Contrasting slab decarbonation patterns control deep carbon cycling efficiency in hot vs. cold subduction zones

Abstract

Subducting slabs play a crucial role in deep Earth carbon cycling, yet the carbon-cycling efficiency of different subduction zones remains debated. Here, we systematically compare decarbonation characteristics in the extremely cold Tonga and hot Colombia subduction zones using observation-constrained thermo-petrological modeling, integrated with volcanic degassing and arc geochemistry. Two contrasting slab decarbonation patterns are identified: efficient “focused” decarbonation in hot subduction zones versus sluggish “diffuse” decarbonation in cold subduction zones. Hot Colombia exhibits a high decarbonation efficiency >15%, dominated by metamorphic carbon release within a narrow depth range of ∼75–115 km and supplemented by partial melting-induced decarbonation. In contrast, cold Tonga shows a much lower decarbonation efficiency of ∼1% by dissolution-dominated carbon removal over a broad depth range of ∼75–310 km. These findings suggest a contrasting efficiency of different subducting slabs for deep carbon cycling and volcanic degassing in present-day and ancient subduction systems.

Focused versus diffuse decarbonation styles are predicted for the hot and cold subduction zones, respectively.

INTRODUCTION

Deep carbon cycling in subduction zones is a fundamental process that may regulate Earth’s long-term climate (1) and therefore influences its habitability (2). Slab decarbonation governs the partitioning of subducted carbon between transport into the deep mantle and surface carbon release through arc volcanism, thereby forming a critical link between Earth’s deep and surface carbon reservoirs. Among the factors influencing slab decarbonation, the thermal structure of subduction zones is widely recognized as a primary control (3, 4). Circum-Pacific subduction zones, the most prominent seismic and volcanic belts on the Earth, exhibit contrasting thermal structures between the western and eastern Pacific, as identified by their slab thermal parameter [Φ = slab age × convergence rate × sin (slab dip) in kilometers (5, 6); data S1]. Western Pacific subduction zones are generally cold, with a mean Φ of 5685 km (Fig. 1A), whereas most eastern Pacific subduction zones are predominantly hot, with a mean Φ of 780 km (Fig. 1B). Such a first-order difference in thermal structure may control slab decarbonation around the Pacific Ring of Fire, potentially influencing volcanic CO₂ degassing (7) and regional environmental conditions (8). In particular, volcanic CO₂ outflux from hot subduction zones in the eastern Pacific (∼17 Mt. year⁻¹) is about two to three times that from cold subduction zones in the western Pacific (∼7 Mt. year⁻¹) [(9); data S1]. Although these volcanic CO₂ outfluxes may reflect contributions from both slab devolatilization (9) and the remobilization of crustal carbonates (10, 11), slab decarbonation is generally considered to represent a major CO₂ source in subduction zone settings (9). However, slab decarbonation behavior and its consequence for volcanic degassing efficiency under these contrasting thermal conditions remain unclear.

Fig. 1. Thermal states of Pacific subduction zones, and geological settings of the cold Tonga and hot Colombia subduction zones.

(A and B) Thermal parameters [Φ = slab age × convergence rate × sin (slab dip) in kilometers] of representative subduction zones in western Pacific (A) and eastern Pacific (B) (with the cut-off value of 8000 km). The thermal parameters are derived from (6); trench lines following (116). (C) Geological setting of Tonga subduction zone, where the Pacific Plate subducts westward beneath the Indo-Australian Plate. (D) Geological setting of Colombia subduction zone, where the Nazca Plate subducts eastward beneath the South American Plate. Light orange boxes show the study areas. AA’ and BB’ are cross sections of Tonga and Colombia subduction zone, respectively. Gray dashed lines mark the slab-surface depth contours in kilometers (35). Convergence rates and seafloor ages are marked by white boxes. Convergence rate is calculated as the sum of subducting plate velocity and trench retreat velocity in the direction perpendicular to the trench, following the definition in (13). Volcano information is from Global Volcanism Program website (https://volcano.si.edu/). HTHH, Hunga Tonga-Hunga Ha’apai; Ruiz, Nevado del Ruiz; Huila, Nevado del Huila.

In the circum-Pacific subduction system, the Tonga subduction zone in western Pacific and the Colombia subduction zone in eastern Pacific represent end-member cases of extremely cold and hot thermal conditions, respectively. Both subduction zones are characterized by intense volcanic eruptions and persistent degassing, providing geodynamically advantageous settings for dissecting the characteristics of slab decarbonation. Tonga is an ocean-ocean subduction zone where the old Pacific Plate [101 Ma; (12)] subducts rapidly [16 cm year⁻¹; (13)] (Fig. 1C). Colombia is an ocean-continent subduction zone characterized by slow subduction [4 cm year⁻¹; (14)] of the young Nazca Plate [13 Ma; (12)] (Fig. 1D). The 2022 Hunga Tonga-Hunga Ha’api (HTHH) eruption [Volcanic Explosivity Index = 6; (15)] in the active Tonga intra-oceanic arc produced a >58-km-high plume penetrating the mesosphere (16), triggering global atmospheric shock waves (17), and further changing stratospheric composition (18) to induce rapid ozone depletion (19). Meanwhile, the Colombia arc, one of the most active continental arcs on the Earth, hosts volcanoes with among the highest annual CO₂ emissions in the Andes, exemplified by Nevado del Huila and Galeras, which are reported to persistently release an average of 862 and 495 tons of CO₂ per day, respectively (7, 20). Moreover, the influence of crustal carbonate remobilization on volcanic CO₂ degassing in these two subduction zones is expected to be minimal (11), making them particularly suitable for isolating the effect of slab thermal structure on slab-derived carbon release. Therefore, dissecting the decarbonation characteristics in these two representative subduction zones is crucial not only for assessing the contrasting deep carbon cycling in the western and eastern Pacific, but also for understanding volcanic-climatic feedbacks and associated regional geohazards.

