Main

In addition to its familiar cycling between the terrestrial biosphere and atmosphere, carbon moves from microfossils on the seafloor to erupting volcanoes and deep diamonds, in a cycle driven by plate tectonics. Subduction links surface biological processes with the deep Earth, creating a planet suffused with the signature of life. The fate and flux of carbon vary from trench to trench, as every subducting slab delivers to the mantle a singular mix of organic and inorganic carbon that heats and pressurizes at a particular rate determined by its age and convergence speed. The inherent reactivity of carbon in each input package and the pressure–temperature–time path experienced by each slab drive chemical reactions that separate carbon into two parts: one that is carried deeper into Earth and one that returns to the surface. Here we discuss (I) the geological happenstance that delivers carbon to deep sea trenches today, (II) the reactions and movements in the subduction zone that mobilize carbon into fluids and melts, (III) the carbon that continues deep into the mantle to form molten carbonates and diamonds, and (IV) the carbon that escapes such a deep descent to be eventually emitted from arc volcanoes (Fig. 1).

Fig. 1: The deep carbon cycle.
figure 1

(I) Carbon input to subduction zones from the sedimentary (yellow), oceanic crust (orange) and mantle (red) layers in the downgoing plate (dark red arrow). (II) Decarbonation reactions in the subduction zone release carbon to upward-moving fluids and melts (blue arrows). (III) Deeply subducted carbon, potentially forming diamonds. (IV) Emission of recycled carbon to the surface via arc volcanism.

Although excellent reviews exist on the global consequence of the deep carbon cycle over Earth’s history1,2,3,4,5,6, our focus here is on the variability in current subduction systems, where carbon inputs and outputs can reveal the physicochemical processes occurring in the subduction zone that ultimately drive the directionality of the cycle. We highlight end-member subduction zones with distinct inputs or pathways, such as Tonga (little sedimentary carbon) and Cascadia (hot slab), each presenting a different natural recycling experiment worthy of focused study. We also show how recent advances in both laboratory and computational approaches now lead to predictions of how and when carbon mobilizes into upward-moving fluids or remains in solid form in the downgoing plate. Finally, we present recent evidence from satellites, portable instruments and melt inclusions that suggests that most of the carbon in arc magmas is recycled from the subduction zone.

The global carbon flux perspective

Earth exhales carbon at volcanoes. Current estimates place the total volcanic outgassing rate at 79 ± 9 (uncertainty 1σ) megatons of carbon per year (Mt C yr−1), with similar amounts emitted at mid-ocean ridges and subduction zones, and the greatest quantities emitted diffusely in intraplate volcanic regions (Box 1). Uncertainties in estimates of volcanic emission rates are quickly declining owing to long-term satellite observation7 and direct sampling of volcanic plumes and diffuse emanations8,9,10. The spatial variability in volcanic carbon is enormous, however, and currently six volcanoes are estimated to control the global budget of direct emissions (Nyiragongo, Popocateptl, Etna, Ambrym, Bagana and Aoba; together contributing more than 5 Mt C yr−1)11. The largest volcanic emissions, however, may be emanating diffusely from calderas and faults10. Although current anthropogenic fluxes (about 9,500 Mt C yr−1)12 dwarf volcanic ones, the long-term fluxes of carbon to the atmosphere have been dominated by volcanic sources over most of Earth’s history. The rate of supply of mantle carbon is governed by tectonic rates of plate spreading and convergence, as well as the plume flux from the deeper mantle, punctuated by large igneous provinces that supply bursts of CO2 in short periods of time.

Deep-Earth carbon fluxes are not a one-way street, however. There is an equally substantial, highly uncertain and spatially variable carbon flux that enters the mantle with oceanic plates at subducting zones. This carbon is derived largely from the ocean, in the form of carbonate shells and remains of marine organisms, as well as carbonate in the oceanic lithosphere. Terrestrial organic carbon is also washed onto the seafloor by large rivers. The fate of subducting carbon has a profound effect on Earth’s evolution, and depends on the efficiency with which carbon is returned to Earth’s surface by devolatilization reactions in the subducting slab and by upward transport in magmas that supply volcanic arcs. We will refer to this return flux of carbon to Earth’s surface as ‘recycling’. A recycling efficiency of 0% means that none of the carbon that is subducted returns to Earth’s surface, but instead is sequestered in the mantle for geologically significant durations.

