fromtabulations in an appendix in his dissertation.

UCODE_2005 allows weighting of data points based onobjective or subjective judgments. We decided to weight alldata points as equally reliable (weight of 1), except in thecase of two high-pH data points from Hsi (2). Duringoptimization, it was discovered that inclusion of these two

data points resulted in unrealistic parameter estimates. Thesepoints are discussed below. Therefore, the data points wereall but eliminated from the simulations by assigning theman extremely low weighting compared to the other points.

Initially, a “bulk” fitting procedure was attempted thatemployed UCODE to simultaneously match all the data tothe four primary surface complexation constants. However,depending upon the starting values selected for the uranylcarbonate surface complexes either the uranyl monocar-bonate (Hfo_wOUO2CO3

-) or dicarbonate surface complex(Hfo_wOUO2(CO3)2

-3) dominated the upper pH region andthe other complexation constant dropped in value by severalorders of magnitude; this had been previously noted in earlierwork (21). It was felt that the two uranyl carbonate surfacecomplexes provided a better overall approach to fit the data.As a result, a stepwise optimization procedure was adopted.The steps undertaken were (1) optimize strong and weak siteuranyl surface complexation constants, (2) evaluate uranylhydroxide surface complex on weak sites, (3) optimize uranylmonocarbonate surface complex on weak sites, (4) optimizeuranyl dicarbonate surface complex on weak sites, and (5)use additional data sets as a fit robustness check.

(1) Estimation of Hfo_sOUO2+ and Hfo_wOUO2

+ Com-plexation Reactions. The two data sets of Hsi (2) preparedwithout carbonate were used to estimate the K1

int and K2int

values. Most of the data points had final pH values less than7.0, so the data provided a good representation of the low-pH adsorption edge. The two data points with the highestpH values were removed because they were far above thelow-pH adsorption edge, and they do not appear to representcomplexation related to the K1

int and K2int reactions. The data

presented by Hsi were reported as final uranium concentra-tions in solution, rather than percentages, with ranges of2-4 orders of magnitude. Both data sets used an initial uranylconcentration of 1 × 10-5 mol/L. One data set used an HFOconcentration of 1 g/L ferrihydrite, while the second data setused a lower HFO concentration of 0.15 g/L ferrihydrite.Uranyl surface complexation constants for the strong andweak sites (i.e., the K1

int and K2int reactions) were simulta-

neously fit to the two data sets. In the initial re-evaluationof the data a trial and error visual fit of the 1 g/L ferrihydritedata set produced an excellent fit using a log K1

int value of4.15 and log K2

int value of 2.05 (21). Using the UCODE_2005/PHREEQC setup, and the two data sets, changed the valuesto 3.736 and 2.534 respectively. Use of two data sets withdifferent HFO concentrations provides a more robust anddefensible fit than the previous fit to a single data set. Theseparameters are poorly constrained individually because ofthe similarity of the mass action equations for K1

int and K2int

(Table 2), but they do help the overall fit by steepening the

TABLE 1. Summary of Experimental Data Conditions Used in This Evaluationa

researcher(group or publication)

seriesnumber

U(VI)total M

ferrihydriteamount

carbonatecondition

electrolyte(M)

number ofdata points

Hsi, 1981; Hsi andLangmuir, 1985

1 1.0 × 10-5 1 g/L ferrihydrite no carbonate 0.1 NaNO3 292 1.0 × 10-5 0.15 g/L ferrihydrite no carbonate 0.1 NaNO3 113 1.0 × 10-5 1 g/L ferrihydrite closed 0.01 M C(IV) 0.1 NaNO3 124 1.0 × 10-5 1 g/L ferrihydrite closed 0.001 M C(IV) 0.1 NaNO3 15

Waite et al., 1994;Payne, 1999

5 1.0 × 10-6 1 mmol/L Fe [0.089 g/L Fe(OH)3] open 1.0 × 10-3.5 atms CO2(g) 0.1 NaNO3 406 1.0 × 10-6 20 mmol/L Fe [1.78 g/L Fe(OH)3] open 1.0 × 10-3.5 atms CO2(g) 0.1 NaNO3 207 1.0 × 10-5 1 mmol/L Fe [0.089 g/L Fe(OH)3] open 1.0 × 10-3.5 atms CO2(g) 0.1 NaNO3 188 1.0 × 10-4 1 mmol/L Fe [0.089 g/L Fe(OH)3] open 1.0 × 10-3.5 atms CO2(g) 0.1 NaNO3 179 1.0 × 10-6 1 mmol/L Fe [0.089 g/L Fe(OH)3] open 1.0 × 10-2.0 atms CO2(g) 0.1 NaNO3 15