Although previous studies qualitatively revealed that hot thermal conditions can enhance slab decarbonation (3, 4, 21, 22) and quantitatively estimated the global slab decarbonation outflux (9, 23, 24), predicted outfluxes in Tonga and Colombia vary substantially and generally exceed observed volcanic outfluxes (see Results and Discussion). Thus, quantitative effects of extreme thermal conditions on slab decarbonation in active present-day subduction zones remain debated. One challenge arises from the fact that slab decarbonation is governed by multiple mechanisms and is strongly coupled with slab dehydration (4, 21, 25). Although several decarbonation mechanisms have been proposed, including metamorphic decarbonation (3, 26), carbonate dissolution (27, 28), partial melting of carbonated rocks (29), and redox reactions of carbonates (30), previous thermo-petrological models considered these mechanisms either in isolation (9, 23) or only in partial combination (24), leaving unresolved questions about the controlling mechanisms under extreme thermal conditions. Specifically, it remains unclear whether CO₂ release in cold Tonga versus hot Colombia subduction zones is dominated by metamorphic decarbonation (i.e., decomposition of carbonate minerals) or by carbonate dissolution (i.e., fluid-mediated dissolution and transport of carbonates). In addition, it remains uncertain whether partial melting of carbonated rocks, particularly sediments, contributes substantially to slab CO₂ release, potentially via in-situ slab melting (31) or diapir-induced melting (32). Another challenge for accurately quantifying slab decarbonation lies in the fact that each subduction zone is inherently unique and exhibits substantial variability in its carbon inputs and outputs (33, 34). Addressing this geologic complexity requires site-specific modeling that integrates realistic constraints on both carbon inputs and outputs, such as bulk-rock compositions and thicknesses of carbonate-bearing rocks within the slab, convergence rate and slab age, volcanic CO₂ emissions and arc geochemistry.

Here, focusing on the Tonga and Colombia subduction zones, we assimilate observational data into open-system thermo-petrological models to quantitatively evaluate both metamorphic decarbonation and carbonate dissolution under contrasting thermal conditions, and to further analyze the effect of partial melting on slab decarbonation. We also compare predicted carbon outfluxes with observed volcanic emissions, and evaluate decarbonation sources and mechanisms based on geochemical data of volcanic rocks. Our results reveal two contrasting decarbonation patterns under extreme thermal conditions. Hot Colombia exhibits an efficient “focused” decarbonation pattern, characterized by intense carbon release within a narrow depth range, whereas cold Tonga exhibits a sluggish “diffuse” decarbonation pattern with lower decarbonation flux and efficiency over a much broader depth range. These contrasting decarbonation patterns play critical roles in deep carbon cycling and arc degassing in both present-day and ancient subduction-zone systems, and could further influence variations of the global carbon cycle and climate evolution from the Archean to present day.

RESULTS AND DISCUSSION

Thermally-controlled contrasting metamorphic decarbonation and carbonate dissolution

To realistically evaluate decarbonation characteristics in the Tonga and Colombia subduction zones, we construct thermo-petrological models constrained by their observed geometrical, geodynamical and geological parameters. Two trench-perpendicular cross-sections, intersecting active volcanoes (i.e., profiles AA’ and BB’ in Fig. 1), are selected for subsequent comparisons with the observed volcanic degassing (supplementary text S1). The geometry of each subduction zone is constrained by the Slab2 (35) and CRUST1.0 (36) databases. In both regions, four carbonate-bearing lithologies are present in the subducting slab, i.e., sediment, basalt, gabbro, and serpentinite. These lithologies are further subdivided into eight compositional layers due to their variations in bulk-rock compositions, particularly volatile contents. The thicknesses of these layers are mainly constrained by geophysical and oceanic-drilling data, and the bulk-rock compositions are derived from oceanic-drilling and dredge samples (table S1; supplementary text S2 to S3; data S2).

We use a coupled thermo-petrological model to simulate dehydration and decarbonation in an open system by integrating pre-calculated thermodynamic look-up tables into the subduction thermal model (fig. S1). The thermal structure is calculated by the finite element code PGCtherm2D (37). The mineral-bound H₂O and CO₂ capacity and carbonate mineral abundance of the compositional layers are calculated using Perple_X software (38). Aqueous solubilities of carbonate minerals are constrained by the Deep Earth Water (DEW) model (39) and experimental studies (40). Detailed modeling methods and limitations are described in the Materials and Methods section, with formulations and numerical procedures provided in supplementary text S4.

Distinct steady-state thermal structures

Tonga and Colombia subduction zones exhibit contrasting thermal structures, with the Colombia slab being substantially hotter than the Tonga slab (Fig. 2, A and B). To quantify their thermal contrast, pressure-temperature (P-T) paths along slab surface, slab Moho, and the bottom of sub-crustal serpentinite layer are extracted from the numerical models (Fig. 2, C and D). Above the maximum depth of decoupling [MDD = 75 km; (41)], the slab-surface temperature difference between Tonga and Colombia remains below 80°C. Below the MDD, the slabs undergo strong near-isobaric heating induced by slab-driven corner flow in the mantle wedge in both subduction zones, which causes a transition from low thermal gradients in the forearc to high thermal gradients in the sub-arc atop subducting slabs. In Colombia, slower subduction of a young plate enhances slab heating, with temperatures reaching 983°C at 330 km depth, nearly 200°C higher than the slab-surface temperature of 796°C reached in Tonga at the same depth. Within the slab interior, the temperatures along the Moho and the bottom of sub-crustal serpentinite layer in Tonga are systematically cooler than those in Colombia (Fig. 2, C and D), primarily due to the older slab in Tonga and its reduced heating from the mantle wedge. Specifically, at the same depth of 330 km, the slab Moho in Tonga is 334°C cooler than that in Colombia, and the bottom of serpentinite layer is nearly 434°C cooler (Fig. 2, C and D).

Fig. 2. Predicted steady-state thermal states of the Tonga and Colombia subduction zones.

(A and B) Thermal structures of Tonga (A) and Colombia (B) subduction zones. Gray lines represent isotherms. The maximum depth of decoupling (MDD) between the slab surface and the overlying mantle wedge is set at 75 km in the model (41), separating the stagnant mantle wedge (“cold nose”) from the convective mantle wedge. White dashed lines mark the slab surface, the overriding-plate Moho and the “cold nose.” (C and D) Representative pressure-temperature (P-T) paths along slab surface, slab Moho, and the bottom of sub-crustal serpentinite layer in the Tonga (C) and Colombia (D) subducting slabs. In the model, the Moho depth is 5.3 km in Tonga slab, underlain by a 24 km-thick sub-crustal serpentinite layer. In Colombia, the Moho depth is 6.7 km, underlain by a 5 km-thick sub-crustal serpentinite layer (see table S1 and supplementary text S2). The purple bands show the region of rapid heating near the MDD.