A low recycling efficiency could have planetary consequences. For example, Earth’s surface has a distinctively high H/C ratio, about an order of magnitude higher than Earth’s chondritic building blocks13. Is this due to more efficient subduction recycling of H2O than that of C over Earth’s history4? The deep subduction of reduced organic carbon amounts to a loss of electrons and therefore an increase in the oxidative potential of Earth’s surface1. Has a low recycling efficiency of organic carbon contributed to Earth’s oxygen-rich surface14? Or is the oxidation of the mantle the net effect of subduction15, given the higher input flux of carbonate than that of organic carbon? What is the separate fate of these oxidized and reduced forms of carbon?

A high recycling efficiency creates a direct connection between subducting carbon and CO2 supply to the atmosphere. For example, there may have been greater subduction of carbonate at certain times in Earth’s history, such as the Mesozoic era, when the seafloor of the shallow Tethys ocean subducted. Did this lead to higher volcanic CO2 emissions and therefore greater warming16,17? Does the climate system respond to CO2 precipitation in the oceanic crust as a sink or source of carbon, and how is this process modulated by plate spreading rates, ocean bottom temperatures and subduction delivery18,19? What are the long-term climate consequences of the fact that subduction of carbonate is a relatively recent development, tied to evolution of calcifying marine microorganisms20? Participation of carbonate subduction in the deep carbon cycle is arguably a recent phenomenon5.

It is therefore critical to understand the efficiency of carbon subduction recycling, which is determined by two general approaches. One is to balance the input (subducted) and output (arc volcanic) fluxes of carbon globally2,3. The most recent estimates indicate about 30% recycling, although the uncertainties are very large in several fluxes (Box 1). Weak correlations may exist between decarbonation efficiency and slab age21. The other approach is to characterize the physicochemical processes in subduction zones that drive recycling, such as fluid and melt generation and transport, coupled with carbon solubility and reaction kinetics, given different carbon feedstocks and pressure–temperature paths. Here we focus on the latter approach and trace carbon through the recycling process, first by considering how and when carbon is deposited on and precipitated in subducting seafloor.

Subducting carbon

Although the ocean is full of carbon—dissolved as bicarbonate and in the shells and bodies of marine organisms—very little carbon makes it to the deep seafloor to be conveyed to a trench. The carbon that enters subduction zones includes calcium carbonate and reduced organic carbon that exist within the sedimentary, oceanic crust and mantle layers of the incoming plate. Each trench makes unique selections from the carbon menu.

Starting with the lowermost layer, mantle peridotite that forms the bulk of the subducting lithosphere readily hydrates and carbonates if exposed to seawater, forming carbonated serpentinites22. However, most peridotite usually resides at least 6 km beneath the seafloor and is not in direct contact with seawater. Faulting and fracturing are necessary to bring mantle rocks to the sea floor or seawater to the mantle. Carbonated serpentinites may form near spreading centres or near trenches. Near spreading centres, extensional faulting is linked to hydration and carbonation reactions, as well as to the precipitation of magnesium and calcium carbonate veins in mantle peridotites22. Carbon isotopes in oceanic peridotites reflect mixing between seawater-derived carbonate and reduced carbon23. Serpentinization is most pronounced at slow-spreading mid-ocean ridges, but most subducting lithosphere is not formed there, because of the low plate production rate and the preferential existence of subduction zones and fast-spreading crust in the Pacific. Oceanic carbonated serpentinites therefore make a minor contribution to the global carbon input flux to subduction zones3, although they may be locally important. The volcanism at the South Sandwich margin, where slow-spreading crust is preferentially consumed, is notable for having some of the highest 11B/10B ratios among island arcs, a feature that could be derived from the high 11B/10B ratio that is typical of seafloor serpentinized peridotites24. The South Sandwich margin may thus represent a rare case, in which subducting carbon predominantly resides in oceanic carbonated serpentinite (Box 2).

