Lyngby, Denmark Received 26 August 2011; accepted in revised form 14 July 2012; available online 22 July 2012 Abstract Three surface complexation models SCMs developed for, respectively,
Trang 1Surface complexation modeling of groundwater arsenic mobility:
Results of a forced gradient experiment in a Red
River flood plain aquifer, Vietnam Søren Jessena,⇑, Dieke Postmab, Flemming Larsenb, Pham Quy Nhanc,
a
Department of Geography and Geology, University of Copenhagen, 1350 Copenhagen, Denmark
b
Department of Geochemistry, Geological Survey of Denmark and Greenland (GEUS), 1350 Copenhagen, Denmark
c
Department of Hydrogeology, Hanoi University of Mining and Geology (HUMG), Hanoi, Vietnam
d
Department of Applied Physics, Graduate School of Engineering, Osaka University, Osaka, Japan
e
Research Centre for Environmental Technology and Sustainable Development (CETASD), Hanoi University of Science (VNU),
Hanoi, Vietnam
f Department of Environmental Engineering, Technical University of Denmark, 2800 Kgs Lyngby, Denmark
Received 26 August 2011; accepted in revised form 14 July 2012; available online 22 July 2012
Abstract
Three surface complexation models (SCMs) developed for, respectively, ferrihydrite, goethite and sorption data for a Pleis-tocene oxidized aquifer sediment from Bangladesh were used to explore the effect of multicomponent adsorption processes on As mobility in a reduced Holocene floodplain aquifer along the Red River, Vietnam The SCMs for ferrihydrite and goethite yielded very different results The ferrihydrite SCM favors As(III) over As(V) and has carbonate and silica species as the main compet-itors for surface sites In contrast, the goethite SCM has a greater affinity for As(V) over As(III) while PO4 and Fe(II) form the predominant surface species The SCM for Pleistocene aquifer sediment resembles most the goethite SCM but shows more Si sorption Compiled As(III) adsorption data for Holocene sediment was also well described by the SCM determined for Pleisto-cene aquifer sediment, suggesting a comparable As(III) affinity of HoloPleisto-cene and PleistoPleisto-cene aquifer sediments A forced gradient field experiment was conducted in a bank aquifer adjacent to a tributary channel to the Red River, and the passage in the aquifer
of mixed groundwater containing up to 74% channel water was observed The concentrations of As (<0.013 lM) and major ions
in the channel water are low compared to those in the pristine groundwater in the adjacent bank aquifer, which had an As con-centration of3 lM Calculations for conservative mixing of channel and groundwater could explain the observed variation in concentration for most elements However, the mixed waters did contain an excess of As(III), PO4and Si which is attributed to desorption from the aquifer sediment The three SCMs were tested on their ability to model the desorption of As(III), PO4 and
Si Qualitatively, the ferrihydrite SCM correctly predicts desorption for As(III) but for Si and PO4it predicts an increased adsorption instead of desorption The goethite SCM correctly predicts desorption of both As(III) and PO4
but failed in the prediction of Si desorption These results indicate that the prediction of As mobility, by using SCMs for synthetic Fe-oxides, will
be strongly dependent on the model chosen The SCM based on the Pleistocene aquifer sediment predicts the desorption of As(III), PO4 and Si quite superiorly, as compared to the SCMs for ferrihydrite and goethite, even though Si desorption is still somewhat under-predicted The observation that a SCM calibrated on a different sediment can predict our field results so well suggests that sediment based SCMs may be a feasible way to model multi-component adsorption in aquifers
Ó 2012 Elsevier Ltd All rights reserved
0016-7037/$ - see front matter Ó 2012 Elsevier Ltd All rights reserved.
http://dx.doi.org/10.1016/j.gca.2012.07.014
⇑ Corresponding author Tel.: +45 51370693.
E-mail address: sjessen1976@gmail.com (S Jessen).