Morrison, Spangler andTripathi, 1995

10 1.0 × 10-5 1.0 g/L Fe(OH)3 closed 0.001 M C(IV) 0.1 NaNO3 1511 2000 µg/L (8.4 × 10-6) 0.52 g/L Fe(OH)3 closed 0.0195 M C(IV) 0.1 NaNO3 9

Wazne et al., 2003 12 4.2 × 10-6 1.42 × 10-3 M Fe [0.126 g/L Fe(OH)3] closed 0.01 M C(IV) 0.01 NaCl 16Fox et al., 2006 13 1.0 × 10-6 1.00 × 10-4 M Fe [0.0089 g/L Fe(OH)3] open 1.0 × 10-3.37 atms CO2(g) 0.01 NaNO3 9

14 1.0 × 10-6 1.00 × 10-3 M Fe [0.089 g/L Fe(OH)3] open 1.0 × 10-1.7 atms CO2(g) 0.005 NaNO3 7

a Open systems use fixed carbon dioxide pressures specified in the table, while closed systems use a total carbonate [C(IV)] concentration for all pH values.

9262 9 ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 43, NO. 24, 2009

low pH region adsorption edge, also the two reactions wereincluded to maintain consistency with the original DLMassumptions.

(2) Evaluation of Hfo_wOUO2OH Surface Complex. Thetwo high-pH data points from the first Hsi data set areproblematic (Figures 1 and 3), as they do not show theincrease in dissolved uranyl predicted by the initial model.It is possible that some small amount of atmosphericcarbonate may have entered the system or that schoepite(�-UO3 ·2H2O) (1) formed as the pH values were adjusted,and the schoepite did not have time to dissolve before themeasurements were completed. In the initial modeling toestimate the log K1

int and K2int values, these two data points

were removed from the data set. However, as these pointsalso represented the highest pH data for the noncarbonatesystem, and similar data are present in sulfate-bearingnoncarbonate solutions used by Morrison et al. (8) (notincluded in this evaluation), it appeared that the two pointsmight represent high-pH conditions for carbonate-free watersand we decided to evaluate the significance of theHfo_wOUO2OH surface complex. Much of the fit relies onthese two suspect data points to obtain an optimized valuefor the uranyl hydroxide surface complexation constant (K5

int).For this optimization, log K1

int and K2int were fixed at their

values optimized in the initial calculations (3.736 and 2.534),using the WATEQ4F database (Figure 3).

Hfo_wOH + UO2+2 + H2O ) Hfo_wOUO2OH + 2H+ K5

int

The importance of the Hfo_wOUO2OH surface complexis slight at best. It is discussed mainly for completeness ofpresentation, and for the rare occasion that it is required fora high-pH noncarbonate solution. It relies upon the fittingto only two uncertain data points and inclusion of the surfacecomplex in any model tends to result in poorer overall fits,which would necessitate refitting of the other parameters.We do not recommend that this reaction be included in anygeneral compilation database.

(3) Estimation of Hfo_wOUO2CO3- and (4)

Hfo_wOUO2(CO3)2-3 Complexation Reactions. Most of the

experimental setups for carbonate-bearing solutions wereperformed at or near atmospheric CO2 pressures of 10-3.5

atmospheres (Table 1). A few experimental setups were rununder different effective pressures, either because theyrepresent closed systems (those not open to a fixed CO2

pressure) or were run at higher CO2 pressures. No attemptwas made to try and adjust CO2 pressures to compensate forpossible changes caused by degassing.

