Pervasive subduction zone devolatilization recycles CO₂ into the forearc

The study addresses a central question in global geochemical cycling: what fraction of CO2 subducted with oceanic lithosphere is released at shallow depths versus transported into the deep mantle? Prior models disagree widely, from nearly complete decarbonation to minimal carbon loss at forearc-subarc depths. The key uncertainty is the nature and extent of aqueous fluid infiltration in subducting slabs, which strongly lowers carbonate breakdown temperatures and enhances decarbonation. Owing to sparse direct observations of fluid-rock interaction in exhumed slabs, predicted deep carbon subduction fluxes span many orders of magnitude, leaving unresolved whether the mantle is gaining or losing carbon over geologic time. The authors aim to provide field-based constraints on fluid conditions and decarbonation efficiency using the exceptionally preserved Cycladic Blueschist Unit (CBU) of Syros and Tinos, Greece, integrating petrography, isotopes, bulk chemistry, and thermodynamic modeling to quantify CO2 loss at forearc depths and its implications for mantle carbon balance.

Previous thermodynamic models of subducted sediments and oceanic crust diverge largely due to assumptions about fluid mobility. Closed-system models predict water-poor evolution and limited decarbonation, whereas open-system models with fluid infiltration predict water-rich conditions and enhanced carbonate breakdown. The stability of carbonate is highly sensitive to the presence of H2O, with reaction temperatures depressed by hundreds of kelvin under water-bearing conditions. Global estimates of deeply subducted carbon vary from ~0.0001 to 52 Mt C/yr, reflecting the lack of observational constraints. Studies on subduction thermal structures and devolatilization pathways highlight that fluid sources such as dehydrating serpentinites and metasediments can supply large volumes of water to adjacent lithologies. Isotope systematics of metamorphic carbonates and evidence for channelized fluid flow in subduction settings further motivate evaluating open-system behavior. The CBU has been widely used to derive inferences about subduction processes, including dehydration pulses, reactive flow, and slab-mantle interactions, making it a suitable natural laboratory to test decarbonation models under well-constrained P-T conditions.

