Sources and sinks of CO2 and CH4 in siliciclastic subterranean estuaries

Anthropogenic production of greenhouse gases (GHGs) has intensified the need to constrain estimates of natural atmospheric sources from both terrestrial and marine systems. Estuaries are known sources of carbon dioxide (CO2) and methane (CH4); however, less is known about GHG dynamics in subterranean estuaries (STEs). We evaluate CO2 and CH4 dynamics in three proximal STEs bordering Indian River Lagoon, Florida, where groundwater flows through siliciclastic sediments with minor carbonate mineral contents. Although the three STEs have similar mineralogical and flow characteristics, CO2 and CH4 concentrations vary by orders of magnitude. Nonconservative mixing of both gases is observed, and CH4 is generally produced while CO2 is sequestered. The extent of methanogenesis is linked to the redox potential of inflowing groundwaters, as well as degree of CH4 oxidation, which results mostly from anaerobic oxidation of methane. Methane concentrations vary by orders of magnitude, and stable isotopic signatures suggest differences in the microbial production pathway between sites. CO2 is sequestered due to the production of alkalinity relative to dissolved inorganic carbon, which occurs both through rapid CaCO3 dissolution at the shoreline as low‐pH groundwater from the siliciclastic aquifer interacts with carbonate minerals in lagoon sediments, as well as redox reactions, particularly sulfate reduction and denitrification. These results demonstrate a high variability in CO2 and CH4 concentrations, and thus fluxes, even among geographically constrained and hydrogeologically similar STEs. Although STEs are sources of both CO2 and CH4 to surface waters, the variability of production and consumption complicates global estimates of GHG fluxes from STEs.

Estuaries represent the interface between freshwater and salt water and are sites of net remineralization of terrestrial organic carbon, which generates large fluxes of CO 2 from water to the atmosphere (Abril and Borges 2004;Cai 2011). Analogous freshwater-saltwater mixing zones exist at the interface between coastal aquifers and adjacent saline pore waters, known as subterranean estuaries (STEs; Moore 1999), which contribute both freshwater and solutes to the coastal zone via submarine groundwater discharge (SGD). Although the impact of SGD on surface water carbon budgets has received relatively little attention compared to nutrients and metals (e.g., Slomp and Van Cappellen 2004;Kroeger and Charette 2008;Spiteri et al. 2008;Roy et al. 2010;Whelan et al. 2011;Johannesson et al. 2011), several factors suggest that SGD may be an important source of greenhouse gases (GHGs). Similar to surface estuaries, STEs are net heterotrophic, but lack the CO 2 sink of photosynthesis that reduces CO 2 fluxes from surface estuaries. Additionally, isolation from the atmosphere allows the development of anaerobic conditions that may drive organic matter remineralization toward methanogenesis. Methanogenesis is the terminal remineralization reaction for organic matter and converts organic matter to CO 2 and methane (CH 4 ). Like CO 2 , CH 4 is a GHG but has approximately 25 times the warming potential of CO 2 and thus relatively smaller CH 4 fluxes have disproportionate impacts on warming and C cycling.
Although GHG fluxes from STEs could be significant, gases may undergo additional chemical transformations due to enhanced biogeochemical processing in the freshwater-saltwater mixing zone (e.g., Moore 1999). Remineralization of organic carbon in fresh groundwater should generate CO 2 and CH 4 , but biogeochemical reactions within STEs may modify their concentrations. Several studies have found SGD to be a significant source of carbon to surface waters (Cai et al. 2003;Liu et al. 2012;Liu et al. 2017). However, little consensus exists as to whether STEs are sources or sinks of CO 2 because mineralogical control of remineralization reactions may alter pore-water buffering capacity, which controls the speciation of dissolved inorganic carbon (DIC) and hence CO 2 concentrations. Changes to the buffering capacity of water depends on the relative impacts of diagenetic reactions to DIC and alkalinity (Alk), and are expressed as difference between concentrations expected from *Correspondence: ajpain@ufl.edu This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited. mixing and the concentrations resulting from reactions of DIC (ΔDIC) and alkalinity (ΔAlk). The relative amount of these changes, expressed as the ratio of ΔAlk : ΔDIC, may alter the CO 2 buffering capacity of water, where higher values for the ratio indicate a greater buffering capacity and thus more CO 2 may be sequestered as HCO 3 − . ΔAlk : ΔDIC ratios should reflect the stoichiometry of the predominant reactions controlling dissolved carbonate chemistry (e.g., Table 1) and thus may be used to infer which reactions control dissolved carbonate chemistry (Cai et al. 2003;Liu et al. 2017).
CH 4 fluxes from STEs have received relatively more attention than CO 2 fluxes because CH 4 is used as a quasi-conservative tracer of SGD due to its typically high concentrations in groundwater compared to surface water (Cable et al. 1996;Corbett et al. 2000;Dulaiova et al. 2010). However, use of CH 4 as a tracer requires understanding of its sources and sinks within STEs, for example, methanogenesis and methanotrophy (e.g., Schutte et al. 2016). Furthermore, CH 4 production occurs via multiple microbial pathways (CO 2 reduction and acetoclastic methanogenesis; Whiticar and Schoell 1986), which produce distinct stable isotopic signatures of CH 4 . Little is known of the microbial pathway by which CH 4 is produced in STEs, although current understanding suggests that CO 2 reduction is favored in settings with higher salinity and lower labile organic carbon availability (Whiticar and Schoell 1986;Megonigal et al. 2005). As both salinity and organic carbon quantity and quality vary within and among STEs and control redox gradients, both the quantity of CH 4 and pathway by which it is produced should vary in STE settings.
