Edinburgh Research Explorer Trace metals in CO2-rich Green River Springs, Utah, USA: an analogue for engineered storage

Sedimentary rocks with high natural CO 2 concentrations provide invaluable analogues for the long-term engineered storage of CO 2 . Some previous studies have reported high trace metal concentrations in sandstone aquifers exposed to CO 2 , a cause for concern should stored CO 2 leak into underground sources of drinking water. However, the intensively studied Jurassic sandstone aquifer in the San Rafael – Green River (Utah, USA) area has trace metal concentrations that are within USA Environmental Protection Agency’s limits for drinking water. Exceptions are As which is plausibly introduced into the aquifer by saline brines external to the aquifer, and salinity which largely is. This shows that CO 2 in aquifers does not inevitably cause trace metal contamination. CO 2 - water-rock batch experiments elucidated the controls on the trace metal concentrations. After the addition of CO 2 , the experiments reproduce well Cu, Cd, Hg, Ni and Zn, with less good agreement for Cr and Pb although these are still low compared to drinking water standards. Major cations used as fingerprints for mobilisation mechanisms suggest that the trace metals are largely derived by desorption, possibly from grain-coating Fe-oxides, rather than by the dissolution of mineral phases. Possible exceptions are Pb and Ni, plus As which is derived from saline brines.

Drinkable water is a valuable resource that must be protected in the event that CO2 is stored underground below aquifers that are, or could be, used as underground sources of drinking water.Some published experiments have exposed samples of aquifer sandstones to CO2-rich waters and have suggested that high levels of toxic metals may be released (e.g.Little and Jackson, 2010).
Hence, here we present trace metal analyses of both natural porewaters and experimental brines to assess the trace-metal loads associated with CO2 -rich porewaters on the Colorado Plateau, USA.
The focus is on eight potentially toxic metals (As, Cd, Co, Cu, Hg, Ni, Pb, Zn) which are the subject of legislation within the European Union.Concentration limits for most of these metals are also set by the USA Environmental Protection Agency (EPA) though Ni and Zn are not subject to enforceable regulation (Environmental Protection Agency 2022A; Table 1).Zinc is listed in the EPA Secondary Drinking Water Standards (Environmental Protection Agency 2022B) due to it imparting a metallic taste to the water.However, these are guidelines to assist public water systems in managing their drinking water for aesthetic considerations and are not legally enforceable.
Two concentration levels are defined for each potential contaminant (Environmental Protection Agency 2022A).The Maximum Contaminant Level Goal (MCLG) is the level of a contaminant below which there is no known or expected risk to health.MCLGs allow for a margin of safety but are nonenforceable public health goals.Maximum Contaminant Level (MCL) is the highest level of a contaminant that is allowed in drinking water; MCLs are set as close to MCLGs as feasible using the best available treatment technology and taking cost into consideration.MCL's are legally enforceable standards in the USA.The secondary MCL (for Zn) is not legally enforceable.The salinity of the drinking water (expressed as total dissolved solids or TDS) is also regulated by the EPA as a secondary drinking water standard, with an upper limit of 500 ppm.
The area around the town of Green River on the Colorado Plateau in Utah, USA, has been extensively studied as an analogue for engineered CO2 storage, as CO2-rich springs and a cold-water 'geyser' (Crystal Geyser) are accessible.The regional hydrogeology is well described by Kampmann et al. (2009).The springs are largely sourced from the Navajo Sandstone which forms the regionally extensive Lower Jurassic 'N-aquifer' along with the Wingate Sandstone and Kayenta Formation (Kampmann et al., 2009).The aquifer is mostly supplied by meteoric water at outcrop in the San Rafael Swell (a broad anticline) to the west of the springs.Additionally, brine enters the aquifers via faults extending down to a Palaeozoic aquifer, formed by Lower Pennsylvanian and Mississippian limestone and dolomite formations (Hanshaw and Hill, 1969).Salinity in the brine is derived from the largely evaporitic Pennsylvanian Paradox Formation (Wilkinson et al., 2009).The erupted CO2 is mostly derived from crustal sources, with a 1-20% contribution from the mantle, with the CO2 already mixed within the Paleozoic aquifer; Tenmile geyser is an exception with 16 to 99% mantle CO2 (Wilkinson et al., 2009).Based on changes in pH, injection of CO2 into the Jurassic aquifer occurs close to the Green River Airport Well and near Small Bubbling Spring (Kampmann et al., 2009).The Green River Airport well, sourced from exploration drill-hole Grand Fault 14-24, emits water from the Navajo Sandstone and is important as it is the spring closest to the recharge area of the aquifer (Kampman et al., 2014).It contains CO2 but no detectable brine from deeper formations (Kampman et al., 2009).At Crystal Geyser, a well-studied cold water 'geyser', the majority of the water is derived from sandstones of the Jurassic Navajo (53 -65%) and Entrada formations (45 -33%), with only 1-2% Paradox brine based on major element geochemistry (Han et al., 2017) or approximately 10 % compared to approximately 10 -20% for the other springs based on stable isotope analysis (Wilkinson et al., 2009).Previous work has concentrated on the geochemical evolution of the subsurface brines, especially Kampman et al. (2009Kampman et al. ( , 2012Kampman et al. ( , 2014)), and the origin of the spring waters.This is not repeated here, where the mobilisation of potentially toxic trace metals is the main focus.
The aim of the study is to determine which trace elements are mobilised by CO2 in this area, and by extension in sites of engineered CO2 storage, including any associated leakage.Both natural water samples and laboratory experiments are used to attempt to determine the source and controls on these elements.No analyses have been previously reported for most of the elements discussed here.

