Cold-water corals in the Subpolar North Atlantic Ocean exposed to aragonite undersaturation if the 2Â Â°C global warming target is not met

The net uptake of carbon dioxide (CO2) from the atmosphere is changing the ocean’s chemical state. Such changes, commonly known as ocean acidification, include a reduction in pH and the carbonate ion concentration ([CO3 ]), which in turn lowers oceanic saturation states (Ω) for calcium carbonate (CaCO3) minerals. The Ω values for aragonite (Ωaragonite; one of the main CaCO3 minerals formed by marine calcifying organisms) influence the calcification rate and geographic distribution of cold-water corals (CWCs), important for biodiversity. Here, high-quality measurements, collected on thirteen cruises along the same track during 1991–2018, are used to determine the long-term changes in Ωaragonite in the Irminger and Iceland Basins of the North Atlantic Ocean, providing the first trends of Ωaragonite in the deep waters of these basins. The entire water column of both basins showed significant negative Ωaragonite trends between − 0.0014 ± 0.0002 and − 0.0052 ± 0.0007 per year. The decrease in Ωaragonite in the intermediate waters, where nearly half of the CWC reefs of the study region are located, caused the Ωaragonite isolines to rapidly migrate upwards at a rate between 6 and 34 m per year. The main driver of the decline in Ωaragonite in the Irminger and Iceland Basins was the increase in anthropogenic CO2. But this was partially offset by increases in salinity (in Subpolar Mode Water), enhanced ventilation (in upper Labrador Sea Water), and increases in alkalinity (in classical Labrador Sea Water, cLSW; and overflow waters). We also found that water mass aging reinforced the Ωaragonite decrease in cLSW. Based on these Ωaragonite trends over the last three decades, we project that the entire water column of the Irminger and Iceland Basins will likely be undersaturated for aragonite when in equilibrium with an atmospheric mole fraction of CO2 (xCO2) of ~880 ppmv, corresponding to climate model projections for the end of the century based on the highest CO2 emission scenarios. However, intermediate waters will likely be aragonite undersaturated when in equilibrium with an atmospheric xCO2 exceeding ~630 ppmv, an xCO2 level slightly above that corresponding to 2 C global warming, thus exposing CWCs inhabiting the intermediate waters to undersaturation for aragonite.


Introduction
The global ocean has absorbed ~30% of the anthropogenic CO 2 (i.e., C ant ; CO 2 from human activities) released into the atmosphere since the industrial revolution (Friedlingstein et al., 2019;Gruber et al., 2019) and will likely sequester ~85% of C ant emissions on the time scales of several thousands of years (Archer et al., 2009). The global oceanic C ant uptake has mitigated atmospheric CO 2 concentration increases due to C ant emissions, and thereby climate change, but has also resulted in an unprecedented rapid long-term shift in the ocean's chemical state known as ocean acidification (OA; e.g., Raven et al., 2005;Gattuso et al., 2014;Williamson and Widdicombe, 2018;Doney et al., 2020). OA refers to the reduction in oceanic pH caused primarily by uptake of atmospheric CO 2 (IPCC, 2011), which in turn reduces the carbonate ion concentration ([CO 3 2− ]) and the ocean's capacity to take up CO 2 (i.e., oceanic buffer capacity or Revelle factor; Bates et al., 2014). Such changes in ocean chemistry may have direct and indirect consequences for marine life and ecosystems (particularly calcifying organisms; e.g., Riebesell et al., 2000;Kroeker et al., 2013;Pörtner et al., 2014;Mostofa et al., 2016;Doney et al., 2020).