To evaluate potential ranges of GHG fluxes from STEs, we assess sources and sinks of CO 2 and CH 4 at three siliciclastic STEs bordering Indian River Lagoon, Florida. The sampling sites are within 50 km of one another and have similar environmental and hydrogeologic characteristics. These similarities thus allow us to assess GHG dynamics in STEs without confounding variables such as flow rates and flow paths. We use salinity-based conservative mixing models to evaluate impacts of reactions on CO 2 and CH 4 concentrations within the STEs. We first assess the controls of methanogenesis and methanotrophy to determine the role of STE biogeochemical processing on CH 4 budgets. We then assess CO 2 concentrations and evaluate mineralogical and redox controls on STE carbonate speciation (Table 1). We compare ΔAlk : ΔDIC ratios in the three sampled STEs with those expected from various reaction stoichiometries (Table 1) in order to evaluate which reactions could control CO 2 concentrations in siliciclastic STEs. These results allow us to evaluate the potential role of SGD on GHG production and sources from coastal systems.

Methods
Indian River Lagoon is located on the Atlantic coast of central Florida and spans approximately 250 km of coastline in three hydrological subunits: Mosquito Lagoon, Indian River Lagoon, and Banana River Lagoon (BRL; Fig. 1). Three STEs in the Indian River and BRLs were included in this study, Eau Gallie North (EGN), BRL, and Riverwalk Park (RWP). Sediments in these STEs are siliciclastic ranging from fine sand to clays, which allow groundwater seepage to occur at rates Table 1. Impact of redox pathways and biogeochemical reactions on alkalinity and DIC.

Equation Reaction
ΔAlkalinity : ΔDIC ranging from 0.02 to 0.9 m 3 d −1 m −1 of shoreline at EGN (Martin et al. 2007). This slow seepage makes STE salinity gradients static over timescales of days to weeks. Seasonal variation in lagoon water salinity and fresh groundwater head is known to cause fluctuations in seepage face width (Roy et al. 2013), and storm-driven saltwater intrusion events can alter seepage face salinity for several months (Smith et al. 2008). We collected pore-water samples during three sampling campaigns: May and September of 2015 and May of 2016. Samples were collected from multilevel piezometers (Martin et al. 2007) that had previously been installed at EGN, RWP, and BRL sites (Fig. 1). Piezometers were installed in 2004 at EGN and during May 2014-September 2015 at RWP and BRL. At EGN, sampled piezometers were installed at 0, 10, and 20 m offshore (EGN-0, , at RWP were 10, 20, and 35 m offshore , and at BRL were 1, 11, 21, and 45 m offshore (BRL-1, BRL-11, BRL-21, and BRL-45; piezometer schematic illustrated in Fig. 2).

Sample collection
Pore-water samples were pumped to the surface through 0.5 cm diameter flexible poly(vinyl chloride) tubing attached to multilevel piezometer ports. Piezometers are constructed with multiple (4 to 8) well screenings at depths ranging from 7 cm to 2.5 m below the sediment-water interface. Tubing leads from the screened intervals to the surface and is sampled by pumping water using a peristaltic pump which samples pore waters at increasing depths below the sediment-water interface (Fig. 2). The tubing was connected to an overflow cup in which a YSI Pro-Plus sensor was installed and used to monitor salinity, temperature, pH, oxidation-reduction potential (ORP), and dissolved oxygen (DO). DO was measured using a polarographic membrane that has an interference from hydrogen sulfide, which prohibited measurement of DO concentrations when hydrogen sulfide was present. Once all parameters were stable, water was filtered through 0.45 μmol L −1 trace-metal grade Geotech medium capacity disposable canister filters into sample vials. Samples for cations and anions were collected in high-density polyethylene bottles; cation samples were preserved with trace-metal grade nitric acid (pH < 2), whereas no preservative was added to anion samples. Samples for nutrients were filtered into 15-mL polypropylene containers and frozen until analysis. DIC samples  were filtered through 0.2 μmol L −1 filters directly to the bottom of 20 mL Qorpac glass vials and allowed to overflow until sealed tightly with no headspace.
Redox-sensitive solutes, Fe(II) and hydrogen sulfide, were measured on 0.45-μmol L −1 filtered water in the field immediately after pumping from the multisampler tubing using colorimetric methods. Fe(II) was measured in triplicate using the ferrozine method (Stookey 1970). Water was sampled from the flowing pore-water stream, and 1 mL of ferrozine was immediately added to 10 mL of sample and agitated to mix. The absorbance at 560 nm was measured after 5 min of reaction with a Hach DR 890 portable colorimeter. Blanks were treated identically to the samples and measured before each sample. Blanks of distilled water were subtracted from sample absorbance and were lower than 0.001 absorbance units. Measured absorbance values were converted to concentrations with laboratory calibrations prepared with Fe(II) standards, utilizing an aliquot of the prepared ferrozine that was used in the field. Sampling for hydrogen sulfide occurred simultaneously with the Fe sampling and H 2 S was measured immediately in triplicate using the methylene blue method according to U.S. EPA methods outlined in Hach (2013). Precision is reported at 0.1 mg L −1 .