Contrasting dehydration characteristics

The thermal contrasts directly control the distinct dehydration behaviors, including dehydration depths and magnitude of slab water outflux (42–44). In the cold Tonga slab, the low T-P ratio along the subduction direction results in dehydration across a broad pressure range (∼2.3–9.7 GPa; fig. S2). Consequently, major dehydration begins near the 75 km depth of the MDD and extends to ∼310 km depth, with sediment, basalt, and gabbro dehydrating sequentially from shallow to greater depths (Fig. 3, A and C). Serpentinite remains stable under the cold slab geotherm, as temperatures are insufficient to break down its hydrous phases (fig. S2). The total dehydration flux of subducting Tonga slab is 46.75 Tg Myr⁻¹ m⁻¹ (Table 1). In contrast, the overall hotter Colombia slab experiences intense shallower dehydration within a narrow pressure range (∼2.3–3.8 GPa; fig. S2). Major dehydration begins at the MDD and is completed by ∼115 km depth (Fig. 3, B and D). Within this narrow depth range, sediment, basalt, gabbro, and serpentinite all undergo intense dehydration. The total dehydration flux in Colombia is 29.24 Tg Myr⁻¹ m⁻¹ (Table 1).

Fig. 3. Dehydration evolution within the Tonga and Colombia subducting slabs.

(A and B) Spatial distributions of mineral-bound and released water within Tonga (A) and Colombia (B) slabs. [(Aa) and (Ba)] Residual mineral-bound H₂O content within the slab. [(Ab) and (Bb)] Mass of free aqueous fluid released via metamorphic reactions. [(Ac) and (Bc)] Total H₂O content in rock system after fluid infiltration, defined as the sum of residual mineral-bound water and free aqueous fluid from the underlying lithologies. Red triangles represent sub-arc depth. White dashed lines represent lithological boundaries among sediment, basalt, gabbro and serpentinite layers. Slab-surface depth is defined as the vertical distance from sea level to the surface of subducting slab; in-slab depth is the perpendicular distance from the slab surface into the slab interior (see fig. S3). (C and D) Cumulative dehydration flux as a function of slab-surface depth in Tonga (C) and Colombia (D), with the corresponding pressure scale shown on the bottom axis. Blue and red regions mark the main dehydration depths of Tonga and Colombia slabs, respectively.

Table 1. Thickness, initial H₂O and CO₂ contents, and modeled H₂O and CO₂ fluxes for major lithological units in the Tonga and Colombia subduction zones.

Subduction zone	Lithology	Compositional layer	Thickness (km)*	Initial content (wt.%)†		Influx (Tg Myr−1 m−1)‡		Outflux (Tg Myr−1 m−1)‡
				H2O	CO2	H2O	CO2	H2O	CO2
Tonga	Sediment	Sediment	0.10	8.38	0.51	4.42	0.27	3.11	0.12
	Basalt	Upper volcanic zone	0.20	6.25	1.90	18.30	3.90	18.29	0.33
		Lower volcanic zone	0.20	3.21	1.19
		Transition zone	0.10	2.37	0.49
		Sheeted dike	0.80	1.67	0.09
	Gabbro	Shallow gabbro	1.65	1.26	0.16	25.95	2.70	25.35	0.06
		Lower gabbro	2.25	1.26	0.11
	Serpentinite	Serpentinite	24.00	2.00	0.25	253.44	31.68	0.00	0.00
	Total		29.30	/	/	302.11	38.55	46.75	0.51
Colombia	Sediment	Sediment	0.25	5.17	22.49	1.75	7.61	0.90	0.26
	Basalt	Upper volcanic zone	0.45	4.14	0.36	8.84	0.72	8.84	0.72
		Lower volcanic zone	0.45	2.31	0.14
		Transition zone	0.30	2.37	0.49
		Sheeted dike	1.75	1.67	0.09
	Gabbro	Shallow gabbro	1.45	1.26	0.16	5.97	0.62	5.97	0.56
		Lower gabbro	2.05	1.26	0.11
	Serpentinite	Serpentinite	5.00	2.00	0.25	13.53	1.69	13.53	0.04
	Total		11.70	/	/	30.09	10.63	29.24	1.57

^*The thicknesses of subducting compositional layers within slabs are mainly constrained by multiple geophysical methods and oceanic drilling data (see supplementary text S2 for details).

^†Bulk-rock compositions of slab compositional layers are constrained by ocean-drilling and dredge samples, normalized to 100 wt.% (see supplementary text S3 and table S1 for details).

^‡Influx and outflux calculations are described in supplementary text S5.

Contrasting decarbonation characteristics

Based on the coupled water-carbon thermo-petrological model, the decarbonation depths, fluxes, efficiencies, and controlling mechanisms under these two extreme thermal conditions are evaluated (Fig. 4; table S2). Metamorphic decarbonation extends from the MDD to ∼130 km depth in Tonga (Fig. 4, Ab and Ca), whereas it is localized between the MDD and ∼95 km depth in Colombia (Fig. 4, Bb and Da). Within these depth ranges, rapid slab heating and intense dehydration pulses promote metamorphic decarbonation (fig. S4). This is consistent with previous studies (22, 45), which show that fluid infiltration during slab dehydration lowers the onset temperature of metamorphic decarbonation. Carbonate dissolution is spatially coincident with the dehydration depth ranges (Fig. 3) and the stability fields of carbonates (fig. S5). In Tonga, carbonate dissolution extends from the MDD to ∼310 km depth (Fig. 4, Ac and Cb), whereas it is restricted between the MDD and the much shallower depth of ∼115 km in Colombia (Fig. 4, Bc and Db) where serpentinite dehydration is complete. Furthermore, the total decarbonation flux in Colombia reaches 1.57 Tg Myr⁻¹ m⁻¹, approximately three times higher than that in Tonga (0.51 Tg Myr⁻¹ m⁻¹; Fig. 4E; Table 1). The higher flux is accompanied by a substantially greater decarbonation efficiency: ∼15% in Colombia vs. ∼1% in Tonga. The two subduction zones also exhibit contrasting principal controlling mechanisms for decarbonation. In cold Tonga, carbonate dissolution dominates the decarbonation processes, accounting for ∼65% of the total carbon outflux; in contrast, metamorphic decarbonation prevails in hot Colombia, accounting for ∼80% of its outflux. Beyond these primary characteristics, the decarbonation behaviors of individual lithologies vary considerably (Fig. 4F; table S2). As the main contributors of decarbonation (Fig. 4, C and D), basalt and gabbro undergo nearly complete decarbonation in Colombia (100% for basalt; 90.9% for gabbro), but exhibit much lower decarbonation efficiency in Tonga (8.3% for basalt; 2.3% for gabbro). The decarbonation efficiency of sediment is lower in Colombia (3.4%) than in Tonga (45.3%), primarily due to the distinct bulk-rock compositions in these two subduction zones. Colombia sediment is composed of carbon-rich limestone and chalk containing ∼22.5 wt.% CO₂ in the bulk composition, whereas Tonga sediment is carbon-poor pelagic clay with ∼0.5 wt.% CO₂ (table S1). The high carbon content in sediment would enlarge the stability fields of mineral-bound carbon (fig. S4) and consequently impede metamorphic decarbonation (9). It is worth noting that the subcrustal serpentinites undergo very limited decarbonation in both Tonga and Colombia (0% and 2.1%, respectively), mainly due to their low temperatures and the enlarged stability fields of carbon-bearing phases associated with their higher magnesium contents relative to oceanic crust (fig. S4). Similar low decarbonation efficiency of serpentinite has also been documented in previous thermo-petrological models (22).