www.elsevier.com/locate/gca
Geochimica et Cosmochimica Acta 98 (2012) 186–201
Trang 21 INTRODUCTION
The contamination of groundwater with arsenic (As) is
an issue of continued concern in Southeast Asia (Fendorf
et al., 2010) Although the mechanisms involved in the
mobilization of As are still debated (Polizzotto et al.,
2006; Seddique et al., 2008), most researchers ascribe the
mobilization of As to the reductive dissolution of
As-con-taining Fe-oxides or Fe-oxyhydroxides, henceforth
collec-tively called Fe-oxides (Nickson et al., 1998, 2000;
McArthur et al., 2001; Dowling et al., 2002; Harvey
et al., 2002; Swartz et al., 2004; Postma et al., 2007,
2010) The electron donor for this process is organic
car-bon, the reactivity and availability of which is the overall
limiting factor for As mobilization (Postma et al., 2007;
Neumann et al., 2009) As a result, the concentration of
As in the groundwater will depend on sediment properties
like the organic carbon present, a possible input of DOC,
and also on the flow path of the groundwater For example,
the sediments in the Mekong delta are 5000–8000 years old
and contain little reactive organic carbon (Polizzotto et al.,
2008) In this aquifer, high As groundwater is found
prefer-entially along flow paths spreading downstream of organic
rich ponds At the other end of the scale, Postma et al
(2007)studied a younger, 400–500 year old, aquifer along
the Red River where As mobilization occurs at shallow
depth throughout the aquifer because of the presence of
reactive sedimentary organic carbon
Once arsenic has entered the groundwater, the next
important question becomes the mobility of As during
transport in the aquifer In a reduced aquifer, As can
ad-sorb onto (mixed valence) Fe-oxides, micas, phyllosilicates
and carbonates (Lin and Puls, 2000; Harvey et al., 2002;
Dowling et al., 2002; Horneman et al., 2004; Chakraborty
et al., 2007; Sø et al., 2008; Charlet et al., 2011).Polizzotto
et al (2006)showed that more than 15% of the total As
content in the sediment of a reduced aquifer from
Bangla-desh could be leached with anoxic deionized water,
indicat-ing the presence of a significant pool of mobile As This
supports the arguments by Stute et al (2007) for the
groundwater As content to be dependent on the extent of
flushing
Previous assessments of the mobility of As in aquifers
have been based on determining the As partitioning
be-tween solid and solution by measuring adsorbed As on
the sediment, using various extraction methods, and
dis-solved As in the groundwater (BGS and DPHE, 2001;
Swartz et al., 2004; van Geen et al., 2008; Nath et al.,
2009; Itai et al., 2010) For Holocene aquifer sediments
from Bangladesh,BGS and DPHE (2001)report an
appar-ent distribution coefficiappar-ent between adsorbed and dissolved
As, K0
d, of 2–6 L/kg, whilevan Geen et al (2008)found a
trend close to K0
d4 L/kg Both studies indicate a high As
mobility in aquifers However,Itai et al (2010)found,
like-wise for Holocene aquifer sediments from Bangladesh,
much higher values, K0
d-As(III) 7.3–46 L/kg and K0
d-As(V)
21–561 L/kg, indicating a far lower mobility
The above range of K0
dvalues for As among different studies could be caused by the influence of other substances
competing for surface sites, particularly because As is a
trace substance in the water and therefore is likely to take
up only a minor fraction of the available surface sites It fol-lows that the influence of competing ions for surface sites, and thereby the composition of the groundwater, is ex-pected to have an important influence on the As mobility
BGS and DPHE (2001), Swartz et al (2004), Appelo
et al (2002)andPostma et al (2007)all used the general-ized two layer Dzombak and Morel (1990) surface com-plexation model (SCM) for ferrihydrite to assess the partitioning of As between water and sediment in the pres-ence of competing ions Indeed, the K0
dvalues calculated in this way spread over a very wide range (Smedley and Kinni-burgh, 2002) depending on the water compositions and the assumptions made It remains to be seen how these calcu-lated K0
dvalues relate to the actual retardation of As under
in situ aquifer conditions Moreover, in reduced Holocene aquifers goethite might be a better model for sedimentary Fe-oxides than ferrihydrite (Postma et al., 2007, 2010), so the use of a SCM for goethite could be more relevant An advanced and flexible SCM for goethite is available as a charge distribution-multisite complexation (CD MUSIC) model (Hiemstra and Van Riemsdijk, 1996; Stachowicz
et al., 2006, 2007, 2008) Using the CD MUSIC model,
Stachowicz et al (2008)successfully modeled the competi-tive sorption of PO4, Ca2+, CO3and As(III) on goe-thite To our knowledge the CD MUSIC model has not yet been applied to As-contaminated aquifers
The above two SCMs have been developed for synthetic Fe-oxides Predicting As mobility in aquifers by application
of a SCM for a synthetic phase would comply with the com-ponent additive approach (Davis et al., 1998) This approach assumes that the controlling sorbent(s) of a natural sedi-ment can be adequately represented by the analog synthetic phase(s), while the complexity added in the natural setting
is of minor importance However, one may question to what extent models for synthetic Fe-oxides give a reason-able description of the processes taking place in aquifers where the Fe-oxides in the sediment may be impure, much more heterogeneous and also may interact with clays and organics in a complex way (Hiemstra et al., 2010) An alter-native strategy could be to use a SCM fitted to sorption data for the competitive adsorption of solutes to the surface
of the aquifer sediment.Davis et al (1998)termed this the general composite approach, which acknowledges that the sorption properties of a sediment may be too complex to
be quantified in terms of the contributions by individual synthetic phases to adsorption The general composite ap-proach has been tried byStollenwerk et al (2007)who mea-sured dual-sorbate competitive sorption of As(III), As(V),
PO4, Si and CO3on a Pleistocene oxidized aquifer sed-iment from Bangladesh.Stollenwerk et al (2007)developed
a hybrid SCM in which the measured sorption properties
on the Pleistocene sediment were combined, for remaining species and electrostatic effects, with the D&M database for ferrihydrite