Near trenches, the subducting plate deforms and fractures because of bending. The resulting faults have been seismically imaged to penetrate into the mantle of the incoming plate, and seismic velocities decrease towards the trench25, leading to speculation that the extent of hydration due to ingress of seawater could consume an ocean every billion years26,27. However, outer-rise serpentinites have never been sampled, and other factors besides hydration—such as fracturing and anisotropy—may explain this reduction in seismic velocity. The extent of carbonation in unsampled outer-rise serpentinites is also unknown, and fluid pathways longer than 5 km may lead to low fluid-to-rock ratios and low carbon transport into the mantle section of the downgoing plate23. Indeed, electrical-resistivity imaging does not support extensive reaction of seawater with the mantle section of the incoming plate at the Central America trench28. Most global flux estimates for carbonated peridotite are low (Box 1).

In contrast to peridotite carbonation, there is abundant evidence for pervasive carbonation of the near-ridge oceanic crust. Seawater-derived fluids circulate predominantly in the higher-permeability upper-crustal volcanic section, leading to low-temperature precipitation of carbonate minerals. Because these carbonates are largely seawater-derived, they are isotopically heavy (δ13C ≈ 0‰, where δ13C is the deviation of the ratio 13C/12C relative to that of Pee Dee belemnite), although some biotic and abiotic processes also lead to CO2 reduction and precipitation of isotopically light carbon (δ13C < −20‰29,30; Fig. 2a). Nonetheless, the dominant form of carbon in altered oceanic crust (AOC) is calcium carbonate (calcite and aragonite) that precipitates in veins and vugs, as has been found in samples from a small number (about 15) of drill sites. Carbon uptake occurs near the ridge axis in crust that is 20 Myr old, but surprisingly, AOC older than 80 Myr has higher carbonate content31. This may be due to higher bottom-water temperatures in the Cretaceous period promoting greater abiogenic carbonate precipitation32. Although low in carbonate, young AOC (less than 10 Myr old) is isotopically light owing to the intense bio-alteration of young crust30. Thus, an important prediction for carbon inputs is that old plates (for example, Marianas and Tonga) will have greater AOC carbonate concentrations and higher average δ13C, whereas young plates (for example, Cascadia and Central America) will have little AOC carbonate with lower δ13C (Fig. 2a).

Fig. 2: Carbon isotopes in subducted input, diamonds and arc volcanic gas.
figure 2

a, Subducted carbon input, including that in the upper volcanic layer of AOC older than 10 Myr (refs 30,102), carbonate sediments from ODP site 765 (ref. 41), total organic carbon, Corg, in Bengal Fan sediments103. b, Diamonds, including 319 eclogitic diamonds (orange; from ref. 30) and Juina SLD (green; ref. 104). c, Volcanic output per kilometre of arc from ref. 92, where error bars are one standard deviation about the mean value. Yellow shading encompasses mantle values. Diamond populations appear skewed towards light carbon isotopes74, whereas arc volcanic C is skewed towards high δ13C values92. It is unclear whether this reflects preferential recycling of carbonate to the arc and preferential subduction of more refractory organic carbon to the deeper mantle source of diamonds, or whether ancient diamonds reflect more-reduced sources in Earth’s past14.

The sedimentary layer that is deposited on top of the oceanic crust contains dramatically different forms of carbon than the largely inorganic precipitates of the oceanic crust and peridotite. Sediments are the graveyard of marine organisms and the resting place of terrestrial organic remains. Organisms that grow a carbonate shell, such as nannoplankton coccoliths and bottom-dwelling foraminifera, are the richest source of carbon deposits on the seafloor. For example, a 100-m section of nannofossil ooze may contain as much carbon as the entire oceanic crust below it (using average values given in ref. 3). Marine sediments also contain the organic remains of marine and terrestrial organisms. Although most sediments have less than 1 wt% organic carbon, deep-sea fans can dominate the input flux at some margins. For example, a 1.5-km section of terrigenous turbidites with 0.35 wt% organic C (Fig. 2a) contains more carbon than the average oceanic crust. The balance of marine carbonate (δ13C = 0‰ to +3‰) versus organic carbon (δ13C = −22‰ to −27‰) has a very large effect on the isotopic composition of subducting carbon (see ref. 33 and Fig. 2a). Carbonate sediments can approach 100% CaCO3 and therefore may contain more than 10 times the carbon of a sediment rich in organic carbon (1% C; Fig. 2a). This is offset by a greater flux of organic carbon-bearing sediments approaching trenches (in thick fans), so the proportion of organic to inorganic carbon subducted globally may be about 20% (ref. 34).

