Although CO2 is generally the most abundant dissolved gas found in submarine hydrothermal fluids, it is rarely found in the form of CO2 liquid. Here we report the discovery of an unusual CO2-rich hydrothermal system at 1600-m depth near the summit of NW Eifuku, a small submarine volcano in the northern Mariana Arc. The site, named Champagne, was found to be discharging two distinct fluids from the same vent field: a 103°C gas-rich hydrothermal fluid and cold (<4°C) droplets composed mainly of liquid CO2. The hot vent fluid contained up to 2.7 moles/kg CO2, the highest ever reported for submarine hydrothermal fluids. The liquid droplets were composed of ∼98% CO2, ∼1% H2S, with only trace amounts of CH4 and H2. Surveys of the overlying water column plumes indicated that the vent fluid and buoyant CO2 droplets ascended <200 m before dispersing into the ocean. Submarine venting of liquid CO2 has been previously observed at only one other locality, in the Okinawa Trough back-arc basin (Sakai et al., 1990a), a geologic setting much different from NW Eifuku, which is a young arc volcano. The discovery of such a high CO2 flux at the Champagne site, estimated to be about 0.1% of the global MOR carbon flux, suggests that submarine arc volcanoes may play a larger role in oceanic carbon cycling than previously realized. The Champagne field may also prove to be a valuable natural laboratory for studying the effects of high CO2 concentrations on marine ecosystems.
While considerable effort has been devoted to exploring for submarine hydrothermal activity along the global mid-ocean ridge (MOR) system where tectonic plates are diverging, very little is known about the distribution and intensity of similar submarine activity on volcanic arcs where plates converge. de Ronde et al.  made the first systematic study of hydrothermal activity in this tectonic setting along the southern Kermadec Arc. Their water column plume surveys showed that a substantial portion of the submarine volcanoes there are hydrothermally active [de Ronde et al., 2001]. Furthermore, in contrast to MOR activity that is mainly confined to the depth range of 2000–2500 m, the Kermadec Arc volcanoes were found to be introducing hydrothermal effluent at a wide variety of ocean depths, ranging from 100 to 1800 m.
During February–March 2003, as part of the Submarine Ring of Fire (SROF) project funded by NOAA's Ocean Exploration Program, the R/V Thomas G. Thompson conducted a comprehensive survey of submarine hydrothermal activity along a second volcanic arc, the Mariana Arc from 13.5°N to 22.5°N [Embley et al., 2004]. Plume surveys were conducted in the water column above ∼50 Mariana Arc submarine volcanoes using a CTD/rosette system that included an Eh sensor for measuring in-situ oxidation-reduction potential. A total of 70 CTD casts were completed, and discrete water samples were collected for analysis of a variety of hydrothermal tracers, including 3He, CH4, CO2, H2S, Fe, Mn, pH, and suspended particles. The analysis of these samples showed that 12 of the Mariana Arc submarine volcanoes surveyed had active hydrothermal discharge. Of these, 8 were new sites and 4 were volcanoes previously known to be hydrothermally active [see Embley et al., 2004]. The newly discovered activity included that on NW Eifuku, a small volcanic cone located at 21.49°N, 144.04°E that rises to a depth of ∼1535 m below sea level (Figures 1 and 2). NW Eifuku is the deepest in a cluster of 3 volcanoes that includes the larger neighboring volcanoes Daikoku and Eifuku. The water column samples collected over NW Eifuku in 2003 had excess concentrations of 3He, CO2, CH4, Fe, and Mn, as well as pH, light backscattering (suspended particle), and Eh anomalies, all confined to the depth range of 1490 to 1620 m [Lupton et al., 2003; Resing et al., 2003].
2. Discovery of the Champagne Site
The second phase of the SROF project consisted of a follow-up expedition aboard the R/V T.G. Thompson in March–April 2004 employing the remotely-operated vehicle (ROV) ROPOS to directly explore and sample a selected group of the Mariana Arc submarine volcanoes. Three ROV dives were devoted to exploring NW Eifuku. Approximately 8 hours into the first dive, at a depth of 1604 m, ROPOS discovered a remarkable hydrothermal field (later named Champagne) with small white chimneys discharging buoyant milky fluid. Subsequent surveys with the ROV located several additional sites of hydrothermal discharge on NW Eifuku, although the most intense venting was found at the Champagne site ∼80 m WNW of the volcano summit. The summit of NW Eifuku was mapped with an Imagenex scanning sonar on ROPOS (Figure 2c), following the methods described by Chadwick et al. . The high-resolution bathymetry shows that the Champagne vent field lies in the steep headwall of a gravitational slope failure that cuts across the top and SW side of the volcano [Chadwick et al., 2004]. Although there were few vent animals right at the Champagne site, an extensive biological community was found within the surrounding few hundred meters, including mussels, shrimps, crabs and limpets.
