The global carbon dioxide (CO2) flux from subaerial volcanoes remains poorly quantified, limiting our understanding of the deep carbon cycle during geologic time and in modern Earth. Past attempts to extrapolate the global volcanic CO2 flux have been biased by observations being available for a relatively small number of accessible volcanoes. Here, we propose that the strong, but yet unmeasured, CO2 emissions from several remote degassing volcanoes worldwide can be predicted using regional/global relationships between the CO2/ST ratio of volcanic gases and whole-rock trace element compositions (e.g., Ba/La). From these globally linked gas/rock compositions, we predict the CO2/ST gas ratio of 34 top-degassing remote volcanoes with no available gas measurements. By scaling to volcanic SO2 fluxes from a global catalogue, we estimate a cumulative “unmeasured” CO2 output of 11.4 ± 1.1 Mt/yr (or 0.26 ± 0.02·1012 mol/yr). In combination with the measured CO2 output of 27.4 ± 3.6 Mt/yr (or 0.62 ± 0.08·1012 mol/yr), our results constrain the time-averaged (2005–2015) cumulative CO2 flux from the Earth’s 91 most actively degassing subaerial volcanoes at 38.7 ± 2.9 Mt/yr (or 0.88 ± 0.06·1012 mol/yr).
Volcanism is the primary mechanism through which carbon (C) stored in the deep Earth1,2 is transferred to surface environments to feed C exchanges in the atmosphere-ocean-biosphere system3. Over geological time, volcanic CO2 emissions have been a key control on atmospheric-oceanic CO2 levels4,5,6,7,8, ultimately regulating evolution of climate and life on our planet9,10.
The global volcanic CO2 flux in modern Earth remains inadequately known11,12 and, ironically, is less constrained for subaerial volcanoes than for the less-accessible mid-ocean ridges, for which the 3He flux13 or the CO2/Ba ratio14 proxies have successfully been applied. Direct volcanic CO2 observations at subaerial volcanoes are technically challenging from both ground11,15 and space16 due to the large atmospheric CO2 burden, and thus remain limited in number17,18. The volcanic CO2 flux can be quantified indirectly by combining simultaneous acquisitions of UV-sensed sulphur dioxide (SO2) fluxes11,15,19,20 and gas compositions (CO2/SO2 ratios), but gas observational networks are still in a developing stage21,22, resulting in sparse and incomplete gas catalogues23,24. CO2 flux data have so far been obtained11,15 for only <60 of the several hundred currently degassing Holocene volcanoes25. CO2 flux records are continuous enough only for a few (<10) volcanoes where permanent instrumentation is operating26,27,28,29, while sparse results (one or a few campaign-style measurements at most) are available for the remaining ~50. In addition, scarce or even no information exists for several top-ranking degassing volcanoes30 in remote regions of the world (e.g., Vanuatu31, Papua New Guinea, the Solomon arc, and the Sunda-Banda arc in Indonesia32,33). Attempts to extrapolate available measurements to all the subaerial degassing volcanoes have been problematic11,23 and require use of questionable statistical approaches34,35. Estimates of the global volcanic CO2 flux thus vary widely, from 66 to 540 Mt/yr11,23.
Ideally, refining the volcanic CO2 inventory would require a comprehensive record comprising simultaneous composition/emission measurements for all the currently active strong volcanic gas emitters globally. The top-degassing volcanic targets during 2005–2015 (Table 1) have recently been identified30 from satellite-based observations of the SO2 flux using the Ozone Mapping Instrument (OMI). Carn et al. (ref.30) identified 91 volcanoes, listed in Table 1, releasing SO2 at rates above the OMI detection limit of 16 tons/day. Gas CO2/ST ratios (where ST is Total Sulfur, corresponding to SO2 in these strongly degassing magmatic-volatile emitting volcanoes) are available for 57 out of these 91 volcanic sources36, from which SO2 fluxes can straightforwardly be converted into CO2 fluxes (Table 1). The remaining 34 volcanoes, however, have so far been impossible to characterise for gas composition, owing to their remoteness and/or poor accessibility, leaving their CO2 fluxes unconstrained.
Here, we explore an alternative approach of indirectly inferring the CO2/ST ratio signature of these “unmeasured” volcanoes, and ultimately their CO2 flux, based on the (far more commonly measured) trace element compositions of their erupted volcanic rocks. Volcanic arc gas CO2/ST ratios and whole-rock trace element ratios (e.g., Ba/La or Sr/Nd ratios) are globally linked36, as both volatiles and fluid/melt-mobile elements (e.g., Ba and Sr) are sourced from fluids delivered from dehydration/melting of subducting slab sediments and altered ocean crust37,38,39,40,41,42. Based on their gas vs. whole-rock associations, arc volcanoes cluster into three Groups36. Group 1, which includes C-poor arc volcanoes (gas CO2/ST ratios ≤2), are thought36 to be sourced by a mantle wedge source contaminated by C-poor slab fluids (derived from either terrigenous sediments or altered oceanic crust). Group 2 volcanoes are assumed to inherit their C-richer (2≤ CO2/ST ratios ≤4) gas composition from incorporation into the mantle wedge of slab fluids derived from melting of carbonated sediments. Group 3 (CO2/ST ratios >4), finally, includes C-rich arc gases, supporting the involvement of an additional crustal C contribution (de-carbonation/assimilation of upper crustal limestones43,44).