A key parameter for understanding how changes in [CO 3 2− ] affect marine calcifying organisms is the seawater saturation state for calcium carbonate (CaCO 3 ) minerals (i.e., Ω; Eq. 1): where Ω is defined as the ratio between the product of the in situ concentrations of calcium ions ([Ca 2+ ]) and [CO 3 2− ] and their expected concentrations when the solution is in equilibrium with a particular CaCO 3 mineral phase such as calcite (Ω calcite ) or aragonite (Ω aragonite ) parameterized through the apparent solubility product (K sp ′ ). Seawater is in equilibrium with a particular CaCO 3 mineral phase when Ω = 1, supersaturated when Ω > 1, and undersaturated when Ω < 1 (e.g., Zeebe and Wolf-Gladrow, 2001). Since variability in [Ca 2+ ] in the open ocean is relatively small and very closely related to variations in salinity (Riley and Tongudai, 1967;Millero, 1995), Ω is primarily controlled by [CO 3 2− ] and K sp ′ . The Ω is lower in the deep ocean than in the upper ocean due to the dependence of K sp ′ on pressure and to a lesser extent on temperature, in combination with the total dissolved inorganic carbon (DIC) build-up from the remineralization of organic matter and the temperature effects on the CO 2 equilibrium constants. This fact, combined with the high [CO 3 2− ] of the upper ocean due to its higher pH, leads to generally supersaturated surface waters and undersaturated deep waters. Below the saturation horizon (Ω = 1), CaCO 3 becomes thermodynamically unstable and tends to dissolve. However, CaCO 3 can dissolve even though Ω > 1 due to biologically-mediated processes (Milliman et al., 1999). Cold-water coral (CWC) reefs are biodiversity hotspots, with the highest diversity occurring at 200-1000 m depth (Roberts et al., 2009). Aragonite is the main CaCO 3 mineral in CWC skeletons, for species such as Lophelia pertusa or Madrepora oculata (Roberts et al., 2009). The geographic distribution of CWCs, with over 95% of all presently known reefs located above the aragonite saturation horizon (ASH; Ω aragonite = 1) in preindustrial times, suggests that waters supersaturated for aragonite may be critical for their existence (Guinotte et al., 2006). In the North Atlantic Ocean, the Atlantic Meridional Overturning Circulation (AMOC) with deep-water formation in the Greenland, Iceland, Norwegian, and Labrador Seas causes the ASH to be the deepest in the global ocean at more than 2000 m depth (e.g., Feely et al., 2004;Jiang et al., 2015). A chemically-favorable deep-water environment allows a broad distribution of CWC communities in the North Atlantic Ocean ( Fig. 1; Guinotte et al., 2006;Tittensor et al., 2010). A corollary is that the AMOC drives a large deep-water penetration of C ant in the North Atlantic Ocean Khatiwala et al., 2013;DeVries, 2014), leading to a substantial shoaling of the ASH by 80-400 m since preindustrial times (Wallace, 2001;Á lvarez et al., 2003;Feely et al., 2004;Tanhua et al., 2007;Pérez et al., 2013Pérez et al., , 2018. By the end of the century, further uptake of C ant is projected to shoal the North Atlantic Ocean ASH depth by more than 2000 m under the IS92a 'business-asusual' scenario (Orr et al., 2005) and reduce the pH by more than 0.2 units over 21% of the seafloor area below 500 m depth in the North Atlantic sector based on RCP8.5 (Gehlen et al., 2014). Such shoaling of the ASH and pH reduction will alter the chemical environment of intermediate and deep waters and expose the majority of the CWC reefs to undersaturated (potentially corrosive) waters for aragonite. Therefore, CWC biomes are predicted to be one of the first deep marine ecosystems affected by OA (Raven et al., 2005;Guinotte et al., 2006;Turley et al., 2007;Maier et al., 2009;Roberts et al., 2009).
