After standard data quality control, 7120 profiles from 43 BGC-Argo floats in the western North Pacific (20–50∘ N, 120–180∘ E) between 2008 and 2019 were selected (Fig. 2). All of these profiles contained measurements of temperature, salinity, pressure, and dissolved oxygen (DO, µmol/kg). The upper 1000 dbar of the ocean was sampled in each profile, and the typical profiling interval was between 5–10 d, with the floats parking at 1000 dbar depth in between. The typical vertical sampling frequency was every 5, 10, and 50 dbar for depth intervals of 0–100, 100–500, and 500–1000 dbar, respectively. Some floats were set with daily profiling and higher vertical frequency (e.g., every 2 dbar) for specific purposes.

All BGC-Argo variables were vertically smoothed with a three-bin running average to remove sharp noises or spikes (Llort et al., 2018). Two key variables, apparent oxygen utilization (AOU) and potential spicity (π), were derived from the direct measurements. Specifically, AOU is defined as the difference between saturated oxygen concentration (Osat) and DO, and Osat is estimated from temperature and salinity (Garcia and Gordon, 1992). AOU is a proxy for water mass age which reflects the microbial respiration of dissolved and particulate organic matter (Sarmiento and Gruber, 2006). Potential spicity referenced to the surface pressure is calculated from pressure, temperature, and salinity following Huang et al. (2018). Seawater is a two-component system. The water mass anomaly is commonly analyzed in terms of the (potential) temperature and salinity anomaly, and isopycnal analysis is also widely used. By definition, the temperature and salinity anomaly on an isopycnal surface is density compensated; thus, the water mass anomaly on an isopycnal surface is commonly described in terms of another thermodynamic variable, which is called spice, spiciness, or spicity. Over the past decades, there have been different definitions of such a thermodynamic variable; however, a most desirable property of such a thermodynamic function is that it is orthogonal to the density. Recently, Huang et al. (2018) proposed a potential spicity function (π) by the least-squares method, which is practically orthogonal to the potential density, with the root mean square of the angle deviation from orthogonality at the value of 0.0001∘. Therefore, combining density and spicity gives rise to an orthogonal coordinate system. It is the thermodynamic variable we used in this study, which allows differentiating water masses with distinct thermohaline properties but similar density. In addition, potential density (σ) referenced to the surface pressure was derived from pressure, temperature, and salinity based on the thermodynamic equation (TEOS-10; McDougall and Barker, 2011); and MLD was estimated based on a threshold (0.05 kg/m3) of the difference in density from a near-surface value (i.e., at 10 dbar) (Brainerd and Gregg, 1995). All these derived variables were calculated for each of the 7120 profiles.

In addition to the BGC-Argo float data, satellite data of daily sea level anomalies (SLAs) and daily geostrophic velocity anomalies (u′ and v′) between 1993 and 2018 were also processed. The geostrophic velocity anomalies were used to calculate the eddy kinetic energy (EKE) as EKE=12u′2+v′2. These data were used to identify the spatial relationship between surface mesoscale circulation and the float profiles.

Subduction associated with eddy pumping is a recognized important contributor to the transfer of carbon and other materials from the surface euphotic layer to the ocean interior (McGillicuddy, 2016; Bord et al., 2019), but investigating the spatial distributions, physical dynamics, and biogeochemical consequences of these episodic small-scale processes is difficult. The BGC-Argo program provides an exceptional data resource for this purpose (Claustre et al., 2020; Chai et al., 2020), but detecting subduction signals where differences among water masses are small is challenging.