Fig. 4. Decarbonation evolution within the Tonga and Colombia subducting slabs.

(A and B) Spatial distribution of mineral-bound and released carbon within the Tonga (A) and Colombia (B) slabs. [(Aa) and (Ba)] Residual mineral-bound CO₂ content within slab. [(Ab) and (Bb)] Mass of released CO₂ via metamorphic decarbonation. [(Ac) and (Bc)] Mass of released CO₂ via carbonate dissolution. White dashed lines represent lithological boundaries among sediment, basalt, gabbro and serpentinite layers. (C and D) Cumulative decarbonation flux as a function of slab-surface depth in Tonga (C) and Colombia (D), with the corresponding pressure scale shown on the bottom axis. [(Ca) and (Da)] Metamorphic decarbonation profiles. [(Cb) and (Db)] Carbonate dissolution profiles. (E) Total decarbonation fluxes and efficiencies of carbonate-bearing lithologies. (F) Decarbonation efficiencies of individual lithological layers.

Considering that hydrous and carbonated fluids released from the subducting slab may react with the overlying mantle wedge to form a serpentinite layer above the slab interface (46), we further assess its potential for water and carbon sequestration. In these additional models, a 1-km-thick carbonated serpentinite layer is imposed atop the Tonga and Colombia slabs, assuming the same bulk-rock composition as sub-crustal serpentinite. Owing to the elevated temperature induced by mantle-wedge corner flow, this layer becomes nearly anhydrous below the MDD in both regions (fig. S6). Carbon sequestration, however, differs between these two subduction zones (fig. S6). In cold Tonga, the basal ∼300 m of this supra-crustal serpentinite layer retains carbon primarily due to its relatively low temperature; whereas in hot Colombia, no carbon is preserved in this layer below the MDD. These results suggest that the slab–mantle interface in cold Tonga may sequester a certain amount of slab-derived carbon, whereas the released carbon is efficiently transferred upward into the mantle wedge in hot Colombia. Furthermore, because temperature increases upward from slab surface to the interior of mantle wedge (Fig. 2, A and B), an even thicker supra-crustal carbonated serpentinite layer would not increase the carbon sequestration. Beyond its role in carbon storage, a carbonate-bearing lithological layer along the base of mantle wedge may act to weaken and lubricate the slab–mantle interface due to the relatively low rheological strength of carbonate minerals compared to olivine (47–49). This behavior is analogous to hydrous serpentinite layers (50) and may influence slab–mantle coupling and subduction dynamics.

Slab decarbonation in comparison with volcanic CO₂ degassing

Volcanic degassing is a primary pathway by which slab-derived carbon returns to Earth’s surface (33). This provides a natural benchmark for evaluating model predictions of slab decarbonation (51). Recent advances in ground- and satellite-based gas monitoring enable quantitative estimates of volcanic CO₂ emissions, offering essential constraints on carbon output budgets in subduction zones (7, 20, 52). To establish observational constraints, we compile ground- and satellite-measured CO₂ outfluxes from active volcanoes in Tonga and Colombia arcs that have eruptive records from the 1970s to 2024 [https://so2.gsfc.nasa.gov/measures.html; (7, 52–57)]. Three different degassing types (51), i.e., eruptive emissions, diffuse emissions, and passive degassing are considered in our compilation (Fig. 5, A to C; data S3). The total volcanic CO₂ outflux in each study area (Fig. 1, C and D) is obtained by aggregating emissions from all the volcanoes and the degassing types (see the Materials and Methods section). When multiple measurements are available for a given volcano and degassing type, two approaches are adopted: (i) selecting the most recently published data (the “Latest” group), and (ii) averaging all the available observations (the “Average” group). Furthermore, model-predicted decarbonation outflux (kt yr.⁻¹) is derived by multiplying the modeled slab CO₂ outflux (Tg Myr⁻¹ m⁻¹; Fig. 4E) by trench length, with associated uncertainties in trench length estimates (see supplementary text S1 and fig. S7).

Fig. 5. Comparison between the observed volcanic CO₂ outflux and the model-predicted slab CO₂ outflux in Tonga (TON) and Colombia (COL) subduction zones.

(A to C) Compiled volcanic CO₂ outfluxes via different degassing types for specific volcanoes: (A) eruptive emissions, (B) diffuse emissions, and (C) passive degassing. Data are sourced from (https://so2.gsfc.nasa.gov/measures.html) and (7, 52–57). Squares and circles represent satellite- and ground-based measurements, respectively. Filled symbols represent CO₂ outfluxes derived by multiplying SO₂ outflux with molar CO₂/SO₂ ratio, whereas open symbols correspond to direct measurement of CO₂ outfluxes. Each error bar is labeled with the data source (left) and the observation period in years (right). (D) Comparison between volcanic observations with model-predicted decarbonation outfluxes. Green and orange regions represent the ranges of model-predicted CO₂ outfluxes from the Tonga and Colombia slabs, respectively, in this study. Gray bars are model-predicted outfluxes from previous studies (9, 23, 24). Diamonds and triangles represent observational data based on the “Latest” and “Average” strategies, respectively. The values of model-predicted and observed CO₂ outfluxes are marked in the diagrams. Details are shown in the Materials and Methods section and data S3.