In order to assess the mobility of As under field condi-tions, we have carried out a forced gradient field experi-ment, where pumping causes low As river water to intrude into an adjacent high As aquifer The induced changes in water chemistry were interpreted with the aid
Trang 3of the above presented three SCMs, i.e., two models
devel-oped for synthetic Fe-oxides, ferrihydrite and goethite, and
one model developed from adsorption data to an aquifer
sediment The overall objective is to evaluate and discuss
the application of these models to the interpretation of
aquifer data We thereby aim to obtain a better quantitative
understanding of the mobility of As in the S.E Asian
aquifers
2 MATERIALS AND METHODS
2.1 Field experiment
2.1.1 Instrumentation of the field site
Three pumping wells (Figs 1 and 2) were constructed by
rotary drilling (Ø170 mm) in May–June 2007 and equipped
with an Ø140 mm PVC casing telescoping into an 8 m long,
Ø90 mm screen The combined pumping rate from the three
wells was 33.6 m3/h, as determined by daily readings of
in-line flow meters The pumped water was discharged 50 m
downstream in the channel Observation wells were
con-structed, in May 2006, by water-jet drilling and equipped
with an Ø60 mm PVC casing and a 1 m long screen A
quartz sand filter pack and a clay seal were installed around
the screen and casing, respectively Immediately after
com-pletion, the well was pumped to remove the water affected
by the drilling The top end of the PVC-casing was sealed
by a rubber cap, to prevent the entrance of surface water
during flooding When the channel stage was high, the
PVC casing was extended with a flexible spacer with a
PVC thread, and the casing was then elongated with one
meter sections of Ø60 mm PVC, each threaded at both
ends The thread connections were tested to be waterproof
at pressure differences of >3 m water column The top end
sections were sealed by rubber caps Sampling could then be
conducted from a small boat Water levels in the channel and observation wells N1 and N4 were recorded by data loggers with an estimated uncertainty of 3–5 cm
2.1.2 Water sampling and field analysis Sampling was conducted with a submersible Grundfos MP1 pump and Ø10 mm polyethylene (PE) tubing Before sampling, three well volumes of water were removed with the pump positioned in the upper part of the borehole The pumping rate was then decreased, and the pump was lowered to a position at the top of the screen Temperature,
O2, pH, and electrical conductivity (EC) were measured by probes mounted in a flow cell connected directly to the sam-pling tube The measurements were carried out with a WTW Multi197i instrument using a WTW Tetracon 96
EC probe, a WTW SenTix 41 pH electrode and for dis-solved O2 a WTW EO 196-1.5 electrode Samples for
CH4were injected directly from the sampling tube through
a butyl rubber stopper into a preweighed evacuated glass vial (Labco 819W), which was immediately frozen in an up-side down position thereby trapping the gas phase in the head space above the sampled water Samples for all other parameters were collected in 50 mL polypropylene (PP) syringes and filtered through 0.2 lm cellulose acetate filters into 20 mL PE vials Aqueous As(V) and As(III) were sep-arated by passing freshly collected sample through first a 0.2 lm membrane filter and then a disposable anion ex-change cartridge at a rate of approximately 6 mL/min using
a syringe The cartridges contained 0.8 g aluminosilicate adsorbent that selectively adsorbs As(V) but not As(III) (Meng and Wang, 1998) The combined syringe, filter and cartridge were flushed five times with N2before processing
a sample Samples were acidified by 0.5 vol.% HNO3 supra-pur As(V) was calculated as the difference between As(III) and As-total
Fig 1 (a) Location of the field site on the banks of the Red River 30 km upstream of Hanoi The dotted lines indicate channels that are fully connected to the main river course during the rainy season but may become disconnected during the dry season The location marked N indicates were the forced gradient experiment was carried out (b) Situation sketch of the field site for the forced gradient experiment, at location N.
Trang 4Fe(II), phosphate and dissolved sulfide were measured
spectrophotometrically using a Hach DR/2010 instrument
and the ferrozine (Stookey, 1970), molybdate blue and
methylene blue methods (Cline, 1967) with detection limits
of 0.2, 1.1, and 0.5 lM, respectively Alkalinity was
deter-mined by Gran-titration (Stumm and Morgan, 1981)
Sam-ples for Si, Na+, Ca2+, Mg2+, and K+were preserved by
adding 0.5 vol.% suprapur HNO3and refrigerated Samples
for NH4+, NO3, Cland SO4were frozen immediately
after sampling Channel water samples were collected 2–
4 m away from the shoreline and the electrodes were
im-mersed directly into the water at a depth of 10–20 cm
2.1.3 Laboratory analysis
Cations were analyzed by flame AAS on a Shimadzu
AAS 6800 Arsenic was determined on the same instrument
using a HVG hydride generator and a graphite furnace
An-ions were analyzed by ion chromatography on a Shimadzu
LC20AD/HIC-20ASuper NH4+ and Si were determined
spectrophotometrically using respectively the nitroprusside
and the ammonium molybdate methods Head space CH4
was determined on a Shimadzu GC-14A with a 1 m packed
column (3% SP1500, Carbopack B) and a FID detector
The aqueous CH4 concentration was calculated using
Henry’s law The detection limits were: As 0.013 lM,
Mn2+0.91 lM, Ca2+0.50 lM, NH4+ 5.6 lM, Si 0.2 lM,
NO30.8 lM, SO42.1 lM, and CH41.9 lM
2.2 Elution of arsenic from the sediment
The sediment consisting of grey to dark grey fine sand
was collected by Postma et al (2007), using a sediment
corer, at the nearby H-transect (Fig 1) in the anoxic part
of the Holocene aquifer at 15.3 m depth This and other
sediments from the H-transect have been characterized by
Postma et al (2007, 2010) In a N2 filled glove box, 12 g
wet sediment (8.6 g dry weight) was transferred from the
core into a 100 mL septum bottle Added to the bottle
was an aliquot of a carefully deoxygenated 10 mM
NaHCO3solution, prepared from boiled water, and
equili-brated with a 5% CO2/95% N2gas mixture, corresponding