In addition to the vent fluid discharge at Champagne vent, droplets coated with a milky skin were rising slowly from the seafloor around the chimneys (Figure 3). The droplets were later determined to consist mainly of liquid CO2, with H2S as a secondary component. The seafloor area of active CO2 droplet flux was characterized by pumice and whitish/yellowish sulfur-rich material. The droplets were sticky and adhered to the ROV like clumps of grapes, although they did not tend to coalesce into larger droplets (Figure 3e). The film coating the droplets was assumed to be CO2 hydrate (or clathrate) which is known to form whenever liquid CO2 contacts water under these P, T conditions [Sloan, 1990]. Liquid CO2 should be buoyant at the depth of the Champagne site, since it has a density less than seawater at depths shallower than about ∼2600 m [Brewer et al., 1999]. At NW Eifuku, droplets percolated out of crevices in the seafloor, and we did not observe the formation of small hydrate pipes as noted at the JADE site in the Okinawa Trough [Sakai et al., 1990a]. The flux of liquid CO2 droplets increased dramatically whenever the seafloor was disturbed by the ROV. This observation is consistent with the presence of a layer of liquid CO2 beneath the surface capped by an impeding layer of CO2 hydrate (see Figure 4). Thus any penetration of the hydrate cap releases the buoyant liquid CO2 beneath. These observations are similar to those reported by Sakai et al. [1990a], who discovered venting of liquid CO2 in the Okinawa Trough back-arc basin. A comparison of the video from both sites indicates a higher flux of CO2-rich droplets at NW Eifuku compared to the JADE site (see Figure 3 caption).
In October–November of 2005 we had a second opportunity to collect samples at NW Eifuku during cruise NT05-18 aboard the R/V Natsushima. During the Natsushima cruise, the ROV Hyper-Dolphin completed 6 dives on NW Eifuku, 2 on the volcano flanks and 4 on the summit area. During this expedition, the Hyper-Dolphin collected additional samples of both the vent fluids and liquid droplets at the Champagne site.
Samples of the Champagne vent fluid were collected in special gas-tight, all-metal bottles constructed of titanium alloy. The bottles, which have an internal volume of ∼150 ml, were initially evacuated. After the connecting lines were flushed, the bottle inlet was opened using a hydraulic actuator, and then hydrostatic pressure quickly forced the vent fluid sample into the bottle. At the end of the ROV dive, the samples were processed on board the ship using a high vacuum extraction line equipped with a low temperature (−60°C) trap and an all-metal bellows pump (Figure 5). The sample was first dropped from the gas-tight bottle into an evacuated glass flask containing ∼1 g of sulfamic acid. The acid lowers the pH of the fluid, thereby aiding in the extraction of CO2 and other dissolved gases. The bellows pump was then used to pump the exsolved gases through the drying trap into a calibrated volume. After the pumping was completed, the total amount of gas was measured using a high precision capacitance manometer attached to the calibrated volume. Then splits of the dry gas were sealed into glass ampoules. For rare gas measurements, the ampoules were constructed of type 1720 or 1724 aluminosilicate glass with low helium permeability. During the 2004 R/V Thompson cruise, our extraction line had the capability of handling about 1.5 l of total gas, which proved to be inadequate for some of the very gassy vent fluid samples. Thus for most gas-rich samples it was necessary to carry out the extractions in multiple steps. While this provided an accurate assessment of total gas content, the multiple step extraction fractionated the samples, making them unsuitable for gas composition or isotopic measurements. For the 2005 R/V Natsushima cruise, the extraction line was fitted with an additional tank increasing the calibrated volume to 11 l, thereby allowing us to extract the samples in one step.
For analysis of dissolved species, additional samples of vent fluid were collected in non-gas-tight PVC pistons with pressure relief valves at the top to capture the water component. Careful measurements using a temperature probe integral with the sampler inlet gave temperatures of 103°C for the most vigorous vents, although several other vents in the area were discharging fluids at temperatures between 11 and 68°C. Careful probing into the seafloor where the liquid droplets were forming found temperatures <4°C, consistent with the existence of CO2 in the liquid or hydrate state.
Sampling of the liquid droplets proved to be even more challenging. On one of the 2004 ROPOS ROV dives we collected about 0.5 l of the liquid CO2 droplets in an inverted plastic cylinder normally used for collection of sediment cores (Figure 6b), and observed the droplets as the submersible ascended to the surface at the end of the dive (Figure 6c). This was similar to an experiment conducted by Sakai et al. [1990a] in the Okinawa Trough. We were able to first observe the continuous conversion of liquid CO2 into white “sherbet-like” hydrate in the cylinder. Then as the submersible passed through ∼400 m depth (at ∼4°C), we observed rapid conversion of both liquid CO2 and hydrate into gaseous CO2. This is precisely the pressure depth at which this phase transition was expected, thus confirming our hypothesis that the droplets were composed mainly of liquid CO2 (Figure 7). The plastic collection cylinder as well as some of the ROV camera face plates suffered permanent damage as a result of contact with the corrosive liquid droplets.