Gas samples were collected via headspace extractions according to methods outlined in Repo et al. (2007). Unfiltered water was pumped into the bottom of 500 mL bottles until they overflowed and immediately capped with rubber stoppers fitted with two 3-way inlet valves. Sixty milliliter of water was extracted from one inlet and replaced with 60 mL of ultrapure N 2 gas. Bottles were shaken for 2 min to equilibrate headspace gas with water, and headspace gas was extracted and immediately injected into 60-mL glass serum bottles that were evacuated immediately before the sample introduction. Samples were stored at room temperature until analysis within 1 week of collection. Method check standards were collected by injecting gases of known concentrations of CO 2 and CH 4 into evacuated vials and treated identically to samples.

Laboratory methods
Gas samples were analyzed for CO 2 and CH 4 concentrations and δ 13 C-CO 2 and δ 13 C-CH 4 on a Picarro cavity ringdown spectrometer. Carbon isotopic compositions are reported in reference to Vienna Pee Dee Belemnite. We report δ 13 C-CO 2 values from the May 2016 sampling trip only because δ 13 C-CO 2 values from previous sampling trips were impacted by an interference from hydrogen sulfide. To remove this interference, sample gas from the May 2016 sampling trip was passed through an in-line elemental copper scrubber before analysis (Malowany et al. 2015).
Anion and cation concentrations were measured on an automated Dionex ICS-2100 and ICS-1600 Ion Chromatograph, respectively. Error on replicate analyses was less than 5%. DIC concentrations were measured on a UIC (Coulometrics) 5011 CO 2 coulometer coupled with an AutoMate Preparation Device. Samples were acidified, and the evolved CO 2 was carried through a silver nitrate scrubber to the coulometer where total C was measured. Accuracy was calculated to be AE 0.1 mg L −1 based on measurement of check standards. Dissolved Si concentrations were analyzed on a Seal AutoAnalyzer III. Error on check standards was less than 10%.
Data processing-dissolved gas concentrations and isotope ratios Conversion to molar units from the headspace concentrations followed the methods outlined in Bastviken et al. (2004). To solve for the moles of gas originally dissolved in solution, we first converted measured gas concentration (ppm) in the headspace to moles: where n g equals the moles (n) of gas in the gaseous phase, P x is the measured partial pressure of CH 4 or CO 2 (atm), V g is the volume of headspace gas (liter), R is the common gas constant (0.0821 L atm K −1 mol −1 ), and T is the temperature (K) of water at the time of collection. The number of moles of gas dissolved in the aqueous phase (n aq ) is calculated by: where C aq is aqueous concentration, V aq is aqueous volume (500 mL minus 60 mL replaced by headspace gas to give a total volume of 440 mL), and K H is Henry's Law constant (mol L −1 atm −1 ). The value of K H depends on temperature according to Sander et al. (2011). We used this relationship to calculate K H of CO 2 and CH 4 for each water sample based on the water temperature at the time of sampling. The concentration of dissolved gas in water samples (C aq ) was then calculated as the sum of the number of moles of gas in aqueous and gaseous phases divided by the aqueous volume: CH 4 oxidation was calculated using the isotopic method outlined in Mahieu et al. (2008) and Preuss et al. (2013). The fraction of oxidized methane (f ox ) in an open system is given by: where δ E is the measured δ 13 C-CH 4 value for each pore-water sample, δ P is δ 13 C-CH 4 of produced methane, α ox is the oxidation fractionation factor, and α trans is a fractionation factor resulting from transportation of CH 4 . While the exact value of Pain et al. Sources and sinks of CO 2 and CH 4 δ P is unknown, diagenetic alteration of δ 13 C-CH 4 values through oxidation or transport only enrich δ 13 C-CH 4 signatures, therefore the value of δ P is take as the most depleted δ 13 C-CH 4 signature per STE site, assuming it has little or no diagenetic alteration. Literature-reported values for α ox range between 1.003 and 1.049. We calculate the fraction of oxidized methane with the largest fraction factor (α ox = 1.049; Mahieu et al. 2008), which yelds the minimum amount of CH 4 oxidation required to explain the observed variations in δ 13 CH 4 and thus is a conservative estimate for CH 4 oxidation. Literature-reported values for α trans vary from 1 for advectiondominated systems to 1.0178 for diffusion-dominated porous media (de Visscher et al. 2004;Mahieu et al. 2008;Preuss et al. 2013). Based on Roy et al. (2011), we assume that transport is advection dominated and thus assume α trans = 1. The concentration of oxidized methane (CH 4(ox) ) is derived by solving the set of equations: CH 4ðoxÞ = f ox * CH 4ðproducedÞ ð15Þ and substituting Eq. 14 into Eq. 15 to eliminate CH 4-produced yields.
Modeling We used concentrations of major cations and anions, pH, temperature, and DIC concentrations to model the alkalinity and speciation of carbonate ions in PHREEQc using the PHREEQc database (Parkhurst 1995). Alkalinity was estimated from the charge balance of the model input (Parkhurst 1997). We also used major elemental composition, pH, temperature, and Si concentrations to model saturation indices of calcite and quartz. The saturation index (SI) is calculated as SI = log (IAP/K sp ), where IAP is the ion activity product and K sp is the solubility product for calcite and quartz minerals.