• Field sampling and petrology: >600 carbonate-bearing hand samples collected from Syros and Tinos (Cycladic Blueschist Unit), spanning four protolith groups: carbonate-bearing siliciclastic rocks, limestones, heavily altered basaltic volcanics, and more lightly altered basaltic oceanic crust. Of these, 211 were analyzed for carbon and oxygen isotopes; 56 representative samples underwent detailed major, minor, and trace element analyses and thermodynamic modeling. Samples were selected to minimize retrogression.
• Chemical analyses: Major elements measured by XRF on fused discs; minor elements by XRF on pressed pellets (SGS Labs). Loss-on-ignition (LOI) corrected for Fe3+/Fetotal (0–0.15) from modal mineralogy. Water contents (0.01–3.8 wt%) constrained by detailed thin-section point counting for key representatives of each lithology; remaining LOI assigned to CO2 (1.5–43.8 wt%). Linear regressions within lithologic groups were used to estimate H2O for non-point-counted samples.
• Stable isotopes: Carbon and oxygen isotopes of carbonate minerals measured at Yale (Thermo MAT 253 KIEL IV and Thermo DeltaPlus XP with GasBench II); precision better than ±0.1‰. Results reported as δ13C (VPDB) and δ18O (VSMOW), representing bulk carbonate per sample.
• Thermodynamic modeling: Theriak-Domino with an internally consistent dataset and activity-composition models for relevant minerals; full COH fluid mixing implemented with appropriate equations of state for H2O and CO2. Calculations include P–T pseudosections and pressure–activity of CO2 diagrams.
• Back-projection of initial volatile content: For each sample, the precursor CO2 content was estimated by calculating equilibrium mineralogy at low-grade conditions (310 °C, 0.8–1.0 GPa) with an excess COH fluid across a wide range of XCO2 (0.0001–1). The initial composition was taken as the composition of the solid minerals at these conditions. Sensitivity to P, T, and XCO2 is generally minimal; for 10 sensitive samples, the minimum plausible initial CO2 was selected to yield conservative minimum ΔCO2 values. Assumptions: (1) equilibrium with a CO2-bearing fluid at low grade; (2) carbonate present at peak P–T implies equal or greater carbonate pre-metamorphism; (3) no major element metasomatism other than volatile loss (stoichiometric carbonate dissolution would increase true CO2 loss).
• Forward modeling of peak conditions and fluid infiltration: Two end-member scenarios were modeled at peak conditions: (1) fully closed system at 525 °C, 1.5 GPa; (2) open system with an excess COH fluid of fixed XCO2 determined by matching observed mineral assemblages to pressure–aCO2 constraints at peak P–T, yielding the equilibrium fluid composition recorded by each sample. Maximum potential decarbonation was estimated by modeling peak mineralogy in the presence of a very water-rich fluid (XCO2 = 0.0001) and comparing resulting carbonate to the back-projected initial.
• P–T paths: Two linear geotherms considered for cumulative degassing: Path A through 525 °C, 1.5 GPa to 650 °C, 1.8 GPa; Path B through 550 °C, 2.0 GPa to 650 °C, 2.4 GPa.
• Flux construction: A global-average subducted crustal stack was assembled (sediments + altered oceanic crust), using lithologic proportions, densities (2700 kg/m3 for sediments; 3000 kg/m3 for mafic), and lithology-specific average initial CO2 contents and observed ΔCO2: limestone C_i=0.39, ΔCO2=0.032; carbonate-bearing siliciclastics C_i=0.19, ΔCO2=0.51; heavily altered volcanics C_i=0.16, ΔCO2=0.69; altered basaltic crust C_i=0.06, ΔCO2=0.46. Weighted averaging yields total observed CO2 loss fraction ≈0.41. Forward-modeled additional loss to 650 °C was added to obtain total forearc release.

• Widespread aqueous fluid infiltration: Equilibrium fluid conditions inferred from mineral assemblages are consistently water-rich, with average aCO2 ≈ 0.06 ± 0.07 (2σ) corresponding to XCO2 ≈ 0.006 at subduction conditions. Closed-system models predict significantly higher aCO2 and cannot reproduce the observed water-rich conditions, requiring open-system infiltration.
• Isotopic evidence: Carbonate δ13C ranges from −4.9‰ to +3.7‰ (VPDB) and δ18O from +10.1‰ to +30.8‰ (VSMOW). Over 70% of samples are lighter in C and O than typical seafloor carbonates (δ13C ~ +1 to +4‰; δ18O ~ +26 to +31‰), consistent with external fluid infiltration, decarbonation-driven fractionation, and/or reduced carbon presence.
• Decarbonation magnitudes: Observed ΔCO2 across rocks ranges from 0% to 91% by mass. Very pure marbles (~40 wt% CO2) show little loss; carbonate-poor to intermediate, especially mixed carbonate–siliciclastic rocks and altered volcanics, record extensive decarbonation. The mass of CO2 released peaks for intermediate-composition rocks; pure limestones and carbonate-poor basalts contribute less per unit mass.
• Global forearc CO2 release: Using lithologic weighting representative of subducted sediments + altered oceanic crust, observed data indicate ~40–41% of subducted CO2 is released by peak forearc conditions. Forward modeling to 650 °C predicts an additional ~25% release, totaling ~65% CO2 loss within the forearc. Most decarbonation occurs in a sharp pulse near 500–550 °C.
• Residual deep flux: Approximately ~35% of initial slab CO2 remains to subarc depths; the model predicts <~2 Tmol CO2/yr is subducted past the forearc globally, comparable to arc volcanic CO2 emissions. Decarbonation reactions alone cannot liberate this CO2 at subarc conditions; other processes (e.g., sediment melting, congruent dissolution, diapirism) are implicated.
• Forearc storage and atmospheric flux: The calculated forearc CO2 release exceeds many estimates of arc volcanic CO2 output, implying substantial CO2 storage in forearc lithospheric mantle (carbonate precipitation) and/or diffuse degassing, consistent with independent observations of forearc carbon sinks.
• Robustness to thermal structure: Similar total degrees of decarbonation are predicted along different P–T paths, with colder geotherms primarily delaying the depth of release rather than reducing total forearc loss, provided water-bearing fluids are available.

