The basal regions of continental ice sheets are gaps in our current understanding of the Earth's biosphere and biogeochemical cycles. We draw on existing and new chemical data sets for subglacial meltwaters to provide the first comprehensive assessment of sub-ice sheet biogeochemical weathering. We show that size of the ice mass is a critical control on the balance of chemical weathering processes and that microbial activity is ubiquitous in driving dissolution. Carbonate dissolution fueled by sulfide oxidation and microbial CO2 dominate beneath small valley glaciers. Prolonged meltwater residence times and greater isolation characteristic of ice sheets lead to the development of anoxia and enhanced silicate dissolution due to calcite saturation. We show that sub-ice sheet environments are highly geochemically reactive and should be considered in regional and global solute budgets. For example, calculated solute fluxes from Antarctica (72–130 t yr−1) are the same order of magnitude as those from some of the world's largest rivers and rates of chemical weathering (10–17 t km−2 yr−1) are high for the annual specific discharge (2.3–4.1 × 10−3 m). Our model of chemical weathering dynamics provides important information on subglacial biodiversity and global biogeochemical cycles and may be used to design strategies for the first sampling of Antarctic Subglacial Lakes and other sub-ice sheet environments for the next decade.
 Chemical weathering beneath the ice sheets may have a global impact if the balance of dissolution processes is similar to that beneath valley glaciers. Recent works have hypothesized that the release of bioavailable iron from sediments in melting icebergs [Raiswell et al., 2006, 2008, 2009] and in aqueous form in the glacial runoff [Föllmi et al., 2009] may be an important stimulant of ocean primary productivity, with potential negative feedback effects on climate via CO2 drawdown. This iron is derived from sulfides present in the bedrock, which is continually exposed by subglacial comminution and, subsequently, is oxidized to produce reactive Fe(III) oxyhydroxides. Intense subglacial physical erosion also promotes high rates of detrital phosphorus export from glaciers with the potential to sustain downstream productivity [Föllmi et al., 2009]. Others have suggested that anoxia generated beneath the continental ice sheets may promote the production of methane after periods of ice advance and carbon burial. This is then released during glacial retreat with the potential to stimulate greenhouse warming [Skidmore et al., 2000; Weitemeyer and Buffett, 2008; Wadham et al., 2008].
 Syntheses of glacial chemical weathering to date have used the chemical composition of bulk runoff from valley glaciers to infer chemical dissolution processes beneath ice sheets [Sharp et al., 1995; Anderson et al., 1997; Hodson et al., 2000]. We argue that these studies failed to fully capture the variability of biogeochemical processes in subglacial environments, particularly long-residence time systems, because (1) mixing processes skew the chemical signature toward the larger-volume dilute channelized component and (2) subglacial meltwater chemistry becomes modified in ice-marginal environments [Brown et al., 1994]. Data from valley glaciers are difficult to scale up to ice sheets, where meltwater residence times and the degree of isolation from the atmosphere are very different. Pure subglacial meltwaters have been sampled for chemical parameters from just a few glaciers worldwide because of the logistical constraints of obtaining samples. We draw on the full spectrum of these data sets and combine them with new ice sheet chemical data sets to provide the first assessment of biogeochemical processes beneath ice sheets.
2. Subglacial Data Sets and Data Analysis
 We use subglacial chemical data from eight ice masses differing in scale (Table 1 and Appendix A for unpublished methods), inferring that subglacial meltwater residence time scales approximately with the size of the ice mass and the degree of isolation from surface environments. The Haut Glacier d'Arolla (Switzerland, 6.3 km2) and Engabreen (Norway, 40 km2) are temperate glacier end-members. They have ice that is at the pressure melting point throughout during summer, and they have well-developed subglacial drainage systems. Midre Lovénbreen (5 km2) is their “polythermal” counterpart, where the surface and margins of the glacier are below the pressure melting point, but parts of the bed are temperate. Finsterwalderbreen (Svalbard) is a medium-sized (∼40 km2) polythermal system, where surface cold ice reduces surface-to-bed hydrological connectivity, prolonging residence times. The largest and most isolated systems are the Bristol Glacier (Greenland Ice Sheet), Lake Vostok (East Antarctica: estimated residence time = 32,000 years; Thoma et al., 2008) and the Kamb and Bindschadler ice streams (KIS and BIS, respectively, West Antarctica). On Greenland, surface meltwater penetrates the bed in the ablation zone. In the Antarctic cases, there is no surface meltwater penetration to the ice sheet bed. Hence, all meltwater derives from basal melting of the ice sheet base due to geothermal and frictional heating. This is anticipated to result in low meltwater fluxes and prolonged meltwater residence times. We also use Lake Vostok accretion ice data, derived from two separate data sets of waters that have frozen to the base of the Antarctic Ice Sheet above Lake Vostok [DeAngelis et al., 2004] (Appendix A for unpublished methods). As for all data sets, we analyze the associations between ions rather than absolute ion concentrations. This minimizes issues associated with (differential) ion rejection during freezing for Vostok ice and the fact that the true lake water concentrations are consequently much higher than those in the accretion ice. Hence, although there may be considerable scatter in plots of one ion versus another, lines of best fit give the average chemical composition of the accretion ice regardless of freezing effects on ion ratios.