In Tonga, the “Average” observed volcanic CO₂ outflux is 90 ± 38 kt year⁻¹, and the “Latest” estimate yields 84 ± 49 kt year⁻¹ (Fig. 5D). In contrast, Colombia has a much higher outflux of 1081 ± 173 kt year⁻¹ (“Average”) or 665 ± 165 kt year⁻¹ (“Latest”) (Fig. 5D). We first compare these compiled volcanic CO₂ outfluxes with model-predicted CO₂ outfluxes calculated from previous studies [(9, 23, 24); table S3; data S3]. Both the predictions based solely on carbonate dissolution [389 ± 57 kt year⁻¹ for Tonga, 1082 ± 307 kt year⁻¹ for Colombia; (23)] and solely on metamorphic decarbonation [153 ± 22 kt year⁻¹ for Tonga, 1496 ± 424 kt year⁻¹ for Colombia; (9)] markedly exceed the observed volcanic CO₂ outfluxes in both subduction zones. Moreover, previous estimates from the integrated metamorphic-dissolution models [80 ± 12 kt year⁻¹ for Tonga, 1839 ± 522 kt year⁻¹ for Colombia; (24)] reproduce the Tonga volcanic outflux but substantially overestimate the Colombia outflux. These discrepancies between previous estimates may be due to the simple, generic bulk-rock composition and thickness of the slab layers used in these models.

The present model is constructed using the specific observational data; it predicts slab CO₂ outfluxes of 110 ± 16 kt year⁻¹ in Tonga and 530 ± 150 kt year⁻¹ in Colombia (Fig. 5D; table S3; data S3). Model predictions are generally comparable to the observational trends and magnitudes, indicating that the hot Colombia slab releases several times more CO₂ than the cold Tonga slab. Specifically, for Tonga, the model-predicted outflux slightly exceeds the observed volcanic flux. This discrepancy could be attributed to potential underestimation of submarine volcanic emissions and partial carbon sequestration within the overlying mantle wedge. Extensive discharge of CO₂ and SO₂ from submarine hydrothermal vents in the Tonga arc has been widely documented (58–60), where direct release of volatiles into seawater potentially leads to underestimation of volcanic CO₂ flux measured by satellite ultraviolet (UV) instruments (15). In addition, slab-derived carbon may be redistributed within the slab–mantle wedge system through fluid transport (61, 62), and may be partially retained within the thin layer (∼300 m) above the slab interface in Tonga as predicted by our model (fig. S6). Similar carbon sequestration processes may also occur in the forearc through carbonate precipitation or biologically mediated carbon uptake (63). For Colombia, the model-predicted slab CO₂ outflux is ∼20% lower than the “Latest” volcanic CO₂ flux, but ∼51% lower than the “Average” estimate. This implies that metamorphic decarbonation and carbonate dissolution alone cannot reconcile the observed volcanic outflux, underscoring the role of additional carbon sources or decarbonation mechanisms. Carbon and helium isotopic signatures (δ¹³C, ³He/⁴He) of volcanic and fumarolic gases in the Colombia arc closely resemble mid-ocean ridge basalt (MORB) values (11, 64), effectively ruling out additional carbon from overriding continental crust. Our compiled geochemical data of volcanic rocks indicate that the excess volcanic CO₂ flux is likely attributed to an additional partial melting-induced decarbonation (see next section for details).

Arc geochemical constraints on decarbonation sources and mechanisms

Geochemical signatures in volcanic arcs preserve fingerprints of slab-derived fluids and melts involved in arc magmatism, providing valuable insights into slab decarbonation processes. Here, we compile isotopic and trace element data of volcanic rocks in the Tonga and Colombia arcs from the GEOROC geochemical database (https://georoc.mpch-mainz.gwdg.de/georoc; data S4). The data are filtered to minimize the effects of fractional crystallization and crustal assimilation using criteria including SiO₂ < 60 wt.%, Mg^# > 40, Eu_N/Eu^* in the range of 0.85–1.15, and Rb/Sr < 0.1 [Mg^#, Eu_N, Eu^* calculated following (65)]. The filtered dataset exhibits minimal signature of crustal assimilation, as evidenced by the lack of systematic correlations between Pb and Sr isotopes, Th/Nb and Ba/Th with SiO₂ content [figs. S8 and S9; methods following (66)]. Previous studies further support limited crustal assimilation in both arcs. In Tonga, δ¹⁸ O and Hf isotopes preclude substantial crustal contamination (67). In Colombia, trace elements (e.g., Th, U, Ba) show no correlations with crustal thickness (68), and xenoliths from the overriding continental crust contain much lower Th and La contents compared to Colombia volcanic rocks (69).

We first use lead isotopes to identify the dominant slab-derived components contributing to arc magmatism. In ²⁰⁶Pb/²⁰⁴Pb versus ²⁰⁸Pb/²⁰⁴Pb diagram, Tonga volcanic rocks show a trend toward Pacific MORB, suggesting dominant contributions from subducted oceanic crust rather than sediment (Fig. 6A). In contrast, Colombia volcanic rocks plot near sediments from Panama Basin and Caribbean Sea (Fig. 6B), consistent with a mixed sedimentary component due to seawater exchange prior to closure of the Central American Seaway (70). The lead isotope trends thus indicate substantial sediment contribution to the magma source beneath Colombia arc. To further resolve the properties of slab-derived components, we examine trace element ratios that can distinguish between altered oceanic crust (AOC) fluid and sediment melt (Fig. 6, C and D). Tonga volcanic rocks display low La/Sm and Th/Yb but high Ba/Th and Rb/Nb ratios, consistent with a dominant contribution from AOC-derived fluids. In contrast, Colombia volcanic rocks show high La/Sm and Th/Yb coupled with low Ba/Th and Rb/Nb, indicating the dominance of a sediment melt contribution. These interpretations are further supported by isotopic and elemental proxies. In Tonga arc, the signature of AOC-derived fluid are widely recognized, as evidenced by isotopic composition ²⁰⁶Pb/²⁰⁴Pb (71), δ¹¹B (72), δ^138/134Ba (73), and δ^98/95Mo (74) and element ratios Ba/La, Sr/Nd and U/Th (20). In Colombia arc, sediment involvement is supported by anomalously high values of Th(U)/La, Nb/Ta and Dy/Yb and analysis of Sr-Nd-Pb-Hf isotope systematics (68, 70).

Fig. 6. Lead isotopes and trace element ratios of volcanic rocks from the Tonga and Colombia arcs.

(A and B) ²⁰⁶Pb/²⁰⁴Pb versus ²⁰⁸Pb/²⁰⁴Pb diagrams for Tonga (A) and Colombia (B) arcs. Data are derived from the GEOROC database (https://georoc.mpch-mainz.gwdg.de/georoc; data S4 and references therein). The compositions of Pacific MORB (117, 118), Pacific sediment (119), Nazca MORB (120), Panama Basin sediment (68), and Caribbean Sea sediment (121) are plotted for comparison. (C) Ba/Th versus La/Sm, and (D) Rb/Nb versus Th/Yb diagrams for Tonga and Colombia arcs. The composition of N-MORB is from (122).