Although sedimentary carbon has the potential to dominate global input fluxes, it may be entirely absent from some subducting sections. Indeed, the odds are stacked against carbon burial, as most carbonate dissolves and organic carbon oxidizes in the water column before reaching the seafloor. The ocean’s cold and corrosive bottom waters are particularly challenging to carbonate survival. The calcite compensation depth (CCD), which marks the transition between carbonate-bearing and carbonate-absent sediments, is about 5,000 m deep in today’s oceans. The CCD increases locally if the carbonate flux is high, as occurs in regions of high biological productivity, but it was generally shallower (less than 3,500 m) earlier in the Cenozoic era35 and the Cretaceous period36. Much of the oceanic crust subducting today is old (average age of about 70 Myr)37 and the combination of a shallower CCD and thermal subsidence with age means that carbonate is rare on the seafloor near trenches. For example, essentially zero sedimentary carbonate is subducted along the Tonga, Central Aleutian and Kuriles–Kamchatka trenches. On the other hand, abundant carbonate is subducted at the Central American margin, where the seafloor is beneath regions of high biological productivity, and at the New Zealand margin, where the seafloor is shallow38 (Box 2).

Organic carbon is also consumed in the oxic ocean and in the sediments themselves by microbially mediated reactions, so its preservation in sediments requires rapid supply and burial. These conditions are met beneath regions of high biological productivity and in deep-sea fans, where rivers deliver high fluxes of carbon-bearing sediment to the ocean from regions of active uplift and erosion39. On the other hand, vast expanses of the ocean are deserts owing to low biological productivity and to the challenge of survival in the harsh oxic sediments40. Seafloor currently subducting in the western Pacific spent most of its lifetime traversing the central gyres and is therefore devoid of organic carbon; essentially no organic carbon is subducted at Tonga or Honshu34. By contrast, turbidite sediments in the Bengal and Indus fans, which derive from and fringe India’s collision zone with Asia, constitute the largest fluxes of organic carbon into trenches. Other margins with notable piles of sediment containing organic carbon include those of Nankai, Cascadia, Alaska, South Chile and the Southern Antilles34.

Thus, subducting carbon depends on geologic happenstance at locations where a deep-sea fan (high sedimentary organic carbon) or shallow seafloor (high sedimentary carbonate) happen to be near a trench, or where the subducting oceanic plate happens to be created by slow spreading (favouring carbonated serpentinites) or was formed in the Cretaceous period (favouring carbonated oceanic crust). There is exceptionally wide global variation in carbon input to a subduction zone; each downgoing plate has a distinct formation, evolution and sedimentation history (Box 2 and ref. 5), with potentially large along-strike variations41. Global averaging obscures these underlying factors controlling recycling efficiency, which are essential for constructing the long-term history of the deep carbon cycle. This affects not only the amount and distribution of subducting carbon but also its reactivity and isotopic composition, and contributes to heterogeneity in the deep Earth. The former affects the fate of carbon in the subduction zone (discussed in the next section) and the latter serves as a useful tracer for the source of volcanic gases and diamonds (discussed in subsequent sections).

Reactions in subducting carbon

Once subducting lithosphere and sediment, and the carbon that they carry, are transported beyond the trench (point (I) in Fig. 1), the fate of carbon is determined by a game of subtraction. Rising pressure and temperature transform the subducted materials chemically and physically, setting the stage for fractional carbon removal by a mix of processes that determine recycling efficiency.