Table 1. Details of Subglacial Data Sets
Reference If Published
Lake Vostok, Antarctica
Kamb and Bindschadler ice streams, West Antarctic Ice Sheet
 We analyze four primary chemical indices. First and second are the concentrations of SO42− and HCO3− that reflect the relative dominance of several major chemical weathering mechanisms, sulfide oxidation, and the sum of carbonation, carbonate hydrolysis, and carbonate dissolution, respectively [equations (1)–(4)] [Holland, 1978].
Third is the (*Ca2+ + *Mg2+) concentration, which is used as a bulk indicator of carbonate dissolution. Fourth is the concentration of (*Na+ + *K+), indicative of silicate dissolution [Holland et al., 1978]. All cation and sulfate concentrations are corrected for their sea salt– and snowpack-derived components, respectively, using Cl− concentrations and ratios of each species to Cl− in either seawater (cations) or locally sampled snow (SO42−). The snowpack sulfate contributions in the Vostok data are derived from mean ratios in the meteoric ice immediately above the accretion ice (Appendix A). Sea salt ratios are used for all ion corrections in BIS and KIS. The residual crustal-derived component is denoted with an asterisk (*).
 We focus on the gradients and intercepts observed for the lines of best fit for associations of HCO3− versus *SO42− and *Ca2+ + *Mg2+ versus *SO42−. We present the mean ratios of HCO3−/*SO42− and *Ca2+ + *Mg2+/*SO42− for the KIS and BIS pore waters owing to the small number of data points. The primary focus on the associations between ions rather than on the absolute ion concentrations minimizes the issues associated with ion rejection during freezing for Vostok ice and the fact that the true lake water concentrations are consequently much higher than those in the accretion ice. Figure 1 displays the HCO3− versus *SO42− associations for all data sets (excluding the KIS and BIS) and provides an example of the plots used for analysis. A wide range of gradients and intercepts are evident (Table 2), which can be used to identify the systematic trends in the dissolution processes between ice masses of different scale.
Table 2. Associations and Values for Individual Data Sets
HCO3− versus *SO42−
*Ca2+ + *Mg2+ versus *SO42−
Mean *Ca2+ + *Mg2+/*Na+ + *K+
Haut Glacier d'Arolla (all data)
y = 0.980 + 188 (n = 139, r2 = 0.75)
y = 1.96x + 171 (n = 140, r2 = 0.93)
26.0 (n = 152)
y = 0.770 + 205 (n = 126, r2 = 0.84)
y = 1.75x + 188 (n = 126, r2 = 0.97)
24.0 (n = 138)
y = 0.72x + 405 (n = 14, r2 = 0.7)
y = 2.00x + 174 (n = 14, r2 = 0.97)
4.70 (n = 14)
y = 1.13x + 87.3 (n = 10, r2 = 0.94)
y = 2.10 + 97.0 (n = 10, r2 = 0.9)
18.0 (n = 10)
Y = 0.981x + 61.7 (n = 6, r2 = 0.46)
y = 1.80x + 102 (n = 6, r2 = 0.86)
17.4 (n = 6)
y = 0.570x + 153 (n = 25, r2 = 0.89)
y = 1.33x + 172 (r2 = 0.68)
8.65 (n = 25)
y = 1.91x + 151 (n = 22, r2 = 0.8)
y = 1.78x + 2.67 (n = 22, r2 = 0.82)
10.4 (n = 22)
y = 0.868 + 334 (n = 26, r2 = 0.9)
y = 1.71 + 307 (n = 26, r2 = 0.98)
10.1 (n = 26)
y = 0.540 + 569 (n = 13, r2 = 0.92)
y = 1.45x + 444 (n = 13, R = 0.99)
6.70 (n = 13)
y = 0.280x + 686 (n = 8, r2 = 0.27)
1.283x + 619 (n = 8, r2 = 0.94)
11.8 (n = 8)
y = 1.44x + 278 (n = 32, r2 = 0.77)
y = 1.82x + 379 (n = 33, r2 = 0.84)
5.70 (n = 33)
y = 0.660x + 340 (n = 21, r2 = 0.7)
1.34x + 415 (n = 22, r2 = 0.81)
12.0 (n = 19)
Bristol Glacier, Greenland
y = 0.175x + 135 (n = 14, r2 = 0.08)
y = 1.24x + 86.0 (n = 14, r2 = 0.8)
3.40 (n = 14)
y = 0.157x + 1.25 (n = 29, r2 = 0.62)
y = 0.95x + 0.790 (n = 29, r2 = 0.99)
4.95 (n = 5)
Kamb Ice Stream (mean values reported)
0.240 (n = 3)
0.550 (n = 3)
0.970 (n = 3)
Bindschadler Ice Stream (mean values reported)