We here establish systematic gas vs. rock relationships at the scale of individual arc segments and/or groups of volcanoes. These relationships, once set, allow us to predict the CO2/ST ratio for any volcano for which trace-element whole-rock information (but not gas composition) is available. Ultimately, using these predicted CO2/ST ratios in tandem with available SO2 flux information30, we derive CO2 fluxes for all current top-degassing volcanoes and, by summation, a refined inventory of decadal (2005–2015) global CO2 emissions from subaerial volcanism.
CO2 fluxes for the Earth’s best-studied volcanoes
Roughly ~62% of the 91 strongest volcanic SO2 sources globally30 have been characterised for both SO2 flux and (episodically) for volcanic gas compositions (Table 1). CO2 fluxes are thus obtained (see “Methods”) by pairing the OMI-based time-averaged 2005–2015 SO2 fluxes30 with the characteristic (mean) CO2/SO2 ratios in the corresponding high-temperature magmatic gases (data from ref.36 unless otherwise noted). The so-derived CO2 fluxes (Table 1) range from 28 to 15,800 tons/day, and are in reasonable agreement (typically within a factor ≤40%) with the CO2 fluxes estimated using ground-based SO2 flux measurements11,15. We estimate the cumulative CO2 flux from the 57 volcanic sources with “measured” gas compositions by applying a Monte Carlo method (see Table 1) to the dataset. The obtained cumulative “measured” flux is 27.4 ± 3.6 Mt/yr (or 0.62 ± 0.08·1012 mol/yr).
Matching gas and whole-rock trace element compositions
Thirty-four top-ranking volcanic SO2 sources do not have gas compositional records (Table 1). We hereafter refer to such volcanoes without CO2/ST information as “unmeasured” volcanoes.
We thus explore a methodology to predict the characteristic volcanic gas CO2/ST ratio of each of these 34 “unmeasured” volcanoes using their averaged trace-element volcanic rock composition (Table S1). Gas CO2/ST ratios in arc volcanoes exhibit systematic global relationships with slab fluid trace-element proxies (e.g., Ba/La or Sr/Nd ratios) in the corresponding whole-rocks, which are interpreted36 as resulting from a common CO2-Ba-Sr derivation from melting of subducted sediments in the slab40 (variably enriched in CO2; ref.42). These relationships, once set at the scale of individual arc segments (Figs 1 and 2) or volcano Groups (Fig. 3), can now be used to infer the representative volcanic gas CO2/ST ratio signature of the 34 “unmeasured” volcanoes (Tables 1 and S1).
The procedure is illustrated in Fig. 1 and Table 2 for Pacaya volcano as an example (see “Methods”). The initial step involves establishing a CO2/ST vs. Ba/La relationship using data for volcanoes for which both gas and trace element data are available (for the specific Pacaya example, we use gas/whole-rock information for Central American volcanoes, see Table S1a and Fig. 1). As in previous work36, the representative CO2/ST ratios used in Fig. 1 (listed in Table 1 and S1a) are obtained by averaging available results for high-temperature gas samples, in the attempt to reduce the effect of secondary processes (e.g., magmatic S scrubbing during gas-water-rock reactions45) that become substantial at hydrothermal (temperature <400 °C) conditions. Secondly, regression analysis is used to fit the gas vs. trace-element association via either a (i) linear or (ii) logarithmic regression model (Fig. 1; see “Methods”). We focus on the two regression models based on the assumption that linear/logarithmic functions best describe depleted mantle (DMM)-slab fluid mixing in a CO2/ST ratio vs. Ba/La (or Sr/Nd) compositional field36. Finally, the adopted regression function is used to derive a “predicted” gas CO2/ST from available Ba/La whole rock data (Fig. 1). In the specific Pacaya example (Fig. 1 and Table 2), using a linear regression to fit the volcanic gas and DMM data-points (our RM3 regression model, see “Methods” and Table S1), the “predicted” gas CO2/SO2 ratio is 1.4 ± 0.75, well within the magmatic gas range (CO2/SO2 ratio of 1.1 ± 1.0) recently determined46 from plume measurements (Fig. 1).
CO2/ST ratios from individual-arc gas vs. trace-element relationships
Gas vs. rock (trace element composition) associations are initially analysed at the scale of individual arc segments, in the assumption that, at such regional scales, sources and transport pathways of volatiles and trace elements are relatively uniform. In truth, intra-arc variations in thickness, age, thermal properties and composition of the slab and overlying plate47, and in the composition of subducted sediments42, are large enough to impact the mechanisms of magma generation, and thus impart regional trends in volatile48 and trace element49 signatures of erupted magmas. Nonetheless, it is on these individual-arc trends that we rely below. Three arc segments have enough volcanoes measured for both gases and rocks to allow reliable gas vs. rock associations to be established (Fig. 2).
