Metamorphic devolatilization of subducted marine sediments and the transport of volatiles into the Earth's mantle

Volatiles, most notably CO2, are recycled back into the Earth's interior at subduction zones1,2. The amount of CO2 emitted from arc volcanism appears to be less than that subducted, which implies that a significant amount of CO2 either is released before reaching the depth at which arc magmas are generated or is subducted to deeper depths. Few high-pressure experimental studies3,4,5 have addressed this problem and therefore metamorphic decarbonation in subduction zones remains largely unquantified, despite its importance to arc magmatism, palaeoatmospheric CO2 concentrations and the global carbon cycle6. Here we present computed phase equilibria to quantify the evolution of CO2 and H2O through the subduction-zone metamorphism of carbonate-bearing marine sediments (which are considered to be a major source for CO2 released by arc volcanoes6). Our analysis indicates that siliceous limestones undergo negligible devolatilization under subduction-zone conditions. Along high-temperature geotherms clay-rich marls completely devolatilize before reaching the depths at which arc magmatism is generated, but along low-temperature geotherms, they undergo virtually no devolatilization. And from 80 to 180 km depth, little devolatilization occurs for all carbonate-bearing marine sediments. Infiltration of H2O-rich fluids therefore seems essential to promote subarc decarbonation of most marine sediments. In the absence of such infiltration, volatiles retained within marine sediments may explain the apparent discrepancy between subducted and volcanic volatile fluxes and represent a mechanism for return of carbon to the Earth's mantle.

Volatiles, most notably CO 2 , are recycled back into the Earth's interior at subduction zones 1,2 . The amount of CO 2 emitted from arc volcanism appears to be less than that subducted, which implies that a signi®cant amount of CO 2 either is released before reaching the depth at which arc magmas are generated or is subducted to deeper depths. Few high-pressure experimental studies 3±5 have addressed this problem and therefore metamorphic decarbonation in subduction zones remains largely unquanti®ed, despite its importance to arc magmatism, palaeoatmospheric CO 2 concentrations and the global carbon cycle 6 . Here we present computed phase equilibria to quantify the evolution of CO 2 and H 2 O through the subduction-zone metamorphism of carbonate-bearing marine sediments (which are considered to be a major source for CO 2 released by arc volcanoes 6 ). Our analysis indicates that siliceous limestones undergo negligible devolatilization under subduction-zone conditions. Along hightemperature geotherms clay-rich marls completely devolatilize before reaching the depths at which arc magmatism is generated, but along low-temperature geotherms, they undergo virtually no devolatilization. And from 80 to 180 km depth, little devolatilization occurs for all carbonate-bearing marine sediments. In®ltration of H 2 O-rich¯uids therefore seems essential to promote subarc decarbonation of most marine sediments. In the absence of such in®ltration, volatiles retained within marine sediments may explain the apparent discrepancy between subducted and volcanic volatile¯uxes and represent a mechanism for return of carbon to the Earth's mantle.
A premise of our work is that realistic modelling of metamorphic devolatilization of subducted lithologies is only possible on the basis of phase equilibria in chemical systems closely approximating actual bulk compositions. Our studies on metamorphic devolatilization of the other two main carbonate-bearing lithologies involved in subuction zones (ophicarbonates and metabasalts) are considered elsewhere 2,7 . Carbonate is abundant in two main pelagic marine sediment lithologies 8 : (1) siliceous limestones and (2)  For each marine sediment bulk composition, the corresponding phase diagram section ( Fig. 1) was computed as a function of pressure (P) and temperature (T) by free-energy minimization 9 . The thermodynamic database of ref. 10 was used for the properties of all end-member species, and mineral solutions were modelled as described elsewhere 9 . Thermodynamic data for H 2 O, CO 2 and their mixtures were computed from the equation of state given in ref. 11. To track metamorphic devolatilization along the top of subducted slabs, we adopted the geotherms 12 for the subduction zones of northwestern and southeastern Japan (Fig. 1). These geotherms are reasonable approximations for the respective extremal lowtemperature and high-temperature geotherms for subduction zones (S. M. Peacock, personal communication).
Because of compositional degrees of freedom in the crystalline and¯uid phases, the phase diagram sections are dominated by multivariant phase ®elds. Consequently, both mineral modes and compositions vary continuously along geotherms (Fig. 2). Phase relations along geotherms up to a pressure P < 3 GPa change signi®cantly ( Fig. 1) because of intersection with numerous phase ®eld boundaries. However, at P . 3 GPa, the geotherms are subparallel to the phase ®eld boundaries (Fig. 1); consequently, little reaction occurs along geotherms at P . 3 GPa. The differences between these regimes are illustrated in Fig. 2. Accordingly, sig-ni®cant changes in the mineralogy and mineral proportions occur up to ,800 8C (P < 3 GPa ) whereas there is comparatively little variation above ,800 8C.
The¯uid composition (Fig. 2) is controlled by multivariant equilibria involving carbonates and hydrous phases. The rise in the mole fraction of CO 2 , X CO 2 , up to ,750 8C correlates with consumption of carbonates ( Fig. 2), whereas the diminution in X CO 2 above ,750 8C occurs because of aragonite production. To track loss of volatiles we computed the percentage (by weight; wt%) of H 2 O and CO 2 for carbonate-bearing marine sediments as a function of pressure and temperature (Fig. 3). In the lower-pressure half of Fig. 3, the negative P±T slopes of the wt% H 2 O isopleths re¯ect negative slopes of phase ®eld boundaries (Fig. 1). In contrast to isopleths with negative slopes at lower pressures, isopleths are subparallel to geotherms at P . 2±3 GPa (Fig. 3).
Siliceous limestones release about 1 wt% CO 2 and 1 wt% H 2 O along the high-temperature geotherm (Fig. 3c, d). Because less CO 2 and H 2 O are released along lower-temperature geotherms, most of the volatile content of siliceous limestones is retained to depths of 180 km and thus such lithologies would undergo little devolatilization upon subduction. This conclusion is compatible with the existence of ultrahigh-pressure marbles 13 .
In contrast with siliceous limestones, H 2 O-rich lithologies with low initial carbonate contents (that is, clay-rich marls) are predicted to undergo considerably more devolatilization (Fig. 3b). Along the high-temperature geotherm, all of the initial CO 2 (4 wt%) and most of the initial H 2 O (10±11 wt%) is released by 90 km depth (that is, forearcs). For geotherms in the lower-temperature half of the area bounded by the limiting geotherms (Fig. 3b), relatively little CO 2 and H 2 O would be released. For various geotherms in the highertemperature half of the area bounded by the limiting geotherms (Fig. 3b), there are signi®cant differences in the amount of devolatilization.
For the carbonate-bearing protoliths considered here, the geotherms at 80±180 km are subparallel to the H 2 O and CO 2 isopleths (Fig. 3); thus, little or no devolatilization is expected. Consequently, for closed-system behaviour, subducted carbonatebearing marine sediments would not provide a source of volatiles for arc magmatism. However, decarbonation of marine sediments at these depths may be driven by in®ltration of H 2 O-rich¯uids originating from intercalated hydrous pelagic or terrigenous sediments, and/or metabasalts in the subjacent slab. Computed 7 and experimentally determined 14 high-pressure phase equilibria imply that signi®cant proportions of the initial H 2 O in subducted oceanic metabasalts are released under forearcs and subarcs. In®ltration of the evolved¯uid into the overlying subducted sediments would induce decarbonation. But because there are no major dehydratioǹ pulses' in subducted metabasalts under volcanic arcs 7,14 , no corresponding pervasive in®ltration of water from dehydrating meta-       Figure 3 Weight percentages of CO 2 and H 2 O for selected marine sediment bulk rock compositions (see Fig. 1). a, Gloss; b, Antilles; c Marianas; d, Vanuatu. Heavy curved lines are limiting geotherms (see Fig. 1). Values of the initial wt% CO 2 and H 2 O are given in the insets (from ref. 8). The CO 2 and H 2 O contents of the¯uid phase can be determined by subtracting the data in these diagrams from the initial volatile contents of the protoliths.