0.120 (n = 7)
0.600 (n = 7)
1.01 (n = 7)
3. Chemical Weathering Processes Under Ice: A Predictive Framework
 The first reaction to occur on the wetting of the subglacial flour is hydrolysis, predominantly of calcite, which generates ∼220 μeq L−1 of HCO3− and Ca2+ in line with the theoretical solubility of calcite in pure water at 0°C [equation (3)] [Tranter et al., 2002a]. Because this reaction involves the rapid acquisition of HCO3− and *Ca2+ in the absence of *SO42−, the intercept in the plots of HCO3− versus *SO42− should be approximately 220 μeq L−1. We note that the intercept values for Vostok accretion ice are artificially low because of the solute exclusion during freezing (Figure 1 and Table 2). Of the remaining data sets, approximately half display intercepts of >220 μeq L−1 in plots of HCO3− and *Ca2+ + *Mg2+ versus *SO42− (Figure 2a and Table 1). This range of intercepts suggests an additional source of CO2 to meltwaters. Acquisition of solute by carbonation reactions using atmospheric CO2 in subglacial environments is negligible because these systems are out of direct contact with the atmosphere and often have high partial pressures of CO2 because of the proton supply from sulfide oxidation and the inability of meltwaters to degas [Tranter et al., 2002a, 2005; Wadham et al., 1998]. Hence, we infer that the additional HCO3− is acquired via the generation of CO2 during the microbial oxidation of organic carbon in subglacial sediments. A direct and significant association (P = 0.01, r2 = 0.83) between the intercepts in plots of both HCO3− and *Ca2+ + *Mg2+ versus *SO42− implies that most of this excess HCO3− is balanced by a similar quantity of *Ca2+ + *Mg2+ (Figure 2a). A gradient of 0.79, Figure 2a further implies that, on average, 80% of microbial CO2 dissolves calcite, with the remainder dissolving silicates. This is a maximum estimate of calcite dissolution because it does not account for the fact that minor amounts of *Mg2+ and *Ca2+ may be derived from silicates. It is consistent, however, with the reaction kinetics for both dissolution processes [Tranter et al., 2002a]. The amount of microbial CO2 is bedrock substrate-dependent, with the hard rock-bedded glacier systems (Arolla, Greenland, Engabreen, and Greenland Ice Sheet) displaying the lowest intercepts (Figure 2a). Higher intercepts for Finsterwalderbreen and Midre Lovénbreen are consistent with the drainage of subglacial meltwaters (via artesian upwellings) from sediment-rich environments [Wadham et al., 1998], which are believed to dominate significant areas of the glacier bed and result from the erosion of underlying sedimentary rocks. These findings reveal that glaciers based on hard, less reactive bedrock constitute less favorable environments for microbial activity, which is consistent with observations elsewhere that subglacial microbial populations tend to correlate with sediment distributions [Sharp et al., 1999].
 Subsequent to carbonate hydrolysis, subglacial waters may evolve via the oxidation of sulfide minerals continually regenerated by glacial comminution and oxidized microbially under both oxic and anoxic conditions [Tranter et al., 2002a, 2002b; Bottrell and Tranter, 2002; Wadham et al., 2004; Wynn et al., 2006]. Sulfide oxidation provides protons for carbonate and silicate dissolution [equations (5) and (6)]. Sulfate is produced during both coupled sulfide oxidation/carbonate dissolution and sulfide oxidation/silicate dissolution. HCO3− is generated during coupled sulfide oxidation/carbonate dissolution [equation (5)] but not during sulfide oxidation/silicate dissolution [equation (6)].