Field methods
As the Entrada, Wingate and Navajo sandstones are not exposed at the surface close to the sampled springs, they were sampled at the San Rafael Swell, located to the west-southwest of Green River.
Here, the formations outcrop as the limbs on a large and spectacular anticline, before dipping eastwards into first a syncline and then a shallow anticline beneath the Green River springs.Four rock samples were collected at roadside locations adjacent to Interstate Route 70 from a cutting through the eastern limb of the San Rafael Swell, around 18 km west-southwest of Green River (Table 2 and Fig. 1).The stratigraphy of the sediments was worked out using the 1980 edition of the Geologic Map of Utah (1:500,000 scale) and accompanying sheet of representative stratigraphic columns and cross sections.Hammer samples of approximately 10 x 10 x 10 cm in size were taken.
Sample S1 was chosen to include both a red/pink and white sandstone, and the boundary between them (Table 2).The sandstones at each location are relatively homogenous though Sample S3 included a thin (mm-scale) weathered 'crust' of red/pink sandstone, with an underlying white sandstone.Interactive maps with photographs of the outcrops are at https://goo.gl/wLpE1C.

Water Sample Collection
Figure 1 shows the location of the sampled springs.An interactive map with photographs is at https://goo.gl/a6D9un.Sampling of waters from 10 cold water springs was carried out on the 4 -6 th of October 2013.At each spring location, 8 separate water samples were collected and treated (Table 3).The sample containers, syringes and filters were pre-cleaned by the University of Texas (UT) using the following regime.Polypropylene bottles for cations and isotopes were rinsed with deionised (D.I.) water then soaked in micro-filtered water for 2 days; rinsed with D.I. water then soaked in 30% trace metal grade HNO3 for 2 days; and finally rinsed in D.I. water then soaked in 18 M D.I. water for 2 days before drying.Polypropylene bottles for anions, amber glass volatile organic analysis (VOA) vials and syringes were rinsed with D.I. then soaked in micro-filtered water for 2 days; rinsed with D.I. then soaked in 18 M D.I. water for 2 days and then dried.Filters were rinsed with D.I. water then soaked in 18 M D.I. water for 2 days.The sampling procedure for small (< 60 mL) bottles and vials was as follows: a full syringe of water was drawn from the geyser, as close to CO2 bubbling as possible, then passed through the filter and discarded to the ground.This was repeated three times.The water was then passed through the filter into each sample container, rinsed and discarded three times; finally the water sample was passed through the filter to fill each sample container, leaving no headspace.Acids and reagents were added using a pipettor, to the volumes and concentrations in Table 3.The containers were sealed with a cap and Parafilm TM before labelling.
For the batch experiments, water was collected in 1000 mL polypropylene bottles that were precleaned with acid.If the geyser pool was large enough, the bottle was filled by submerging it and the fluid discarded to the ground three times, before filling to the brim and capping it a final time.The cap was sealed with Parafilm™.If the geyser was small, for example Little Bubbling Spring, then the bottle was filled to the brim using the syringe (unfiltered) instead of submerging it.The sampling for mercury (Hg) was carried out based on the method outlined in the UK Government Department for Energy and Climate Change (DECC) sampling guidance issued to North Sea oil and gas operators (Department of Energy & Climate Change, 2014).Samples were therefore collected in glass vials and preserved using 1M sulphuric acid and 1M potassium dichromate.Major cations (Ca, K, Li, Mg, Na, NH4 + ) and anions (Br, Cl, F , NO2 -, NO3 -, PO4 -2 , SO4 2-) of water samples were analysed on two Dionex ICS-1100 Ion Chromatography systems equipped with an AS-AP auto sampler at the Bureau of Economic Geology, University of Texas (UT) laboratories.Samples were diluted by a factor of 10 after collection, in order that Ca and SO4 2-concentrations did not exceed 500 mg/L to conform with the limits of the analytical equipment.Repeat analysis of calibration standards (n = 4) gave the precision of the instrument, 1, as between 0.08% and 5.83% (median = 0.65%).Accuracy data was not provided by UT for the IC analysis.Major and trace cation data was obtained by Inductively Coupled Plasma Mass Spectroscopy (ICP-MS) for the 35 elements Ag, Al, As, B, Ba, Bi, Ca, Cd, Co, Cr, Cs, Cu, Fe, K, Li, Mg, Mn, Mo, Na, Ni, P, Pb, Rb, Sb, Se, Si, Sn, Sr, Th, Ti, Tl, U, V, Zn & Zr.Total dissolved solids (TDS) were calculated from the sum of these data.Mercury analysis was performed by Scientific Analysis Laboratories Ltd using Cold Vapour Atomic Fluorescence Spectroscopy (CV-AFS) in accordance with the recommended procedures of the UK Department of Energy & Climate Change (2014).