Subduction patches below the seasonal and permanent pycnoclines can be identified in vertical profiles by anomaly matrices of temperature, salinity, and dissolved oxygen (DO). Examples of these events are illustrated in time series from the BGC-Argo profiling float (MR2901556), between 28 July and 18 August 2014 (Fig. 4). Positioned on the southern perimeter of the Kuroshio Extension region, the surrounding ocean conditions are less energetic, with fewer eddy activities and small sea level anomalies. During the observation period, the float was trapped in a warm-core eddy (Fig. S2) and moved from the margin of the eddy to the eddy core, as indicated by the gradual increase in depth of the anomalous patch (right panels of Fig. 4). Due to the vertical cruising of the Argo profiler in an environment with velocity shear, it may sample different parts of the same subduction patch, as indicated by slightly different depths, and anomalous spicity and oxygen concentration. Here, intermittent patches of elevated spicity (π), lower AOU, and greater dissolved oxygen are visible in the upper 600 dbar (boxes 1–3, Fig. 4). Potential spicity (π), a parameter dependent on pressure, temperature, and salinity (Huang et al., 2018), is a sensitive indicator of water mass differences. AOU is the difference between the measured dissolved oxygen concentration and its equilibrium saturation concentration in water with the same physical and chemical properties. It reflects the degree of progressive microbial decomposition of organic matter since the water was last at the surface in contact with the atmosphere (Garcia and Gordon, 1992; Sarmiento and Gruber, 2006). Despite this oxygen consumption, these injected waters retain excess net oxygen concentrations relative to the surrounding mesopelagic zone (Fig. 4d).

Llort et al. (2018) successfully identified eddy subduction in BGC-Argo data from the Southern Ocean using anomalies in spiciness (Flament, 2002; Huang, 2011; McDougall and Krzysik, 2015), a parameter derived from a different function of pressure, temperature, and salinity than potential spicity (Huang et al., 2018). However, we found that spiciness frequently missed signs of subduction while misidentifying other signals as subduction, and the 20-bin method used by Llort et al. (2018) significantly dampened the subduction signals in our data. Potential spicity (π) (Huang et al., 2018), on the other hand, greatly improves the ability to distinguish among similar water masses due to its orthogonal coordination with density, a feature that spiciness lacks. This added sensitivity revealed reliable signals of subduction in these BGC-Argo data. The algorithm based on peak detection here shows better capabilities in capturing and quantifying the subduction signals (see Methods, Fig. S1).

For the same subduction event, continuous subduction patches are expected to be identified from the Argo profiles. The discrete anomalous π and AOU signals, highlighted in boxes 1–3 in the example time series (Fig. 4a–d; 31 July, 10 and 12 to 15 August), indicate that they stemmed from distinct subduction events, opportunistically captured by this BGC-Argo float. The first two anomalies (July and early August) each appeared in only a single profile, perhaps indicating a limited spatial scale of these subduction events. In contrast, the mid-August anomaly persisted over four consecutive profiles. We further examined the corresponding time series of temperature, salinity, and potential density and found salinity also showed a similar anomalous signal. As such, we suspect the consecutive subduction patches were most likely from a more sustained or a larger spatial subduction event. It should be noted that the high detection chance of subduction within 21 d was likely given that the float was trapped in a warm-core eddy (Fig. S2), which does not indicate this area is a place where it is easy to observe subduction events (see Sect. 3.2).

We used our peak detection algorithm with the π and AOU data and applied it to all 7120 BGC-Argo profiles (2008–2019) in the western North Pacific (Fig. 5). Our algorithm resolved 335 subduction patches, spread over an unexpectedly large area in the western North Pacific. Overall, subduction patches were identified in 288 profiles (4.0 %) (some profiles have multiple patches at different depths), with approximately 83 % of these being concentrated in the Kuroshio–Oyashio Extension region (Fig. 5a). High (≥6 cm) climatologic sea level anomalies (SLAs) and the corresponding distribution of eddy kinetic energy (EKE) are evidence of the strong energetic ocean processes in this region (Fig. 5a and b). By contrast, far fewer subduction patches were identified in the less energetic region to the south of 29∘ N despite a higher BGC Argo sampling density (Fig. 2b), consistent with eddy-related processes being important for driving these subduction events. Even so, the true frequency of these events across the entire region is certain to have been under-sampled given their small scales relative to the dispersed BGC-Argo float positions.

Discrete signals of subduction were detected throughout the mesopelagic depth range (∼ 100–1000 dbar), with the majority detected below 300 dbar (green and yellow dots in Fig. 5a). The deepest penetrations (≥450 dbar) occurred largely in areas experiencing the highest EKE while the shallowest (100–300 dbar) were largely restricted to areas with lower EKE (Fig. 5a). Based on 16 years of Argo float data (N=1 226 177) in the global ocean, Feucher et al. (2019) found that the depth of the permanent pycnocline differs between the subtropical (i.e., <35∘ N) and subpolar (>35∘ N) regions, with the depth of the permanent pycnocline being 300 and 450 m in the subpolar and subtropical sections of the western North Pacific. Similarly, using the limited BGC-Argo dataset used in this study (Fig. 2), we also found shallower annual maximum MLD in the subpolar section than that in the subtropical section (see Fig. S3). As a result, 56 (16.7 %) and 104 (31.0 %) subduction patches were found to be above and below the depth of the permanent pycnocline (i.e., 450 m) in the subtropical section; and in the subpolar section, 34 (10.1 %) and 141 (42.1 %) subduction patches were above and below the permanent pycnocline (i.e., 300 m). Overall, roughly half (52 %) of the detected subduction signals were below 450 dbar in this region of the western North Pacific, while 22 % penetrated far deeper (up to 800 dbar; Table S2).