The Central American Volcanic Arc (CAVA) CO2/ST vs. Ba/La relationship, obtained from results listed in Table S1a, is illustrated in Figs 1 and 2a. The systematic along-arc trace-element patterns in CAVA volcanic rocks49 (Fig. 2b,c) originate from changes in geometry, age, thermal regime and extent of serpentinization of the subducting Cocos plate slab50. As more recently found36,51, such trace-element variations correlate with those of CO2/ST ratios in high-temperature magmatic CAVA gases. These correlations (e.g., Figs 1 and 2a) have been explained36,51 as resulting from the variable addition of C-Ba-Sr-rich fluids issuing from melting of limestone-bearing slab sediments, with the highest slab-fluid influx being observed in Nicaragua52, where magmatic gases consistently have C-rich (Group 2) affinity (Fig. 2a). In Costa Rica and El Salvador, magmatic gases are typically C-poorer36,51 (Group 1), in line with the lower slab affinity (and more depleted mantle-like signature) of trace-element ratios (Fig. 2). All the CAVA volcanic SO2 emitters of Table 1 have been measured for gas composition (at least for their CO2/ST ratio), except for Guatemalan volcanoes Fuego and Santa Maria. We use the CAVA CO2/ST vs. Ba/La association (of Fig. 2a) to fill this gap of knowledge. Using the RM3 regression model in tandem with mean whole-rock Ba/La ratios (Table S1a and Fig. 2a), we infer CO2/ST ratios of respectively 1.7 ± 0.75 (Fuego) and 1.6 ± 0.75 (Santa Maria).
Our compilation (Table 1) shows that volcanic gas CO2/ST data are available for the majority of the volcanic SO2 emitters in the Northern (NVZ), Central (CVZ) and Southern (SVZ) Volcanic Zones53 of the Andes (Southern America). Very limited gas information is available54 for Ecuadorian volcanoes, however, and here we use the CO2/ST vs. Ba/La association (for South-America: Fig. 2d) to fill this knowledge gap. In the Andes, there is documented evidence in the literature for large along-arc variations in volcanic rock trace-element geochemistry55,56,57,58. Our partial whole-rock dataset, based on the subpopulation of Andean volcanoes listed in Table S1b, demonstrates an overall south-to-north increase in trace-element slab-fluid proxies (Ba/La, Sr/Nd and U/Th; Fig. 2e,f), from Copahue volcano in Argentina (SVZ) to Nevado de Ruiz in Colombia (NVZ). Importantly, the along-arc variations in the volcanic gas CO2/ST ratio scale well with the trace-element variation patterns (Fig. 2d), again suggesting common source processes. The trace-element signature of the three most actively degassing volcanoes today in Ecuador, Tungurahua59, El Reventador60 and Cotopaxi61 (the latter not appearing in the 91 list of top degassing volcanoes30), places Ecuadorian magmatism in an intermediate position between Colombian volcanoes in the NVZ (the richest in Ba and Sr, but also CO2; Fig. 2d) and intermediate C-rich Peruvian volcanoes62 further to the south (in the CVZ). The mean Ba/La ratios, combined with the CO2/ST vs. Ba/La linear regression model displayed in Fig. 2d, constrain the CO2/ST ratio for Tungurahua and El Reventador at 2.5 ± 0.8 and 2.2 ± 0.8, respectively (see Table 1 and S1b). A consistent CO2/ST ratio is inferred for Cotopaxi (2.5 ± 0.8).
The case of Indonesia, which includes the Sunda-Banda and Sangihe-Halmahera arc segments, is particularly problematic (Fig. 2g–k). The large along- and within-arc variations in crustal63 and slab64 structures, combined with heterogeneities in the sedimentary slab input42 (Fig. 4), make it difficult to characterize regional trends in volatile sources. In the Java sector of the Sunda arc, the respective roles of crust and slab in controlling rock65 and gas66 geochemistry are widely debated, with some authors stressing the importance of upper plate assimilation67,68 and others emphasising a slab control69,70,71. The Group 3 signature36 of Merapi and Bromo (Fig. 2g) supports involvement of crustal carbon in Central Java72. South-to-north along-arc trends in gas 3He/4He (decreasing) and CO2/3He (increasing) ratios66 suggest a crustal volatile contribution is also likely in Sumatra, where the crust is especially thick and limestones widely exposed63,67. In contrast, crustal assimilation is supposedly minor (if any) in other sectors, including west and east-Java65, Nusa69,73 Banda74 and Halmahera33. In these segments of the Sunda-Banda and Sangihe-Halmahera arcs75, along-arc variations in He-C isotopes66,76,77, and the sparse high-temperature gas information, suggest variable C delivery from the slab, and thus coexistence of Group 1 and 2 volcanism (Fig. 2g). This is not unexpected, in view of the C heterogeneity in subducted sediments, from terrigenous and C-poor (Sumatra-Java) to pelagic and C-richer (Nusa, east Sunda)42 (Fig. 4). The diverse volatile sources that are possibly involved, in addition to the paucity of gas data, create scatter in CO2/ST vs. Ba/La (Fig. 2g). Only 9 Indonesian volcanoes have been measured for both whole-rock trace element composition and (high-temperature) magmatic gas composition (Table S1c). These CO2/ST vs. Ba/La data can be fitted by either a linear (RM3) or logarithm (RM4) regression model with identical regression coefficients (R2 = 0.52; Fig. 2g). We therefore infer the CO2/ST ratio signature of the “unmeasured” Indonesian volcanoes (Table 1) by averaging the output of the two regression models (Table S1c). The low regression coefficients (Fig. 2g) imply the inferred CO2/ST ratios should be treated with caution, as they require validation/refinement with an improved (more than 9 data-points) gas vs. trace element relationship. We caution, in particular, that the predicted CO2/ST ratios (Table 1) may either over-estimate (for Group 1 volcanoes) or under-estimate (for Group 3 volcanoes) by a factor ~1.3 (the max error in Fig. 2g) the real volcanic gas CO2/ST ratios of “unmeasured” Indonesian volcanoes.