Spring water pH & alkalinity
The pH and alkalinities of spring waters were measured at 10 locations (Fig. 1).pH and temperature of the water in each spring was measured after sampling by dipping a combined portable pH and temperature probe into the geyser, connected to a Thermo Scientific Orion 3 Star pH meter, accurate to ± 0.01 pH and 0.1 °C, pre-calibrated at the University of Texas.After sample collection of 40mL samples, alkalinity was measured by adding 1.6N HNO3 using a digital titrator and titrating to an endpoint pH of 4.50 ± 0.02.The starting pH, temperature and start and end readings on the titrator were entered into the USGS online alkalinity calculator using the fixed endpoint method (http://or.water.usgs.gov/alk/)for each sample.Temperatures were measured as close to the bubbling point of the spring as possible, therefore temperatures should reflect the upwelling water temperature, rather than stagnant water which could be affected by time-of-day and/or seasonal temperature variations.

Batch experiments
Pieces of sandstone were broken off the original field samples using a geological hammer, and these pieces reduced to chips with a jaw crusher.The chips were then disaggregated with a mortar and pestle, being careful not to grind the sample in order to preserve as best as possible any mineral overgrowths and to avoid exposing fresh mineral surfaces.Once disaggregated, 30 ± 0.1 g of sample was added to each flask in which the experiments were to be conducted, described by Wilkinson et al. (2020).The samples were allowed to react with 250 ± 1 mL of Crystal Geyser brine (Na 3270ppm; Cl 3970 ppm) before CO2 was added after approximately 3 months, then allowed to react for approximately another 3 months.The temperature was the ambient for the laboratory, approximately 21 °C; pH was measured more frequently within the first 10 days of the experiment and immediately following the start of CO2 bubbling at day 82 than at other times, to ensure that rapid initial changes in pH were captured in the data.

Results
pH values for all spring locations lie in the range 6.3 -6.8 (Table 4).The lowest pH values were measured at Green River Airport Well (pH 6.31) and Tenmile Geyser (pH 6.34).Alkalinity, measured as the concentration of bicarbonate, is also relatively constant across all the spring locations, with the obvious exception of Green River Airport Well again, which has an alkalinity of 2,335 mg/L, some 35% lower than the next lowest value of 3,589 mg/L at Tenmile Geyser.Water temperatures were also fairly constant, ranging from 13.8 -18.5 °C, with the exceptions of Green River Airport Well (26.6 °C) and Side Seep (22.3 °C).The Side Seep location was stagnant water in a dried up creek, and measurements were taken mid-afternoon, therefore the elevated value of this location is probably not a reflection of the upwelling water temperature.
The spring waters are saline according to the classification of Kharaka and Hanor (2007;Table 5) with total dissolved solids (TDS) generally ranging from 11250 ± 44 mg/L (Crystal Geyser) to 17800 ± 65 mg/L (Torrey's Spring), with a median of 14900 mg/L.The exception is Green River Airport Spring which has a TDS of 3,750 ± 20 mg/L.Concentrations of As, Cu and Zn in Green River Airport Well are generally lower than other springs (Table 5).The remaining of the trace metals of interest (Cd, Cr, Hg & Pb) were not detected above analytical detection limits at any of the springs.Of the other 13 elements analysed (Table 5) Green River Airport Well has generally the lowest concentrations.A notable exception to this is Si which has a higher concentration (32 ± 3 mg / L) than any of the other springs by a factor of 8.

X-ray diffraction (XRD)
The Utah sandstones are dominated by quartz (57 to 90 wt % by XRD; Table 6).Gypsum, hematite and pyrite are below 1 wt.% (Table 6) and these values are indicative only of the probably presence of trace amounts of these minerals.The proportions of each are, however, quite similar across all of the samples.The Entrada formation sandstones are much more carbonate rich (12-13 wt %) than the Wingate and Navajo sandstones (< 1 %).Most of the carbonate is calcite, dolomite and ankerite with only traces of siderite.

Batch Experiments
Initial pH values of the batch experiments are slightly higher than the field-measured pH for Crystal Geyser of 6.7 (Fig. 2 and Table S1), probably due to some CO2 degassing during transport and storage of the samples collected 6 months prior to the start of the experiments.pH values then increase for all 4 batch reaction vessels in essentially the same manner to maximum values of between 8.8 (Entrada S1) and 9.0 (Wingate S3), recorded immediately after CO2 bubbling commenced.There is then a rapid drop in pH for all samples after CO2 bubbling begins to pH values of between 6.1 (Entrada S2) and 6.2 (Wingate S3).pH values then increase slightly for all the samples and maintain a steady mean pH of 6.4 for the remainder of the experiment (Fig. 2).These steady pH values are lower than at the start of the experiment, and therefore lower than the field-measured value at Crystal Geyser (although similar to the Green River Airport Spring and Tenmile Geyser field values; Table S1).