There is a distinct seasonality in subduction, with most (∼ 70 %) signals being observed between March (the maximum) and August (Figs. 6 and S4), with the peaks appearing in early March. Since these anomalous patches can be observed after their formation, there is a delayed period between the peak of formation and the peak of observation.

Ocean eddies are sustained by many processes including small-scale mixed-layer instabilities set up by large-scale atmospheric forcing in winter (Sasaki1 et al., 2014); in particular, the extreme mixed-layer deepening events are attributable to the accumulation of excessive surface cooling driven by synoptic storms in connection with cold-air outbreaks (Yu et al., 2020). Each cooling episode brought by the synoptic storm may lead to the formation of subduction patches due to strong mixed-layer deepening. Thus, subduction patches are primarily formed during winter and spring when deep mixing processes occur. Although only 8.3 % of the total profiles were obtained in March, they accounted for 17.3 % of all observed subduction patches (Fig. S4a); correspondingly, the monthly subduction detection rate (i.e., the number of profiles with identified subduction patches divided by the total number of profiles available) was the highest in March, at ∼ 10 % (Fig. S5). In a pioneering work, Stommel (1979) argued that a demon works in the ocean by selecting the later winter (typically later March in the Northern Hemisphere) water mass properties and injecting them into the subsurface ocean. This mechanism is now called the Stommel Demon in dynamical oceanography (Huang, 2010). The high detection rate of episodic subduction patches in March was consistent with observations of large-scale subduction in this region during late winter because mesoscale and sub-mesoscale eddy activities are prevalent when large-scale subduction occurs (Qu et al., 2002; Qiu et al., 2007; Nishikawa et al., 2010; Liu and Huang, 2012; Zhang et al., 2014; Xu et al., 2014). The March-to-August time frame also coincides with the onset and establishment of warming-induced shoaling of the mixed-layer depth, when winter-subducted waters are less likely to be re-entrained into surface waters by winds (Dall'Olmo et al., 2016; Palevsky and Doney, 2018). Indeed, based on the entire BGC-Argo dataset in Fig. 2, we found that the monthly MLD reached a maximum in February and March and then decreased until August (Fig. S3). It should be noted that, despite the number of subduction patches identified in the time frame of April–August being slightly larger than those in September–December (Fig. S4a), the detection rates did not vary much between these time frames (Fig. S5). In contrast, comparatively few (3.0 %) of the subduction patches were detected in January and February, in which time the detection rates were also low (< 2 %, Fig. S5). Although specific timelines between the observed subduction patches and their formation could not be determined, it is reasonable to anticipate that more energetic winds and the accumulated strong heat loss during mid-winter contributed to the peak in subduction signatures observed in March. However, there were no spatial patterns of the subduction patches detected in each season (Fig. 5c). The current BGC-Argo profiling asset is not sufficient for studying how those subduction patches change on interannual scales.

The Kuroshio–Oyashio Extension zone lies between the subtropical and subpolar gyres in the North Pacific, and it is a recognized hot spot for water mass exchange via eddy transport (Yasuda et al., 1996; Talley, 1997; Joyce et al., 2001; Zhang et al., 2014; Xu et al., 2016) and substantial ocean-to-atmosphere heat flux (Jing et al., 2020). It is not surprising then that the majority of subduction signals were observed in this region in spite of less float coverage (Fig. 5). In the southern area of the Kuroshio Extension, where the subtropical mode water is formed, the MLD could reach 300–400 m between January and March (Fig. S3). Subduction patches are formed after the water parcels are detached from the base of the mixed layer, and they could be identified below the base of the winter mixing layer (Fig. 5c). Large-scale circulation and seasonal variability in the mixed-layer depth here typically result in late-winter subduction of subtropical mode waters (Qiu et al., 2007; Oka et al., 2009; Oka and Qiu, 2012; Xu et al., 2014, 2016), and sharp horizontal density gradients can enhance strong vertical exchanges (Marshall et al., 1993; Hurlburt et al., 1996; Liu et al., 2012; Ma et al., 2017). Rapid heat loss to the wintertime cool, dry continental air masses flowing across the Kuroshio–Oyashio Extension erodes the seasonal thermocline to its maximum depth in February–March (Cronin et al., 2013), the latter portion in which the subduction patches were most frequently observed (Fig. 6).