to the 10 meq/L alkalinity and pH 7 observed in the field at the coring site (Postma et al., 2007) Possible trace levels of oxygen in the CO2/N2gas were trapped by passing the gas through an acetate buffered FeSO4solution The bottle was capped by a 10 mm thick rubber stopper, removed from the glove box, and stripped for glove box-H2by bubbling with the deoxygenated CO2/N2gas The bottle was shaken side-to-side for two hours, and then left in an upright position for settling of the suspended particles for one to 5 days Samples of the supernatant were withdrawn using a needle and a PP syringe, passed through a 0.2 lm syringe filter and stored refrigerated in PE vials After removal of most of the supernatant, the bottle received a new aliquot of 40–90 mL
of the 10 mM NaHCO3solution The mass of samples and amendments was determined gravimetrically This cycle was repeated five times Arsenic was measured in each batch of elutant with the As concentration decreasing for each subsequent step A sub-sample of the sediment was heated to 110°C for 24 h to determine the water content
2.3 Modeling
Surface complexation modeling was carried out using PHREEQC (version 2.16; Parkhurst and Appelo, 1999) For ferrihydrite, we used the two layer SCM byDzombak and Morel (1990)extended with surface species for carbon-ate (Appelo and De Vet, 2003), silicate (Swedlund and Webster, 1999) and Fe(II) (Liger et al., 1999; Appelo
et al., 2002) For goethite, the charge distribution-multisite complexation (CD MUSIC) model (Hiemstra and Van Rie-msdijk, 1996) was used with the surface complexes com-piled for PHREEQC by David Kinniburgh in the code PhreePlot (www.phreeplot.org) The surface complexation reactions, their affinity constants and source are listed in
Table EA-1in the Electronic Annex The capacitances for the 0–1 and 1–2 planes are C1= 0.85 F/m2 and
C2= 0.75 F/m2, respectively (Stachowicz et al., 2008) The performance of the CD MUSIC model in PHREEQC was verified by successfully reproducing the model lines
Fig 2 Cross section of the field site for the forced gradient experiment ( Fig 1 ), showing fine grained overbank deposits (hatched) on top of the sandy aquifer (white area) The position of three 5 m deep hand drillings in the flat channel bottom is indicated; the rightmost intersected presumed lenses of sand are indicated by small dots Well screens are indicated by horizontal dashes The well screens intersect a layer of coarse sand with pebbles indicated by the dotted interval at 7 to 9 masl The annual maximum (wet season) and minimum (dry season) groundwater table and channel water level is indicated.
Trang 5for single-sorbate As(III) adsorption (Fig 4inStachowicz
et al., 2006) and H3AsO3–PO4–Ca2+ triple-sorbate
co-adsorption (Fig 7inStachowicz et al., 2008)
The model for adsorption on Pleistocene aquifer
sedi-ment from Bangladesh, byStollenwerk et al (2007), uses
experimentally determined adsorption constants for
As(III), As(V), PO4 , and HCO3, in combination with
theDzombak and Morel (1990)model for ferrihydrite for
other components and for electrostatic effects The new
reactions were formulated for the Hfo_w sites (termed Sites
inStollenwerk et al (2007)) and added to the PHREEQC
input file overriding the standard Dzombak and Morel
(1990) database The three models will be referred to in
the text as the D&M model, the CD MUSIC model and
theStollenwerk et al (2007)model The standardDzombak
and Morel (1990)database extended with surface species
for carbonate, silicate, and Fe(II) (see above) will be
re-ferred to as the D&M database Aqueous speciations were
for all three models conducted using the wateq4f.dat
data-base, although for the D&M and CD MUSIC models this
database was modified with the aqueous As speciation
con-stants fromLangmuir et al (2006)
2.4 Hydrogeology
The forced gradient experiment was conducted at the
bank of a side channel of the Red River (Fig 1) Here,
the Holocene aquifer consists of fine sand to pebbles
under-neath a surface layer of silty–clayey material and fine sand
(Fig 2) Drillings in the channel bottom (Fig 2) reached a
sand layer after penetrating the silty–clayey unit, suggesting
that the sandy aquifer extends beneath the channel bottom
In the drilling furthest away from the wells, a sandy
se-quence with only few thin clay layers was observed
The surface water-groundwater interaction at the field
site has been described byLarsen et al (2008) During the
dry season the river stage decreases, the channel becomes
disconnected from the Red River (Figs.1a and 2) and
al-most dries out near the end of the dry season (Fig 2) The groundwater flow direction is NE towards the Red River During the rainy season, the Red River stage rapidly increases and the channel and the Red River become con-nected (Fig 2) Fig 3 displays the channel stage and the hydraulic head in observation wells N1 and N4 (Figs 1 and 2) from 25 days before to 55 days after the start of pumping on 7 July 2007 In the period from day 25 to day 34, the hydraulic head in the channel is higher than
in wells N1 and N4, but on day 34 the channel stage de-creases to below the hydraulic head in the wells From day zero to around day 35, N1 has a higher hydraulic head than N4 (Fig 3) indicating flow from the channel towards the pumping wells After 30 days of pumping, the channel stage drops causing the hydraulic gradient to reverse and the groundwater to flow in the opposite direction The nat-ural hydraulic gradient between the aquifer and the channel now overrides the effect of the pumping
3 RESULTS
Table 1displays the composition of the pristine ground-water and the composition of the channel ground-water after it has become flooded by the rising river The groundwater is an-oxic with 2.5 lM As(III) and 0.6 lM As(V), the EC is
784 lS/cm with Ca2+and Mg2+ as the main cations that are charge-balanced by alkalinity The channel water is oxic, contains no As and has a low EC of 168 lS/cm (aver-age for first 9 days of pumping) In terms of aqueous spe-cies, other than As, that may adsorb, one may note the much higher concentration of Fe(II), HCO3 , PO4 and
Si in the pristine groundwater as compared to the channel water.Fig 4displays the changes in concentration during the pumping experiment in observation wells N1 and N4 (Figs 1 and 2) Nine days after the start of pumping, the
EC begins to decrease, indicating the first appearance of
Fig 3 The water level in the channel and in observation wells
N1and N4 ( Figs 1 and 2 ) from 25 days before to 55 days after the
start of pumping on 7 July 2007 During the first 34 days of
pumping the water level in the channel is above that of the wells
and channel water intrudes into the aquifer After day 34, the
channel water level drops quickly to below the groundwater level
and the flow direction is reversed in spite of pumping.