Trace metals in the batch experiments
The composition of the brines extracted from the batch experiments are given in Table S1.
Concentrations of As and Cd were not detected above the analytical detection limits, with the exception of a single Cd data point for the Entrada S2 sample (Fig. 3).Concentrations of Cr, Hg and Pb are also low throughout duration of the experiment (Fig. 3).The addition of CO2 at day 82 appears to mobilise some Cu, Hg, Pb and Zn (and possibly Cr), but these increased concentrations only persist for a maximum of approximately 30 days before returning to pre-CO2 concentrations.
Chromium concentrations for Entrada S1 reduce to lower than pre-CO2 levels.Copper concentrations before CO2 addition appear to be fairly constant and remain within the spring water range.Several days after CO2 bubbling commences, spikes in concentrations occur for the Entrada S1 and Wingate S3 samples, with a maximum value of 49 ± 5 g/L which is above the maximum spring value of 40 ± 3 g/L (Torrey's Spring).However, concentrations then reduce back to lower than pre-CO2 levels, and therefore within the range of the spring values.Concentrations of Ni are below detection limits for the majority of initial 3 months of the experiment, and therefore below the spring samples range of 4.4 ± 0.6 -11.7 ± 0.9 g/L, and remain so until day 97 where concentrations begin to increase for Navajo S4.Entrada S1 follows suit from day 112 and Entrada S2 and Wingate S3 from day 140.Final concentration values on day 166 are around the range of the natural spring waters (Fig. 3).
Zinc concentrations effectively remain constant throughout the experiment, with values at or exceeding the upper end of the springs' range of 6.7 ± 0.3 -37 ± 7 g/L (Fig. 3).The addition of CO2 resulted in some spikes in concentrations for several days following, however final concentrations (20 ± 7 -30 ± 10 g/L) are slightly lower than starting concentrations (28 ± 9 -38 ± 13 g/L).In summary, where concentrations exceed the limit of detection (LOD), concentrations of the trace metal are stable through the experiments, with essentially similar trends for the 4 rock samples.The addition of CO2 to the flasks results in some temporary instability in concentrations of most of these elements, however concentrations stabilise again at similar, or lower, concentrations than pre-CO2.The exception is Ni, where concentrations begin to rise from around 15 days after commencing CO2 bubbling.Concentrations of Cu, Ni and Zn are comparable to those measured in the spring waters, while Cr, Hg and Pb were all mobilised to some extent during the experiments but not detected at all in the spring water samples.The exception is As, where the experiments do not mobilise the element above the limit of detection, despite the high concentrations (above legal drinking water standards) in the natural spring waters.

Major elements in the batch experiments
Concentrations of 8 major elements are presented here: Al, Ba, Ca, Fe, K, Mg, Mn & Sr (Fig. 4).For the most part, concentrations for these elements are very similar in each of the 4 batch samples with concentrations of 5 of these 8 elements falling into two distinct trends.Firstly, after a small initial rise, both K and Mg maintain constant concentrations up to the start of CO2 bubbling, thereafter having a slight initial decrease (Fig. 4) but going on to slowly rise for the remainder of the experiment with bubbled CO2.Concentrations of K remain largely within the range of the spring waters, although by the end of the experiment they have increased to slightly above.Magnesium concentrations for the batch experiments are consistently above the narrow range for the springs (Fig. 4).
Secondly, Ca, Mn and Sr show decreasing concentrations through the initial 3 months of the experiments for all samples.The minima are outwith the spring concentration range (Fig. 4).The addition of CO2 to the flasks results in a rapid increase and then stabilisation of concentrations over a matter of days.The stabilised concentrations are within, or just below, the lower bounds of the spring concentrations.
Barium concentrations are initially within the range of spring concentrations for the first 3 months of the experiment, with the exception of Navajo S4 which initially increases in concentration to 71± 6 g/L before returning to similar concentrations to the other samples by the end of this period.After CO2 bubbling starts, there is a rapid increase in a matter of hours to maximum concentrations for all of the samples, with Navajo S4 again having the highest concentration of 140 ± 10 g/L .This is a similar trend to Ca, Mn and Sr after CO2 bubbling starts, however unlike these elements, concentrations of Ba then decline steadily over the next three months, possibly reaching stable bubbling, which is significantly higher than in the spring samples which are below the detection limit (BDL; 0.18 g/L for ICP-MS).Concentrations then drop by a factor of 10 several days after CO2 bubbling begins and remain at this level (or BDL) for the remainder of the experiment.