Ascertaining the frequency and spatial extent of these lower-latitude episodic events will be important for establishing their overall contribution to the transport of surface waters into the mesopelagic zone, but this goal is challenged by the presently limited distribution of BGC-Argo floats. It may be possible to obtain a first-order estimate of their frequency by linking the subduction signals here to surface-expressed indicators of mesoscale circulation processes. Moreover, our findings suggest that spicity should be adopted more generally in probing BGC-Argo datasets to improve our understanding of the spatial and temporal distribution of subduction processes.

Beyond being a water mass indicator, AOU is a proxy for cumulative net community respiration and a sensitive indicator of carbon export in the upper mesopelagic zone (Emerson et al., 2001; Pan et al., 2014; Catala et al., 2018; Bushinsky and Emerson, 2018). This export comprises remineralized carbon as well as dissolved and slowly sinking particulate organic matter carried by the subducting waters (Stukel et al., 2017). The magnitude of AOU may be used as an indicator of the time since subduction, with the first-order assumption being that the larger-scale processes initiating these subduction events generated similar surface production. Values of ΔAOU at the anomalous peak depth ranged between -10 (the minimal threshold used) and -81 µmol/kg (Fig. 7a). This proxy was highly variable over the space–time domain, similarly to the variations in Δπ (Fig. 7b). In general, 61.7 % of the subduction patches had ΔAOU values in the range of -30 to -10 µmol/kg with the remainder having greater oxygen depletions (i.e., ≤-30 µmol/kg) (Fig. 5b). Water masses subducted below 450 dbar (i.e., the permanent pycnocline) had an average AOU anomaly of -25.7 ± 15.3 µmol/kg.

There was no clear relationship between the depth of subduction and ΔAOU (Fig. 7a), suggesting either surface conditions (e.g., water temperature, primary productivity) were substantially different when the seawater parcels were subducted or these signatures stem from non-systematic differences in the time since subducted waters were last at the surface. On the other hand, it is noted in Fig. 7 that the depth positions of the subduction patches appear to somewhat extend from northeast to southwest and are deeper along the isopycnal surface as illustrated in Fig. 1. This phenomenon is clearly shown when averaging the depth of subduction patches both latitudinally and longitudinally (Fig. S6). Along the latitude, despite a few deep subduction patches identified at 42–43∘ N (at around 550 m), the mean depths of the subduction patches show a clear increasing pattern from latitude 37–42∘ N to a latitude of 32–37∘ N, i.e., 300 m vs. 500 m. However, the depth positions tend to become shallower and shallower south of 32∘ N. Along the longitude, the depth positions generally appear to be deeper from east to west. As such, it is most likely that the subduction that occurred in the northern KE (37–42∘ N) could travel southwestward from shallow to deep depth, and these waters could reach 32∘ N. The increasing depth positions of subduction patches from 26 to 32∘ N tend to suggest the gradually downward movements of the subducted water masses carried by the general trend of the anticyclonic gyre-scale circulation, yet further investigation is needed.

In the subpolar region, for the subduction patches identified above and below the depth of the permanent pycnocline (i.e., 300 m), respectively, the averaged ΔAOU values are -32.9 and -25.8 µmol/kg, averaged ΔDO values are 42.5 and 32.5 µmol/kg, and averaged thicknesses (i.e., vertical extension of the subduction patch) are 127.5 and 126.6 m (Table 1). In the subtropical region, the depth of the permanent pycnocline was deeper (i.e., 450 m) and the subduction patches above and below this layer were associated with a mean ΔAOU of -27.2 and -28.5 µmol/kg, mean ΔDO of 31.2 and 36.4 µmol/kg, and mean thickness of 128.7 and 128.1 m (Table 1). In general, the vertical extension (i.e., thickness) of the subduction patches identified in each layer and in each region did not vary much between 126.6 and 128.7 m. The mean ΔAOU and ΔDO were stronger above the depth of the permanent pycnocline than those below the depth of the permanent pycnocline in the subpolar region, yet the opposite case is shown for the subtropical region, where the mean ΔAOU and ΔDO were weaker above the depth of the permanent pycnocline than those below the depth of the permanent pycnocline. Interestingly, it is noted that the mean ΔAOU and ΔDO values in the subtropical region below 450 m were also weaker than those in the subpolar region above 300 m, which further supports the potential northeast-to-southwest pathway of subducted waters shown in Fig. 7.