The stoichiometry of the coupled sulfide oxidation/carbonate dissolution generates SO42− in molar equivalent ratios of 1:1 with respect to HCO3− and 2:1 with respect to *Ca2+ + *Mg2+. Hence, gradients of ∼1 and 2 in plots of HCO3− versus *SO42− and *Ca2+ + *Mg2+, respectively, reflect complete coupling of sulfide oxidation and carbonate dissolution (via anoxic or oxic mechanisms; Tranter et al. [2002a]), whereas gradients of <1 and <2, respectively, suggest that some protons from sulfide oxidation dissolve silicate minerals, liberating Na+ and K+ [Tranter et al., 2002a]. Most subglacial meltwaters display gradients of <1 and <2, respectively (Figure 2b), suggesting that protons generated from the oxidation of sulfide minerals are indeed used in variable degrees to dissolve silicate minerals. In support of this contention is the direct and significant association (r2 = 0.48, P = 0.01) between mean ratios of *Ca2+ + *Mg2+/*Na+*K+ and gradients of HCO3− versus *SO42− (Figure 2c). Figures 2b and c also indicate that the degree of coupled silicate dissolution and sulfide oxidation is directly related to the scale of the ice mass and to the degree of subglacial isolation. The temperate valley glaciers (Engabreen, Arolla), with well-developed subglacial drainage networks, display gradients close to the theoretical values of 1 and 2, respectively. The middle-scale polythermal system, Finsterwalderbreen, displays low to intermediate gradients of 0.2–0.66 and 1.3–1.4, respectively, while Midre Lovénbreen exhibits a range of values that largely reflect changing hydrological conditions between years. The lowest gradients of HCO3− and *Ca2+ + *Mg2+ versus *SO42− are evident in the four sub-ice sheet meltwaters (0.1–0.2 and 0.6–1.2, respectively). The spatial patterns in mean ratios of *Ca2+ + *Mg2+/*Na+*K+ also map with the scale of the ice mass (Figure 2c), with sub-ice sheet meltwaters displaying the lowest ratios (<5). Estimates of meltwater residence times beneath Antarctica and Greenland are wide ranging (from days in parts of Greenland [Wadham, unpublished data, 2009] to 104 years for Lake Vostok [Thoma et al., 2008]) but are likely to be orders of magnitude higher than for valley glaciers because surface melt inputs are absent over large parts of the ice sheets. The most likely explanation for the enhanced silicate dissolution in these remote, long-residence time systems is the attainment of saturation with respect to calcite, but calcite exhaustion may also occur.
 These assertions suggest that the dominant control on subglacial meltwater chemical composition is the scale of the ice mass. We contend that bedrock lithology exerts a second-order effect. For example, meltwaters from all seven ice masses display significant concentrations of Ca2+ as the dominant cation and HCO3− and SO42− as the primary anions, supporting the assertion that sulfide and calcite minerals are present in most (noncarbonate) bedrock types and dominate chemical dissolution because of their high geochemical reactivity [Anderson, 2007]. The smaller temperate systems, Engabreen and Haut Glacier d'Arolla, have silicate-rich bedrock lithologies, yet their subglacial chemical composition principally reflects the action of coupled sulfide oxidation/carbonate dissolution. Our data do show, however, that bedrock affects meltwater chemistry by controlling the development of a suitable microbial substrate. Softer sedimentary bedrocks are likely to give rise to thicker, more widespread, subglacial sedimentary deposits, which, in turn, provide a suitable habitat for microbial communities. Hence, meltwaters from these environments display higher proportions of microbially generated CO2 and higher intercepts for lines of best fit in the plots of HCO3− and *Ca2+ + *Mg2+ against *SO42− (e.g., Finsterwalderbreen). In contrast, glaciers based on less erodible substrates (e.g., continental shield bedrock) may have more localized and thinner subglacial debris deposits and, hence, less microbial activity.
4. Wider Impacts of Sub-Ice Sheet Chemical Weathering
 The above synthesis can be used as a framework for evaluating the likely evolution of meltwaters beneath ice sheets, including the Antarctic Subglacial Lakes. Our findings have four significant implications. First, microbially generated CO2 is an important driver of chemical weathering in many subglacial environments, as indicated by the high intercepts in the plots of HCO3− and *Ca2+ + *Mg2+ versus *SO42− (Figure 2a). We calculate that this microbially derived HCO3− accounts for up to 19%–65% (mean = 36%, n = 5) of the total HCO3− and that 14%–37% (mean = 22%, n = 5) of the total solute is acquired using microbial CO2 as a proton source (Appendix B). The assertion that microbial CO2 drives significant subglacial chemical weathering counters a previous work, which has suggested that exhaustion of carbonic acid poses a limitation on the dissolution processes beneath glaciers [Anderson, 2007]. Our data further indicate that this microbial effect is greatest for glaciers overlying soft sediments and, therefore, could be important in driving chemical dissolution beneath ice streams and outlet glaciers resting on deformable sediments. This is particularly true in Antarctica, where basal melt is the primary source of O2 and CO2 because there is a limited supply of these gases to the basal environment from surface melting.