Previous studies have focused on the changes in Ω aragonite at local and global scales (e.g., Feely et al., 2004;Orr et al., 2005;Jiang et al., 2015), and the possible environmental drivers of these changes (e.g., Murata et al., 2015;Omar et al., 2016;Turi et al., 2016;Xu et al., 2020). However, none of them focused on changes in Ω aragonite and their drivers in the Irminger and Iceland Basins in the Subpolar North Atlantic Ocean (Fig. 1). The Subpolar North Atlantic Ocean may serve as a "natural laboratory" of the effect of OA on deep-water benthic ecosystems (Andersson et al., 2011) for all the reasons aforementioned (e.g., CWC biomes, substantial C ant uptake and deep-water penetration of C ant , and possible changes in the AMOC system). and lower (dark purple) limbs of the Atlantic Meridional Overturning Circulation in the study region, adapted from Daniault et al. (2016). White crosses, black triangles, blue diamonds, and red squares represent station locations across the Irminger and Iceland Basins (list of cruises in Table S1). Colored dots represent known cold-water coral reefs (Freiwald et al., 2017), where the colour represents their depth (colour scale; in m). BFZ stands for Bight Fracture Zone. Figure produced with Ocean Data View (Schlitzer, 2020). (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.) In this work, we describe long-term Ω aragonite trends in the entire water column of the Irminger and Iceland Basins during 1991-2018, using high-quality oceanic observations of the seawater CO 2 -carbonate system variables collected during thirteen cruises, providing the first published trends of Ω aragonite in the deep waters of the Irminger and Iceland Basins. García-Ibáñez et al. (2016) reported the observed reduction in pH and its drivers in the main water masses of these basins for 1991-2015. This work builds on that study by examining the consequences of the pH reduction on Ω aragonite and the physical and chemical drivers of the Ω aragonite changes, such as freshwater impacts and biogeochemical and transport/mixing impacts. We discuss our estimated Ω aragonite trends in the context of previously reported trends. Finally, we use the observed Ω aragonite trends to estimate the expected changes in Ω aragonite for future increases in atmospheric CO 2 . Those projections are used to infer when deep CWC communities inhabiting the Subpolar North Atlantic Ocean would be exposed to waters undersaturated for aragonite.

Dataset
Thirteen cruises along the same track across the Irminger and Iceland Basins were selected for our synthesis of the Ω aragonite temporal evolution, spanning 28 years from 1991 to 2018 ( Fig. 1; Table S1). Earlier data were extracted from the GLODAPv2.2020 data product, which provides bias-corrected, cruise-based, interior ocean data (Key et al., 2015;Olsen et al., 2020). This historical data was supplemented by hydrographic and seawater CO 2 chemistry data collected during the OVIDE 2018 cruise (Lherminier, 2018).
The accuracy of the GLODAPv2 data product is better than 0.005 for salinity, 2% for inorganic nutrients, 4 μmol kg − 1 for DIC and total alkalinity (TA), and 0.01 for pH . For the OVIDE cruise in 2018, the overall accuracy of nutrients was 1%, 2 μmol kg − 1 for TA, and 0.0014 for pH. For cruises with DIC measurements only, TA was computed according to Velo et al. (2013) procedures, using temperature, salinity, nitrate, phosphate, silicate, and oxygen as predictor parameters. The uncertainty of climatological TA is 5.4 μmol kg − 1 .
For cruises with TA and pH measurements, DIC was computed from TA and pH. These calculated DIC values have an associated average uncertainty of 6.3 μmol kg − 1 , calculated with the CO2SYS errors program , taking into account the uncertainties in the input variables and the default errors for carbonate and borate system constants. Ω aragonite was calculated (at in situ temperature and pressure) from temperature, salinity, DIC, and TA data (plus inorganic nutrients data). The uncertainly in calculated Ω aragonite was 0.077 on average computed with the CO2SYS errors program  and default errors for carbonate and borate system constants and the uncertainties in the input variables. In particular, uncertainly in calculated Ω aragonite was 0.078 for cruises with measurements of DIC and climatological TA, and 0.077 for cruises with measurements of TA and DIC estimated from TA and pH.
All calculations for the carbonate system were conducted using CO2SYS (Lewis and Wallace, 1998) for MATLAB (van Heuven et al., 2011), with the carbonic acid dissociation constants of Mehrbach et al. (1973) as reformulated on the total hydrogen scale by Lueker et al. (2000), the bisulfate dissociation constant of Dickson (1990), and the borate-to-salinity ratio of Uppström (1974).

Trend determination
The temporal changes in Ω aragonite and their potential drivers were determined in the main water-mass layers in the Irminger and Iceland Basins, namely Subpolar Mode Water (SPMW), upper and classical Labrador Sea Water (uLSW and cLSW, respectively), Iceland-Scotland Overflow Water (ISOW) and Denmark Strait Overflow Water (DSOW) (Fig. 2a).