Initial evaluation of the uranium and carbonate systems(21) indicated that models including either a uranyl mono-carbonate or a uranyl dicarbonate surface complex couldboth separately provide acceptable fits to most of theexperimental data. UCODE_2005 calculations were initiallyperformed that allowed for the simultaneous adjustment ofboth uranyl carbonate surface complexes. The two reactionsare

Hfo_wOH + UO2+2 + CO3

-2 ) Hfo_wOUO2CO3- + H+ K3

int

Hfo_wOH + UO2+2 + 2CO3

-2 ) Hfo_wOUO2(CO3)2-3 + H+ K4

int

These initial simulations generally resulted in the programselecting one reaction and increasing the log K value (eitherK3

int or K4int), and reducing the other constant, often to near

zero.Evenwithallthedatausedintheestimates,UCODE_2005’sselection of the uranyl monocarbonate over the uranyldicarbonate surface complex depended upon starting condi-tions. This is related to the nature of the high pH desorptionedge, which is controlled by a combination of factorsincluding, point of zero charge, and the dominance of uranylcarbonate complexes, such as UO2(CO3)3

-4 in solution.TABL

E2.

Sum

mar

yof

UCOD

E_20

05Ge

nera

ted

Resu

ltsto

Dete

rmin

eUr

anyl

Surfa

ceCo

mpl

exat

ion

Cons

tant

son

toHF

O

uran

ylsu

rfac

eco

mpl

exat

ion

spec

ies

nam

ere

actio

nop

timiz

edlo

gK

95%

confi

denc

ein

terv

alno

.of

obse

rvat

ions

no.o

fda

tase

tsav

erag

ere

sidu

al(lo

g)av

erag

ere

sidu

alm

agni

tude

(log)

note

s

stro

ng

site

ura

nyl

K1in

tH

fo_s

OH

+U

O2+

2)

Hfo

_sO

UO

2++

H+

3.73

6W

AT

EQ

4F1.

62to

5.85

382

-1.

77×

10-

21.

42×

10-

1O

nly

no

nca

rbo

nat

esy

stem

su

sed

3.79

2N

EA

2.17

to5.

41-

1.78

×10

-2

1.38

×10

-1

wea

ksi

teu

ran

ylK

2int

Hfo

_wO

H+

UO

2+2)

Hfo

_wO

UO

2++

H+

2.53

4W

AT

EQ

4F1.

79to

3.28

382

-1.

77×

10-

21.

42×

10-

1O

nly

no

nca

rbo

nat

edsy

stem

su

sed

2.50

7N

EA

1.83

to3.

18-

1.78

×10

-2

1.38

×10

-1

ura

nyl

mo

no

carb

on

ate

K3in

tH

fo_w

OH

+U

O2+

2+

CO

3-2)

Hfo

_wO

UO

2CO

3-+

H+

9.03

4W

AT

EQ

4F8.

75to

9.32

8110

3.49

×10

-3

2.49

×10

-1

K5

no

tu

sed

,p

H<6

.33

9.15

0N

EA

8.90

to9.

407.

48×

10-

32.

46×

10-

1K

5n

ot

use

d,

pH

<6.3

39.

014

WA

TE

Q4F

8.67

to9.

353.

40×

10-

22.

70×

10-

1K

5u

sed

,p

H<6

.33

ura

nyl

dic

arb

on

ate

K4in

tH

fo_w

OH

+U

O2+

2+

2CO

3-2)

Hfo

_wO

UO

2(C

O3)

2-3+

H+

15.2

8W

AT

EQ

4F15

.1to

15.5

217

121.

88×

10-

22.

97×

10-

1K

5n

ot

use

d

15.2

8N

EA

15.0

to15

.51.

52×

10-

23.

14×

10-

1K

5n

ot

use

d15

.24

WA

TE

Q4F

15.0

to15

.54.

85×

10-

23.

22×

10-

1K

5u

sed

ura

nyl

hyd

roxi

de

K5in

tH

fo_w

OH

+U

O2+

2+

H2O

)H

fo_w

OU

O2O

H+

2H+

-5.

111

WA

TE

Q4F

-4.

74to

-5.

4840

23.

73×

10-

22.

26×

10-

1K

5ev

alu

ated

bu

tsh

ou

ldn

ot

be

incl

ud

edin

dat

abas

eco

mp

ilati

on

ura

nyl

dic

arb

on

ate

chec

ko

nas

sum

pti

on

sK

4int

Hfo

_wO

H+

UO

2+2+

2CO

3-2)

Hfo

_wO

UO

2(C

O3)

2-3+

H+

15.2

8W

AT

EQ

4F15

.1to

15.5

233

142.

49×

10-

23.