AOC fluid signature in Tonga arc is consistent with model-predicted devolatilization

Our model indicates that slab dehydration and decarbonation in Tonga is primarily sourced from subducting basalt and gabbro, with both processes occurring spatially coincident over a depth range from the MDD to ∼310 km (Figs. 3C and 4C). The release of AOC-derived aqueous fluids can mobilize Large Ion Lithophile Elements (LILEs) into the mantle wedge, producing the characteristic AOC fluid signature recorded in Tonga volcanic rocks (Fig. 6, C and D). Additionally, beyond spatial correlation, slab-derived aqueous fluid and carbon (whether metamorphic or dissolved) exhibit inherent chemical affinity, enabling their potentially coupled transport into magma-generation regions within the mantle wedge (27, 75–81). Thermodynamic calculations demonstrate that slab-released carbon can exist in aqueous fluids in various molecular and ionic forms, depending on temperature, pressure, pH, and oxygen fugacity conditions (75, 76). Furthermore, theoretical models and high-pressure-temperature experiments predict that binary molecular systems H₂O-CO₂ and H₂O-CH₄ can become miscible under the P-T conditions typical of slab devolatilization and the sub-arc mantle-wedge (77, 78, 81). Moreover, fluid inclusions in high-pressure and ultrahigh-pressure metamorphic rocks contain coexisting aqueous fluids and carbon species, including carbon ions (27), CH₄ (79), CO₂ (80), demonstrating the feasibility of coupled water-carbon transport into the mantle wedge. Collectively, these lines of evidence suggest that the AOC fluid signature in Tonga arc is consistent with the model results that both dehydration and decarbonation are mainly sourced from oceanic crust.

Sediment melt signature in Colombia arc indicates a potential contribution from melting-induced decarbonation

The absence of sediment melt signature in Tonga, compared with its clear presence in Colombia, points to a thermally controlled difference in sediment melting and associated decarbonation, whereby partial melting of subducting sediment can release carbon into ascending melts (82), thereby contributing to magmatic CO₂ output (34). Two potential melting mechanisms for subducting sediment are evaluated by combining our model results with experimentally constrained sediment solidi and thermo-mechanical modeling, i.e., in-situ slab melting (31) versus diapir-induced melting (32). To assess the potential role of in-situ sediment melting, we compare the model-derived P-T path of subducting sediment to experimentally constrained solidi. Since the sediment solidus is sensitive to bulk composition, particularly H₂O and CO₂ contents, we extract the H₂O and CO₂ concentrations of sediments from our model results (fig. S10) to select experimental solidi with analogous compositions for comparison (table S4). In Tonga, the cold P-T path of subducting sediment containing ∼2.2 wt.% H₂O and ∼0.4 wt.% CO₂ at ∼75 km depth (fig. S10) does not intersect any plausible solidus (fig. S11), indicating that in-situ melting is unlikely. In contrast, the hotter P-T path of Colombia sediment containing ∼4.3 wt.% H₂O and >22 wt.% CO₂ at ∼96 km depth (fig. S10) intersects the hydrous, carbon-rich solidus (fig. S11), implying favorable conditions for in-situ melting.

For the alternative diapir-induced melting, previous thermo-mechanical modeling (83) has predicted sediment instability time as a function of sediment layer thickness and temperature (fig. S12). The ∼100 m-thick Tonga sediment (table S1) would require temperature exceeding 1000°C (fig. S12), which is not attainable in our model (Fig. 2C). In contrast, the ∼250 m-thick Colombia sediment (table S1) requires ∼817°C for diapir initiation (fig. S12), corresponding to a depth of ∼105 km in our model (Fig. 2D). Moreover, 3D diapir initiation models suggest that slower subduction promotes diapir formation (84). Consequently, the higher temperatures and slower subduction of Colombia sediment are favorable for diapir development. These buoyant diapirs may subsequently ascend into the mantle wedge, where hotter thermal conditions promote partial melting and associated decarbonation (70, 85). Overall, substantial melting-induced decarbonation of Colombia sediment is predicted whether through in-situ melting or diapir-induced melting, whereas melting of Tonga sediment is unlikely; both predictions are consistent with the geochemical observations (Fig. 6). Furthermore, in Colombia, partial melting-induced decarbonation may provide an additional mechanism to reconcile the discrepancy between model-predicted and observed CO₂ outfluxes (Fig. 5D), and could further increase the total decarbonation efficiency (∼15%) predicted from metamorphic decarbonation and carbonate dissolution alone (Fig. 4E). In this case, the preferential partitioning of carbon into melts (82) may result in a slight overestimation of the contribution from metamorphic decarbonation and carbonate dissolution in Colombia sediment (totaling 0.26 Tg Myr⁻¹ m⁻¹); but the total CO₂ outflux is expected to increase due to the partial melting-induced decarbonation. The contribution from Colombia sediment to volcanic CO₂ degassing is supported by the compositions of volcanic gases in the Colombia arc, which show high CO₂/SO₂ (34, 54) and CO₂/³He (64, 86) ratios, consistent with carbon input from carbon-rich subducting sediment (87). Moreover, mantle xenoliths from Colombia include carbonate-bearing pyroxenites and peridotites containing Ca-rich aragonite, dolomite and Mg-calcite, consistent with metasomatic interaction between sediment-derived carbonic agents and the overlying mantle wedge (70, 88). In addition to sediment, the P-T paths of subducting oceanic crust and serpentinite in both Tonga (Fig. 2C) and Colombia (Fig. 2D) slabs generally remain below the experimentally constrained solidi of carbonated oceanic crust (29) and mantle (89), suggesting that partial melting of these lithologies is unlikely to contribute to slab decarbonation.

Deep carbon cycling in hot versus cold subduction zones

By integrating thermo-petrological modeling, volcanic CO₂ emissions, and arc geochemistry, we have quantitatively dissected the contrasting decarbonation characteristics of the Colombia and Tonga subduction zones. As thermal end-members of present-day global subduction systems, these two subduction zones provide a conceptual framework for understanding deep carbon cycling in hot versus cold subduction environments (Fig. 7). Our results identify two distinct decarbonation patterns, i.e., efficient “focused” decarbonation in hot subduction zones versus sluggish “diffuse” decarbonation in cold subduction zones. In hot subduction zones, slab decarbonation exhibits a “focused” decarbonation pattern (Fig. 7A), characterized by high-efficiency, metamorphism-dominated carbon release within a narrow depth range. The high thermal gradient in the subducting slab leads to intense, focused metamorphic decarbonation at shallow depth, while carbonate dissolution plays a secondary role because carbonate solubility remains comparatively low at shallow dehydration conditions. Additionally, the elevated slab temperature in hot subduction zones promotes in-situ or diapir-induced sediment melting, offering an additional mechanism for carbon release from subducting slab. In contrast, cold subductions zones display a sluggish “diffuse” decarbonation pattern (Fig. 7B), in which low-efficiency, dissolution-dominated carbon release occurs over a much broader depth range. The cold subducting slab undergoes limited metamorphic decarbonation near the MDD and contributes modestly to the total carbon outflux. Accordingly, carbonate dissolution becomes the controlling mechanism, extending to greater depths where increasing aqueous solubility promotes continued carbon release.