To assess the impacts due to mixing vs. reactions in the STE, we constructed salinity-based conservative mixing models. Surface saltwater compositions were used for the saline end member. Freshwater end members for conservative mixing models are taken as the average of the freshest samples collected at shoreline piezometers to minimize impacts due to groundwater interacting with lagoonal sediments, which likely differ in composition from solid aquifer material. The use of freshest samples also minimizes mixing between inflowing fresh groundwater and marine salt water, which will impact water chemistry and likely dissolved gas concentrations. The choice of a freshwater end member for each STE site (rather than a common end member for all STEs) also minimizes the impact of heterogeneity in the freshwater aquifer on STE mixing model results. For instance, the dissolved gas concentration of fresh groundwater may vary based on catchment land cover and heterogeneity in aquifer composition, which may impact infiltration rates and the degree of interaction between groundwater and organic soil layers, as well as the terminal electron acceptor concentrations in groundwater (for instance nitrate).
At BRL, the freshwater values were averaged for all sampling campaigns (n = 3; Table 2) For EGN, the freshwater end member was derived from samples collected during the September 2015 and May 2016, because in May 2015, only two water samples were available at the shoreline, and both were more diagenetically altered and saltier than the freshest water samples from the other two sampling times. For RWP, the freshwater end member was the average of the freshest samples collected in May and September 2015, because the shoreline piezometer was damaged and not available to sample after the September 2015 campaign.

Mixing model results
Freshwater end members have lower pH and ORP values compared to saltwater end members for BRL and RWP, but for EGN, the freshwater end member has a higher ORP value than the EGN saltwater end member (Table 2). All freshwater end members have higher concentrations of CO 2 and CH 4 , as well as higher concentrations of DIC compared to saltwater end members. BRL and RWP freshwater end members have higher Table 2. Chemical composition (average AE 1 SD) of freshwater and saltwater end members used in salinity-based conservative mixing models. Freshwater end members are the average of freshest and least diagenetically altered samples from shoreline piezometers at each seepage face at each sampling time. Saltwater end members are the average of lagoon surface water collected at each seepage face at each sampling time.
Site n End member Salinity Gas concentrations vary less in the salt water than freshwater end members: BRL, EGN, and RWP contain 0.2 AE 0.1, 0.1 AE 0.1, and 0.1 AE 0.1 μmol L −1 of CH 4 and 32 AE 13, 53 AE 30, and 40 AE 38 μmol L −1 of CO 2 , respectively. CO 2 and CH 4 are nonconservative with salinity at all STE sites (Fig. 3a,b). CO 2 concentrations are generally lower than those predicted by conservative mixing, while CH 4 concentrations are generally higher. Maximum CH 4 concentrations reach 200 μmol L −1 at BRL and 400 μmol L −1 at RWP and occur in low-salinity portions of the STE, whereas maximum CH 4 concentrations reach 7 μmol L −1 at EGN and occur near a salinity of 20. pH increases with salinity at all STE sites (Fig. 3c). DIC and alkalinity mix nonconservatively, and measured concentrations are generally higher than those predicted by conservative mixing although some alkalinity concentrations are lower than those predicted by conservative mixing, particularly at RWP (Fig. 3d,e). Both the ratios of Alk : DIC and proportion of DIC speciated as CO 2 decrease with salinity at all sites (Fig. 3f,g).
δ 13 C-CO 2 and δ 13 C-CH 4 Cross plots of δ 13 C-CO 2 and δ 13 C-CH 4 signatures indicate production pathways of CH 4 as well as extent of oxidation, because each process fractionates carbon between the two species within a range of fractionation factors (ε c ; Whiticar and Schoell 1986;Whiticar 1999). Here, isotopic compositions of CO 2 and CH 4 indicate that methanogenesis at RWP is consistent with CO 2 reduction, while that produced at BRL and EGN is consistent with acetoclastic methanogenesis, as well as CH 4 oxidation (Fig. 4a). The asymptotes of cross plots of CH 4 concentrations vs. δ 13 C-CH 4 values indicate that the δ 13 C signature of CH 4 produced by microbial methanogenesis (δ P ; Eq. 13) is similar for BRL and EGN at −53‰ and −54‰, but lower at RWP at −78‰ (Fig. 4b).
Estimates of CH 4 oxidation are reported as both a fraction of total dissolved CH 4 (f ox ; Eq. 13) as well as mass of CH 4 oxidized (CH 4(ox) ; Eqs. 14-16). Values of f ox show little relationship with salinity (Fig. 5a), and CH 4(ox) concentrations are highest in low-salinity samples of all sites (Fig. 5b). Values of f ox and CH 4(ox) concentrations reach maxima at negative ORP values (Fig. 6a,b).

Mineral saturation indices
Freshwater end members are at equilibrium or slightly supersaturated with respect to quartz at all sites, while saltwater end members are undersaturated. SI quartz of STE samples are variable with salinity (Fig. 7a). Freshwater end members Fig. 4. δ 13 C-CH 4 compared to δ 13 C-CO 2 (a) and CH 4 concentrations (b). Zones of CO 2 reduction and acetoclastic methanogenesis, and expected change in isotopic compositions due to oxidation are taken from Whiticar (1999). Fields are defined by fractionation factors between CO 2 and CH 4 , ε c , where ε c ffi δ 13 C-CO 2 − δ 13 C-CH 4 .

Pain et al.