Trends in Ω aragonite and their drivers were calculated using the areaweighted average values of interpolated bottle data at dbar resolution while considering the thickness of the layer and the distance between measurements (García-Ibáñez et al., 2016). For Ω aragonite , a pressuredependent variable (e.g., Dickson et al., 2007), a single reference pressure per water-mass layer (Table 1) was defined to remove pressure effects due to varying sampling strategies.
To determine the causes of the temporal changes in Ω aragonite , we assumed linearity to decompose the Ω aragonite change into that associated with each of the potential drivers, using a first-order Taylor-series deconvolution (Kwiatkowski and Orr, 2018): where ΔΩ is the temporal change in Ω aragonite while ΔT, ΔS, ΔTA, and ΔDIC are the corresponding temporal changes in in situ temperature (T), salinity (S), TA, and DIC. We separated the freshwater impacts on DIC and TA from the changes related to variations in ocean internal biogeochemical processes and transport/mixing (e.g., Keeling et al., 2004;Lovenduski et al., 2007;Hauri et al., 2013): where S 0 is the reference salinity, which is set to 35 (Normal Standard Seawater; Millero et al., 2008; very close to the average salinity observed in our dataset, S = 34.94). sDIC and sTA are the DIC and TA values normalized to a salinity of 35 and assuming a non-zero freshwater end-member (Friis et al., 2003). sTA, S, and sDIC correspond to the average values of sTA, S, and DIC during 1991-2018. Regrouping the terms related to the contribution from freshwater forcing on Ω aragonite gives the following equation: We separated ∆sDIC into C ant and natural DIC (C nat ): The sensitivity of Ω aragonite to each driver ( ∂Ω ∂Driver ) was estimated by allowing a change in only one driver according to their observed trend while setting the other drivers to their average value during 1991-2018.
C nat values were determined as the difference between DIC and C ant , with C ant values determined through the back-calculation technique φC T 0 (Vázquez-Rodríguez et al., 2009;Vázquez-Rodríguez et al., 2012;Ríos et al., 2012). This latter method has been used to describe C ant variability along the Atlantic Ocean and reported estimates were in agreement with those determined with other carbon-based and chlorofluorocarbon-based methods (Vázquez-Rodríguez et al., 2009;Fajar et al., 2015;Guallart et al., 2015;Fröb et al., 2018). Overall, the uncertainty of φC T 0 -derived C ant has been reported to be 5 μmol kg − 1 (Pérez et al., 2008;Vázquez-Rodríguez et al., 2009).
The temporal trends in Ω aragonite and their drivers were calculated considering the variability within each layer using a weighted simple linear regression analysis, where the weights were assigned as the inverse of the standard error of the mean (1/σ , where σ is the standard deviation of the samples within each layer and N is the number of samples within each layer) taking into account uncertainties in the input variables.
For further reference, the vertical sections of the mean in situ temperature, salinity, TA, DIC, C ant , and C nat during 1991-2018 are shown in Fig. S1.

Projection of future changes in aragonite saturation states (Ω aragonite )
The response of ocean chemistry to increasing atmospheric mole fraction of CO 2 (xCO 2 ) is non-linear due to the decrease in the oceanic buffer capacity (Revelle factor; e.g., Sarmiento et al., 1995). Therefore, OA effects on the chemical environment can only be approximated as linear for small perturbations, i.e., short periods. To overcome this issue and to project OA effects on Ω aragonite over long periods, the non-linear response of Ω aragonite to the increase in atmospheric CO 2 can be approximated as linear by performing the study in terms of logarithms. Projecting future changes in ocean chemistry relative to atmospheric CO 2 rather than time has the advantage of being independent of the IPCC Shared Socioeconomic Pathways (SSP). The relationship between Ω aragonite and atmospheric CO 2 also allows the computation of the preindustrial Ω aragonite (Table S4).