In 2004, liquid CO2 droplets were also collected by gluing a length of PEEK™ tubing into the plastic cylinder mentioned above and connecting the other end to the inlet of a titanium gas-tight bottle. The ROV again collected about 500 ml of the liquid droplets by holding the plastic cylinder inverted above the buoyant droplet stream. Then the gas-tight bottle was opened, drawing liquid CO2 into the bottle. As a safety precaution, we opened the bottle several times before the ROV surfaced to allow gas to escape and relieve the internal pressure. Because there was a mixture of liquid CO2, hydrate, and water in the bottle, this led to fractionation of the sample gas composition. Furthermore, this liquid droplet sample had to be extracted in multiple steps, leading to further fractionation. However, we were still able to confirm that the droplets consisted of >90% CO2 by volume.
During the 2005 R/V Natsushima cruise, we employed a new method for the collection of the liquid droplets that was very successful. One of us (C. Young) designed a special “droplet catcher” consisting of a conical metal spring that was connected with PEEK™ tubing to a special titanium gas-tight bottle with low (∼10 cc) internal volume (Figure 6d). The spring coil was first filled with liquid droplets by holding it over a stream of CO2 droplets exiting the seafloor (Figure 6e). We knew from previous experience that the droplets are sticky and do not tend to coalesce. The droplets were visible through the spring coils but stayed in place inside the coils. Then the spring coil was compressed against a flat surface on the ROV, thereby expelling most of the excess water between the droplets. Finally, the small volume gas-tight bottle was triggered, drawing the droplet sample into the bottle. During the Natsushima cruise, we were able to collect 4 good samples of the liquid droplets with this technique. Subsequent analysis showed that each sample contained about 5 cc of liquid CO2. This converted to about 7 l of gas at STP in the extraction line, which we were able to handle quite easily with our enlarged calibrated volume.
In both 2004 and 2005, water column samples were collected using a CTD rosette package. Plume identification was accomplished using both a light scatter sensor and an Eh sensor. Samples were collected into Niskin type bottles and sub-sampled for helium isotopes, CO2, and other plume components.
He and Ne concentrations, 3He/4He ratios, and 13C/12C ratios were determined by mass spectrometry, while CO2, CH4, H2, and other gas concentrations were determined by gas chromatography. Total CO2 in the water column plume samples was analyzed by coulometry. Radiocarbon was measured on selected samples at the Center for Accelerator Mass Spectrometry at Lawrence Livermore National Laboratory. Hydrogen sulfide was analyzed on fluid samples collected with the non-gas-tight PVC pistons using the conventional methylene blue method. However, due to possible gas loss from the PVC pistons, these H2S values represent only a lower limit. As an alternative, splits of the gases from selected gas-tight bottle samples were sent to Atmospheric Analysis and Consulting (AAC), Inc., Ventura, CA, for analysis of reduced sulfur compounds by sulfur chemi-luminescence (method ASTM D-5504). In addition, AAC measured CO2 abundances by conventional thermal conductivity gas chromatography on the same samples, thereby producing a measure of the H2S/CO2 ratio.
The analytical results for the 2004 and 2005 samples from NW Eifuku are summarized in Tables 1 and 2. By theory and observation, high-temperature hydrothermal fluids are nearly devoid of Mg, so pure “end-member” fluid compositions normally are estimated by extrapolating to zero Mg [Edmond et al., 1979; Seyfried, 1987; Von Damm, 1990]. However, even though in-line temperatures of ∼103°C were measured during sampling, none of the samples collected at the Champagne site had a Mg concentration less than 43 mmol/kg, although several samples approach this value. As we will explain in the Discussion section, we do not believe that a high temperature zero-Mg end-member exists at the Champagne site. Rather than extrapolate to zero magnesium, which would yield unrealistically high temperatures and CO2 concentrations, we assign the value of 43 mmol/kg Mg to the undiluted 103°C vent fluid that exists at the seafloor. In adopting this approach, we assume that the range of Mg concentrations that we measure is due to entrainment of local seawater during sampling. In Figure 8, vent fluid properties are plotted versus Mg, and mixing lines are shown extrapolating the concentrations to the assumed end-member Mg value of 43 mmol/kg. By extrapolating to this value, we believe we are correcting each sample for seawater dilution during sampling.
Table 1. Gas Compositions for Vent Fluid and Liquid Droplet Samples From NW Eifukua
Units are mmol/kg unless noted otherwise. For the conventional gas chromatograph analyses (CO2, CH4, H2, N2, O2, Ar, CO), the concentrations are precise to about ±5%.
The O2 concentration is not a reliable indicator of air contamination since O2 may have been reduced by reacting with H2S in the ampoule.