Sources and sinks of CO 2 and CH 4 are undersaturated with respect to calcite for both BRL and EGN sites, but supersaturated for RWP, while saltwater end members are supersaturated for all sites (Fig. 7b). SI cal values decrease to below the freshwater end member for some STE samples at BRL and RWP, but are consistently higher than the freshwater end member for EGN (Fig. 7b). Ca concentrations display nonconservative mixing with salinity, and measured concentrations are lower than conservative mixing lines in low-salinity STE samples from BRL and RWP and higher than conservative mixing lines for mid-to high-salinity samples. Ca concentrations of STE samples are consistently higher than the conservative mixing line for EGN at all salinities (Fig. 7c).
Redox gradients ORP decreases with salinity at all sites, and steepest decreases are observed in samples with a salinity less than 5 (Fig. 8a). Dissolved organic carbon (DOC) concentrations decrease with salinity for BRL, but increase with salinity for EGN, and remain relatively constant with salinity for RWP (Fig. 8b). Nitrate concentrations are below detection limit for most samples, but several low-salinity samples have elevated con  samples and reach maximum concentrations of 100 μmol L −1 (Fig. 8e).

ΔAlk : ΔDIC ratios
Deviations between measured concentrations of DIC and alkalinity and conservative mixing lines between freshwater and saltwater end members are expressed as ΔDIC and ΔAlk. We show cross plots of samples at each STE site, which are grouped by salinity: low (< 5), mid (between 5 and 20), and high salinity (> 20). Low to mid-salinity samples at BRL are depleted in DIC relative to conservative mixing lines, although ΔAlk values are near 0 (Fig. 9a). With a few exceptions, other data points, ranging from low to high salinity, plot between the ΔAlk : ΔDIC lines of sulfate reduction and methanogenesis. Low-salinity STE samples at EGN plot between the ΔAlk : ΔDIC lines of CaCO 3 dissolution and sulfate reduction. Other samples plot between the ΔAlk : ΔDIC lines of sulfate reduction and denitrification (Fig. 9b). At RWP, fresh STE samples plot near the ΔAlk : ΔDIC line of CaCO 3 precipitation (Fig. 9c). Other samples (low to high salinity) plot near the ΔAlk : ΔDIC line of sulfate reduction.

Discussion
Observed concentrations of CO 2 and CH 4 in Indian River Lagoon STEs vary by orders of magnitude between sites, as well as along salinity gradients within individual study sites (Fig. 3a,b). The variability observed in freshwater end members (Table 2) is small compared to differences between sites and with salinity ( Fig. 3a,b), which indicates that biogeochemical processing within these STEs is the major control on these GHG concentrations and would have a larger impact on CO 2 and CH 4 fluxes from STEs than variability in fresh groundwater composition. Despite differences in concentrations, we observe that measured CH 4 concentrations in STEs are consistently greater than those predicted from conservative mixing and indicate net production, while measured CO 2 concentrations are consistently lower and indicate net consumption. The similarity in chemical changes between sites suggests that processes controlling CO 2 and CH 4 production and consumption are related between sites, but vary in magnitude. We discuss controls of these processes below, followed by implications of these findings for fluxes of CO 2 and CH 4 in SGD.

CH 4 sources and sinks
Sources of methane to Indian River Lagoon STEs include inland fresh groundwater, as shown by elevated CH 4 concentrations in freshwater end members of both BRL and RWP (Table 2), as well as methanogenesis within the STEs. Concentrations of CH 4 in fresh groundwater end members (51 AE 54 and 62 AE 49 μmol L −1 , respectively) are similar to groundwater concentrations in other studies where CH 4 is used as a tracer of SGD (Cable et al. 1996;Corbett et al. 2000;Dulaiova et al. 2010). However, methanogenesis within BRL and RWP STEs increases CH 4 concentrations by 5-to 10-fold. The low freshwater CH 4 concentrations at EGN (<1 μmol L −1 ) are also enhanced by 5-to 10-fold by methanogensis within the STE, but the maximum concentration (7 μmol L −1 ) is orders of magnitude lower than the other sites (hundreds of micromol per liter at BRL and RWP; Fig. 3b). Therefore, although the majority of CH 4 in STEs is produced within the STEs at each site, the magnitudes of CH 4 sources and sinks vary substantially and result in orders of magnitude variability in STE CH 4 concentrations. These findings challenge the underlying assumptions of studies using CH 4 as a conservative tracer of fresh groundwater inputs via SGD and imply that current estimates of SGD volumes, if based on relatively lower inland groundwater CH 4 concentrations, may be overestimated if additional CH 4 is generated when groundwater flows through STEs. CH 4 may still be useful as a qualitative indicator of SGD if combined with other tracers such as Rn and Ra, which do not undergo biogeochemical modification in sediments or the water column and instead reflect the degree of interaction between sediments and saline or fresh pore waters isolated from direct contact with the atmosphere (Cable et al. 1996;Corbett et al. 2000;Dulaiova et al. 2010). However, because SGD fluxes of Rn and Ra to surface waters may result from either fresh SGD or recirculated seawater, while CH 4 concentrations are typically elevated in fresh groundwater only and predominantly reflect fresh SGD, SGD estimates using Rn and Ra are not directly comparable to those using CH 4 , and knowledge of the sources and sinks of CH 4 in STEs is needed to evaluate the role of CH 4 as a tracer of SGD as well as the role of SGD in coastal CH 4 and carbon budgets.