From our set of observations, we estimated future changes in Ω aragonite through linear extrapolation of the observed relationship between ln(Ω aragonite ) versus ln(xCO 2 ), using mean annual atmospheric xCO 2 data from Mauna Loa, Hawaii (www.esrl.noaa.gov/gmd/ccgg/tre nds/). Our projections assume a thermodynamic equilibrium between fCO 2 in the atmosphere and seawater. As such, they should be taken as an upper bound of the expected changes, primarily because the CO 2 in the ocean lags that in the atmosphere (Orr et al., 2005). Our projections are based on linear extrapolation of the observed trends in Ω aragonite during the 28 years from 1991 to 2018. These trends include variability imparted by inter-annual changes in remineralization rates, ventilation rates, and water mass circulation, which may not progress linearly.  (Schlitzer, 2020).

Changes in Ω aragonite and their potential drivers
Changes in Ω aragonite are important for understanding how OA might impact marine ecosystems. During 1991-2018, upper-ocean and intermediate waters of the Irminger and Iceland Basins were supersaturated for aragonite (Ω aragonite > 1), while waters below 2500 m were undersaturated for aragonite (Ω aragonite < 1) (Fig. 2a). This vertical gradient of Ω aragonite can be explained as a combination of the following processes: (a) the temperature and pressure dependence of K sp ′ (Mucci, 1983); (b) the changes in the TA / DIC ratio created by both biological activity (Carter et al., 2014) and the temperature-dependence of fCO 2 (Jiang et al., 2015); and (c) the temperature-dependence of the carbonic acid dissociation constants (Dickson and Millero, 1987). There were no appreciable differences in the Ω aragonite fields of the Irminger and Iceland Basins despite observed differences in pH (García-Ibáñez et al., 2016). A layer with relatively low pH was observed at ~500 m depth in the Iceland Basin (García-Ibáñez et al., 2016, their Fig. 2c), a feature that is not present in the Ω aragonite distribution due to the temperature-dependence of the chemical speciation of seawater CO 2 -carbonate chemistry species (e.g., Jiang et al., 2019, their Fig. S4). The speciation change combined with the change in K sp ′ with temperature and pressure (Mucci, 1983) buffers the observed change in pH in terms of Ω aragonite .
The study region is an area of low vertical Ω aragonite gradient compared to the tropics and subtropics (e.g., Jiang et al., 2015). The Ω aragonite values in the surface mixed layer (Ω aragonite ≈ 2) are lower than those in subtropical and tropical areas (Ω aragonite ≈ 4) (e.g., Jiang et al., 2015). The ISOW and DSOW layers are already undersaturated or close to undersaturation for aragonite, but they are supersaturated for calcite (another of the main CaCO 3 minerals formed by marine calcifying organisms; not shown), which is about 1.5 times less soluble than aragonite (Mucci, 1983).
Between 1991 and 2018, Ω aragonite below 100 m depth decreased by a mean value of − 0.11 ± 0.05 in both the Irminger and Iceland Basins (Fig. 2b). The upper SPMW layer and the recently ventilated uLSW layer of the Irminger Basin experienced the largest decrease in Ω aragonite , which in general was higher than 0.15. The deep-water Ω aragonite decreased by up to 0.1. The decrease in Ω aragonite during 1991-2018 caused the ASH to shoal by an average of 400 m in the Irminger Basin and by 250 m in the Iceland Basin. The increase in Ω aragonite observed in the upper 100 m is linked to the difference in surface temperature between 1991 and 2018 cruises, which took place in different seasons.