Most subduction patches with strong AOU anomalies were observed between March and August (particularly March; see Fig. S4), after the seasonal mixed layer began to shoal, consistent with expected higher levels of phytoplankton production, which results in a greater degree of respiration in the subducted waters. More respiration means a great degree of oxygen consumption and thus a more negative offset from the surface-saturated concentrations before subduction. Only 0.6 % of the total subduction patches had ΔAOU values of ≤-30 µmol/kg in January and February (Fig. S4b). It should be noted that ΔAOU also strongly depends on the surface water temperature (which determines the solubility of oxygen) when it is subducted. The π anomalies show similar variation patterns with months (peaked in March), with stronger Δπ coupled with stronger ΔAOU (Fig. S4c).

Global ocean inventories of oxygen have been decreasing, and current climate models predict this trend is likely to accelerate over the next century (Oschlies et al., 2018). However, these models suffer from considerable gaps in understanding, one of which is the absence of small-scale transport processes such as the events captured here (Oschlies et al., 2018). The average residual DO enrichment in the subduction patches, defined as the difference in DO concentrations within and adjacent to the subducted waters, was 34.5±19.8 µmol O2/kg, with levels as high as ∼ 88 µmol O2/kg below 450 dbar during March. These differences reflected ∼ 20 % higher oxygen concentrations than in the surrounding mesopelagic waters. The integrated DO enrichment reached a maximum in March (Fig. 6d). Based on these residual excess oxygen concentrations, the oxygen inventory within these features was estimated to be on the order of 64 to 152 g O2/m2 (Eqs. 1 and 2). Specifically, the DO inventories below the permanent pycnocline in the subtropical and subpolar regions were on the order of 64.3–161.5 and 61.2–142.1 g O2/m2, respectively (Table 1). Note that the DO inventories here are the average values of the amount of oxygen injected into the ocean interior by an episodic subduction event. These oxygen inventories may represent a significant source of ventilation to our study region.

Co-injection of oxygen below the permanent pycnocline by eddy pumping has not been given close consideration in previous studies, largely because it is less relevant for high-latitude, oxygen-rich waters. However, weak ocean ventilation in the tropical and subtropical mesopelagic zone is leading to declining oxygen concentrations (Karstensen et al., 2008; Oschlies et al., 2018; Robinson, 2019) and expansion of oxygen minimum zones in many regions of the oceans (Stramma et al., 2008; Breitburg et al., 2018). These episodic, dispersed subduction events likely represent a significant source of ventilation to help offset the de-oxygenation phenomenon and to support the expected climate-driven effects of increasing temperature on the metabolic oxygen demand of mesopelagic organisms (Wohlers et al., 2009). Enriched oxygen supplies into the mesopelagic zone will also influence remineralization rates of sinking particulate organic carbon in the ocean's twilight zone (Buesseler et al., 2007; Steinberg et al., 2008), affecting carbon sequestration timescales. Current global-scale biogeochemical models are too coarse to capture the effect that these sub-mesoscale processes may have on mesoscale oxygen variability (Takano et al., 2018) or to account for this additional oxygen supply. Overall, the intensity of these export events below the permanent pycnocline is remarkable, and they should be adequately considered in biogeochemical models.