For these samples, the gas content was more than our extraction system could easily handle, and some fractionation of the gases occurred. For this reason we are reporting only those gas compositions that are reliable.
Gas-tight bottle sample, H2S analyzed by sulfur-chemiluminescence at AAC.
Samples collected with the piston sampler, H2S analyzed by the conventional methylene blue method.
For the liquid droplets, which are essentially pure gas, we report the CO2 content as a volume%.
Helium isotope ratio expressed as R/RA, where R = 3He/4He and RA = Rair = 1.39 × 10−6. The solubilities of the two isotopes in water are slightly different, with the result that for air saturated water R/RA = 0.98.
53450 ± 3200
38600 ± 520
Champagne, 2nd Site
Champagne, 2nd Site
Turning our attention at first to the 2004 results, based on an extrapolation to 43 mmol/kg Mg, the 103°C Champagne hydrothermal fluid contained a surprising ∼3.0 moles/kg of CO2. This is an order of magnitude higher than any CO2 values previously reported for submarine hydrothermal fluids (Figure 9). As will be discussed later, we believe that this very high CO2 concentration in the vent fluid is the result of subsurface entrainment of liquid CO2 and/or CO2 hydrate. The overall gas composition of the 2004 vent fluid was ∼3000 mmol/kg CO2, ∼12 mmol/kg H2S, <0.2 mmol/kg CH4 and H2, and 0.01 mmol/kg 4He. Although we analyzed for CO, it was for the most part below our detection limit (Table 1). Concentrations of N2, O2, Ar, and Ne are also included in Table 1 as indicators of air or seawater contamination. Two of the samples collected in 2005 (H494-GT4 and H497-GT7) did suffer from air contamination on the basis of their N2 and O2 concentrations, but not enough to compromise the other gas measurements. The Champagne vent fluids have lithium concentrations in the range of 20 to 26 μmol/kg, significantly lower than the ambient seawater concentration (26.5 μmol/kg), and pH ranging from 3.4 to 4.8. The low end of measured pH of Champagne vent fluids is consistent with CO2 buffering in the end-member fluid.
As discussed above, we were not able to collect an unfractionated sample of the liquid droplets in 2004. However, analysis of the one liquid droplet sample that we collected confirmed that the droplets were composed of >90% CO2, with the remaining gas assumed to be H2S.
As discussed above in the Methods section, determining H2S concentrations in these NW Eifuku samples was challenging because of the high gas content of the samples. Shipboard analysis of samples collected with the non-gas-tight PVC pistons shows a roughly linear trend of increasing H2S with decreasing Mg, with H2S reaching approximately 4.5 mmol/kg in the water phase of the least diluted PVC piston samples (Figure 8b). However, because the PVC piston samples have lost significant gas volume in most cases, these are minimum values for H2S in the fluids. One gas-tight sample collected in 2004 (R793-GT5) was analyzed by sulfur chemi-luminescence gas chromatography at Atmospheric Analysis and Consulting, Inc., and this value lies on the mixing line through the highest of the PVC piston results (Figure 8c). This suggests that the end-member concentration (∼12.5 mmol/kg H2S) derived from this mixing line (the red mixing line) represents the best estimate for the 2004 Champagne fluid composition. In all of the samples analyzed at AAC, H2S was the only sulfur species detected.
During the return visit to NW Eifuku in 2005, while the Champagne site still had a constant flow of vent fluid and liquid droplets, there seemed to be slightly less activity than observed in 2004. During our first visits to the Champagne site in 2004, the vent fluid was discharging through several small white chimneys (see Figures 3a, 3b, and 3c). We were surprised to find that these chimneys, which were destroyed by the ROV during the 2004 sampling, had not re-grown during the intervening 18 months. As shown in Tables 1, 2, and 3 and in Figure 8, on the basis of “end-member” extrapolations, compared to 2004 the CO2 concentration was lower in 2005, accompanied by higher H2S, lower He, and higher C/3He ratios. One of the major accomplishments of the 2005 Hyper-Dolphin dives was the successful collection of 4 uncontaminated samples of the liquid droplets (see Methods section above). The liquid droplets had a gas composition of ∼98% CO2, <0.01% CH4 and H2, ∼6 ppm He, and ∼0.8% H2S (see Tables 1 and 3). While this composition is similar to that of the vent fluids, the liquid droplets collected in 2005 have lower H2S/CO2 and higher He/CO2 ratios than the 2005 vent fluids (Table 3).
Table 3. Estimated End-Member Compositions, Based on Extrapolating to a Mg Concentration of 43 mmol/kg
Helium isotope ratio expressed as R/RA where R = 3He/4He and RA = Rair = 1.39 × 10−6. The solubilities of the two isotopes in water are slightly different, with the result that for air saturated water R/RA = 0.98.