The results provide a direct, field-calibrated constraint on subduction zone decarbonation, demonstrating that open-system infiltration of water-rich fluids is pervasive and critical to facilitating carbonate breakdown at forearc depths. This addresses the longstanding debate by showing that a large fraction (~40% observed; ~65% including modeled to 650 °C) of slab CO2 is released before reaching subarc depths, ruling out massive deep subduction of most carbon under typical conditions. The pronounced lithologic control on ΔCO2 implies that carbon fluxes off the slab are highly sensitive to the proportions of sediments, altered volcanics, and carbonate rocks entering trenches, leading to spatial and temporal variability among subduction zones. While the overall efficiency of forearc decarbonation appears robust across plausible thermal structures, the timing and depth of peak release depend on the geotherm. The inferred forearc CO2 output exceeding arc volcanic emissions suggests significant storage within the overriding plate, aligning with emerging evidence for carbonated forearc mantle and surface forearc sinks. The relatively small residual slab CO2 flux to subarc depths implies that additional mechanisms beyond simple decarbonation (e.g., melting, dissolution, diapirism, or forearc mantle transport) must operate to supply CO2 to arcs where slab signatures are detected. The modern imbalance between mantle outgassing (~5–7 Tmol/yr) and subduction of CO2 past the forearc (<~2 Tmol/yr) further implies that Earth's convecting mantle may be undergoing long-term carbon depletion.

This study integrates comprehensive field observations, isotopic data, whole-rock chemistry, and thermodynamic modeling from the Cycladic Blueschist Unit to quantify subduction zone decarbonation. It demonstrates that water-rich fluid infiltration is ubiquitous and indispensable for driving metamorphic decarbonation, leading to release of ~40% of slab CO2 by peak forearc conditions and ~65% by 650 °C, with a sharp release peak near 500–550 °C. Only ~35% of slab CO2 is likely retained to subarc depths, indicating that subduction is generally inefficient at transporting CO2 deep into the mantle. The findings imply substantial forearc carbon storage and a potential long-term net loss of carbon from the convecting mantle. Future research should: (1) quantify the role and extent of congruent carbonate dissolution at greater depths; (2) better constrain fluid sources, pathways, and channelization in different subduction regimes; (3) assess variability in lithologic inputs and thermal structures globally to refine carbon flux budgets; and (4) directly evaluate the fate of forearc-released CO2 (storage vs. degassing) and its linkage to arc volcanic outputs.

• Representativeness: The Cycladic Blueschist Unit likely reflects relatively warm subduction; although modeled P–T paths suggest similar total decarbonation under colder regimes, timing/depth may differ. Exhumed complexes may not capture the full spectrum of subduction conditions.
• Process coverage: The approach does not quantify stoichiometric carbonate dissolution, melting, or diapirism, which could add to CO2 release at greater depths; thus ΔCO2 estimates are conservative minima.
• Modeling assumptions: Back-projection assumes equilibrium with a CO2-bearing fluid at low grade, equal or greater pre-metamorphic carbonate than at peak, and no major-element metasomatism beyond volatile loss. Deviations could alter initial CO2 estimates.
• Fluid composition constraints: Two samples had unconstrained equilibrium fluid activities due to simple mineralogy. More broadly, inferring XCO2 from assemblages depends on thermodynamic models and activity-composition relations.
• Isotope interpretations: Low δ13C and δ18O values can reflect multiple processes (external fluid infiltration, decarbonation fractionation, reduced carbon), introducing ambiguity without mineral-scale or in-situ analyses.