 Second, the oxidation of both organic carbon and sulfide minerals strips oxygen from meltwaters, creating the potential for a diversity of redox conditions at ice sheet beds and highly anoxic conditions in the ice sheet interior. This corroborates what is already known about valley glaciers, where sulfides are oxidized anoxically by chemolithoautotrophs, with nitrate or Fe(III) as terminal electron acceptors [Bottrell and Tranter, 2002; Wadham et al., 2004, 2007; Wynn et al., 2006]. Our data compilation implies that sulfide oxidation continues to be important under ice sheets. Certainly, if SO42− present in the KIS and BIS waters is derived from sulfide oxidation, it is impossible to generate the concentrations of sulfate observed (53 meq L−1) without driving the system anoxic (e.g., 300–400 μeq L−1 of SO42− can be produced from oxygen-saturated snowmelt at 0°C [Tranter et al., 2002a]). Hence, it is likely indicated that organic carbon and sulfide minerals are being oxidized using alternative oxidizing agents to the atmospheric oxygen. If this is the case, then ice sheet beds may be a host to diverse syntrophic communities of microbes, adapted to local bed Eh/pH conditions, which are, in turn, perpetuated by this very microbial activity. This analysis lends strong support to the contention that sub-ice sheet environments may have been sites of methanogenesis during periods of glacial advance and organic carbon burial and were important sources of methane to the atmosphere during ice wastage [Skidmore et al., 2000; Weitemeyer and Buffett, 2008; Wadham et al., 2008]. It also supports recent contentions that bioavailable nanoparticulate iron (oxyhydroxides) can be generated beneath ice sheets in a similar way to that beneath valley glaciers and that continual regeneration of sulfide minerals by comminution keeps pace with oxidation processes [Raiswell et al., 2006, 2008, 2009]. This is consistent with the amplification of 16S rRNA gene sequences that are closely related to species known to be involved in sulfur and iron oxidation from subglacial waters from Bench Glacier, AL [Skidmore et al., 2005]; KIS basal sediments [Lanoil et al., 2009]; and Vostok accretion ice [Christner et al., 2006]. It lends powerful support to the idea that sediments released from ice sheets, via icebergs or proglacial sediment deposition, may have the potential to stimulate ocean primary production [Raiswell et al., 2006, 2008, 2009; Föllmi et al., 2009].
 Third, our data show that as scale and, hence, residence time/isolation increase, silicate mineral dissolution dominates as meltwaters attain saturation with respect to calcite. Because silicate minerals contain trace amounts of nitrogen (e.g., in micas and feldspars) and potassium (in feldspars and biotite) [Holloway and Dahlgren, 2002], their enhanced dissolution can supply subglacial environments with the nutrients required to sustain microbial life. The high concentrations (30 meq L−1) of *Na+ and *K+ in the KIS and BIS pore waters are consistent with enhanced silicate dissolution [Skidmore et al., 2010] and suggest that chemical weathering beneath Antarctic Ice Streams is highly aggressive, driven by sulfide oxidation and microbially produced CO2. Consistent with these assertions is the morphology of sand grains sampled from the base of the Whillans Ice Stream (B), which are dominated by chemical weathering microfeatures similar to those visible on particles that have been crushed and treated with HF acid [Tulaczyk et al., 1998]. Enhanced silicate weathering also help explain the abundance of clay mineral assemblages (e.g., kaolinite and smectite) recovered in marine sediment cores located close to Antarctic ice stream outlets [Ehrmann et al., 1992] that have previously been inferred to reflect long-distance transport of preweathered material [Ehrmann et al., 1992]. This assertion is supported by the presence of kaolinite and smectite in Whillans Ice Stream sediments [Tulaczyk et al., 1998]. If this is the case, there are implications for inferring past glacial influences on marine sedimentary record (i.e., changes in transport versus variations in in situ chemical dissolution).
 Fourth, our data reveal that sub-ice sheet environments are highly reactive and dynamic reservoirs of water and solute, with the potential for export to downstream environments. Numerous studies have investigated the global significance of valley glaciers and northern hemisphere ice sheets for global solute fluxes and CO2 drawdown [Anderson et al., 1997; Sharp et al., 1999; Hodson et al., 2000; Tranter et al., 2002a, 2002b], yet the export of solute from Antarctica remains unquantified, apart from a single study focused on dissolved iron export [Statham et al., 2008]. Figure 2 indicates KIS and BIS pore water features at the end of a spectrum of meltwater chemical compositions, clustering with the other ice sheet meltwaters. These data provide a unique opportunity to estimate potential solute fluxes from the Antarctic Ice Sheet for the first time because they represent the chemistry of sub-ice stream meltwaters before their discharge to the ocean, e.g., via outburst events linked to subglacial lake drainage [Fricker et al., 2007; Goodwin, 1988].