Eddy-associated pumping is also one of several processes contributing to net global ocean carbon export (McGillicuddy, 2016; Boyd et al., 2019), but its importance is generally thought to be comparatively small because the relatively shallow penetration leads to shorter carbon sequestration times (Lévy et al., 2001; Karleskind et al., 2011a, b; Omand et al., 2015; Nagai et al., 2015; Boyd et al., 2019). That is, much of the carbon “exported” to the upper mesopelagic zone over spring and summer is returned to the atmosphere by deep winter mixing. At higher latitudes, where eddy pumping has been most studied, subduction must extend up to > ∼ 1000 m to reach below the permanent pycnocline (Palevsky and Doney, 2018; Boyd et al., 2019). However, the permanent pycnocline in the western North Pacific is much shallower – on the order of ∼ 300–450 dbar (Qiu and Huang, 1995; Feucher et al., 2019) – and most of the observed subduction signals here extended far below this depth (Tables 1 and S2). Thus although the subduction depths shown here are similar to those observed at higher latitudes, they represent much longer carbon sequestration timescales than those previously associated with eddy pumping (Boyd et al., 2019). As such, in addition to oxygen exports, the observed subduction patches seem to also transport large amounts of carbon into the ocean interior particularly below the permanent pycnocline. However, the lack of carbon measurements on the BGC-Argo floats used in this study impeded us from quantifying the carbon inventory within the subduction patches.

Because the BGC-Argo profiler only captures snapshots of subduction events, it is impossible to quantify the vertical transporting rate, which is needed to quantify export fluxes, of subduction from the BGC-Argo float data alone. Alternatively, the lifetime of subduction patches could be used to infer subduction rates, yet due to the dynamics and episodic characteristics of eddy subduction, there are currently no estimates of how much time these water masses maintain differentiated properties in the mesopelagic zone for, and there are numerous physical and biogeochemical processes influencing them.

The AOU, DO, and π anomalies were integrated within the study domain over the year to assess the extent of subduction in the western North Pacific (Fig. 6, Table 2). π anomalies were divided into negative or positive Δπ – i.e., π being greater or less than that in surrounding waters – which can suggest their modes of formation. Negative Δπ would correspond with the subduction of colder and less saline waters, such as along the edges of cyclonic eddies, while positive Δπ would be associated with the eddy pumping of warmer-core, anticyclonic eddies. The subduction patches were clearly dominated by negative Δπ, and more negative Δπ values corresponded with much larger ΔAOU and ΔDO values (Fig. 6, Table 2), suggesting they were associated with cyclonic, cold-core, upwelling-dominated eddies that have higher oxygen solubilities, higher nutrient flux to the surface, and thus higher plankton production. Conversely, the association of lower ΔAOU and ΔDO values with positive Δπ values would align with the lower oxygen solubility, nutrient flux, and plankton production expected for warmer-core, downwelling, anticyclonic eddies. Moreover, the majority of deep intrusions had negative Δπ values (Fig. 6, Table 2) consistent with colder waters following deeper isoclines. In contrast, anticyclonic eddies would push warm, lower-oxygen, and less-biomass-containing waters to shallower depths. These findings suggest that tracking the activity of cyclonic eddies in regions with shoaling permanent pycnoclines (Chelton et al., 2011; McGillicuddy, 2016) may be particularly important for quantifying these deeper subduction processes.

The findings here indicate that eddy-associated subduction is an important mechanism driving oxygen enrichment below the permanent pycnocline across the western subtropical Pacific region, particularly near the Kuroshio Extension (KE). Moreover, the abundance of these discrete, small-scale subduction events is almost certainly under-sampled in the BGC-Argo dataset. The frequency of this subduction is expected to vary as the KE oscillates between two dynamic states – quasi-stable and unstable – linked to the Pacific Decadal Oscillation (PDO) or North Pacific Gyre Oscillation (NPGO) (Di Lorenzo et al., 2008). When quasi-stable, the KE jet shifts north and generates less eddy activity than the unstable, highly meandering, southward KE jet, which reduces eastward transport and sharply increases eddy kinetic energy (Qiu and Chen, 2010; Lin et al., 2014). Superimposed on these KE oscillations has been an increase in the ratio of cyclonic to anticyclonic eddies associated with a climate-driven intensification of tropical storms in the western Pacific and the multidecadal trend of acceleration in Kuroshio flow (Zhang et al., 2020), suggesting that the importance of eddy-associated subduction processes in this region has been increasing and may continue to increase in the future. This linkage needs to be considered in designing future ocean observation programs and the modeling of global biogeochemical cycles to adequately capture the damping effects that eddy-associated subduction may exert on increasing atmospheric CO2 and de-oxygenation in the tropical and subtropical ocean.