Discussion
Regionally, meteoric water enters the N-aquifer system in the San Rafael Swell and flows east, the first sampling point being the Green River Airport Well (Fig. 1).Although this well has high CO2 levels, it has no detectable content of deep-sourced saline brine that ascend from the underlying Carboniferous strata (Kampmann et al., 2009), a relatively low salinity and low concentrations of Cu, Zn and As (Table 5).Hence, the N-aquifer has essentially 2 zones: from the outcrop in the San Rafael Swell to the first brine input after the Green River Airport Well, and beyond here.This model is supported by the behaviour of Cl-, a conservative tracer.A plot of Cl-vs Na+ shows that the springs lie close to a mixing line with the Ismay brine (Spangler, 1992) but with variable proportions of meteoric water and brine.
There is a general correlation between the distance along the aquifer flow-path beyond the Green River Airport Well and salinity (Fig. 5), suggesting that there are multiple points of entry for the brines into the N-aquifer, and not a single point of entry which would result in constant Clconcentrations beyond the entry point, or even the 2 entry points identified by Kampmann (2009) as close to the Green River Airport Well and near Small Bubbling Spring.Using this porewater model, trace metals in the porewaters of the N-aquifer may hence be derived either from the Paradox brines, or from the N-aquifer rock matrix.There is evidence that at least some of the trace metal load in the aquifer waters is derived from the Paradox brines, although no analysis for this is known.
The springs further down the flow path than the Airport Well all have higher salinity, a detectable content of brine from deep sources (1-2% in Crystal Geyser according to Han et al., 2017;  (Fig. 6).It is hence probable that the deep-sourced brines are introducing much of the As and Cu.
Nickel concentrations in the Green River Airport Well are comparable to the other springs so there is no evidence that Ni is being significantly added by the deep brines (Fig. 6).
If the Paradox brines were of constant trace-element composition and no other processes were controlling trace metal concentrations, then there should be strong correlations between Cl and both As and Cu.For As, this is the case for 6 of the data points (Fig. 7) but 4 others have much lower As than would be predicted.This could indicate that some of the Paradox brines have low As concentrations, however there is no obvious spatial grouping of the low-As springs.However, As can be absorbed onto the aquifer rocks, see below where there is evidence that this is occurring in the batch experiments.The Cu data correlate well for all 8 out of 10 of the data points (Fig. 8).Torry's Spring and Little Bubbling Spring have higher than expected values but are not close spatially.The zinc data are too variable to discern any trend with distance and lack any obvious spatial pattern, again perhaps suggesting that the Paradox brines are of variable composition or that adsorption is affecting concentrations.The remaining trace metals (Cd, Cr, Hg, Pb) are BDL in all springs, so are either not present in significant concentrations in the deep brines or are present but are absorbed by the aquifer rock before the fluids arrive at the springs.Given the lack of spatial correlation between wells that do not fit the simple two-fluid mixing model, then spatially variable adsorption is perhaps the simplest explanation.
An analysis of the major element geochemistry of the fluids sampled at the springs, compensated for the input of Paradox brine, allowed Kampmann et al. (2009;their Eqn. 13) to describe the effects of the CO2 input for the brine-affected zone by the following reaction: plagioclase + K-feldspar + gypsum + dolomite + H + + HCO3 − = kaolinite + smectite + calcite + quartz It seems reasonable to assume that a similar reaction is occurring in the zone that lacks brine and free-phase CO2, driven by CO2 derived from the atmosphere and any soil processes.Hence, trace metals could be introduced into the aquifer waters by dissolution of any of plagioclase, K-feldspar, gypsum or dolomite.Given that the concentration of CO2 associated with atmosphere and soil processes in a semi-arid environment will be much lower than associated with free-phase CO2, then any trace metal input from the above reaction is likely to be much less prior to the input of high CO2 concentrations than afterwards.Kampmann et al. (2009) corrected the major element concentrations in the spring waters for the input of the Paradox brine, however it is not possible to accurately correct the trace metal concentrations using the same method, as the only published analysis of the brine (Spangler, 1992) does not include the appropriate metals.It is possible to this it is concluded that the trace metals are not derived by the dissolution of feldspars, gypsum or dolomite (with the possible exception of Cu).Neither can they be derived by the dissolution of kaolinite, smectite, calcite or quartz as these are modelled as precipitating (Kampmann et al., 2009).
There is however evidence for the dissolution (and possibly precipitation) of an Fe-bearing phase that is not modelled by Kampmann et al. (2009).Concentrations of Fe vary greatly both between individual sampling locations and between different studies.The highest concentrations were reported by Kampman et al. (2014) at Champagne Geyser (14.9 mg/L) and for this study at Little Bubbling Spring (14.8 mg/L).These measured concentrations are similar to concentrations in leaching experiments using red sandstones from Utah, of up to 15 mg/L, using an H2S-equivalent reducing agent (thioacetamide) with CO2 (Purser et al., 2014).Using CO2 with CH4, a maximum of 3 mg/L Fe 2+ was recorded, and < 1 mg/L with only CO2 (Purser et al. 2014, their Figure 3), which is consistent with low concentrations obtained with the experiments for this study (maximum 0.2 mg/L; mode 0.009 mg/L; Table S1).By analogy with the experiments of Purser et al. (2014) the Fe is derived from Fe-rich grain coatings which are principally hematite.Experiments using the Mount Simon sandstone did find evidence for the dissolution of an iron-bearing phase (in this case pyrite; Dávila et al., 2020) but the study did not report trace metal concentrations.For the Green River springs, however, none of As, Cu or Ni correlate with Fe (R 2 < 0.25 in all cases), arguing against a simple dissolution / precipitation control for these elements.Wigley et al. (2012) speculated that trace metals are mobilised from Fe-oxide grain coatings but are re-deposited at a reaction front, limiting their overall mobility in the migrating fluids.Significant differences in Fe concentrations in individual springs between published studies may indicate changes in redox conditions in the Naquifer with time, which can be explained as changes in the proportion of H2S and / or methane to CO2.
A test can be made for whether carbonate dissolution is contributing to concentrations of Ca, Mg and alkalinity in the Green River springs.A plot of Ca 2+ and Mg 2+ equivalent concentrations (meq/L) against HCO3 -(meq/L), with a 1:1 relationship representative of a stoichiometric relationship between the cations and anion is indicative of carbonate dissolution (Stallard and Edmond, 1983).(Heath et al. 2009;Kampman et al. 2009, 2014and Shipton et al. 2004).The 1:1 stoichiometric relationship line is plotted for comparison.The data from Heath et al. (2009) and Shipton et al. (2004) clearly do not fit the 1:1 trend.The Kampman studies and this study's data both cluster around the 1:1 relationship and display a quasi-linear relationship.The conclusion therefore is that carbonate dissolution is occurring, however this is not the sole supply of Ca, Mg and HCO3 -to the groundwaters.In contrast, Kampman et al. (2009)  The relatively high salinity of the springs (TDS from 3,750 to 17800 mg/L) exceeds the EPA secondary drinking water standard of 500 ppm by a factor of at least seven and in some cases a factor of over thirty.Hence, the spring waters are not deemed potable, and it could be argued that the remaining EPA drinking water standards cannot be meaningfully applied to these waters.However, the origin and controls of the trace metals may be applicable to CO2-rich waters that are potable, and, in the unlikely event of a leak of engineered CO2 into an underground supply of drinking water, it will be standards such as those of the EPA which will be used to determine what remediation is required.
Trace metal concentrations in the CO2-rich springs are significantly below legal limits for USA drinking water (except for As) despite previous reports of high trace metal loads in aquifers with high CO2 contents (e.g.Little and Jackson, 2010).Overall, trace metal concentrations are low in water samples from 10 CO2-rich springs and where present, their source or sources are uncertain.For As, this is plausibly introduced into the aquifer in solution in the Paradox brines, as above and hence may not be related directly to the high CO2 concentrations.