 We focus our analysis on the peripheral fast-flow areas of the Antarctic Ice Sheet, where basal melting is indicated by accelerated ice flow (>10 m yr−1) and the presence of active subglacial lakes [Smith et al., 2009]. Here, meltwater is likely to have a clear path to the ice margin without undergoing basal freeze-on during export [Smith et al., 2009]. We calculate water and solute fluxes from (1) the “Ice Sheet Interior, ” where velocities are 10–100 m yr−1 [Bamber et al., 2000] and assuming mean basal melt rates of 1.6 mm yr−1 [Joughin et al., 2004], and (2) the “Ice Streams, ” where ice velocities are >100 m yr−1 [Bennett, 2003]. We define two types of ice stream according to the pattern of basal melting. The Ross Ice Streams are a well-known example of what we term here as “low-melt rate ice” (LMR) streams, where basal melt rates are uniformly <10 mm yr−1. We calculate total solute fluxes from the grounded sector of the Ross Ice Streams, including ice sheet, tributaries, and ice stream components, using published basal melt rates and catchment areas [Joughin et al., 2004]. Pine Island Glacier (PIG) is an example of what we term “high-melt rate” (HMR) ice streams, where a proportion of the bed exhibits unusually high basal melt rates (>10 mm yr−1 [Joughin et al., 2003]). We calculate solute fluxes from the grounded sector of PIG catchment. Here, we use basal melt rates for the strong bed area of 500 mm yr−1 (estimated as ∼2.5% of the total area of PIG [Joughin et al., 2003]) and assume that basal melt rates over the rest of the PIG catchment are similar to mean values for the Ross Ice Streams [Joughin et al., 2004]. Because the relative proportion of LMR versus HMR ice streams in Antarctica is unknown, we create two scenarios to predict what the overall fast-flow ice stream contribution to Antarctic solute fluxes might be. Areas of fast-flow (>100 m yr−1) presently account for 20% of the total grounded ice sheet area (calculated from Bamber et al. ). Scenario A assumes that LMR ice streams account for 90% of the fast-flow area, and HMR comprise the remaining 10%. In scenario B, LMR and HMR ice streams have equivalent areas (i.e., 50% of the total). In both scenarios, we assume that both types of ice stream have rates of solute removal similar to the average values reported for the Ross Ice Streams and PIG (Table 3; values of 10.3 and 63 t km−2 yr−1, respectively). Total maximum and minimum estimate solute fluxes are derived for Antarctica by adding the interior ice sheet contribution (Table 3, row 1) to fluxes calculated in scenario A (minimum estimate) and scenario B (maximum estimate).
Table 3. Calculations of the Potential Solute and Water Fluxes from Antarctica
1. “Antarctic Interior” (ice velocity 10–100 m yr−1)
1.6 × 10−3
2. Ross Ice Streams (LMR)
3 × 10−3
3. Pine Island Glacier (2.5% area = HMR)
1.5 × 10−2
4. Scenario A: Ice Streams
3.8 × 10−3
5. Scenario B: Ice Streams
9.0 × 10−3
6. Ice Streams + “Interior” (1 + 4)
2.3 × 10−3
7. Ice Streams + “Interior” (1 + 5)
4.1 × 10−3
 We use the mean chemical composition (4.1× 10−3 t m−3) of the KIS and BIS pore waters as an estimate of the total dissolved solute concentration in meltwaters draining the East and West Antarctic Ice Sheets. Water fluxes (m3) are calculated from basal melt rates integrated over the appropriate area of Antarctica (Table 3), with specific annual water discharge (in m) derived by dividing the annual water flux by the catchment area. The total solute flux (in t yr−1) is the product of the total water flux (in m3) and the solute concentration (in t m−3). Weathering rates (in t km−2 yr−1) are calculated by dividing the total solute flux by the catchment area (in km2).