As the terminal metabolic process of organic matter remineralization, methanogenesis occurs when other terminal electron acceptors have been depleted (Froelich et al. 1979). It is therefore more likely to occur in freshwater than salt water, which contains high concentrations of sulfate, because methanogens are readily out-competed for chemical substrates by sulfate-reducing microbes (Whiticar 1999). The redox potential of inflowing fresh groundwater should therefore exert an initial control on CH 4 fluxes in SGD, because methanogenesis may only occur if organic carbon supplies in freshwater are high compared to terminal electron acceptor abundances. While the fresh groundwater end members at BRL and RWP are reducing and support methanogenesis, potentially due to factors such as land use or aquifer solid composition that  Table 1 are displayed: (i) Fe oxidation (Eq. 3), (ii) CaCO 3 dissolution (Eq. 6), (iii) sulfate reduction (Eq. 4), (iv) denitrification (Eq. 2), (v) methanogenesis (Eq. 5), (vi) aerobic respiration (Eq. 1), (vii) Fe/sulfide oxidation (Eqs. 7 and 8), and (viii) CaCO 3 precipitation (Eq. 6).
would impact the redox potential of inflowing groundwater, fresh groundwater at EGN has positive ORP values (Table 2) and low organic carbon content (Fig. 8b), leading to lower CH 4 concentrations compared to BRL and RWP (Table 2). Despite limited methanogenesis in the freshwater end member of EGN, additional CH 4 is generated in more saline portions of the STE (Fig. 3b) but is likely inhibited by relatively high concentrations of sulfate. In contrast, the majority of methanogenesis occurs in fresher portions of the BRL and RWP sites, and likely results from the initial low redox potential of groundwater combined with higher availability of organic carbon in freshwater portions to drive further remineralization reactions. Additional methanogenesis within STEs at BRL and RWP may result from more intense microbial activity driven by increases in the availability of dissolved organic carbon, as has been shown to occur in other STE sites (Suryaputra et al. 2015) and in particular at Indian River Lagoon (Pain et al., In press).
In addition to variability in the concentrations of CH 4 produced in Indian River Lagoon STEs, stable isotopic signatures of CH 4 suggest that its microbial production pathway may differ between sites. Cross plots of δ 13 C-CH 4 and δ 13 C-CO 2 provide evidence of a difference in methanogenesis pathway because data from RWP plot predominantly in the region typical of CO 2 reduction, whereas data from BRL and EGN plot in the acetoclastic region, as well as that produced by CH 4 oxidation (Fig. 4a). Acetoclastic methanogenesis at BRL and EGN is supported by similarity in their δ P signatures: δ 13 C-CH 4 at BRL and EGN are at −53‰ and −54‰, respectively, whereas the δ 13 C-CH 4 of CH 4 produced at RWP is −78‰ and may reflect a greater proportion of CO 2 reduction ( Fig. 4b; Whiticar 1999).
Variations in the predominant methanogenesis pathways are often attributed to salinity or organic carbon quality, and acetoclastic methanogenesis is more typically observed in freshwater environments with greater availabilities of labile organic carbon substrates (Whiticar and Schoell 1986;Megonigal et al. 2005). Salinity is not likely the driver of the differences in pathways between sites here, because although δ 13 C-CH 4 signatures suggest different methanogenesis pathways at BRL and RWP, both occur in fresh portions of STEs (Fig. 3b). Organic carbon substrates may vary in quality, although previous assessment of dissolved organic carbon quality at these sites indicates similar shifts in organic matter quality along STE salinity gradients (Pain et al., In press). The observed difference in methanogenesis pathway may therefore be related to other factors, such as the structure of the microbial community or nutrient availability (Megonigal et al. 2005). While further investigation of methanogenesis pathways is beyond the scope of this study, our findings highlight that both the concentrations as well as methanogenesis pathway in STEs may be more variable than previously considered and may complicate estimates of fluxes and stable isotopic compositions of CH 4 derived from SGD. However, the ubiquitous production of CH 4 at STE sites suggests that current estimates of fluxes of CH 4 via SGD may be underestimates if based on inland groundwater CH 4 concentrations.
Once produced, CH 4 may be consumed through microbial oxidation to CO 2 through both aerobic and anaerobic pathways (e.g., anaerobic oxidation of methane [AOM]). Here, CH 4 oxidation appears to decrease CH 4 concentrations across the entire range of salinity (Fig. 5a). The impact of this oxidation on CH 4 concentrations, and therefore on SGD fluxes, can be examined by converting f ox to mass of CH 4 oxidized (Eqs. 14-16). The magnitude of the CH 4 oxidation sink is greater in lowsalinity samples of BRL and RWP, where CH 4 concentrations are highest (Fig. 5b). Additionally, the greatest extent of CH 4 oxidation, both in terms of f ox and CH 4(ox) , is greatest at negative ORP values (Fig. 6a,b), indicating that oxidation is likely anaerobic.
Both the production of CH 4 via methanogenesis and consumption via methane oxidation alter concentrations of CH 4 in STEs and therefore will alter fluxes via SGD. The magnitude of CH 4 sources is greater than sinks in the STEs sampled here, leading to measured CH 4 concentrations that are greater than that expected by conservative mixing (Fig. 3b). Our results suggest that several generalizations may be made about the controls of CH 4 sources and sinks in STEs. The magnitude of the CH 4 source is likely related to groundwater redox potential. Methanogenesis may be a dominant organic matter remineralization pathway as reducing water flows through fresher portions of STEs where inhibition from seawater sulfate is minimized. The magnitude of the CH 4 oxidation sink may be limited by groundwater residence time, or sulfate and oxygen gradients. Specifically, long groundwater residence times (~195 d) between the zones of methanogenesis and discharge at the sediment-water interface have led to near-complete consumption of groundwater-derived CH 4 in other settings, implying that increases in groundwater residence time may be associated with lower CH 4 fluxes via SGD (Schutte et al. 2016). CH 4 oxidation may also be chemically limited because O 2 is required for aerobic CH 4 oxidation while AOM is frequently coupled to sulfate reduction (Megonigal et al. 2005). The rate of delivery of O 2 to STE sediments may therefore control the extent of aerobic CH 4 oxidation and would be limited in these STEs, because oxygen is rapidly reduced within the first few millimeters of the sediment-water interface. AOM would vary across the salinity gradient, because seawater is the principal source of sulfate in STEs (Megonigal et al. 2005). Consequently, this and other work (e.g., Schutte et al. 2016) suggest that oxidation modifies CH 4 fluxes via SGD.