Fig. 7. Contrasting decarbonation characteristics in hot versus cold subduction zones.

(A) Efficient “focused” decarbonation in hot subduction zone and (B) sluggish “diffuse” decarbonation in cold subduction zone. Colored areas represent the decarbonation regions via different mechanisms. Colored arrows represent CO₂ outfluxes via different mechanisms, with arrow size representing the flux magnitude.

The different principal controlling decarbonation mechanisms, i.e., metamorphic decarbonation in hot settings and carbonate dissolution in cold settings, are also recorded in exhumed oceanic metamorphic terranes. For example, multi-scale petrological and geochemical studies indicate pervasive fluid-mediated carbonate dissolution in ancient cold subduction zones, including Alpine Corsica (90), Zermatt-Saas in western Alps (91), Zagros in southern Iran (92), and the Cycladic subduction complex in Greece (28). The thermodynamic modeling of metasediment, metabasalt and ophicarbonate in cold subduction zones such as Corsica and Zermatt-Saas (93, 94) show that the extent of metamorphic decarbonation is limited. These studies of exhumed oceanic metamorphic terranes support the “diffuse” decarbonation patterns in cold subduction zones. Notably, in the Cycladic subduction complex in Greece (28), both metamorphic decarbonation and carbonate dissolution have been documented (28, 95), suggesting that it may represent a transitional regime in which both decarbonation mechanisms operate concurrently.

Globally, present-day cold subduction zones are mainly distributed in the western Pacific, and hot subduction zones in the eastern Pacific (Fig. 1, A and B). Based on our thermo-petrological models, subducting slabs in the eastern Pacific subduction zones may experience more intense carbon release than those in the western Pacific, a consequence that is corroborated by volcanic degassing observations (9). Volcanic CO₂ outflux in the eastern Pacific is about 2–3 times higher than in the western Pacific (data S1). In terms of decarbonation mechanisms, our results imply that carbonate dissolution prevails in western Pacific subduction zones, whereas metamorphic decarbonation predominates in eastern Pacific. In addition, eastern Pacific subduction zones are more likely to undergo partial melting-induced decarbonation. Furthermore, our modeling results reveal that both hot and cold subduction systems are characterized by overall limited decarbonation efficiency with ∼15% in hot Colombia and ∼1% in cold Tonga. It indicates that a substantial fraction of subducted carbon could be retained and transported into the deeper mantle in modern global subduction zones. In particular, the carbonated serpentinite layer acts as an efficient carbon sink with very low decarbonation efficiency in all thermal conditions, which is consistent with recent petrological observations (96).

On geological time scales, Precambrian subduction zones are generally considered to be hotter than present-day settings (97). Based on our results, hot subduction zones tend to favor more efficient decarbonation, potentially limiting carbon transport into the deep mantle. As the upper mantle and subduction zones cool, enhanced carbon transport into the deep mantle may contribute to a decrease in atmospheric CO₂ and influence Earth’s habitability, although plant-mediated silicate weathering carbon sink should also play an important role (98, 99). Observation-based, targeted thermo-petrological models not only explain the contrasting volcanic CO₂ degassing characteristics in western and eastern Pacific subduction zones, but also provide quantitative constraints on present-day and ancient subduction-zone deep carbon cycling and volcanic degassing. These insights provide a basis for understanding atmospheric and climate evolution on geological time scales in relation with evolving subduction thermal regimes from the Archean to present day (97).

MATERIALS AND METHODS

Thermal structure of subduction zones

The thermal structure of subduction zone is calculated using the PGCtherm2D, a steady-state two-dimensional finite element model (37, 100) that has been benchmarked with comparable thermal models (101). The model consists of the subducting slab with prescribed convergence, a stagnant mantle wedge (the “cold nose”), a convective mantle wedge with a temperature- and stress- dependent rheology, and an overriding plate (fig. S1). Following (41), we assume the maximum depth of decoupling between the subducting slab and overlying mantle wedge to be 75 km. Slab convergence rates are prescribed to be 16.0 cm year⁻¹ for Tonga (13) and 4.1 cm year⁻¹ for Colombia (14). Slab ages are set to be 101.2 Ma for Tonga and 13.2 Ma for Colombia (12) to constrain the trench-side vertical thermal boundary according to the GDH1 plate cooling model (102). A typical back-arc heat flow of 65 mW m⁻² (6) is imposed for both subduction zones to constrain back-arc-side vertical boundary. Frictional and viscous heating along the slab–mantle interface is included following previous studies (37, 103, 104). The apparent coefficient of friction (μ′) are chosen to be 0.07 for Tonga and 0.03 for Colombia, consistent with a previous study (103). Other model parameters are listed in table S5. The details of governing equations, rheological parameters are similar to (103, 104).

Thermodynamic look-up tables

The mineral-bound H₂O capacity, CO₂ capacity and carbonate mineral abundance of individual compositional layers are calculated using Perple_X software [version 7.1.6; (38, 105)] with the internally consistent thermodynamic dataset of Holland and Powell (106). The bulk-rock compositions of sediment, basalt and gabbro are normalized to Na₂O-CaO-K₂O-MnO-FeO-MgO-Al₂O₃-SiO₂-H₂O-CO₂ system, and that of serpentinite is normalized to CaO-FeO-MgO-Al₂O₃-SiO₂-H₂O-CO₂ system (table S1; supplementary text S3; data S2). The activity-composition models for solid minerals and H₂O-CO₂ fluid used in this study are listed in table S6. All phase diagrams cover 0–1200°C and 0.05–10 GPa, discretized into 1000 × 1000 nodes with resolutions of 1.2°C and 9.95 MPa. Considering the influence of variable water content on slab decarbonation in open system, CO₂ capacity and carbonate mineral abundance are calculated under different H₂O contents ranging from 0 to 13 wt.% at 0.5 wt.% intervals. The calculated phase diagrams for H₂O and CO₂ capacity are shown in figs. S2 and S4.