During transit to sub-arc depths, three main processes subtract carbon from the slab: mechanical removal, metamorphic decarbonation and melting (Box 3). Frontal accretion or underplating of sediment during the initial stages of subduction may remove a substantial amount of carbon41. The dynamic environment at the top of the slab can result in mixing and removal of slab material. Complex tectonic mixtures of lithologies—mélanges—are important in many exhumed subduction complexes, but to first order they are subject to the same set of processes as their constituent lithologies, and their presence or absence cannot be predicted for modern subduction systems, so they are omitted here. Likewise, the extent of subduction erosion of forearc material is debated, as is its carbon content34,42,43. Mechanical removal reflects off-scraping and diapiric flow of large rock masses44,45. Because this is chiefly evident from past subduction complexes exhumed to the surface in the geologic record, only rarely is it possible to establish where and how mechanical removal operates today. A prominent instructive exception is in the Mariana forearc46.

The removal of metamorphic carbon from the slab depends on the kindness of water. Carbonate minerals and graphite/diamond are stable on their own up to very high temperatures and pressures; in the absence of any other processes or materials, such carbon is extremely refractory and its recycling efficiency is low. However, subducted lithologies also carry hydrogen and oxygen bound in minerals, and the rising pressure and temperature drive their release to create a free aqueous fluid that is buoyant and reactive with carbon. The fluid strips carbon from these rocks, carries it along its path and deposits it downstream in rocks or melts as a response to changing conditions. Traditional thinking focused only on molecular forms of carbon in the fluid, such as CO2, and concluded that its recycling efficiency is low, except perhaps in the hottest subduction systems47,48,49. Driven by emerging experimental and field studies, recent work also considers the roles of the oxidation state, pH and other rock constituents and dissolved species, generally leading to greater recycling efficiency3,50,51,52,53,54. Although carbon solubilities may contribute to substantial redistribution, they are probably insufficient on their own to account for the removal of all carbon in every instance.

The recycling efficiency can be substantially boosted by melting, especially at the slab top. Graphitized organic matter has very low solubility in sediment melts14, and modern subduction geotherms yield temperatures that are typically too cool to trigger melting of subducted carbonate-bearing peridotite, AOC and sediment at sub-arc depths in the absence of a free water-rich fluid55,56,57. However, water-rich fluids, especially those sourced from dehydrating peridotites at sub-arc depths, can trigger melting, especially in the sedimentary layer. Melts derived from carbonate-bearing AOC and sediment can remove substantial carbon58,59. Moreover, in many subduction systems, the slab top beneath the arc is near conditions at which there is complete miscibility between sediment melt and aqueous fluid60, making continuous dissolution the operative process. Such fluids have great transport capacity, although the role of carbon in them remains poorly understood. Carbonate-rich melts can be produced by the final stages of dehydration of subducted mantle lithosphere beneath the arc front61.

Any carbon remaining after these processes may be transported in the slab or possibly in the superjacent mantle wedge3. At relatively shallow depths the overlying lithospheric mantle is static, but at a certain depth mantle wedge peridotite kinematically couples to the slab and is dragged downwards62. Thus, the downward-dragged mantle wedge represents an additional carbon sink created by the subduction process.

Much past debate focused on isolating a single mechanism as the primary pathway for carbon subtraction. Here we emphasize that it is more useful to recognize that several mechanisms can operate in parallel or in series, and that each global segment experiences them in unique proportion and extent. Considering the different carbon inputs to the global subduction system (Box 2), Box 3 describes the operation of the three distinct processing mechanisms for carbon removal. Metamorphic decarbonation is likely to operate in all subduction zones, although the extent to which this removes carbon varies with thermal structure and hydration extent. In the cold Tonga system, it is probably inefficient. The hot Cascadia subduction system stands in stark contrast, with substantial carbon loss by devolatilization/dissolution and melting of predominantly reduced sedimentary carbon. The Central American sediments, rich in biogenic carbon of both oxidized and reduced forms, may fractionate during devolatilization and melting, with preferential recycling of soluble, oxidized carbonate compared to less-soluble organic carbon. The slow-spreading crust feeding the South Sandwich system is expected to be dominated by carbonated peridotite in a young, warm slab. The degree to which mechanical diapiric transport operates today is not known; however, one setting in which it may have been important in the recent past could be the Alpine subduction system. Melting was unlikely as there was no volcanic arc. Though carbon inputs were likely high in Tethyan Mesozoic oceanic crust, the combination of diapirism and dissolution/dehydration could have driven substantial carbon recycling. These examples highlight the conspicuous expressions of individualism in Earth’s subduction zones that become obscured by global averaging.