Table 1 The composition of pristine groundwater (boring N1 ( Fig 1 b) on day 9 after initiation of pumping) and of channel water (average for first 9 days).
Parameter Groundwater
(N1, day 9)
Channel water (Avg day 1 to 9)
Unit
Alkalinity 8.80 1.53 meq/L
Trang 6water derived from the channel The curves for EC are
shifted by approximately 5 days between the well closest
to the channel (N1) and the well (N4) closest to the
pump-ing well (Fig 2) The hydraulic gradient reverses on day 34
(Fig 3) where a minimum EC of 340 lS/cm is observed in
both N1 and N4 (Fig 4) Thereafter, the EC increases
again, but this time first in N4 and thereafter in N1 The
EC in the two observation wells has returned to the initial
values on day 54 The EC of the channel water increases
after day 34, and approaches that of the groundwater This
reflects the discharge of groundwater to the channel, which
becomes of increasing importance subsequent to day 34,
when the hydraulic head of the aquifer started to exceed
that of the channel stage (Fig 3) During the flooding
sea-son in the year after the observations inFigs 3 and 4were
made, with no pumping, the EC in N1 did not shown any
decrease, indicating that the observed changes are due to
the pumping and not a natural phenomenon For most
ma-jor ions, the changes in concentration follow those in EC
(Fig 4andEA-1) Even though the channel water is oxic,
the groundwater in N1 and N4 remained anoxic, and
meth-anic, during the whole experiment As(III) follows the
gen-eral trend with the lowest concentrations occurring near
day 34 but for As(V) it is hard to identify a clear trend in
the scatter of the data While the pH in the channel water
was close to 7.7, the pH in N1 and N4 remained very close
to 7.0 ± 0.1, as in the pristine groundwater
To separate the effects of mixing and chemical reactions
on the water chemistry, conservative mixing calculations were carried out using channel- and groundwater inTable 1
as endmembers The fraction of channel water mixed into the groundwater, fchannel, was calculated using alkalinity, which at neutral pH is close to the HCO3 anion concentra-tion, as inert component (Appelo and Postma, 2005):
fchannel¼ðmalk;sample malk;groundwaterÞ
ðmalk;channel malk;groundwaterÞ; ð1Þ where m denote concentration Alternatively EC could be used instead of alkalinity in Eq.(1); in practice the resulting
fchannel are barely distinguishable from those calculated using alkalinity Using fchannel the concentration expected for conservative mixing for each solute, i, is calculated from:
mi;mix¼ fchannel mi;channelþ ð1 fchannelÞ mi;groundwater ð2Þ The results of the mixing calculations are included as the lines in Fig 4 The maximum fraction of channel water in the waters sampled in N1 and N4 is 0.74.Fig 4shows that for most components the variation is very well described by the mixing lines The observation that conservative mixing between the channel water and groundwater may explain most of the variation indicates that the system has a high dispersivity Rather than a simple displacement of ground-water by the intruding channel ground-water, it appears that
chan-Fig 4 Changes in water chemistry in observation boreholes N1 (filled symbols) and N4 (open symbols) and in the channel (crosses) For location see Fig 1 The water composition calculated for conservative mixing between river- and groundwater is indicated for N1 by the solid line and N4 by the broken line.