00×

10-

1K

5n

ot

use

d-

this

run

was

ate

sto

fro

bu

stn

ess

toad

dit

ion

alh

igh

-pH

dat

a

VOL. 43, NO. 24, 2009 / ENVIRONMENTAL SCIENCE & TECHNOLOGY 9 9263

Ultimately, it was decided to include both reactions, but toestimate the reaction constants separately by continuing thestepwise approach used to evaluate the K1

int, K2int, and K5

int

reaction constants.To estimate the log K3

int value, data were selected with pHvalues less than 6.33 from the initial 10 data sets thatcontained carbonate. The low-pH data were used becauseuranyl monocarbonate surface complex would dominate inlower pH systems, this set of calculations essentially correctsthe low pH adsorption edge for the presence of a uranylcarbonate surface complex. The optimization was performedtwice: once with K5

int and once without. In each case, K1int,

K2int and K5

int (when included) were fixed at their previouslyoptimized values, and only K3

int was estimated.To estimate the uranyl dicarbonate surface complexation

constant value (log K4int), all 217 data points from the first

12 data sets (i.e., not including the two sets from Fox et al.(9)) listed in Table 1 were used. This optimization was alsoperformed twice: once with K5

int and once without. In eachcase, the log K1

int, K2int, K3

int, and K5int (if included) values

were fixed at their optimized values, and only K4int was

estimated. These fitting calculations essentially fit to the highpH desorption edges, which as mentioned previously isactually fairly insensitive to the selection of the specific uranyland carbonate bearing surface complex. Figure 4 shows thefinal fit when compared to one example data set. The figurealso includes the distribution of the four selected uranyl-bearing surface complexes. Other fits are included in the SI.

(5) Fit Robustness Check. Finally, to test the robustnessof the optimized fits, two previously unused data sets (thetwo Fox et al. (9) data sets in Table 1) were appended to thepreviously used data compilation. Because these data setsonly included carbonate system data with pH > 6.33, onlyK4

int was re-evaluated. This final simulation used all 233 datapoints from the 14 data sets. The optimized K4

int value wasidentical to the one estimated using the previous simulations,providing additional confidence in the optimized valuesobtained from earlier simulations.

Previously we had demonstrated (21) that the twocarbonate surface complexes (Hfo_wOCO2

- andHfo_wOCO2H) had little impact on the extent of uranyladsorption. Waite et al. (6) also made a similar observation.This is probably because most of the experimental conditions

were limited to the relatively low atmospheric value of 10-3.5

atmospheres for CO2 gas. These two complexes become moresignificant at higher partial pressures of CO2 gas and in somesolute transport model scenarios.

Table 2 summarizes the optimized fit results for both theWATEQ4F and NEA database model sets. In addition, Table2 provides three metrics describing the uncertainty associatedwith the fits: (1) 95% confidence interval ranges, (2) averageresidual, and (3) average magnitude of residual. Thesesummary statistics are automatically generated by UCODE_2005. Smaller 95% confidence interval ranges generallyindicate less uncertainty associated with the optimized result.Average signed residuals and residual magnitudes indicatehow well the model fits the data; thus, both metrics shouldbe close to zero if there is a good fit to data. Differences inthe UCODE_2005 selected constants between the twodatabases are not statistically different (Figure 5).

For comparison, the same 14 experimental data sets weresimulated using the original two D&M constants plus thetwo carbonate surface complexes (see SI). The resultingaverage residual (0.2) was approximately an order of mag-nitude larger than the values listed in Table 2. The averageresidual magnitude was also larger (0.5). Thus, the newconstants result in a significantly improved fit to theexperimental data.

DiscussionFigure 6 compares the model-predicted concentrations tothe measured concentrations. Overall the agreement betweenthe modeled and experimental data is good with the exceptionof a few outliers, which we attribute to setup of the originalmeasurements. In addition, much of the data was obtainedfrom plots and tables that show percent adsorbed, ratherthan final concentrations in solution, and data presented inthis fashion can produce reporting limit issues. This hasresulted in generally poorer fits in the middle pH range whereadsorption is maximized. Examination of the individual fits(see SI) indicates that generally the better fits are for theadsorption and desorption edges, whereas mid-pH rangestend to show poorer agreement. Reporting limit truncationis probably responsible for the five data points whosemeasured values are 1.0 × 10-9 moles per kilogram of water(mol/kgw), which represents 99.9% adsorption of an initially1.0 × 10-6 mol/kgw solution.