In parallel, the aqueous solubilities of major carbonate minerals (calcite/aragonite, dolomite and magnesite) are adopted from a recent study that integrates experimental data with the Deep Earth Water (DEW) model (39, 40). These solubility data are originally determined over 300–800°C and 0.5–5 GPa, and are here extrapolated to a broader P-T range of 0–1000°C and 0–10 GPa using a spline method (fig. S13). The extrapolated values are capped at the maximum solubilities reported in (23).

Thermo-petrological modeling

We have developed a coupled thermo-metamorphic-dissolution model, HEPs_vol, that operates in an open system for deep carbon cycling (107, 108). The model integrates slab thermal structure with thermodynamic look-up tables to sequentially simulate slab dehydration and decarbonation. In this modeling framework, decarbonation is represented by two distinct processes: metamorphic decarbonation, involving the decomposition of carbonate minerals, and carbonate dissolution, involving the transfer of carbon into aqueous fluids. The slab geometry is illustrated in supplementary text S2 and fig. S1, with a uniform spatial resolution of 50 meters in both horizontal and slab-normal directions.

Dehydration is first calculated by integrating the P-T conditions and P-T-dependent H₂O capacity of corresponding compositional layers. The content of released free water from each rock cell is computed as the difference in mineral-bound water content between adjacent rock cells along the subduction direction. The mass of released free water from each rock cell is then obtained by multiplying its free water content by the mass of each rock cell, assuming a rock density of 3300 kg m⁻³, following previous thermo-petrological studies (24, 43, 109). Fluid infiltration is assumed to migrate upward with pervasive flow following the approach of (9, 21). The total H₂O content in each cell is updated by combining infiltrated free water from underlying lithologies with the residual mineral-bound water within the current rock cell. The total mass of free H₂O in each rock column is calculated by summing contributions from the base of the column to the slab surface. Metamorphic decarbonation is subsequently estimated by integrating the P-T paths with P-T-water-dependent CO₂ capacity of corresponding compositional layers (107). Carbonate dissolution is quantified by multiplying the aqueous solubilities of carbonate minerals that are stable under local P-T-water conditions by the mass of free H₂O, also accounting for the saturated solubilities of carbonate minerals (108). The content and mass of released CO₂ from each rock cell are calculated following the same method as for H₂O. Finally, accounting for the infiltration of carbon-bearing fluids, the released CO₂ mass of each column is obtained by summing values from the base to the slab surface. Detailed formulations and numerical procedures are provided in supplementary text S4.

Thicknesses of eight compositional layers used in our models are mainly constrained by geophysical and oceanic drilling data (table S1; supplementary text S2; data S2). As both Tonga and Colombia margins are tectonically erosive (110), trench sediments are expected to be largely subducted with limited accretion. Accordingly, we adopt the drilling-constrained thicknesses as a first-order estimate of the subducting sediment layer. The serpentinite thickness in Colombia is estimated based on the 2D kinetics-limited reactive-flow model (111) due to the lack of detailed geophysical constraints. It is noteworthy that sensitivity tests with variable serpentinite thickness from 4 to 10 km do not change the main model predictions that metamorphic decarbonation dominates in the Colombia subduction zone (fig. S14).

Compiling observed volcanic CO₂ degassing

Volcanic CO₂ emissions from the Tonga and Colombia arcs since the 1970s are compiled from both ground-based and satellite-based measurements, including eruptive emissions, diffuse emissions, and passive degassing [data S3; (7, 52–57); https://so2.gsfc.nasa.gov/measures.html]. To improve the accuracy of degassing estimates, we incorporate recently published long-term observational datasets. For example, satellite-based passive degassing data spanning 2005 to 2024 are derived from OMI, OMPS, TROPOMI observations using an inverse-variance weighting approach (112). Ground-based passive degassing data span the period 2005–2017, with a gap-filling method applied to compensate missing intervals (56). Eruptive emissions are derived from satellite observations covering 1978 to 2023.

Among the compiled datasets, some studies report direct CO₂ flux measurements, whereas others provide only SO₂ fluxes. In the latter case, CO₂ flux is estimated using the molar CO₂/SO₂ ratio (34) following: ${\mathrm{CO}}_2 \left(\text{mass}\right)={\mathrm{SO}}_2\left(\text{mass}\right)\times \frac{{\mathrm{CO}}_2}{{\mathrm{SO}}_2}\left(\text{in molar}\right)\times \frac{44}{64}$. The molar CO₂/SO₂ ratios are either measured directly at high-temperature fumaroles or estimated from the trace element ratios (e.g., Ba/La, Sr/Nd) of volcanic rocks based on established correlations with CO₂/SO₂ (20, 34). To calculate the total volcanic CO₂ outflux from each subduction zone, we aggregate emission data across the three degassing types and across all individual volcanoes within the study area. For cases where multiple measurements are available for a given volcano and degassing type, two strategies are applied: (i) selecting the most recently reported value (“Latest” group), and (ii) computing the arithmetic mean of all available measurements (“Average” group). Standard deviations are calculated following the approaches of (7, 20, 51). The details of the calculation are provided in data S3.

Model limitations

The limitations of our modeling approach are summarized here. Firstly, the fluid dynamics of released aqueous fluids has been simplified to assume vertical migration via pervasive flow within the slab in our model (9, 21). In reality, fluid transport is controlled by fluid buoyancy, permeability/fracturing, and compaction-pressure gradients (113), and would ideally require two-phase or even multi-phase flow modeling. Secondly, carbonate solubility data are extrapolated beyond their experimentally or thermodynamically constrained P-T range (40), introducing additional uncertainty at pressures exceeding 5 GPa or temperatures above 800°C. Thirdly, our modeling framework focuses on volatile release from the subducting slab, specifically dehydration and decarbonation processes. It does not explicitly account for carbonation and melting processes within the mantle wedge, nor for the remobilization and assimilation of crustal carbonates in the overriding plate. Finally, the effects of supercritical fluids on slab decarbonation are not incorporated (114, 115), because quantitative constraints on carbon partition coefficient between supercritical fluids and rocks are still lacking. Despite these limitations, the model provides a valuable and state-of-the-art framework to quantify metamorphic decarbonation and carbonate dissolution in present-day subduction zones.

Supplementary Materials

The PDF file includes:

Supplementary Text S1 to S5

Figs. S1 to S14

Tables S1 to S6

Legends for data S1 to S4

References

Other Supplementary Material for this manuscript includes the following:

Data S1 to S4