CO2/ST ratios from Group-based gas vs. trace element relationships
Several of the “unmeasured” (for gas) volcanoes in Table 1 are sited in arc segments for which insufficient gas/rock information is currently available to establish individual-arc associations (as those analysed in Fig. 2). In order to derive information on their CO2/SO2 ratio gas signature, we use the global relationship between CO2/ST and Ba/La in Groups 1–2 volcanoes (ref.36) (Fig. 3).
The majority of the remaining “unmeasured” (for gas) volcanoes in Table 1 are sited in arc segments for which available deep sea drill holes point to the lack of C-rich lithologies (limestones) in the subducted sediment succession42 (Fig. 4). Trench sediments poor in C have been identified in the segment of the Pacific Ring of Fire (Fig. 4) that stretches from Aleutians-Kuril-Kamchatka to the N/NW to Marianas/Japan/Philippines further south (10 “unmeasured” volcanoes in total – see Table 1). Where high-temperature gas information is available, a CO2-poor (Group 1) signature of volcanic gases36 has typically been observed in such carbonate-poor trenches (Fig. 4), matching well the small sedimentary slab C input42. Sediments are similarly C-poor (e.g., prevailingly terrigenous and biosiliceous42) in the Tonga and South Sandwich arcs (3 “unmeasured” volcanoes; Fig. 4). We therefore assign to Group 1 all the “unmeasured” (for gas) arc volcanoes fed by carbonate-poor trenches. Group 1 volcanoes exhibit little change in gas CO2/ST ratios with increasing Ba/La (Fig. 3a). This implies either (i) limited C delivery from the slab in the absence of carbonated sediments (e.g., that fluids/melts delivered by terrigenous sediments, altered oceanic crust and/or serpentinite are not major C sources36), or (ii) that slab C and S are added to the mantle wedge in 1:1 to 4:1 proportions at most (Group 1 volcanoes typically have CO2/ST ratios ~3–4 times higher than the DDM). The lack of dependence on Ba/La (Fig. 3a) means that we can prudently use the measured Group 1 CO2/ST ratio average (1.2 ± 0.5; see Table S1d) for all the “unmeasured” (for gas) Group 1 volcanoes in Table 1.
Group 2 volcanoes are, by definition36, those having CaCO3-rich sediments in their trenches. These volcanoes typically have more C-rich volcanic gas composition (CO2/ST ratio >2 but ≤4) and exhibit stronger, steeper correlation between gas CO2/ST and trace element ratios (Fig. 3a). These Group 2 volcanoes are located in high biological productivity zones close to the tropics, where sediments are increasingly biogenic in nature and/or where seafloor is shallow enough (above the calcite compensation depth, CCD) to support carbonate deposition42 (Fig. 4). Of the few remaining “unmeasured” (for gas) volcanoes in Table 1, those in the Papua New Guinea-Solomon-Vanuatu arc segment are thus potential candidates for Group 2. The Papua New Guinea-Solomon arc sectors (Fig. 4) are a particular challenge because no gas samples are available, and no deep sea drill holes have been placed in the seafloor of the Solomon Sea, seaward of their trenches. Likewise, there are few relevant piston cores to provide any seafloor samples. Our inferences are thus based on seafloor depth, assumptions about the regional CCD, and drill sites in other, nearby southwest Pacific marginal seas. At DSDP Site 63, in the East Caroline Basin north of New Britain, carbonate lithologies were encountered throughout the entire section, from the Quaternary to the middle Oligocene basaltic basement78. This site, at 4472 m water depth, has thus been above the CCD over its entire history. Similarly, drilling at DSDP 287 (4653 m water depth), in the Coral Sea south of Papua New Guinea and east of the Solomon Islands, intercepted abundant carbonate lithologies through most of the sedimentary section to its lower Eocene basement79. Given that the water depths of the Solomon Sea are predominantly <4500 m seaward of the New Britain, Solomon and Northern Vanuatu trenches, we expect this seafloor to have been above the CCD for much of its history as well, and thus to be delivering carbonate-rich sediment to these subduction zones. Based on the above, we consider it very likely that “unmeasured” volcanoes in the Papua New Guinea- Northern Solomon-Vanuatu arcs belong to Group 2. We use therefore the CO2/ST ratio vs. Ba/La global association for Group 2 volcanoes (see Fig. 3a) to predict (based on trace element information) CO2/ST ratios ranging from 2.1 ± 0.7 to 2.7 ± 0.7 for these volcanoes (Tables 1 and S1e). We note that the two “measured” volcanoes in the central and southern Vanuatu arc (Bembow on Ambrym Island, and Yasur on Tanna Island) both exhibit Group 1 gas affinity (CO2/ST of 1.5–1.6), implying that the predicted C-richer gas signature for northern Vanuatu volcanoes requires validation from measurements.