Batch experiments
The batch experiments all used brine collected from Crystal Geyser as the initial aqueous fluid, therefore each experiment had 2 sources of trace metals, the brine and the rock samples.The results show that there was transfer of metals both from the rock to the brine, and the brine to the rock, discussed below.

Reactions prior to the addition of CO2
For some metals, the concentrations recorded during the experiments are much lower than the preexperiment porefluid concentrations (As < 0.02 vs 39 g/L; Ni mostly < 0.04 vs 8 g/L though maximum values are comparable).Assuming that the metals were not lost to adsorption onto the vessels used for transport or the glass walls of the batch reaction vessels, then they must have been absorbed onto the surfaces of the rock grains.For As, the addition of CO2 did not increase concentrations, but for Ni concentrations increased into the range observed in the spring waters, at least proving in this case that the low concentrations prior to CO2 addition were not an artifact (Fig.

3).
For As, a potential explanation of the apparent adsorption of the metals and the very low (BDL) concentrations of As in the batch experiments even after the addition of CO2 is that the rock samples used in the experiments have very low (mobile) As concentrations.The samples are from the San Raphael swell, i.e. the recharge area for the N-aquifer system and lie on the regional flow path before the input of the deep-sourced brines, and hence before the As input apparently associated with the Paradox brines (see above).Any mineral surfaces that are capable of adsorbing As will be 'undersaturated' with the metal as they have not been exposed to high concentrations, and hence will be capable of absorbing As from solution.Arsenic concentrations are very low (BDL) in the experiments since the As is not derived from the rock matrix, and is hence it is not present to be leached from the rocks during the experiments.The same argument cannot be made for Ni, as this is not added to the aquifers by the brines, as above, but must be derived from the rock matrix.It is possible that much of any previously adsorbed Ni in the San Rafael rock samples has been desorbed and removed by meteoric water leaching, leaving mineral surfaces available for adsorption.For other metals, the concentrations in the batch reactions exceed those in the initial fluid (Cr 0 -3 vs < 0.3 g/L; Zn 25 -50 vs < 0.2 g/L; Pb 0 -5 vs < 0.035 g/L).These metals must have been derived from the rock samples.
The concentration of Ca, Mn and Sr all declined in the pre-CO2 phase of the experiments, from values at or close to the range of the springs, to values below this.This implies the precipitation of a Ca-bearing phase, and given that Mg does not follow the same trend, then the phase is presumably calcite rather than dolomite.This is associated with a pH change of approximately 7 to 9, which will cause the reaction of much of the HCO3 -to CO3 2-, which may have been caused by a gradual degassing of the Crystal Geyser water in the experiments prior to the introduction of the CO2.