FIGURE 3. Comparison between Hsi’s (2) experimental data ofuranyl adsorption based upon 10-5 m U(VI) sorbed onto 0.15(triangles) and 1 g/L (lined X’s) of ferrihydrite in a 0.1 NaNO3solution with no carbonate with DLM calculations (solid lines)using updated log K1

int and log K2int values for the WATEQF

database. Dotted lines show model predictions with the uranylhydroxide surface complex (log K5

int term). Dashed lines showmodel fits using constants estimated in conjunction with NEAdatabase that exclude the uranyl hydroxide surface complex.

FIGURE 4. Comparison between Payne’s (5) experimental data(vertically lined X’s) with DLM calculations assuming foursurface complexation reactions for uranyl (K1

int through K4int)

using the WATEQ4F database. The distribution of the differentsurface complexes are also shown. Total sorbed concentrationsdeveloped by model calculations shown by the solid line.Based upon 10-6 m U(VI) sorbed onto 1 mmol of HFO in a 0.1NaNO3 solution at a CO2 partial pressure of 10-3.5 atmospheres.

9264 9 ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 43, NO. 24, 2009

This study was primarily designed to provide correctionsto the original DLM database as it applies to uranyladsorption. To maintain as much consistency with theoriginal DLM as possible certain assumptions wereretained, such as using only monodendate type surfacecomplexes, and limiting high affinity sites to cationadsorption reactions. Inclusion of uranyl carbonate surfacecomplexes into the original DLM database corrects anoversight with this database, and improves the fitting ofuranyl sorption in carbonate-bearing systems. The agree-ment between the final model selected parameters andthe overall data set of 233 points as represented in 14 datasets developed by five different research groups indicatesthat the experimental efforts show a consistency that

supports the general application of this revised model indescribing uranyl adsorption.

The modeling presented in this paper has shown thatthe previously published and widely used uranyl sorptionconstants in the original DLM compilation result inunrealistic predictions of uranium fate and mobility inthe environment. This is environmentally significantbecause the previous constants tend to overpredicturanium sorption under low-pH conditions. Consequently,resultant models were not environmentally conservativeand could result in inaccurate predictions. The re-evaluation conducted here corrects that problem.

AcknowledgmentsWe thank MWH Americas, Inc. and GeoTrans, Inc. insupporting this work.

Supporting Information AvailableAdditional details are available. This material is availablefree of charge via the Internet at http://pubs.acs.org.

Literature Cited(1) Langmuir, D. Aqueous Environmental Geochemistry; Prentice

Hall: Upper Saddle River, NJ, 1997.(2) Hsi, C. K. D. Sorption of uranium (VI) by iron oxides. Ph.D.

Dissertation, Colorado School of Mines, Golden, CO,1981.

(3) Hsi, C. K. D.; Langmuir, D. Adsorption of uranyl onto ferricoxyhydroxides: applications of a surface complexation sitebinding model. Geochim. Cosmochim. Acta 1985, 49, 2423–2432.

(4) Payne, T. E.; Lumkin, G. R.; Waite, T. D. UraniumVI adsorptionon model minerals. In Adsorption of Metals by Geomedia;Jenne, E. A., Ed.; Academic Press: New York, 1998; pp75-97.

(5) Payne, T. E. Uranium (VI) interactions with mineral surfaces:controlling factors and surface complexation modeling. Ph.D.Dissertation, University of New South Wales, Sydney, NSW,1999.

(6) Waite, T. D.; Davis, J. A.; Payne, T. E.; Waychunas, G. A.; Xu, N.Uranium(VI) adsorption to ferrihydrite: Application of a surfacecomplexation model. Geochim. Cosmochim. Acta 1994, 57, 2251–2269.

(7) Wazne, M.; Korfiatis, G. P.; Meng, X. Carbonate effects onhexavalent uranium adsorption by iron oxyhydroxide. Environ.Sci. Technol. 2003, 37, 3619–3624.

(8) Morrison, S. J.; Spangler, R. R.; Tripathi, V. S. Adsorption ofuranium(VI) on amorphous ferric oxyhydroxide at high con-centrations of dissolved carbonate(IV) and sulfur(VI). J. Contam.Hydrol. 1995, 17, 333–346.