Trang 7nel water enters through highly permeable layers and
be-comes mixed with the groundwater residing in the adjacent
layers However,Fig 4also shows that for As(III), As(V),
PO4and Si there are distinct differences between
concen-trations calculated for conservative mixing and the
mea-sured values, although the meamea-sured values for As(V)
appear quite scattered The calculated values for Fe(II)
are also somewhat different from the measured values
These differences must be due to chemical reactions taking
place in the aquifer
Mobile As, present in the anoxic aquifer sediment, was
determined by repeated elution with a 10 mM NaHCO3
solution Only As(III) was found to be present in the
elu-tant.Fig 5displays the As concentration in the elutant
ver-sus the cumulative amount of eluted As It shows a
near-linear relation which extrapolates to a concentration of
mo-bile As(III) of 7.7 nmol/g sediment The continuous
de-crease in the As concentration in the repeated elution
steps does not indicate the dissolution of a phase that
con-trols the aqueous As concentration by a mineral
equilib-rium It is more consistent with As being desorbed from
the sediment surface
4 DISCUSSION
4.1 The modeled speciation of adsorbed arsenic
4.1.1 Site density normalization
When applying a surface complexation model to a
sedi-ment, assigning a plausible surface site density is an
impor-tant, though non-trivial, step For example, surface site
densities in previous studies have been calculated from
the amount of 1 M HCl-extracted Fe (Swartz et al., 2004)
or fitted to water chemistry data (Postma et al., 2007) In
the present study we elute As from the sediment (Fig 5),
and thereby obtain the amount of As adsorbed in
equilib-rium with the groundwater composition at the point of
sed-iment sampling (Table 2 in Postma et al (2007)) Under
in situ conditions, the mobile amount of As(III) of 7.7 nmol As/g sed (Fig 5) corresponds to 7.7 nmol As/g 6183 g/
L = 47.6 lmol adsorbed As(III) per liter of contacting groundwater, assuming a porosity of 0.3 and grain density
of 2.65 g/cm3 By varying the amount of sorbent in the SCMs (grams of ferrihydrite, goethite or Stollenwerk
et al (2007)’s Pleistocene sediment), the number of sites reactive towards As(III) in the CD MUSIC, D&M and
Stollenwerk et al (2007) models was normalized, so that each SCM produced the same in situ adsorbed As(III) con-centration of 47.6 lmol/L groundwater when at equilib-rium with the water composition at the point of sediment sampling The resulting concentration of surface sites for which a surface reaction with As(III) is defined is for the
CD MUSIC model 0.17, for the D&M model 0.28, and for the Stollenwerk et al (2007) model 0.51 lmol sites/g sediment; these respective site concentrations were applied
in all model simulations presented inFigs 7–13 (Further details are available in the Electronic Annex.) Accordingly, the range of the concentration of surface sites that poten-tially may adsorb As(III) in the three models is quite small For comparison, the model predictions made byBGS and DPHE (2001) used 0.74 lmol sites/g For the sediment eluted byPolizzotto et al (2006), Swartz et al (2004)fitted
a sorption capacity of 0.11 lmol sites/g (calculated using a
Fig 6 Adsorption of As on Bangladesh aquifer sediments in the absence of competing anions, except for the Stollenwerk et al (2007) data points for Pleistocene sediment which were measured in
a solution containing 70 mg/L Ca2+, 24 mg/L Mg2+and 194 mg/L
Cl at pH 6.8 The solid lines are predicted by the SCM by
Stollenwerk et al (2007) , derived from these and additional experiments Also included are adsorption data from Itai et al (2010) for As(III) on three different Holocene sediments measured
at pH 7.3 in a 10 mM MOPS (3-morpholinopropanesulfonic acid) buffer solution The data points measured by Nath et al (2009) are for sorption of As(V) in 1 mM NaNO 3 at pH 7.5 or 7.7 on Holocene sediment.
Fig 5 Desorption of As from aquifer sediment from the nearby
H-transect ( Fig 1 ) by repeated equilibration with 10 mM NaHCO 3
solutions The As concentration in each subsequent batch of
extractant is shown on the X-axis The cumulative amount of
eluted As is shown on the Y-axis Only As(III) was found to
desorb The extrapolated content of desorbable As is 7.7 nmol/g.
Trang 8porosity of 0.3 and a grain density of 2.65 g/cm3) Much
lower surface site concentrations of 0.010 and 0.016 lmol
sites/g were applied by Appelo et al (2002) and Postma
et al (2007), respectively Because of competitive sorption
with other solute ions, only the high surface site
concentra-tions allow for the presence of the several nanomoles of
ad-sorbed As per gram of sediment (Fig 5)
4.1.2 SCM for aquifer sediment
In principle, models developed according to the general
composite approach are site specific We therefore need to
assess whether it is reasonable to apply the Stollenwerk
et al (2007)model for Pleistocene aquifer sediment to data
from our reduced Holocene aquifer setting Fig 6 shows
the adsorption data for As(V) and As(III) of Stollenwerk
et al (2007), measured on Pleistocene aquifer sediment
from Bangladesh in the absence of competing anions The
results show much stronger adsorption for As(V) than for
As(III) The solid lines shown inFig 6 are predicted by
the model by Stollenwerk et al (2007) Itai et al (2010)
measured the adsorption of As(III) on three Holocene
aqui-fer sediment samples from Bangladesh and Nath et al
(2009) measured the adsorption of As(V) on Holocene
aquifer sediment from West Bengal, India Data from these
studies are included inFig 6 Interestingly, the adsorption
data for As(III) on Holocene sediment byItai et al (2010)
plot close to the As(III) adsorption data for Pleistocene
sed-iment byStollenwerk et al (2007) This suggests that the
affinity of Holocene and Pleistocene sediments for As(III)
is comparable A similar conclusion was reached by Itai
et al (2010)based on discrete K0
dmeasurements on both Holocene and Pleistocene sediments For As sorption on
Holocene sediments, Itai et al (2010)found significantly
higher K0
dvalues for As(V) than for As(III), in agreement
with the results ofStollenwerk et al (2007)for Pleistocene
sediment The results for As(V) sorption on Holocene
sed-iments byNath et al (2009)(Fig 6) indicate less sorption