Table 3 indicates that the “Antarctic Interior” does not contribute significantly to total global solute fluxes (<1%) and displays low overall rates of chemical weathering (6.6 t km−2 yr−1; Table 3). However, these initial calculations exclude solute export from ice stream corridors, where basal melt rates are generally higher [Joughin et al., 2003, 2004]. Our calculated solute fluxes from the Ross Ice Streams (LMR type) and PIG (HMR type) are just a fraction of a percent of the world total, yet rates of solute export from PIG are approximately three times the world average because of the high basal melt rates [Joughin et al., 2003]. We calculate that if 10%–50% of Antarctic Ice Streams were the HMR type, ice stream solute fluxes could be between 39 and 93 × 106 t yr−1, respectively (0.7%–1.7% of the global total [Holland, 1978]). These figures rise to 72–126 × 106 t yr−1 (1.3%–2.3% of the global total) with the inclusion of solute fluxes from the Antarctic Interior, which is a similar order of magnitude to values reported from large world rivers (Amazon = 153 × 106 t yr−1 [Gaillardet et al., 1999]). It is notable that our calculated rates of solute export are several orders of magnitude higher than those for all other glacial and nonglacial systems for the same specific discharge [Anderson, 2007]. The potential to significantly affect global weathering rates is likely to be small, but solute and water fluxes to coastal areas of Antarctica could be of local and regional significance. For example, our calculated total meltwater flux is approximately a third of the annual runoff from the Greenland Ice Sheet [Box et al., 2006], while water fluxes from the Ross Ice Streams are of a similar order to those generated by basal melting beneath the Ross Ice Shelf [Shabtaie and Bentley, 1987]. Because this subglacial water is likely to be solute-rich (e.g., mean alkalinity concentrations for KIS/BIS pore waters are more than double average values of ∼2300 μeq L−1 reported for the Ross Sea [Bates et al., 1998]), it could constitute a significant source of solute and alkalinity to Antarctic coastal waters. Better quantification of these processes and their downstream effects awaits more comprehensive data regarding solute concentrations and basal melt rates under ice streams.
 In conclusion, ice mass size exerts a critical control on the balance of subglacial biogeochemical processes. We contend that ice sheet beds are driven to anoxia by the microbial oxidation of sulfide minerals and organic carbon, creating a spectrum of Eh/pH microhabitats for microbial colonization and optimal conditions for methanogenesis in the ice sheet interior. Enhanced silicate mineral dissolution liberates key nutrients critical for the survival of microbial communities in these oligotrophic systems, and sulfide oxidation continues to drive dissolution despite the anoxic conditions and potential exhaustion effects. We estimate that solute fluxes from Antarctica are similar to those of the largest rivers on our planet, and rates of chemical weathering are high for the specific discharge. Consequently, export of this solute to marine embayments could have significance for regional water and solute budgets. These data provide just a glimpse of the new insights into ice sheet biogeochemistry that are certain to emerge during the next decade, as proposed drilling campaigns [Wadham et al., 1998, 2004] delve for the first time into wet-based ecosystems beneath ice sheets.
A1. Engabreen: Fall 2004, Spring 2006, and Fall 2006
 Samples were collected directly from subglacial intake pipes bored in bedrock, which feed meltwaters from the glacier bed. Each sample was filtered immediately through 0.45-μm cellulose nitrate filters and stored in polyethylene bottles in the dark for up to 6 months before analysis. The major base cations (Na+, K+, Ca2+, and Mg2+) and strong acid anions (SO42− and Cl−) were determined by ion chromatography on the Dionex 4000i Ion Chromatograph System. The precision and accuracy of analyses were within 10%. Alkalinity was determined from the negative deficit of charge balance.
A2. Greenland: Bristol Glacier 2000
 Water samples were collected from near the main portal of Bristol Glacier at ∼11.00 h daily during a subglacial outburst flood from day of year 179.7 (27 June) through 188 (8 July) in 2000. All water samples were filtered immediately after collection through a 0.45-μm cellulose nitrate membrane filter, using a handheld vacuum pump and Nalgene filter unit. Filtrate for analysis of the major cations and anions was decanted into 20-mL scintillation vials, prerinsed three times with the filtrate before filling. Analysis of the major cations, anions, and dissolved Si was conducted 4–6 months after collection. The major cations and anions were determined by ion chromatography on the Dionex 4000i Ion Chromatograph System. Alkalinity (predominantly HCO3−) was determined from the negative deficit of charge balance. The precision of SO42−, Cl−, Ca2+, Mg2+, Na+, and K+ analyses was ±0.9%, ±7%, ±0.6%, ±0.2%, ±0.7%, and ±20%, respectively. The accuracy of determinations was −1.6%, +5.9%, −4.8%, −6.6%, −1.9%, and 12.6%, respectively.
A3. Vostok Accretion Ice
 Ice cores from 1686, 2334, 2758, 3081, 3537, 3548, 3567, 3569, 3570, 3572, 3573, 3575, 3576, 3578, 3579, 3581, 3582, 3585, 3588, 3591, and 3605 m were obtained from the U.S. National Ice Core Laboratory (Denver, CO), and cores from 3612 and 3622 m were obtained from the Laboratoire de Glaciologie et Géophysique de l'Environment (LGGE) in Grenoble, France. Ice cores were stored in sealed polyethylene tubing at −25°C until analysis. Ice cores used for major ion determinations were cut with a band saw, housed with at −10 °C walk-in freezer, and lightly scraped (at least 0.5 mm from each surface) with a clean razor blade to remove any possible contamination and condensation frozen onto the surface. Only a clean latex glove touched the ice, and new gloves were worn for each sample. Ice samples were transferred to a BioGard vertical laminar flow hood (Model B6000-1; Baker Company) and warmed to room temperature (∼20°C) until the surface began to melt (∼5 minutes). The ice was then held with clean stainless steel forceps and rinsed thoroughly with 300 mL of 0.2-μm-filtered Barnstead nanopure water. The rinsed ice was placed in a prerinsed Ziploc freezer bag and was allowed to melt completely at room temperature. Samples for major ions were collected from the bags with plastic pipettes and transferred to prerinsed HDPE bottles and stored frozen at −25°C until analysis.