The Ω aragonite shows significant decreasing trends in all the watermass layers of both basins between 1991 and 2018 ( Fig. 3; Table S2), mainly driven by changes in sDIC, generally led by the anthropogenic component ( Fig. 4; Table S2), as was the case for the changes in pH (García-Ibáñez et al., 2016), as expected (Xue et al., 2020). Water-mass layers of both the Irminger and Iceland Basins gained DIC as atmospheric CO 2 increased, with convective processes providing an important pathway for transferring DIC to intermediate and deep layers (e.g., Pérez et al., 2008;Fröb et al., 2016). The main differences between the drivers for the changes in Ω aragonite and pH are found in the influence of temperature and TA. García-Ibáñez et al. (2016) found that the main driver counteracting the pH decrease was TA. In this work the influence of TA as a counteracting factor of OA is lower due to a different separation of the physical and biogeochemical drivers performed in this work, highlighting that the contribution of TA to pH changes was mainly through changes in salinity. Temperature changes have opposite effects on Ω aragonite and pH as a result of the temperature effect on the acid-base equilibrium of the CO 2 -carbonate system (Dickson and Millero, 1987), with increases in temperature causing a decrease in pH but an increase in Ω aragonite . Therefore, temperature increase dampens the decrease in Ω aragonite caused by OA but reinforces the decrease in pH.

Changes in Ω aragonite in mode waters (SPMW)
The upper layers, SPMWs, present the highest rate of Ω aragonite decrease for 1991-2018 ( Fig. 3; Table S2). Our rates of Ω aragonite change in the SPMW layers of the Irminger (− 0.0052 ± 0.0006 yr − 1 ) and Iceland (− 0.0049 ± 0.0015 yr − 1 ) Basins are comparable with those in surface waters of the Irminger Sea time-series (Fig. 5a, green upward-Fig. 3. Temporal evolution of the average aragonite saturation state (Ω aragonite ) in the water-mass layers (Table 1)  , where σ is the standard deviation of the samples within each layer, and N is the number of samples within each layer), taking into account uncertainties in the input variables. Trends ± standard error of the estimate (in yr − 1 ) are also given. All trends are statistically significant at the 99% level (p-value <0.01).
pointing triangle, IrS) and the Iceland Sea time-series (Fig. 5a, navy star), as well as at 200-500 m depth in the Iceland Sea and Norwegian Basin (Fig. 5a, white squares, IcS and NB, respectively), and in North Atlantic Central Waters of the North-East Atlantic Ocean (Fig. 5a, lightblue downward-pointing triangle, NACW). However, our rate of Ω aragonite decrease of the SPMW layers is half those at 0-200 m depth in the Iceland Sea and Norwegian Basin (Fig. 5a, yellow squares). Note that the water masses in the Nordic Seas (Iceland Sea and Norwegian Basin) are not SPMW but Arctic Ocean waters (Olafsson et al., 2009), which have a higher rate of acidification (Fransner et al., 2020). Besides, the changes in Ω aragonite for upper-ocean waters in the North Atlantic Subpolar Gyre region are subject to significant variability depending on the seasons and periods considered (Leseurre et al., 2020, their Fig. 4).
Comparing our results with those for time-series stations in the Atlantic Ocean outside our study region, our trends for mode waters are around half those reported for the surface layer at the Subtropical Atlantic time-series stations ESTOC (European Station for Time series in the Ocean at the Canary Islands; Fig. 5a, green upward-pointing triangle) and BATS (Bermuda Atlantic Time-series Study; Fig. 5a, pink leftpointing triangle), but comparable with those at CARIACO (CArbon Retention In A Colored Ocean; Fig. 5a, green upward-pointing triangle).
Our Ω aragonite rates of the SPMW layers are slightly lower than those found in surface waters of the Central North Pacific time-series station HOT (Hawaii Ocean Time-series) and of the subpolar South Pacific Munida Time-Series Transect (Fig. 5a, green upward-pointing triangles), but more than four times lower than in the upper 400 m of the subtropical South Pacific (Fig. 5a, black asterisk).