Reactions after the addition of CO2
The addition of CO2 to the experiments reversed the trends seen in Ca, Mn and Sr, implying that the newly precipitated calcite was at least partly re-dissolved (Fig. 4).With only a single exception (Mn for sample S4) the initial concentrations were not attained, so that the dissolution was not complete.The dissolution was rapid (a few days at most), after which concentrations of Ca, Mn and Sr remained constant or increased only slowly.None of the trace elements show similar trends, so that neither the precipitation of calcite nor the subsequent dissolution appeared to have controlled trace element concentrations.Correlations between the 8 trace metals and Ca, Mg and Fe are either weak or non-existent, supporting the notion that carbonate dissolution and precipitation are not strong controls on the concentrations of trace metals in the batch experiment fluids.
A half of the trace elements (Cr, Cu, Pb and Zn) show relatively high concentrations in the days after the introduction of the CO2, but in all cases decline to values similar (Pb) or lower (Cr, Cu, Zn) than before the addition of CO2.The trace metals with temporary increased concentrations upon addition of CO2 were generally those which have an association with carbonates (Cu, Pb and Zn but not Ni), but this may be coincidence, as above.It may be significant that the transient spikes in concentration coincide with a period of anomalously low pH (Fig. 11), which was subsequently buffered back to the range of values recorded within the natural springs.A pH control on absorption / desorption on mineral surfaces is therefore possible (e.g.Hsia et al., 1994).These transient spikes in concentration would have little or no effect upon a natural system, where water flow rates might transport dissolved metals for distances that would be very small compared to the dimensions of the natural aquifer system.
The only trace elements that show convincing and sustained increases in concentration upon addition of CO2 are Ni and Pb (Fig. 3).Nickel is commonly cited in the literature as being enhanced with the addition of CO2 to batch experiments with calcareous sandstones and sands, similar to the samples used in this study (Little and Jackson, 2010;Lu et al., 2010 andCahill et al., 2013).However, the gradual increase in Ni does not correlate with the sudden increase in Ca after CO2 additionthough this may be the re-dissolution of (low Ni) calcite precipitated in the pre-CO2 stage of the experiments.It is notable that Ca does increase slowly for the remainder of the experiments after the sudden increase associated with the addition of CO2.It is possible that this slower dissolution is liberating Ni, though not from the neo-formed (pre-CO2) calcite but from pre-existing (authigenic) calcite.Similar arguments apply to Pb, which is also recorded as associated with carbonate mineral dissolution (e.g.Wilkinson et al., 2020).Lead also occurs in K-feldspars (Smith and Brown, 1988), and there is a steady increase in K after the addition of CO2 which could indicate K-feldspar dissolution.
Unlike the example reported by Wilkinson et al. (2020), where unfeasible concentrations of Pb would have been required in the feldspar to account for the observed solution compositions, here the calculated Pb concentrations are on the low side of the range expected (10-1000 ppm; Smith and Brown, 1988).Hence, the dissolution of K-feldspar is a plausible source of the Pb seen in the experiments after the addition of the CO2, and there is no significant correlation between Pb and pH (Fig. 11) so that desorption from mineral surfaces seems relatively unlikely.
Whereas Cd and Cr were found to be somewhat mobile with CO2 in previous studies (Terzi et al., 2014), their mobility in this study was low, especially after the addition of CO2 for Cr.Mercury responded initially to the CO2 bubbling with increased concentrations, particularly for Entrada S2, but within days had returned back to BDL.The effect of CO2 for Hg appeared therefore to be a temporary mobilisation, but which was not sustained beyond a few days, possibly due to volatilisation, i.e. loss from the reaction vessel.Alternatively, as Hg concentrations are high when pH is less than approximately 7 (Fig. 11) this could be interpreted as evidence for pH-dependant absorption / desorption, as above.
Concentrations of Fe in the experiments are low (0.005 -0.177 mg/L) compared with concentrations found in the spring waters (1.2 -14.8 mg/L).Iron mobility is low where it is oxidised to Fe 3+ .
Concentrations of Fe are higher in the spring waters most likely due to reducing conditions caused by dissolved methane (Wigley et al., 2013) or hydrogen sulphide (e.g., Purser et al. 2014).Purser et al. (2014) demonstrated that significant Fe 2+ (3 -15 mg/L) could be leached from Utah sandstone collected in the same area, using these reducing agents.

Conclusions
The concentrations of trace metals are low in the Utah spring waters (except for As) despite the presence of free-phase CO2.All the waters (except for As and a single exception for Hg) meet USA EPA standards for drinking water except for total salinity which is largely derived from externally sourced (Paradox Formation) brine.Concentrations of 4 of the trace metals (Cd, Cr, Hg, Pb) are consistently below analytical detection limits.This shows that CO2 in aquifers does not inevitably cause trace metal contamination.
The detectable trace elements in the spring waters are largely from brines sourced from the underlying Paradox Formation (As, Cu) or the rock matrix (Ni).For As, the experiments mobilised very little of the metal as it is not sourced from the rock, which may in fact be capable of absorbing (and hence immobilising) As.For Ni, the dissolution of carbonate minerals is a possible source, though the experiments show that pH-dependant adsorption is also a potential control.None of the trace metals perfectly fit a simple mixing model between Paradox brine and aquifer recharge, suggesting either that the Paradox brine is of variable composition, or that some of the trace elements (especially As) are adsorbed onto mineral surfaces within the aquifer.

Figure Captions
://www.lyellcollection.orgby University of Edinburgh on Aug 23, 2022 ://www.lyellcollection.orgby University of Edinburgh on Aug 23, 2022 ://www.lyellcollection.orgby University of Edinburgh on Aug 23, 2022 ://www.lyellcollection.orgby University of Edinburgh on Aug 23, 2022 ://www.lyellcollection.orgby University of Edinburgh on Aug 23, 2022 from https://www.lyellcollection.orgby University of Edinburgh on Aug 23, 2022 end of the experiment.With the exception of Navajo S4, the batch concentrations lie within the range of the spring values.Of the remaining 3 elements, Fe does not appear to exhibit any systematic trend, although at the commencement of CO2 bubbling, there are temporary slight increases in concentrations for Entrada S1 and Navajo S4, to values of 60 ± 20 g/L and 110 ± 40 g/L, respectively, followed by a general stabilisation and systematic increase in Fe concentration.Maximum batch Fe concentration values (180 ± 70 g/L) are also significantly lower than the spring range of 1,190 -20,600 g/L.Finally, Al was detected at concentrations fluctuating between 140 ± 50 g/L and 600 ± 200 g/L prior to CO2 10 -20% according to Wilkinson et al., 2009) and higher concentrations of As and Cu than the Airport Well ://www.lyellcollection.orgby University of Edinburgh on Aug 23, 2022 ://www.lyellcollection.orgby University of Edinburgh on Aug 23, 2022 calculate relative trends of element concentrations as these are independent of the Paradox brine composition -hence the graphs in Fig. 9 have no absolute concentration (y-axis) units and are only relative values.Comparing the data for As, Cu and Ni to the modelled reaction from Kampmann et al. (2009; Fig 9 lower right) is apparent that there is no correlation between the extent of the reaction and the concentrations of the trace metals for As or Ni, though Cu is more equivocal.From