MARIANAS
basalts is expected in subarcs. Barring extensive in®ltration of externally derived¯uids, our study implies marked devolatilization under forearcs (for clay-rich marls with high-temperature geotherms) or retention of H 2 O and CO 2 to depths well beyond subarcs (for siliceous limestones in all geotherms and clay-rich marls with low-temperature geotherms). Accordingly, most of the initial CO 2 and H 2 O in subducted marine sediments will not be released beneath volcanic arcs. This inference is consistent with both the de®ciency in the amount of CO 2 released from arc volcanoes compared to the amount of CO 2 contained within subducted carbonates (Table 1) and with the imbalance between subducted versus expelled H 2 O (ref. 1).
Our equilibrium analysis implicitly assumes that there is no signi®cant kinetic overstepping and metastability of metamorphic reactions. Although signi®cant disequilibrium has been suggested for the transformation of anhydrous oceanic basalts and gabbros to eclogites 15 , the catalytic effect of H 2 O (ref. 15) implies that equilibrium is more likely in dehydrating systems such as subducted sediments.
Melting is an alternative mechanism for release of volatiles from subducted sediment. Recent experiments using marine red clay 16 suggest that sediment melting does not occur for the geotherms that we consider here. However, because metastable starting materials (for example, red clay) are unsuitable models for subduction-zone metamorphism and melting, con®rmation of this conclusion requires experiments with more realistic initial mineral assemblages. Dissolution of minerals in supercritical¯uids remains a possible, albeit largely unquanti®ed, alternative mechanism for devolatilization.
As shown in Fig. 2,¯uids produced by metamorphism of subducted marine sediments are H 2 O-rich. Consequently, expulsion of such¯uids to the overlying mantle wedge would not substantially affect the P±T conditions of melting (solidus) of the mantle wedge compared to those expected in the presence of a pure H 2 O¯uid.
Devolatilization of subducted sediment could contribute to seismicity along the tops of subducted slabs. The continuous nature of devolatilization is compatible with the spread of earthquake hypocentres along individual subduction zones 17 . However, correlation of slab seismicity with metamorphic devolatilization of subducted sediments needs to consider the marked differences in devolatilization for different bulk compositions and geotherms. Spring temperatures in temperate regions have increased over the past 20 years 1 , and many organisms have responded to this increase by advancing the date of their growth and reproduction 2±7 . Here we show that adaptation to climate change in a longdistance migrant is constrained by the timing of its migratory journey. For long-distance migrants climate change may advance the phenology of their breeding areas, but the timing of some species' spring migration relies on endogenous rhythms that are not affected by climate change 8 . Thus, the spring migration of these species will not advance even though they need to arrive earlier on their breeding grounds to breed at the appropriate time.

M
We show that the migratory pied¯ycatcher Ficedula hypoleuca has advanced its laying date over the past 20 years. This temporal shift has been insuf®cient, however, as indicated by increased selection for earlier breeding over the same period. The shift is hampered by its spring arrival date, which has not advanced. Some of the numerous long-distance migrants will suffer from