Validity of whole rock trace element proxy for CO2/ST
Our predicted CO2/ST ratios stand on the assumption that gas compositions are linked to trace element compositions of their source magmas at either regional (Fig. 2) or global (Fig. 3) scale. Implicit in establishing such relationships is that gas (CO2/ST) and trace-element (Ba/La) whole-rock tracers are inherited by the same processes at their source, and are similarly conserved during magma ascent, decompression and degassing/eruption36. For Ba/La, a link has been made between signatures of arc rocks and subducted sediments at corresponding tranches80, so that this and other trace element ratios are commonly used as slab-fluid proxies for charactering the mantle source of magmas81,82. Both elements exhibit incompatible behaviour during magma differentiation, so that the source-inherited ratios are essentially conserved during magma evolution, at least for the mafic to intermediate (andesitic) magma compositions considered here (as outlined in the Method).
The behaviour of volatile components CO2 and S is obviously complex during the generation and evolution of slab fluids and mantle-derived magmas83. Not only are slab sources and processes only partially understood for C and S12,39, but these volatile species will be selectively extracted from melt and partitioned into the vapour phase according to their melt solubilities (that dependent in a complex fashion on magma T-P-X-redox conditions), upon magma decompression and differentiation84,85. One may thus argue that degassing-related fractionations, for which abundant model86,87,88, experimental89 and observational90 evidence exists, act as to render the CO2/ST ratios in both degassed melt (preserved in melt inclusions in phenocrysts) and exsolved vapour (discharged as volcanic gases) unrepresentative of the mantle source compositions, and thus unlinked91 to Ba/La or other trace element proxies.
Where sufficient data exists (e.g., Figs 2a,d and 3a), however, the CO2/ST vs. Ba/La correlations appear systematic and statistically significant, and we consider unlikely that these associations are purely accidental. Our regional/global associations here, thus, implicate that the time-averaged CO2/ST ratios of volcanic gases ultimately reflect the volatile ratios in the parental (un-degassed) melt, and in the mantle source. To reconcile this with the well-established degassing-driven CO2 vs. ST fractionations, we observe that, at least at mafic systems, comparison between measured and modelled (from numerical simulation of magma degassing paths using volatile saturation codes86,87,88) gas CO2/ST ratios typically imply equilibrium pressures (e.g., pressures of final gas-melt segregation) of 0.1–5 MPa during quiescent degassing activity29,36,84,85,92. Thus, at least during non-eruptive periods, during which the majority of the volcanic gas observations in the literature are taken, observations and models both indicate very shallow (a few hundred meters below the magma-air interface) gas segregation from the convecting feeding magmas93,94. If shallow closed-system degassing conditions85,94 prevail, then the magmatic gas phase released as volcanic gas during open-vent activity does represent an integral of volatiles exsolved from melt during most (P > 5 MPa) of the magma decompression path. This released magmatic gas is thus very similar in composition to the source and parental melt volatile signature, irrespective of its hydrous (for arc volcanoes) or more H2O-poor (for non-arc systems) nature20,93. The short-lived (days to weeks) pulses of CO2-rich gas, seen prior to eruption of mafic arc volcanoes27,28,29,84,92, imply somewhat deeper (typically, ~10–30 MPa) last gas-melt equilibration, but yet suggest closed-system is maintained up to rather shallow levels in a magmatic plumbing system, at least during quiescence. During basaltic explosive activity, deeper gas segregation is implied by gas observations95,96, but such eruptive degassing contributes only a minor fraction of the total degassing budget, which is dominated by passive emissions93.
The lack of a systematic correlation between volcanic gas CO2/ST ratios and SO2 fluxes (Fig. 5) further supports the idea that the former are not significantly affected by variable extents of magma degassing and gas-melt separation depths at various volcanoes. In mafic systems, the SO2 flux is a proxy for the rates of magma degassing in a volcano’s shallow (<3 km) plumbing system93. As such, at least in principle, shallow magma ascent and decompression should be tracked by increasing SO2 flux and decreasing CO2/ST ratios in the surface gas output26, a relationship that is not observed in our global dataset (Fig. 5). The SO2 flux-independent, distinct CO2/ST distributions of Group 1, 2 and 3 volcanoes (see Fig. 5) suggest, instead, that source signature, rather than degassing, ultimately controls the longer-term, time-averaged volcanic gas compositions. We caution that CO2/ST ratio volcanic gas compositions may become less source-related in intermediate to silicic systems, where the gas output is often buffered by gas-melt equilibration in crustal, vapour-saturated magma reservoirs97,98,99,100. It is thus possible that part of the scatter in our gas vs. trace-element associations (Figs 2 and 3) is caused by the intermediate (andesitic) systems included in our dataset. Silicic systems have intentionally been excluded from our compilation.