than found for the Pleistocene sediments Nath et al (2009), however, did not determine the adsorption of As(III) to their sediment, so a direct comparison of As(III) and As(V) sorption is not available It is also obvious that sorption properties must differ among sediments, depend-ing on mineralogy, surface area, organic matter content, etc But in this case the results inFig 6suggest that these differences are not very big and therefore it is reasonable
to test the Stollenwerk et al (2007)model for Pleistocene aquifer sediments on our Holocene aquifer system in order
to compare the behavior of a natural sediment with that of synthetic Fe-oxides predicted by the D&M and CD MUSIC models
4.1.3 Comparison of modeled surface speciation The three SCMs, normalized in the previous section, were used to calculate the composition of the sediment sur-face in equilibrium with the pristine groundwater (Table 1) The results of these calculations are shown inFig 7, for the surface sites that are able to adsorb As(III), and are surpris-ingly different The groundwater contains about five times
as much As(III) compared to As(V) (Table 1) The CD MUSIC model for goethite, however, shows a higher sur-face concentration for As(V) as compared to As(III) indi-cating a much stronger sorption of As(V) than of As(III)
In contrast, the D&M model for ferrihydrite shows stronger sorption for As(III) than for As(V) Also in terms of ions competing for sites with As on the surfaces of the synthetic Fe-oxides, ferrihydrite and goethite, the results are very dif-ferent For ferrihydrite, the D&M model calculates as the main surface complexes, apart from protonated and depro-tonated sites, bicarbonate (52%) and silica (27%) while phosphate surface complexes constitute only 5% For goe-thite, however, the CD MUSIC model calculates that 50% of the sites are occupied by phosphate, with Fe(II) sur-face complexes as the second most important species (30%) while both silica and bicarbonate cover less than 1% It
Fig 7 The surface speciation at equilibrium with pristine groundwater ( Table 1 ), calculated for the sites that are able to adsorb arsenic, with the D&M model for ferrihydrite, the CD MUSIC model for goethite and the model for Pleistocene aquifer sediment by Stollenwerk et al (2007) , using the normalized total site concentration of, respectively, Hfo_w = 0.28 (D&M model), Goe_uni = 0.17 (CD MUSIC model), and Site = 0.51 ( Stollenwerk et al (2007) model) lmol sites/g of sediment The field for each element in the pie diagram may cover several surface species The field denoted “H” indicates the sum of all protonated or deprotonated surface sites.
Trang 9should further be noted that in the CD MUSIC model,
As(III) binds almost exclusively (99%) to the surface as
a ternary As(III)-Fe(II) complex, Goe_uniOAs(OH)3Fe+0.5
(Hiemstra and van Riemsdijk, 2007), which covers 2% of
the surface sites
The surface speciation calculated with theStollenwerk
et al (2007)model (Fig 7) resembles most that of the
goe-thite SCM Similar to goegoe-thite, the surface affinity for
As(V) is much stronger than for As(III) Phosphate is again
the most important ion competing for surface sites with a
41% coverage, while bicarbonate complexes are
insignifi-cant On the other hand, silica complexes cover 18% of
the surface sites and in this respect it resembles more the
ferrihydrite surface It is furthermore conspicuous that
pro-tonated and depropro-tonated surface sites are more important
in theStollenwerk et al (2007) model as compared to the
two other SCMs In the SCM byStollenwerk et al (2007)
the dissociation constant for surface hydroxyls, pKa2, is
nearly two orders of magnitude larger than in the D&M
model, and consequently deprotonated surface sites are
important (8% coverage) in the former model, while
negli-gible in the latter
4.2 Sensitivity analysis
The previous section has shown that the different SCMs
produce very different results in terms of the amount of
As(III) and As(V) adsorbed and the predominant ions
com-peting for surface sites Because the relations between the
aqueous solutions and the surface complexes are complex
and non-linear, it is not easy to perceive how the different
models would react towards changes in aqueous
composi-tion Therefore, a number of sensitivity tests were carried
out with the different models in order to test their behavior
with the range of groundwater compositions that are
ob-served in S.E Asia
In the first test, the distribution coefficients were
calcu-lated for As(III) and As(V) as a function of the
concentra-tion with the three models (Fig 8), using the pristine
groundwater composition (Table 1) and the normalized
surface site concentrations given in the previous section,
but varying the As(V) or As(III) concentration For ferrihy-drite, the D&M model predicts that As(III) sorbs about twice as strongly as As(V) However, for goethite the CD MUSIC model predicts that As(V) adsorption is more than three times stronger than for As(III) In the Stollenwerk
et al (2007)model for aquifer sediment, the affinity of the surface for As(V) is even higher The As(III) and As(V) concentrations in the pristine groundwater (Table 1) are indicated on the isotherms, inFig 8, except for As(V) in the Stollenwerk et al (2007) model, which is completely out of range The results obtained with the different models have major implications for the amount of mobile As, aqueous plus adsorbed, that is predicted to be present in the system This is particularly apparent for As(V) where the low groundwater concentration of 0.6 lM relates to a low mobile pool of As(V) in the D&M model, while the
CD MUSIC model and particularly theStollenwerk et al (2007) model suggest that a lot of surface bound, poten-tially mobilizable As(V) is present in the system, even when the aqueous concentration is low Finally, the large differ-ences in the calculated distribution coefficients have a major bearing on the retardation and thereby on the mobility of As(III) and As(V) Retardation is defined as R = 1 + Kd where Kd= [sorbate concentration in mol/L]/[solute con-centration in mol/L] and therefore dimensionless (Appelo and Postma, 2005) All three models show a retardation for As(III) in the narrow range R = 9–20 (Kd= 8–19;
Fig 8), while for As(V) the D&M model predict it to be very mobile (R = 5–10; Kd= 4–9) while the two other SCMs predict As(V) to have a decisively low mobility (R = 28–204; Kd= 27–203)!