 A “control” ice core, prepared by freezing 0.2-μm-filtered nanopure water in a polycarbonate tube and treated in exactly the same way as the Vostok ice core samples, contained very low levels of major ions, close to or below the detection limit (∼0.1 μeq L−1), and indicated that background particulate matter contamination represented <20% of the lowest concentration of particulates observed in any of the samples. The migration of the kerosene based drilling fluid into the core was analyzed for selected samples by gas chromatography-mass spectrometry and shown to be negligible.
 Ion chromatography was used for major ion determinations. Samples were run through the Dionex DX-120. The eluent flow rate was set to 1.2 mL min−1 for both anion and cation analyses. A large sample loop, ∼400 μL, was used for all samples and standards. A Dionex IonPac CS12A analytical column (4 × 250 mm) and a CG12A guard column (4 × 50 mm) were used for determining the cations. The eluent was 0.13% methanesulfonic acid solution. An ultra cation self-regenerating suppressor was used. A Dionex IonPac AS14 analytical column (4 × 250 mm) and an AG14 guard column (4 × 50 mm) were used for anion analysis. The eluent was a 1.0 mM NaHCO3 and 3.5 mM Na2CO3 solution. An ultra anion self-regenerating suppressor was used.
A4. Midre Lovénbreen (1997–2000)
 Samples emerging from upwelling outlets were collected by hand in 1997, 1999, and 2000 and were filtered immediately through 0.45-μm cellulose nitrate filters. All samples were stored airtight in prerinsed 60-mL polyethylene Nalgene bottles and refrigerated awaiting further analysis. Anions of chloride, sulfate, and nitrate were determined on return to the United Kingdom using the Dionex DX 100 Ion Chromatograph System. On the basis of the repeat analysis of reference standard materials, precision was quoted as 0.72%, 3.35%, and 1.19% relative standard deviation (RSD) for nitrate, chloride, and sulfate, respectively. For the 1997 and 1999 samples, total alkalinity (predominantly present as bicarbonate) was determined by titrating with 1 mmol acid (HCl) to an end point pH of 4.5 using BDH mixed indicator. Precision calculated from repeated analysis of the samples was quoted as 5% RSD. In 2000 and 2002, alkalinity was determined from the change balance deficit. Calcium and magnesium were analyzed by atomic absorption spectroscopy. Precision estimates from repeat analysis of the reference standard materials of comparable concentrations to the samples under analysis were calculated as 0.27% and 2.46% RSD for calcium and magnesium, respectively.
Appendix B:: Microbial CO2 Calculations
 We calculate the percentage contribution of microbially derived HCO3− (HCO3−Microbial) to total HCO3− for glacier data sets where the intercept of the line of best fit for plots of HCO3− versus *SO42− was >220 μeq L−1 and, hence, cannot be explained by calcite hydrolysis. Hence, this calculation was performed for the glaciers Finsterwalderbreen (1994 and 1997), Midre Lovénbreen (2000 and 2002), and Engabreen (spring 2004). We express the excess HCO3− (above the theoretical value of 220 μeq L−1 for calcite hydrolysis) as a percentage of the mean HCO3− for each glacier data set [equation (B1)], where the mean HCO3− is calculated by averaging the HCO3− concentrations for each temporally discrete sampling point.
We also calculate the total amount of solute generated via microbial CO2 (assuming that each mole of microbially derived HCO3− is balanced by the molar equivalent of cation charge). This quantity is expressed as a proportion (%SoluteMicrobial) of the total mean solute concentration (soluteMEAN) for the different glaciers [equation (B2)], where total mean solute concentrations are derived in a similar manner to the HCO3− mean concentration.
This excess “microbial” HCO3−, on average, accounts for 36% of all HCO3− in meltwaters. When added to the equivalent cation charge and expressed as a percentage of total crustal mean solute concentrations, we estimate that, for this data set, 14%–37% of solute is acquired using microbial CO2.
 This paper was written by J. L. Wadham while on a Phillip Leverhulme Award fellowship. Data have been compiled from 15 years of international research, funded by a variety of sources, including Natural Environment Research Council, United Kingdom; the National Science Foundation, Arlington, VA; the Royal Society; Leverhulme Trust; and the European Union, Brussels, Belgium.