The helium in both the Champagne vent fluid and in the liquid droplets had an isotopic ratio of R/RA = 7.31 ± 0.05, a value typical of subduction zone systems (R = 3He/4He and RA = Rair) [Poreda and Craig, 1989; Hilton et al., 2002]. During the 2003 water column surveys over NW Eifuku, helium samples were collected that allowed an estimate of the end-member helium isotope ratio based on the co-variation of [3He] versus [4He]. This estimate gave R/RA = 7.25 ± 0.4, in remarkably good agreement with direct measurements of the vent fluids (see Figure 10). Furthermore, the C/3He ratio estimated from measurements of ΣCO2 and [3He] in the water column plumes differed only by 10% from that determined from the vent fluids (Figure 10c). This suggests that reliable estimates of the ratios of certain vent fluid properties can be made from samples of the overlying water column plumes, even though these plumes typically contain only 0.1% or less of the pure vent fluid.
Isotopic analysis of the CO2 in the Champagne vent fluid gave δ13C = −1.75‰, while the carbon in the liquid droplets was slightly heavier (δ13C = −1.24‰). The C/3He ratio ranged from 1.6 to 9.7 × 1010 for the Champagne vent fluids, and from 1.4 to 1.9 × 1010 for the liquid droplets (Table 2). A sample of the CO2 from the Champagne site 103°C vent fluid was analyzed for radiocarbon at the CAMS facility at Lawrence Livermore National Laboratory. The result was Δ14C = −998.7‰, corresponding to an age of 53450 ± 3200 years, or a fraction of modern carbon of only 0.0013 (see Table 2). Analyses of the 68°C vent fluid and of the liquid CO2 droplets yielded similar results. Thus the carbon in the Eifuku CO2 is “dead” (age ≥ 50,000 years).
In order to put the very high CO2 concentration of the Champagne vent fluids in perspective, Figure 9 compares the end-member CO2 concentrations at NW Eifuku with those at the Okinawa Trough and at various mid-ocean ridge hydrothermal sites. Although Sakai et al. [1990a] observed liquid CO2 venting at the JADE site in the Okinawa Trough, the 320°C vent fluid at the JADE site contained only 200 mmol/kg of CO2. As shown in Figure 9, the end-member CO2 concentrations at the Champagne site are an order of magnitude higher than any values reported for other hydrothermal fluids, and 100 times higher than average values at MOR systems. The structure of the hydrothermal system at NW Eifuku is clearly different from that at the JADE site, where high-temperature, zero-magnesium fluids are produced in a reaction zone with low water/rock ratio [Sakai et al., 1990a, 1990b].
The CO2 concentration at NW Eifuku is even more remarkable when it is compared against the CO2 solubility at these P, T conditions. The solubility of CO2 in seawater at 160 bars, 100°C is ∼1.0 mole/kg [Wiebe and Gaddy, 1939; Takenouchi and Kennedy, 1964], much lower than the concentrations we measured in 2004. The most plausible explanation for the apparent super-saturation of CO2 is that the Champagne vent fluid is entraining small amounts of liquid CO2 and/or CO2 hydrate as the hot vent fluid penetrates the layers of CO2 liquid and hydrate that we propose exist beneath the seafloor (Figure 4). Incorporation of only 6% by volume of liquid CO2 into the vent fluid would increase the CO2 concentration from 1 to 2.7 moles/kg, and this liquid CO2 would likely not be visible as a separate phase in the vent fluid stream.
The 2005 vent fluid samples had lower CO2 concentrations and different relative proportions of dissolved gases compared to the earlier 2004 collections. In fact, none of the 2005 water samples had CO2 concentrations above the 1.0 mole/kg solubility of CO2 at the conditions at the Champagne site. Furthermore, as discussed above, the 2005 vent fluid samples had higher end-member H2S/CO2 ratios, and higher C/3He ratios compared to the 2004 samples (Table 3). For whatever reason, the liquid droplets have lower H2S/CO2, lower C/3He, and are heavier in δ13C compared to the vent fluids. This difference between the vent fluid compositions in 2005 versus 2004 may be due to temporal changes in the degree of entrainment of liquid CO2 and/or hydrate into the rising vent fluid. Our results indicate that all of the vent fluid samples are actually a mixture of the pure “subsurface” hydrothermal fluid combined with varying amounts of entrained liquid CO2 or CO2 hydrate. The end-member compositions estimated for the 2005 samples (see Table 3) may thus represent a form of the Champagne vent fluid relatively un-contaminated with entrained liquid CO2, while the gas-rich 2004 vent fluids had more entrained liquid CO2. For the 2004 vent fluid samples with ∼2.7 moles/kg CO2, most of the CO2 came from the entrained liquid droplets and/or hydrate, while most of the H2S and He was already dissolved in the hot fluid before it reached the near surface liquid CO2 layer.