The good match between our predicted and measured CO2/SO2 ratios at Pacaya volcano (Fig. 1) also support, although indirectly, the validity of our gas vs. trace element associations. In addition to Pacaya, recent airborne gas measurements54 at Tungurahua and Cotopaxi volcanoes in Ecuador have found CO2/SO2 ratios (in the 2 to 2.5 range) fully overlapping our predicted range (2.5 ± 0.8; Table 1). These successful tests provide confidence in the robustness of our predicted CO2/ST ratios. We caution that, in order to validate our methodology further and reduce the scatter in gas vs. trace element scatter plots (e.g., Fig. 3g), gas observations should be prioritized in remote, unexplored volcanoes in Papua New Guinea, Sandwich Islands, Solomon Islands, Sumatra, east Sunda-Banda, and north-Vanuatu. In some of these arc segments (e.g., Sumatra, Sunda), crustal C may be involved63,66,67, in which case our predicted CO2/ST ratios may underestimate the actual magmatic gas ratio (by a factor up to ~1.5–2). We also advise that, since only high-temperature (SO2-dominated) gas data are used to establish our gas vs. trace-element associations (Figs 2 and 3), our predicted CO2/ST ratios are representative of the magmatic gas signature, irrespectively of whether or not hydrothermal processes are acting to alter the actual and total gas volcano emissions. For example, the hydrothermal (H2S-rich) gas emissions from Marapi volcano in Sumatra have measured CO2/ST ratios of 20.5 ± 1.1 (Table 1), well distinct from what we would predict (CO2/ST ratio of ~2.6) using the whole-rock Ba/La (19 ± 3; Table S1c) and the Indonesian gas vs. trace-element relationship (Fig. 2g). As such, discrepancy between measured and predicted CO2/ST ratio at any other hydrothermal volcano may lead to apportioning the fraction of S lost to (or C produced by) the hydrothermal system. While we believe that hydrothermal processing should be the exception rather than the rule for the satellite-sensed volcanoes here, we ultimately anticipate our predicted CO2/ST ratios (Table 1) will require revision and upgrading as new high quality gas data become available for newly measured volcanoes.
One important aspect to consider is that our regional/global associations (Figs 2 and 3) are based on averaging trace element information for rocks erupted during decades to millennia of volcanic activity. As such, the CO2/ST ratios predicted from such associations should be viewed as long-term means over a volcano’s lifespan, rather than the instantaneous measurements as obtainable by direct gas observations. These “geologic” gas CO2/ST ratios may thus serve, when combined with measured S content in mafic glass inclusions, to estimate the initial CO2 content in parental, un-degassed melts, and eventually in the sub-arc mantle. Both are similarly poorly contrained101,102 due to pre- and post-entrapment loss to vapour of poorly soluble CO2.
A decadal global CO2 flux budget
Our predicted CO2/ST ratios are converted into CO2 fluxes (Table 1) by assuming ST = SO2 and scaling to the OMI-based mean SO2 fluxes for the 2005–2015 period30. We focus on the OMI satellite dataset owing to advantages brought by its global and coincident observations, but yet observe that quantitatively similar results would be obtained using ground-based SO2 flux observations instead15. The predicted CO2 fluxes range from 57 tons/day (Kanlaon volcano in the Philippines) to 6200 tons/day (Bagana volcano in PNG) (Figs 4 and 6). Uncertainty in the derived CO2 fluxes (see Table 1, column N) is based on propagation of the respective errors on SO2 flux (column G) and predicted CO2/ST ratios (column I).
The total cumulative CO2 emissions from the 34 “unmeasured” volcanoes (those with no measured gas information available) would thus be ~11.4 ± 1.1 Mt/yr (~0.26 ± 0.02·1012 mol/yr), thus adding an additional ~34% to the cumulative “measured” mean CO2 emissions in 2005–2015 (27.4 ± 3.6 Mt/yr; Table 1). Finally, our extrapolated (measured + predicted) CO2 flux budget is 38.7 ± 2.9 Mt/yr (or 0.88 ± 0.06·1012 mol/yr). It is important to notice that our approach, in which the CO2/ST ratio signature of each volcano is independently evaluated, leads to far better constrained CO2 budget (7% uncertainty at 1 SD) that would be possible using any “averaged” volcanic CO2/ST ratio proxy (as has been often attempt in past studies). For example, scaling the mean global SO2 flux (23 ± 15 Mt/yr) to the mean volcanic CO2/ST ratio (2.7 ± 3.6) (all data from Table 1) would lead to a global CO2 flux of 62 ± 92 Mt/yr (e.g., 148% uncertainty at 1 SD).
Based on our results, we infer that 6 strongly degassing volcanoes with time-averaged (2005–2015 means) CO2 fluxes of ≥ 5000 tons/day dominate the global CO2 budget (Figs 4 and 6). One of these (Bagana, PNG) is an “unmeasured” volcano and would not have been identified as a top CO2 emitter without the proxy approach developed here. It is interesting to observe that while the SO2 global budget is dominated by the Group 1 volcanoes (accounting for 13 out of the 28 strongest volcanic SO2 sources; Fig. 6), the CO2 global budget is predominantly determined by the CO2-enriched arc volcanoes in Group 2 (13 out of 28) and Group 3 (5 out of 28, with 2 - Popocatepetl and Etna - in the top-5 list) (Fig. 6). Two continental rift volcanoes (Nyiragongo and Nyamuragira) and two within-plate volcanoes (Kilauea and Erebus) also appear in our top-28 list of volcanic CO2 emitters (Fig. 6).