Fig. 10
Fig.10shows the data for this study along with those obtained and calculated from the literature modelled the precipitation of calcite, but the dissolution of Ca-feldspars.Trace metal concentrations, where detected in the Green River springs, correlate poorly with major elements in most cases.While Pearson's correlation coefficients for a number of element pairs are indicative of strong relationships, further examination shows that these are often skewed by individual high or low values.Only Ca and Ni show a strong positive linear correlation.Nickel concentrations may therefore be related to carbonate dissolution.
://www.lyellcollection.orgby University of Edinburgh on Aug 23, 2022 ://www.lyellcollection.orgby University of Edinburgh on Aug 23, 2022 ://www.lyellcollection.orgby University of Edinburgh on Aug 23, 2022 With the exception of Fe, Pb and As, the concentration of trace metals in batch experiments are within the range, or close to, the concentrations found in the springs.Fe is probably affected by the oxidation state of the fluids in the experiments, rendering it comparatively insoluble compared to the subsurface brines.The addition of CO2 to batch experiments reduces experimental pH values to just within the range of values for the springs in the Green River area, indicating the capacity of the rock to buffer pH.Trace metal concentrations do not correlate well with any of the major elements including Ca, Mg and Fe (for carbonates).Therefore, where present, trace metal concentrations in the batch experiments are unlikely to be due to mineral dissolution (with the possible exception of Ni and more certain exception of Pb), and more likely the result of ion exchange / desorption processes, most plausibly involving Fe-oxide grain coatings.

Fig. 1 .
Fig. 1.Location map of outcrop and spring water sampling areas in the San Rafael Desert, near Green River, Utah.Clustered samples are expanded for clarity, top left and bottom left for outcrop rock samples and spring samples, respectively.

Fig. 2 .
Fig. 2. pH values for the batch reactions.CO2 flow commences at day 82.Shaded area indicates the field-measured pH values of the Green River spring waters.

Fig. 3 .
Fig. 3. Concentrations of trace metals for the batch experiments.Dashed vertical line at day 82 represents start of CO2 flow.Shaded areas indicate the range of concentrations detected across 10 Green River spring locations (Figure 1).No shading present indicates element concentrations fall below analytical detection limits.

Fig. 4 .
Fig. 4. Concentrations of elected major elements for the batch experiments.Dashed vertical line at day 82 represents start of CO2 flow.Shaded areas indicate the range of concentrations detected across 10 Green River spring locations (Figure 1).No shading present indicates element concentrations fall below analytical detection limits, except Fe where the springs' range is significantly greater than the batch concentrations.

Fig. 5 .
Fig. 5. Chloride concentration versus distance from the airport well.

Fig. 6 .
Fig. 6.Arsenic and Cu concentrations for the Airport Well (dark circles) are low compared to the other springs, but Ni lies within the range of the other spring water analyses.

Fig 7 .
Fig 7. Concentrations of As vs Cl for the spring waters.See text for interpretation.

Fig. 8 .
Fig. 8. Concentrations of Cu vs Cl for the spring waters.See text for interpretation.

Fig. 9 .
Fig. 9. Trace element concentrations in the N-aquifer versus distance from the Airport Well show no similarity to the modelled dissolution of K-feldspar modelled by Kampmann et al., 2009) for As or Ni, though some resemblance for Cu.The plotted concentrations are corrected for the input of Ismay brine using the method of Kampmann et al. (2009) and are relative values as the trace metal concentrations in the brine are unknown.The lower right figure shows calculated changes in feldspar abundance versus distance from the Airport Well, redrawn from Kampman et al. (2009).

Fig. 10 .
Fig.10.Spring water analyses plot close to a 1:1 line for carbonate vs Ca plus Mg in this study and those ofKampmann (2009Kampmann ( , 2014)), supporting carbonate mineral dissolution as a source of alkalinity.

Fig 11 .
Fig 11.Scatter plot of pH and Utah batch concentration data.Arsenic not shown as concentrations all BDL.Error bars are 2 of the mean of standard reference materials.

Table Captions Table 1 .
Concentration limits in USA drinking water, EPA National Primary Drinking Water Regulations

Table 2 .
Rock sample details

Table 3
Spring sampling bottle types, volumes, type of subsequent analysis, and chemical additives.*Volatile Organic Analysis vial.

Table 4 .
Field measurements of spring waters.

Table 5 .
Spring water composition by ICP-MS.*Concentrations in mg/L.**Hg analysed by ICP-MS at University of Edinburgh.Blank concentration values are < LOD, or not reported.

Table 6 .
Composition of the rock samples by XRD.Downloaded from https://www.lyellcollection.orgby University of Edinburgh on Aug 23, 2022