In many cases, diffuse hydrothermal fluids are located near high-temperature fluids or their chemistry indicates that they are dilutions of high-temperature fluids, with overprinting low-temperature reactions [Edmond et al., 1979; Butterfield and Massoth, 1994; Butterfield et al., 1997, 2004; Sedwick et al., 1992]. That is not the case at NW Eifuku. If we were to extrapolate the temperature and fluid composition to a zero-magnesium value, the results would be nothing like a fluid produced in a high-temperature water/rock reaction zone. For example, a zero-Mg extrapolation at NW Eifuku would yield temperatures of 500–600°C, CO2 concentrations of 10–20 mol/kg, and zero Li concentration. The implication of this is that we are not dealing with a high-temperature aqueous system, but with a high-temperature CO2 system, entraining some water that undergoes incomplete reaction to remove some seawater magnesium and extract some elements from the rock. CO2 migrating upward from a gas pocket in a magma chamber must cool as it ascends through volcanic rock and may entrain small amounts of seawater (Figure 11). Water and CO2 are immiscible at 500 bars at temperatures below 310°C [Takenouchi and Kennedy, 1964; Bowers, 1991] and separate into a CO2-rich vapor and a water-rich liquid. As pressure increases, immiscibility of H2O-CO2 occurs at lower temperatures [Bowers, 1991]. In a system dominated by the flux of hot CO2 from a magma chamber, the penetration of water into the core of the system will be limited at typical hydrothermal temperatures (up to ∼350°C) due to the immiscibility. If seawater is not first heated by hot rock (and the water chemistry at the Champagne site indicates minimal high-T water/rock reaction), then the outer portions of the CO2 column will be in the P-T region of H2O-CO2 immiscibility. As long as the flux of hot CO2 from the magma chamber and cooling in the pathway to the seafloor is maintained, the penetration of water into the CO2 –rich zone is inhibited. The presence of a gas hydrate phase at temperatures below 10°C may also inhibit penetration of water into the sub-seafloor CO2-rich zone.
Some insight into the origin of the high CO2 concentrations at NW Eifuku can be gained from the isotopic composition of the CO2 and the relation of CO2 to 3He. The δ13C of the Champagne vent fluids (−1.75‰) is much heavier than the δ13C = −13 to −4‰ typical for carbon in MOR vent fluids [Kelley et al., 2004]. The NW Eifuku CO2 is also heavier than that reported for the Mid-Okinawa Trough (−5.0 to −3.7‰) [Sakai et al., 1990a, 1990b], falls at the heavy end of the range reported for arc volcanoes in general (−7 to −2‰) [Sano and Williams, 1996; van Soest et al., 1998], and at the lighter end of the range for marine carbonates (−2 to +1‰) (Figure 12) [Hoefs, 1980]. The C/3He ratio for the Champagne vent fluids and liquid droplets (1.3 to 9.4 × 1010) is similar to that reported for the Mid-Okinawa Trough [Sakai et al., 1990a, 1990b], but an order of magnitude higher than the average value of 2 × 109 found in MOR vent fluids [Resing et al., 2004]. These δ13C and C/3He values indicate that the majority of the carbon flux originated from marine carbonates incorporated into the melt as part of the subduction zone melting process rather than from mantle carbon. Using the method outlined by Sano and Marty  based on δ13C and C/3He values, and taking sedimentary organic matter to have a δ13C value of −30‰ as did Sano and Marty , we estimate that the NW Eifuku CO2 was derived 88% from marine carbonates, 9% from mantle carbon, and 3% from sedimentary organic matter. If instead we assume that the δ13C of sedimentary organic matter is −20‰, then we calculate 87% from marine carbonates, 9% mantle, and 4% from sedimentary organics. These fractions are similar to those observed at subaerial arc volcanoes [Hilton et al., 2002]. The fact that the radiocarbon is “dead” suggests that the CO2 flux is mainly derived from subducted carbonates incorporated into the melt at depth and that local sediments are not responsible.
It is a simple matter to show that the extremely high concentrations of CO2 at NW Eifuku cannot be easily derived from either water/rock reaction or from dissolution of putative carbonates within the volcanic edifice. On the first count, it has already been shown [Butterfield et al., 1990; Sedwick et al., 1994] that by extracting all of the CO2 from 1 kg of MORB, assuming the maximum reported level of 8 mmol CO2/kg rock [Dixon et al., 1988], into 1 kg of water (a typical water/rock ratio), the total CO2 concentration in the aqueous phase would not exceed ∼8 mmol/kg. However, according to Wallace , undegassed arc magmas contain ∼3000 ppm of CO2. Using the same water/rock ratio of 1, this would produce only 68 mmol/kg of CO2 in the hydrothermal fluid, still far below the 900–2700 mmol/kg we observe. Furthermore, the Champagne vent fluids have lithium concentrations lower than the ambient seawater concentration, indicative of minimal high-temperature water-rock reaction (a high water/rock ratio). Thus it appears to be impossible to extract enough CO2 from basaltic or andesitic lava by water/rock interaction to reach the levels of CO2 found in the Champagne vent fluids, or even in some MOR vents (e.g., Axial Volcano on the Juan de Fuca Ridge or 9°N East Pacific Rise). On the second count, the low pH of Champagne vent fluids is inconsistent with calcium carbonate dissolution. In addition, dissolution of carbonates is self-limiting unless there is an additional source of acid to drive more dissolution. In that case, the calcium concentration would also be very high, but it is lower than seawater concentration in the Champagne vent fluids. We conclude that the CO2 at this site must be derived from magma degassing, as there is no other plausible source.