Carbon beyond the arc: diamonds and more

Carbon that survives beyond sub-arc depths is exceptionally challenging to investigate. In the absence of a fluid phase, even trace carbonate remaining in slab lithologies can trigger melting that may be small in volume but is geochemically important. Most slab geotherms will eventually intersect conditions at which carbonated crust melts63,64, producing a magma highly enriched in carbonate known as carbonatite, which has extremely high mobility, very similar to that of water65,66. Such melts will quickly migrate into the surrounding mantle, where vastly different chemical conditions lead to reactions that transform carbon into new and more resistant forms such as diamond. The durability of diamond allows it to act as a long-term store of information about carbon mixing back into the mantle via subduction.

There are two geodynamically distinct reservoirs of subducted carbon in the mantle: the deep lithospheric keels beneath continents and the more voluminous, convecting upper mantle, transition zone and upper part of the lower mantle. Most diamonds are derived from the volumetrically minor lithosphere; relationships to subducted carbon exist but are difficult to quantify because of the complex history and great age of lithospheric diamonds. However, rare sub-lithospheric diamonds (SLD) appear to be more directly linked to deeper subduction processes extending into the transition zone and beyond. Study of these diamonds is transforming our understanding of the deep carbon cycle.

Key to the origin of SLD is the strong chemical contrast between the subducting crust and the ambient convecting mantle. This difference is not just in bulk composition but also, more importantly, in oxidation state. Whereas most crustal rocks are derived from Earth’s oxidized surface reservoirs, the mantle is relatively reducing, and becomes more so with depth. Driven chiefly by changes in the stability of iron-rich minerals with depth67,68,69,70,71, this leads to conditions that stabilize an iron-rich metallic phase and, at depths greater than about 140 km, produce diamond as the stable form of carbon.

SLD are identified by virtue of the mineral inclusions that they hold, which signal a depth greater than that of the deepest lithospheric keels (>200 km). SLD are younger than lithospheric diamonds, show highly complex growth histories and display strong chemical links to subducted crust72,73. Carbon isotope data for SLD are quite variable, extending to very low δ13C values consistent with carbon derivation from subducted sediment or altered basalt30,72,74 (Fig. 2b).

If SLD are consistently associated with subducted carbon, then how do they form and how are they transported to the surface? The mineral inclusion assemblages in SLD may show peridotite affinity, but they dominantly host eclogitic mineral assemblages75. Traditionally, inclusion ‘affinities’ are assumed to identify the host rock in which diamond grew. This view implies a narrow set of rock types with a restricted range of chemical characteristics that make it difficult to explain the wide variations observed in SLD. However, an emerging alternative model posits that SLD and their inclusion phases coprecipitate from the fluids and melts derived from subducted crust, so that diamond and its inclusions are products of crystallization in the ambient mantle during reactive flow of melt migrating away from the slab. Low degrees of interaction yield coprecipitating phases that are more similar to the source eclogite, whereas higher degrees of interaction yield an assemblage that is more representative of the ambient mantle. Such a model predicts that inclusion assemblages are chemically linked although they appear to reflect different host rocks, and that there should be straightforward elemental and isotopic patterns consistent with the evolution of conditions from the eclogite source to the ambient mantle. This model better explains the highly enriched character of inclusion minerals, which are otherwise anomalous if they are direct samples of the slab76. It also explains C–O stable isotopic systematics77.

SLD are but a small fraction of diamonds erupted in the kimberlites in which they are found, and diamond-bearing kimberlites are but a small fraction of all kimberlite magmas, which are themselves a minute but important component of global magmatism. As noted earlier, the vast majority of diamonds come from the subcontinental mantle lithosphere. Like SLD, they are best interpreted as metasomatic in origin, and some show evidence for the involvement of subducted carbon, but they differ from SLD in their great age (extending to more than 2.5 Gyr). A wide range of possible mechanisms form lithospheric diamonds, including mixing of mantle and subduction-related fluids, precipitation from metallic melts and mechanical mixing of diamonds formed in subducted slabs with mantle before capture in the lithosphere. As with the happenstance of carbon subduction, the record of subducted carbon found in diamonds is also a highly random sample of material heterogeneously distributed in space and time; it is not in control of its own fate and it depends on largely independent processes to be mined from the deep.