Fig 9displays the adsorption of As(III) and As(V) as a function of pH, using the same aqueous composition and surface properties as before but calculated for varying
pH For ferrihydrite, the D&M model indicates that As sorption as a function of pH is complex At a pH higher than 8.5, sorption of both As(III) and As(V) strongly de-creases For As(III), Hfo_wH2AsO3 is the only surface complexed species and this species becomes less stable as
pH decreases For As(V) the surface species Hfo_wH
2-AsO4, Hfo_wHAsO4 and Hfo_wOHAsO43 are
sequen-Fig 8 Adsorption isotherms calculated for the groundwater in Table 1 , while varying the As(III) or As(V) concentration, with the D&M model for ferrihydrite, the CD MUSIC model for goethite and the model for Pleistocene aquifer sediment by Stollenwerk et al (2007) The squares indicate the measured As(III) and As(V) concentrations ( Table 1 ) As(V) in the Stollenwerk et al (2007) model runs out of range For normalized site densities refer to the caption of Fig 7
Trang 10tially formed from low to high pH Because the
Hfo_wHAsO4 species has a comparatively lower
forma-tion constant there is a depression in the As(V) sorpforma-tion
curve near pH 6 For goethite, the CD MUSIC model
pre-dicts that in the range pH 6.5–10 the ternary As(III)–Fe(II)
surface complex Goe_uniOAs(OH)3Fe+0.5 mediates a
strong As(III) adsorption Outside this pH range, As(III)
sorption becomes weak with the species
Goe_uni-OAs(OH)2 0.5 as most important at lower pH and
(Goe_uniO)2AsOH at higher pH For As(V), close to
100% remains adsorbed up to pH 8, while adsorption
rap-idly decreases towards higher pH In theStollenwerk et al
(2007)model, adsorbed As(III) is close to 100% above pH 7
with Hfo_wAsO3 2as the only important surface species,
which becomes unstable below pH 7 The strong affinity
of the surface for As(V) in this model, is also displayed in
the sorption behavior as a function of pH since close to
100% of As(V) stays adsorbed over the entire pH range
The pristine groundwater (Table 1) has a pH of 6.98 and
similarly most waters in As-bearing aquifers in S.E Asia
have near neutral pH However, asFig 9illustrates, in
par-ticular the CD MUSIC and theStollenwerk et al (2007)
models predict that even small decreases in pH may
signif-icantly lower As(III) adsorption
InFig 7, phosphate was found to adsorb very strongly according to both the CD MUSIC model for goethite and the Stollenwerk et al (2007) model for aquifer sediment, while the D&M model for ferrihydrite predicts much less
PO4 adsorption In groundwater from our field site the
PO4 concentration is only 8 lM, but for Bangladesh groundwaterSwartz et al (2004)reported a PO4 concen-tration in the range 26–68 lM Accordingly, it can be antic-ipated that variations in the phosphate concentration of the groundwater may have a strong influence on the adsorption
or desorption of As Fig 10shows the results of model runs, where PO4 (as H3PO4) was either added or removed from the system, while keeping pH fixed at 6.98 Zero on the X-axis corresponds to the measured groundwater con-centrations for As(V), As(III) and PO4 (Table 1) The varying PO4 and derived As concentrations are plotted
on the Y-axis The amount of PO4added was adjusted
to reach an aqueous PO4 concentration between 60 and
70 lM, corresponding to the Bangladesh situation reported
bySwartz et al (2004) In the case of ferrihydrite, the D&M model calculates that 300 lM PO4 must be added to in-crease the aqueous PO4concentration from 8 to 68 lM The difference (300–68 = 232 lM) is the PO4 that is ad-sorbed on the surface Because of PO4 adsorption, the
Fig 9 The effect of pH on the adsorption of As(III) and As(V) calculated for the groundwater in Table 1 (while varying the pH) with the D&M model for ferrihydrite, the CD MUSIC model for goethite and the model for Pleistocene aquifer sediment by Stollenwerk et al (2007) For normalized site densities refer to the caption of Fig 7
Fig 10 The effect of the aqueous PO 4 concentration on the adsorption of As(III) and As(V) calculated for the groundwater in Table 1 by adding or removing PO 4 from the system The amount of PO 4 added or removed in the model calculations with the D&M model for ferrihydrite, the CD MUSIC model for goethite and the model for Pleistocene aquifer sediment by Stollenwerk et al (2007) is shown towards right or left on the X-axis At zero on the X-axis the pristine groundwater concentrations of As(III), As(V) and PO 4 can be read from the Y-axis For normalized site densities refer to the caption of Fig 7