Although it is clear that the venting at the Okinawa Trough and NW Eifuku Champagne sites locally produces very high CO2 concentrations, the question remains as to how significant the overall carbon flux is on a global scale. We estimated the volume flux of liquid CO2 droplets at the Champagne site by examining video collected during the 2004 ROV dives. We first estimated that there were about 300 droplet streams rising from the 10 m2 area of the Champagne site, and that each stream contained 2 droplets/s, each with an average diameter of ∼1.5 cm. This gives a total liquid CO2 flux of ∼1 liter/s or 1 kg/s (assuming a density of 1g/cm3), equivalent to ∼23 moles CO2/s. Using a similar method, we estimate the CO2 flux from the Champagne hot vents to be ∼ 0.5 mole/s or only about 2% of the liquid CO2 flux. The CO2 flux from the liquid droplets at the Champagne site (8 × 108 moles/yr) approximately equals the combined carbon flux from all of the Endeavour Ridge vent fields on the Juan de Fuca Ridge [Rosenberg et al., 1988; Lilley et al., 1993], or about 0.1% of the global MOR carbon flux which is estimated at 0.5–2.0 × 1012 moles/yr [Resing et al., 2004]. The carbon flux from the Champagne site is also about 0.1% of the global CO2 flux from subaerial arc volcanoes, estimated at ∼1.6 × 1012 moles/y [Hilton et al., 2002]. Although these flux estimates for NW Eifuku are admittedly only accurate to a factor of 2 or so, this carbon flux is surprising, since NW Eifuku is a small, young arc volcano and not a major volcanic edifice. Furthermore, the fact that NW Eifuku is a submarine volcano suggests that carbon fluxes based on observations of subaerial volcanoes may have underestimated the global fluxes from arcs. If there are many such submarine sites active along volcanic arcs and back arcs, then there is potential for a significant impact on oceanic carbon cycling. For completeness it should be noted that Hilton et al.  estimated the carbon flux from an average subaerial arc volcano at 2 × 1010 moles/y, about 25 times higher than the carbon flux at NW Eifuku. They arrived at this estimate by normalizing the CO2 flux to the measured SO2 flux at various subaerial arc volcanoes, rather than by direct measurements of the CO2 outgassing rate. Thus the carbon flux at NW Eifuku may be significant for submarine carbon cycling, but not necessarily for the global subaerial carbon flux.
Additional visits and possibly long-term monitoring of the site are required to determine if this high CO2 flux is time dependent. For example, Miyakejima volcano (Japan) underwent a months-long extremely high-volume magmatic degassing event following a caldera collapse in 2000 [Kazahaya et al., 2004]. Recent mass-wasting at the summit of NW Eifuku could have been triggered by movements along what appears to be a NW-SE fracture underlying the volcano (Figure 2c). Deep conduits could be enlarged and new ones opened during this process. Alternatively, long-term degassing during non-eruptive periods on some volcanoes has been tied to endogenous growth by magmatic intrusion [Allard, 1997].
Recently there has been considerable interest in the possible oceanic disposal of fossil fuel CO2 as a means to alleviate the increase of atmospheric CO2 [Brewer, 2000]. One important question concerns the fate of the CO2 after it is introduced into the ocean. Brewer et al.  measured the rate of dissolution of liquid CO2 injected into the ocean at a depth of ∼800 m, and found that 90% of the buoyant CO2 droplets dissolved within 200 m above the injection point. As shown in Figure 10, our water column measurements in the vicinity of NW Eifuku are in basic agreement with the Brewer et al.  results. On several hydrocasts collected over the volcano in 2003 and 2004 we found large excesses in 3He and CO2 that co-varied almost perfectly [Lupton et al., 2003; Resing et al., 2003]. In every case the excess 3He and CO2 was confined to the depth range of 1490 to 1620 m and returned to background values about 150 m above the depth of the NW Eifuku vent fields (Figure 10).