Overall, the total amount of carbon that exists in the form of diamond is unknown. Even if kimberlite eruptions are taken as representative of normal mantle, the overall amount of carbon in diamond is small3. It seems likely that the more common return path for carbon stored in metal carbides and diamond is as a component of magmas derived from the mantle; it is typically so chemically modified that little can be said about the processes preceding melting.

Carbon returned: volcanic gas

Explosive volcanic eruptions—common within the volcanic arcs that form above subduction zones—are driven by the dramatic exsolution of volatiles, including CO2, as magmas rise to the surface. This is due to a strong drop in solubility with decreasing pressure. Magmas can readily dissolve large quantities of carbon as carbonate in the mantle78, but magmas with more than 1 wt% CO2 will be vapour-saturated and degas throughout their entire ascent in the crust79. The standard method used to measure the concentration of magmatic volatiles is to analyse melt inclusions trapped in early-formed minerals. This approach has been successful in estimating the concentrations of H2O, S, Cl and F, but not CO2, owing to its much lower solubility at crustal pressures80. Moreover, as much as 90% of the CO2 in a melt inclusion can be exsolved to a shrinkage bubble during cooling81. Unfortunately, the literature is replete with measurements of CO2 in melt inclusions compromised by degassing or bubble formation. After bubble reconstruction, arc melt inclusions may contain thousands of parts per million of CO282,83, but there is no guarantee that these melts did not lose CO2 by degassing before inclusion formation. Thus, a major outstanding challenge is to determine the CO2 concentration of primary arc magmas (Box 4).

Given that erupted magma has lost its CO2, a more fruitful approach is to measure the lost gas. However, CO2 above the atmospheric background is difficult to measure using remote-sensing techniques. Campaign-style gas measurements show large spatial and temporal variations in CO2 fluxes in different volcanic systems including diffuse degassing from soils, springs and hydrothermal systems84,85. Diffuse degassing is generally strongest for large silicic calderas that erupt infrequently10. Thus, although satellite detection holds promise for the future86, estimating CO2 fluxes from volcanoes is difficult owing to its non-steady-state and non-point-source distribution.

A recent promising approach for the estimation of volcanic CO2 fluxes has been to make use of the ratio of CO2 to sulfur. Relatively inexpensive multicomponent gas analyser systems (such as Multi-GAS) have revolutionized our understanding of CO2/S variations at different volcanoes over their eruption cycles. In addition to their great utility in eruption forecasting87,88, multi-year records from persistently degassing volcanoes show that most gas release occurs passively during the inter-eruptive period89. Integrated CO2/S correlates in some volcanic arcs with ratios of non-volatile trace elements (for example, Sr/Nd or Ba/La) linked to subduction recycling of carbonate and slab fluids9. Arcs with low CO2/S ratio (for example, in the western Pacific) are associated with a lack of subducting sedimentary carbon, whereas those with higher CO2/S ratios are associated with subducting sedimentary carbonate (Box 2). Sometimes magmas intercept limestones in the upper plate and drive decarbonation90, a process recognized from an anomalously high CO2/S ratio (>5) that is not linked to other slab tracers89. CO2/S data have also led to CO2 flux estimates11 when coupled to volcanic SO2 fluxes measured from space7. The biggest volcanic emitters are found where carbonate-rich sediments subduct (Central America, Vanuatu and Papua New Guinea), whereas low-CO2-flux volcanoes occur where no carbonate sediments subduct (Kuriles-Aleutians; Box 2). These studies show clear signs that volcanic CO2 variations reflect varying subducted input, but longer records are needed to understand temporal variations91.