Our extrapolated global CO2 flux of 38.7 ± 2.9 Mt/yr is lower than previous global volcanic CO2 flux estimates in the literature, ranging from 66 to 540 Mt/yr (see ref.11 for a review). Several causes can explain this mismatch.
First, and most importantly, our global volcanic CO2 budget here only includes the contribution from the “strongly degassing volcanoes” that emit SO2 in quantities large enough to be detected from space (by OMI in this specific case30). We therefore admittedly do not take into consideration in our estimate the CO2 contribution from mildly degassing “magmatic” volcanoes (those still emitting SO2, but at levels too low to be resolved by satellites) and from “hydrothermal” volcanoes in which CO2 is emitted in combination with H2S (instead of SO2). Although typically exhibiting weaker surface gas manifestations, compared to the OMI-detected volcanoes characterised here, these magmatic-hydrothermal systems do often exhibit C-rich gas compositions36 (reflecting the extent/mechanism of gas-water-rock reaction with meteoric-hydrothermal fluids45), and do emit CO2 at the ~1000 tons/day level in the most extreme cases17, but most typically in the hundreds of tons/day range15. Considering that several hundreds of volcanoes worldwide are currently undergoing mild magmatic-hydrothermal degassing activity, this emission type could be responsible for the emission of several tens of Mt CO2/yr globally11,15. Also, we do not account for the CO2 output from volcanic lakes103, and diffuse/regional soil CO2 emissions around volcanic systems104, for which more data and alternative extrapolation approaches would be required. We therefore stress our results are not intended to represent total CO2 emissions from global subaerial volcanism, but rather the magmatic CO2 budget fraction contributed by the most actively degassing volcanoes on Earth.
Secondly, the mismatch in the estimated CO2 fluxes (this work and previous studies) derives (at least partially) from the distinct gas datasets used. We here specifically base our CO2 budget calculations on a consistent set of coincident (satellite-based) SO2 flux measurement, taken during a relatively short (decadal) period and with same retrieval/processing technique. In contrast, previous estimates have been hampered by the combination of sparse observations, taken over several decades, and with diverse observational/retrieval techniques. Even volcanoes which are persistently active alternate periods of elevated degassing with phases of reduced activity, and so non-coincident observations (taken over periods spanning several decades) may lead to biases. For example, by combining measurements taken between 1954 and 2011, a cumulative CO2 flux of 59.7 Mt/yr (from 33 measured volcanic gas plumes) was obtained11, or 2 times more than our mean 2005–2015 flux. We also explicitly use CO2/SO2 information for high-temperature magmatic gases only, in contrast with previous efforts23 in which individual arc CO2 emissions have been quantified also considering low-temperature hydrothermal gas samples in which the C-rich composition is not representative of the strongly degassing “magmatic” arc systems. We also cannot rule out that part of the discrepancy is due to our Ba/La approach, which may only represents the sub-Moho magmatic CO2 flux, and not a potentially large44 recycled crustal CO2 flux. Finally, our “measured” CO2 dataset is extrapolated to the total number of “unmeasured” strongly degassing volcanoes by predicting, for each of them, the specific CO2/SO2 ratio gas signature, rather than relying on the assumption that the global CO2 flux population obeys a specific statistical distribution (e.g., the power law distribution105).
Our results implicate that the arc volcano C flux (~8 ± 0.6 Mt C/yr) corresponds to a significant amount (~50%) of the subducted sedimentary carbonate (15 ± 2 Mt/yr; ref.106), but only a relatively small fraction ( < 21%) of the total C input at arc trenches (40–114 Mt C/yr; refs1,12). Thus, either the C input is balanced by “diffuse” C output forms, such as regional aquifers or soil degassing107 in the arc crust, or a substantial fraction of the subducted C is ultimately not erupted, but rather stored either in the lithospheric mantle8 or in the deep mantle1,2.
The SO2 flux compilation30 we rely on in this study includes a list of the 91 top-ranking volcanic SO2 degassing sources in 2005–2015 (Table 1). This set of consistent (identical retrieval/processing technique) and simultaneous (global) measurements has improved upon the shortcomings of previous catalogues108, which combined SO2 fluxes obtained with diverse techniques and in disparate temporal intervals (often differing by several decades).
These SO2 flux data are converted into CO2 fluxes by using either measured or predicted molar CO2/ST ratios. For these strongly degassing volcanoes, ST is assumed to correspond to SO2 throughout, since SO2 detection by satellites implies limited or no interaction with hydrothermal system (and thus trivial reduced S species, such as H2S).