6. Summary and Conclusions
In summary, we have discovered a site at ∼1600 m depth on NW Eifuku, a submarine volcano on the northern Mariana Arc, which is venting droplets of liquid CO2 at an estimated rate of 8 × 108 moles/yr. This is only the second locality where submarine venting of liquid CO2 has been observed, the other being the mid-Okinawa Trough [Sakai et al., 1990a, 1990b]. The Champagne site on NW Eifuku is also venting hot (∼100°C) vent fluid with CO2 contents up to 2.7 moles/kg, far above the solubility (∼1.0 mole/kg) at these P, T conditions. Observations at the site indicate the presence of a subsurface liquid CO2 layer under a capping layer of CO2 hydrate. We attribute the apparent CO2 super-saturation in the vent fluid to entrainment of small amounts of liquid CO2 and/or CO2 hydrate into the ascending vent fluid stream. The liquid droplets are composed of >98% CO2, ∼1% H2S, with only trace amounts of H2 and CH4. The dissolved gases in the vent fluid have a similar composition, with a slightly greater concentration of H2S (∼3%). The δ13C (CO2) and CO2/3He ratios fall in the range typical for volcanic arcs, and indicate that the carbon is derived ∼90% from marine carbonates, the remainder being mantle carbon and sedimentary organic matter. The fact that the radiocarbon is dead (age ≥ 50,000 years) suggests that the source is subducted carbonates incorporated into the melt at depth in the subduction zone and not local carbonates on the volcano edifice.
Sakai et al. [1990a, 1990b] explained their observations in the mid-Okinawa Trough in terms of separate CO2-rich and H2O-rich fluids that formed as the result of magma chamber degassing. They also discussed subsurface boiling as a possible mechanism for generating these two phases. It is clear that separate cold CO2-rich and hot H2O-rich fluids exist at the same site in close proximity at NW Eifuku. At the vent site, we envision a mechanism in which hot water is venting through an area of liquid CO2 and CO2 hydrate and entrains these, generating hot fluids with CO2 contents higher than predicted by the limits of CO2 solubility. In contrast to the 320°C fluids found in the Okinawa Trough, the 103°C fluid temperatures at NW Eifuku are ∼250°C below the boiling point at 1600-m depth, and thus shallow subsurface boiling is unlikely. The 103°C fluids do not show signs of intense water/rock interaction, and their low alkali metal content is indicative of a high water/rock ratio. Given that we do not find a zero-magnesium, high-temperature fluid at NW Eifuku, it is impossible to extract enough CO2 from the rock into circulating seawater to form an aqueous fluid saturated with CO2. Instead, the extreme CO2 concentrations at NW Eifuku require direct degassing of CO2 from a magma chamber, cooling and migration to the seafloor, resulting in the generation of the CO2-rich and H2O-rich fluids that we observed. The physical/chemical model we have proposed (Figure 11) differs substantially from the mid-ocean ridge model of extraction of gases from rock by circulating hot water. If our model is correct, then elemental and isotopic fractionations that occur as a result of magma degassing, CO2 condensation, hydrate formation, H2O-CO2 mixing, and phase separation add considerable complexity to the interpretation of gas ratios and isotopic ratios.
The Champagne vent field and the other sites of hydrothermal activity on NW Eifuku clearly merit further study. As mentioned above, NW Eifuku is only the second locality where natural venting of liquid CO2 has been reported, the other being the Okinawa Trough, a back-arc basin environment. At the time of its discovery, the Champagne site was the only arc volcano where the phenomenon of liquid CO2 venting had been found. However, venting of a separate CO2 gaseous phase was recently observed at 3 other submarine arc volcanoes: Nikko volcano in the Mariana Arc [Lupton et al., 2005], and Giggenbach volcano and Volcano 1, both in the Kermadec Arc [Lupton et al., 2005; Stoffers et al., 2006]. Furthermore, to our knowledge liquid CO2 venting has never been found on mid-ocean ridges, suggesting that this type of activity is more prevalent on volcanic arcs and the associated back-arc basins. Experiments are being designed to accurately measure the flux and oceanic dispersal of CO2 at NW Eifuku. In addition to physical and chemical measurements, the hydrothermal sites on NW Eifuku are a valuable natural laboratory for studying the effects of high CO2 concentrations on marine ecosystems.
We thank K. Shepard, K. Tamburri, and the other members of the Canadian ROPOS team, and the captain and crew of the R/V Thompson for their excellent support during the 2004 SROF expedition, and K. Chiba and the other members of the Hyper-Dolphin team and the Captain and crew of the R/V Natsushima for their support during the NT05-18 expedition. We thank S. Merle for help with the figures and Tom Brown and the Natural 14C Group at the Center for Accelerator Mass Spectrometry at Lawrence Livermore National Laboratory. V. Salters, D. Hilton, C. German, and T. Fischer provided constructive reviews of the manuscript. Radiocarbon measurements were supported in part by funding from CAMS through the University Collaborative Research Program. This publication was partially funded by the Joint Institute for the Study of the Atmosphere and Ocean (JISAO) under NOAA cooperative agreement NA17RJ1232, contribution 1154. This work was supported by the NOAA Ocean Exploration Program and by the NOAA VENTS Program. This is PMEL contribution 2843.