CO 2 sources and sinks
Contrasting with observations of CH 4 production at all Indian River Lagoon STE sites, CO 2 concentrations are consistently lower than those predicted by conservative mixing models (Fig. 3a) and indicate net consumption. These changes are accompanied by increases in pH (Fig. 3c), increases in alkalinity above conservative mixing values (Fig. 3e), increases in ratios of Alk : DIC (Fig. 3f), and decreases in the proportion of DIC speciated as CO 2 (Fig. 3g). The ratio of Alk : DIC determines the CO 2 buffering capacity of water, and higher ratios of Alk : DIC correspond with lower proportions of DIC speciated as CO 2 (Denman et al. 2007). Changes to Alk : DIC ratios result from reactions that produce and consume DIC and Alk in different proportions, and the stoichiometries of these reactions are well-known (Table 1). Here, increases in Alk : DIC with salinity ( Fig. 3f) suggest that in situ reactions in STEs cause alkalinity to be generated in greater quantities than DIC as water flows through STE sediments. These changes alter the carbonate speciation of pore waters and leads to the net sequestration of CO 2 as HCO 3 − , thus decreasing SGD fluxes of CO 2 .
Reactions that alter carbonate chemistry may occur in STEs due to both changes in the solid-phase materials available to drive reactions as water flows from terrestrial aquifers through mixed marine and terrestrial sediments, as well as shifts in terminal electron acceptors that may be used to remineralize organic matter. In sediments containing calcium carbonate minerals, CaCO 3 dissolution is frequently the dominant reaction that sequesters CO 2 as HCO 3 − . In the STEs sampled here, changes in mineral saturation indices and nonconservative mixing of Ca suggest that CaCO 3 dissolution occurs and produces alkalinity. This production reflects an increase in CaCO 3 mineral availability between predominantly siliciclastic inland aquifer material (Berndt and Katz 1992) and mixed terrestrial and marine lagoon sediments (Dorsett et al. 2011). Fresh groundwater end members are in equilibrium with siliciclastic inland aquifer material (SI quartz ≥ 0; Fig. 7a), whereas SI cal ≤ 0, reflecting limited interaction with CaCO 3 minerals (Fig. 7b). However, SI cal values increase in STE samples compared to fresh groundwater end members, particularly at shoreline piezometers of BRL and EGN, and Ca concentrations above conservative mixing lines suggest that CaCO 3 dissolution occurs (Fig. 7b,c). These changes suggest that rapid CaCO 3 dissolution occurs when groundwater that is undersaturated with respect to CaCO 3 begins to interact with lagoon sediments. This interaction generates alkalinity in greater proportions than DIC (Eq. 6) thus increasing Alk : DIC ratios (Fig. 3g) and leads to net CO 2 sequestration (Fig. 3a). ΔAlk : ΔDIC ratios of EGN freshwater are consistent with CaCO 3 dissolution as the primary reaction altering carbonate speciation and plot near the ΔAlk : ΔDIC line produced by CaCO 3 dissolution (Fig. 9b). Although CaCO 3 dissolution may alter carbonate chemistry, particularly at the shoreline piezometers of EGN and BRL, the role of CaCO 3 dissolution should be limited when waters are supersaturated or approach equilibrium with respect to CaCO 3 . This is the case for most STE samples apart from shoreline piezometers of BRL and EGN. In some cases, SI cal values near or above 0 and Ca concentrations lower than conservative mixing suggests that some CaCO 3 precipitation occurs, particularly in fresh portions of BRL and RWP (Fig. 7b,  c). Additional reactions are therefore necessary to cause the increases in Alk : DIC ratios observed with increasing salinity, because CaCO 3 precipitation would decrease Alk : DIC ratios (Eq. 6). These reactions may include sulfate reduction, denitrification, and iron reduction, which play an important role in the generation of alkalinity in siliciclastic settings with limited CaCO 3 availability (Table 1; Berner et al. 1970). The alkalinity generated by redox reactions should vary with salinity, because freshwater-saltwater mixing in STEs leads to sharp gradients in redox potential, and each redox reaction produces DIC and alkalinity at distinct ratios (Table 1).