For 57 out of these 91 volcanic SO2 sources, we convert SO2 fluxes into CO2 fluxes, by pairing the former with the characteristic (mean) molar CO2/ST (CO2/SO2) ratios in the corresponding volcanic gases (Table 1). For arc volcanoes, we use the time-averaged molar CO2/SO2 ratios compiled by (ref.36), integrated with novel gas information for eight new targets that have only recently been measured for the first time (see Table 1 for data provenance). Arc volcanoes are ranked in Groups (1 to 3) following the original categorization36. For non-arc volcanoes (here referred as Group 4), we average available volcanic gas information in the literature (see Table 1 for data sources). Note that, for both arc and non-arc, in cases where more than one volcano are listed in the original dataset30 (e.g., Nyiragongo + Nyamuragira) due to insufficient spatial OMI resolution, we averaged the available volcanic gas information for the individual volcanoes, weighting each volcano’s CO2/ST ratio by its ground-based S flux (where available) to obtain a combined CO2/ST ratio for the pair (see Table 1).
Thirty-four out of the 91 top-ranking volcanic SO2 sources30 have never been characterised for volcanic gas composition (Table 1). These include four of the top-ten ranking volcanic SO2 emitters30 (Bagana, Rabaul and Manam in Papua New Guinea, and Aoba in the Vanuatu archipelagos; Fig. 4). To indirectly infer the molar CO2/ST ratio gas signature of each of these 34 volcanoes, we use the averaged (mean) trace-element composition of the corresponding volcanic rocks. To this aim, as in earlier work36, we extract trace-element information (Ba, La, Sr, Nd, U and Th whole-rock concentrations) either from the Earthchem data-portal (http://www.earthchem.org/), or from other sources (for volcanoes that do not appear on Earthchem) (see Table S1). Mafic to intermediate (<55% SiO2) rocks are only considered, same as in other work109. From these, we calculate, for each volcano, the mean (±1 SD) of the Ba/La whole-rock ratios (Sr/Nd and U/Th ratios were also calculated; see Table S1). These ratios, in combination with the gas vs. whole-rock relationships illustrated in Figs 1–3, are finally used, to predict the characteristic volcanic gas CO2/ST ratio signature for each of the 34 “unmeasured” volcanoes.
The procedure is exemplified in Fig. 1 for the Pacaya volcano example. We select Pacaya because the recently obtained gas compositions46 can serve as a test of the methodology. The initial step involves establishing the relationship between CO2/ST gas ratios and whole-rock Ba/La ratios, using data for volcanoes for which both gas and trace element data are available (see Fig. 1; Table S1). The CO2/SO2 vs. Ba/La relationship can be established at the scale of individual arc segments (e.g., Figs 1 and 2), or for volcano Groups36 (Groups 1 or 2) (Fig. 3). For the Pacaya example, we rely on gas/whole-rock information for the well-characterised Central American Volcanic Arc (CAVA; Fig. 1). Secondly, we use regression analysis to fit the gas vs. trace-element association via either a (i) linear or (ii) logarithm regression model (Fig. 1). We find that linear regression yields the best data fit in the majority of the cases (see the Pacaya example, Fig. 1a,b), and this regression model is used throughout unless where indicated (see Table S1). We also find that data fitting is systemically optimised when the DMM composition is included in the fitting procedure (compare Fig. 1a,b), and this option is maintained throughout. Note, however, the method output (e.g., the outputted CO2/ST ratio) is poorly sensitive to this choice (see Table 2). Finally, the adopted regression model function (RM3 in the Pacaya example; Fig. 1 and Table 2) is used to calculate a “predicted” gas CO2/ST from available Ba/La information (Fig. 1). The confidence interval or delta, calculated from the regression line and one standard deviation about the regression, is taken as a proxy for the uncertainty in the predicted CO2/ST ratios. Uncertainty on the predicted ratios, as derived, incorporates (although indirectly) uncertainty/variability in “measured” gas CO2/ST ratios (average uncertainty at 1σ, ~26%) and whole-rock Ba/La ratios (average uncertainty at 1σ, ~16%) (see Table S1). In the specific Pacaya example (Fig. 1 and Table 2), our “predicted” gas CO2/SO2 ratio (1.4 ± 0.75) matches well the recently measured45 magmatic gas range (CO2/SO2 ratio of 1.1 ± 1.0). Our tests show that remarkably similar CO2/SO2 ratios (see Table 2) are obtained using other trace-element slab fluid tracers, such as the Sr/Nd ratio (Fig. 1c). We opt in the following for the Ba/La regression model because (i) La is more frequently available than Nd in the Earthchem dataset for the majority of the volcanoes, and (ii) use of the Sr/Nd ratio requires a priori knowledge of volcano affinity for a specific Group (Group 1 and 2 typically exhibit diverse distributions in a CO2/SO2 vs. Sr/Nd scatter plot; see Fig. 1c). This latter information is frequently not a priori available (see below). The same procedure is applied to all unmeasured volcanoes (Table S1a), and the “predicted” ratios (assumed to correspond to CO2/SO2) are combined with SO2 flux results to ultimately infer the CO2 fluxes (Table 1).
Publisher’s note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
This research was funded by the DECADE research initiative of the Deep Carbon Observatory. We wish to thank E. Liu and anonymous reviewer for helpful comments on an earlier version of the manuscript.