Another useful tool is C isotopic ratios in volcanic gases, which may be lower or higher than mantle values for individual volcanoes. Some gases have low 3He/4He ratios, a signal of upper-crustal contamination92; however, the full range of δ13C values is found in volcanic gases that have mantle-like 3He/4He. Low-δ13C volcanic carbon occurs at margins where low-δ13C organic carbon is subducted (for example, the Aleutians), whereas high δ13C values occur where high-δ13C carbonate is subducted (for example, Central America; Box 2). Globally, volcanic arcs with the highest C fluxes have the highest δ13C, suggesting a dominance of carbonate sources (Fig. 2c). This could reflect preferential recycling of carbonate in the subduction zone. The CO2/3He ratios of arc volcanic gases are consistent with most of the carbon (80–95%) being derived from recycled, subducted sources (Box 4). These constraints, coupled with those from the CO2/S and CO2/Nb ratios, point to primary arc magmas containing up to 1 wt% CO2 (Box 4), an order of magnitude higher than that of average mid-ocean-ridge basalt (MORB; about 1,000 ppm)93. Volcanic-arc data are still sparse, however, and portable mass spectrometers mounted on helicopters94 and field vehicles95 are providing new methods for carbon isotope analysis in remote regions. Carbon isotope measurements in springs are also revealing subducting and upper-plate sources, as well as sequestration in the biosphere96.

Conclusions

Carbon subduction is neither a steady-state nor a globally averaged process. The input fluxes to deep sea trenches are spatially and temporally heterogeneous, as are slab pressure–temperature paths, and thus every subduction zone is a different carbon recycling experiment. In this Review, we have developed three essential viewpoints:

(1) Carbon subduction is geological happenstance. Fixing carbon in oceanic peridotite or burying it in sediments is inherently difficult, as the long path lengths and high solubilities in the deep ocean work against it. Sedimentary carbon is virtually absent from some subduction zones and dominates others (Box 2). Carbonate veining is pervasive in the oceanic crust but its extent varies with crustal age (Fig. 2a). Past (Tethys subduction) or future (Atlantic subduction) carbonate inputs might be larger than those occurring today. At other times, continent collision and erosion deliver organic carbon to deep-sea fans that subduct. Such tectonic events will affect the climate system over long timescales.

(2) Sediments drive the variability in subducting carbon. If the latest flux estimates are accurate (Box 1), then carbonate sediments dominate the subducting flux of carbon by a factor of about two over oceanic-crust and peridotite input fluxes. Recycling of carbonate may explain the high C fluxes and high δ13C values observed in some volcanic arcs (Fig. 2c). Sediments also supply the greatest flux of organic carbon to the solid Earth (Box 1), and the fate of reduced and isotopically light carbon is critical to the oxidation state of Earth’s surface and mantle, and may relate to the low δ13C values of eclogitic diamonds (Fig. 2b). There may be a complementarity to C isotopes, with heavy carbon isotopes reflecting predominantly carbonate recycling beneath arcs and light carbon isotopes representing deep subduction of organic carbon as a source of diamonds.

(3) There is no single recycling efficiency for carbon. The recycling of carbon back to Earth’s surface via arc volcanism depends on the carbon-carrying capacity of fluids and melts in each subduction zone, which in turn are a function of the slab pressure–temperature paths, the pH and redox potentials, and the bulk compositions of each system (Box 3). Not only does the input vary for each subduction zone, but so also do the parameters that govern the recycling efficiency. Current global flux balances (Box 1) place carbon recycling efficiency at 25%, but the associated uncertainties are very large. Even if accurate, this number is insufficiently precise, so the global average carries questionable meaning for any particular margin, and its variation with time could be substantial. For example, carbonate recycling could be more efficient than organic carbon recycling, and sedimentary carbon could be more easily liberated than oceanic-crust carbon. Modern subduction zones offer many natural experimental settings (Box 2) to test how initial conditions, such as plate age and organic-to-inorganic ratio, affect recycling efficiency (Box 3). Carbon–helium–sulfur systematics point to 80–95% of arc volcanic carbon deriving from the subduction zone (Box 4), so the signal is large. The stakes are high for the climate system, with subducting carbon acting as a source or sink, depending on its recycling efficiency. Coordinated effort is needed on all fronts—quantifying carbon inputs, modelling chemical transfer in the subduction zone and measuring volcanic outputs and deep diamonds—to constrain the global impacts of subducting carbon.