ΔAlk : ΔDIC ratios indicate that redox reactions control carbonate chemistry to a greater extent than CaCO 3 dissolution in mid-to high-salinity samples, but reactions differ between STE sites. At EGN, where CaCO 3 dissolution is an important process for the water composition at the shoreline, mid-and high-salinity samples reflect a greater importance of sulfate reduction or denitrification (Fig. 9b). Denitrification likely occurs in freshwater samples of EGN, because nitrate concentrations reach 200 μmol L −1 (Fig. 8c), but consumption of all nitrate via denitrification would only yield up to 250 μmol L −1 DIC (Eq. 2). Because ΔDIC values are in the millimol per liter range in the low-salinity samples that contain high nitrate concentrations (Fig. 9b), denitrification likely plays only a limited role in generating excess DIC and alkalinity because of low nitrate concentrations. While maximum HS − concentrations reach 100 μmol L −1 (Fig. 8e), which would only yield a ΔDIC of 200 μmol L −1 according to reaction stoichiometry (Eq. 4), measured concentrations do not reflect the full extent of sulfate reduction, because iron sulfide mineral precipitation is known to occur and may consume nearly all free Fe(II) and HS − . The close proximity of EGN mid-and high-salinity data points to the ΔAlk : ΔDIC line produced by sulfate reduction suggests that it is the predominant redox reaction altering carbonate speciation in higher salinity waters where pore waters are close to equilibrium with respect to CaCO 3 minerals (Fig. 7b).
ΔAlk : ΔDIC ratios at BRL and RWP cluster between the lines produced by sulfate reduction and methanogenesis in the upper right-hand and lower left-hand quadrant of the ΔAlk : ΔDIC plot, indicating net DIC consumption (Fig. 9a,c). Data points between the ΔAlk : ΔDIC ratios of sulfate reduction and methanogenesis may result from the co-occurrence of these processes. Co-occurrence is supported by elevated sulfide concentrations (up to 150 μmol L −1 at BRL and 100 μmol L −1 at RWP) across the observed salinity range, coinciding with elevated CH 4 concentrations in low (< 5) to mid (5-20) salinity samples at both sites.
A number of fresh to mid-salinity samples of BRL and RWP are depleted in DIC relative to conservative mixing (Fig. 9a,c). Some of this DIC consumption may be related to CaCO 3 precipitation, which is supported by some positive SI cal values and Ca concentrations lower than conservative mixing values in freshwater portions of both sites (Fig. 7b,c). For BRL in particular, additional reactions such as Fe reduction would be necessary to produce alkalinity, as data points are disproportionately depleted in DIC compared to alkalinity if DIC consumption is due to CaCO 3 precipitation alone (Fig. 9a). While further investigation is needed to determine the cause of the DIC sink in these waters, our results indicate that both redox reactions as well as shifts in CaCO 3 availability and mineral saturation state drive net CO 2 sequestration in STEs and serve to decrease CO 2 fluxes via SGD. However, despite the CO 2 sequestration in STEs, most of our STE samples have CO 2 concentrations elevated above surface water concentrations, indicating that SGD is a source of CO 2 to the Indian River Lagoon. Exports of alkalinity via SGD may additionally impact surface water CO 2 dynamics by altering surface water DIC : Alk ratios, which determines dissolved carbonate speciation. Here, nonconservative behavior and production of alkalinity in STEs may provide additional alkalinity to surface waters, which may enhance surface water buffering capacity and sequestration of CO 2 as HCO 3 − , as was observed in Liu et al. (2017).

Implications for GHG fluxes via SGD
We observe that reactions in STEs significantly modify the GHG composition of SGD, and lead to net production of CH 4 coinciding with net sequestration of CO 2 . These findings highlight challenges in the estimation of GHG fluxes of SGD, because despite relatively low variation in the concentration of CO 2 and CH 4 in inflowing groundwater, STE reactions lead to orders of magnitude differences in the concentrations of both gases. Extrapolation of SGD GHG fluxes even within the hydrogeologically homogenous and geographically constrained setting of this study therefore must account for a high degree of uncertainty in the estimation of SGD GHG concentrations. Despite differences in concentrations between sites, STE samples contain higher concentrations of both CO 2 and CH 4 compared to saltwater end members (Fig. 3a,b), and SGD is therefore a source of GHGs to coastal zones. The observed variability in CO 2 and CH 4 concentrations resulting from STE biogeochemical processing may preclude large-scale estimates of SGD on the coastal carbon budget.
Although STEs examined in this study are likely similar in hydrogeological characteristics such as groundwater flow rate, residence time in the STE, and composition of sediments, variations in these parameters are likely to impact STE processing of CO 2 and CH 4 concentrations and thus fluxes due to SGD. Residence time may be related to STE hydrogeology, and should be greater when SGD flows through fine grained sediments compared to coarse sand or karstic conduits (Pain et al., In press). Temporal variability due to changes in temperature or STE end member composition may also occur, as well as variation in groundwater residence times due to seasonal differences in recharge and groundwater flow rates (Michael et al. 2005). In the case of CH 4 , increasing STE residence time would likely lead to greater consumption of CH 4 via methanotrophy, and greater residence times may therefore reduce groundwater fluxes of CH 4 . Land use may play an additional role by altering the organic carbon and terminal electron acceptor concentrations and thus the redox potential of inflowing groundwater. The magnitude of the CO 2 sink within STEs should depend on the mineralogical composition of sediments as alkalinity may be produced via calcium carbonate mineral dissolution as well as Fe oxide reduction, and would thus depend on the availability of these mineral phases to generate alkalinity and sequester CO 2 . Given the likely impacts of sediment composition, groundwater flow rates, and residence times of STEs on CO 2 and CH 4 concentrations, CO 2 and CH 4 fluxes due to SGD are likely highly variable but could have important impacts on coastal carbon budgets due to their elevated concentrations with respect to surface waters and nonconservative behavior in STEs.