(9) Fox, P. M.; Davis, J. A.; Zachara, J. M. The effect of calciumon aqueous uranium(VI) speciation and adsorption toferrihydrite and quartz. Geochim. Cosmochim. Acta 2006, 70,1379–1387.

(10) Dzombak, D. A.; Morel, F. M. M. Surface ComplexationModelingsHydrous Ferric Oxide; John Wiley and Sons: NewYork, 1990.

(11) Parkhurst, D. L.; Appelo, C. A. J. User’s Guide to PHREEQC(Version 2)sA Computer Program for Speciation, Batch-Reaction,One-Dimensional Transport, and Inverse Geochemical Calcula-tions; Water Resources Investigation Report 99-4259; USGS:Denver, CO, 1999.

(12) Allison, J. D.; Brown, D. S.; Novo-Gradac, K. J. MINTEQA2/PRODEFA2, A Geochemical Assessment Model for Environ-mental Systems, Version 3.0 User’s Manual; EPA/600/3-91/021;U.S. EPA: Athens, GA, 1991.

(13) Bethke, C. M. Geochemical Reaction Modeling; Oxford UniversityPress: New York, 1996.

(14) Baes, C. F., Jr.; Mesmer, R. E. The Hydrolysis of Cations; WileyInterscience: New York, 1976.

(15) Guillaumont, R.; Fanghahel, T.; Neck, V.; Fuger, J.; Palmer, D.;Grenthe, I.; Rand, M. H. Update on the Chemical Thermody-namics of Uranium, Neptunium, Plutonium, Americium andTechnetium; Elsevier B.V.: Amsterdam, 2003.

(16) Westall, J.; Hohl, H. Comparison of models for the oxide-solutioninterface. Adv. Colloid Interface Sci. 1980, 12, 265–294.

FIGURE 5. Comparison between Payne’s (5) experimental data(vertically lined X’s) with DLM calculations assuming foursurface complexation reactions for uranyl (K1

int through K4int)

using the WATEQ4F database (solid line with open squares).Also included are the DLM fits using the NEA database (dashedline with open circles). Dotted line represents original fit usingLFER-derived constants of D&M uranyl surface complexationreactions. Based upon 10-6 m U(VI) sorbed onto 1 mmol of HFOin a 0.1 NaNO3 solution at a CO2 partial pressure of 10-3.5

atmospheres.

FIGURE 6. Scatter plot of DLM predicted concentrations ofU(VI) with experimentally obtained data (measuredconcentrations). The figure includes all 233 points from 14 datasets and represents the combined work of five differentresearch groups. Modeled results (predicted concentrations)used the four major uranyl surface complexation reactionsdefined in this study (K1

int through K4int) as calculated with the

WATEQ4F database. The series numbers 1-14 correspond tothe data sets listed sequentially in Table 1.

VOL. 43, NO. 24, 2009 / ENVIRONMENTAL SCIENCE & TECHNOLOGY 9 9265

(17) Appelo, C. A. J.; Van der Weiden, M. J. J.; Tournassat, C.; Charlet,L. Surface complexation of ferrous iron and carbonate onferrihydrite and the mobilization of arsenic. Environ. Sci.Technol. 2002, 36, 3096–3103.

(18) Poeter, E. P.; Hill, M. C.; Banta, E. R.; Mehl, S.; Christensen,S. UCODE_2005 and Six Other Computer Codes for UniversalSensitivity Analysis, Calibration and Uncertainty Evaluation;Techniques and Methods 6-A11; USGS: Reston, VA, 2005.

(19) Skold, M. E.; Thyne, G. D.; McCray, J. E. Using UCODE_2005and PHREEQC to determine thermodynamic constants fromexperimental data. Ground Water 2007, 45 (3), 368-373.

(20) Herbelin, A. L.; Westall, J. C. FITEQLsA Computer Program forDetermination of Chemical Equilibrium Constants from Ex-perimental Data (Version 4.0); Report 99-01; Oregon StateUniversity: Corvallis, OR, 1999.

(21) Mahoney, J. J.; Jakubowski, R. T. Assessment of uranyl sorptionconstants on ferrihydritesComparison of model derived con-stants and updates to the diffuse layer model database. InUranium Mining and Hydrogeology; Merkel, B. J., Hasche-Berger, A., Eds.; Springer: Berlin, 2008; pp 919-928.

ES901586W

9266 9 ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 43, NO. 24, 2009