The differences in the rates of Ω aragonite decrease in our study area and in the Nordic Seas (Iceland Sea time-series) and subtropical and tropical regions are expected due to the temperature-dependence of Ω aragonite (e.g., Carter et al., 2014;Takahashi et al., 2014, their Fig. 17;Jiang et al., 2015). The temperature-dependence of Ω aragonite originates from the temperature-dependence of K sp ′ (Mucci, 1983) and the temperature-dependence of [CO 3 2− ], which results from the temperature-dependence of both the carbonic acid dissociation constants (Dickson and Millero, 1987) and fCO 2 that influences the TA / DIC ratio (Jiang et al., 2015). The rates of Ω aragonite decrease in the SPMW layers are mainly driven by changes in sDIC ( Fig. 4; Table S2). ΔΩ sDIC values are largely caused by C ant uptake (solubility pump). The increase in salinity partly counteracts the effect of the sDIC, reducing the Ω aragonite decrease by up to 22%. The salinization of SPMW  likely results from the transport of a higher proportion of saline subtropical waters into the North Atlantic Subpolar Gyre since the mid-1990s (e.g., Flatau et al., 2003;Häkkinen and Rhines, 2004;Böning et al., 2006;Thierry et al., 2008). The impacts of temperature and sTA on the Ω aragonite decrease of the SPMW layers are negligible. The addition of the Ω aragonite changes due to each driver (ΔΩ sum ) reproduces ~80% of the observed Ω aragonite trends (ΔΩ obs ) in the SPMW layers, with no significant difference between ΔΩ sum and ΔΩ obs . The ΔΩ sum does not fully reproduce ΔΩ obs because of temporal variability in Ω aragonite (Fig. 3) as well as in the drivers of ΔΩ aragonite (Fig. S2), and because of the simplification of the first-order Taylor expansion.

Changes in Ω aragonite in intermediate waters (uLSW and cLSW)
During 1991-2018 the Ω aragonite of uLSW decreased by − 0.0036 ± 0.0004 yr − 1 in the Irminger Basin and by − 0.0025 ± 0.0002 yr − 1 in the Iceland Basin ( Fig. 3; Table S2). Our estimates of ΔΩ aragonite for the uLSW layers are twice those reported for water at 1000 m depth for the entire North Atlantic Ocean (Fig. 5b, orange right-pointing triangle). However, the rate of the Ω aragonite decrease of uLSW is lower than that of Arctic Intermediate Waters (500-1000 m depth) in the Iceland Sea and Norwegian Basin (Fig. 5b, yellow squares, IcS and NB, respectively). The rate of the Ω aragonite decrease of uLSW is also lower than that of Mediterranean Water in the North-East Atlantic Ocean (Fig. 5b, light-blue downward-pointing triangle, MW).
The Ω aragonite decrease of the uLSW layers is mainly driven by C ant uptake ( Fig. 4; Table S2), with changes in C nat significantly counteracting the decrease in Ω aragonite in the Iceland Basin. The increase in  (Table 1) of the Irminger (a) and Iceland (b) Basins: in situ temperature (∆Ω T ; green bars), salinity (∆Ω S ; navy bars), salinity-normalized total alkalinity (∆Ω sTA ; light-blue bars), salinity-normalized total dissolved inorganic carbon (∆Ω sDIC ; red bars), salinity-normalized anthropogenic CO 2 (∆Ω sCant ; orange bars), and salinity-normalized natural DIC (∆Ω sCnat ; yellow bars). Note that sDIC = sC ant + sC nat and, therefore, ∆Ω sDIC = ∆Ω sCant + ∆Ω sCnat (Section 2.2). The addition of the Ω aragonite changes from each driver is also given (∆Ω sum ; white bars). Trends are reported in (x 10 − 3 ) yr − 1 , and the error bars represent the standard error of the estimate. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.) Ω aragonite of uLSW due to C nat indicates that the uLSW layer was occasionally ventilated (e.g., Våge et al., 2009;Yashayaev andLoder, 2009, 2016;Kieke and Yashayaev, 2015;Fröb et al., 2016). The impacts of temperature, salinity, and sTA on Ω aragonite trends of uLSW are small-tonegligible (with the counteracting effect of sTA on ΔΩ aragonite of uLSW as small as 2-6%).
For the cLSW layers, the Ω aragonite decreased by − 0.0017 ± 0.0002 yr − 1 in the Iceland Basin and by − 0.0023 ± 0.0002 yr − 1 in the Irminger Basin for 1991-2018 ( Fig. 3; Table S2). Our estimates of ΔΩ aragonite in the cLSW layers are comparable with those in the LSW of the North-East Atlantic Ocean (Fig. 5c, light-blue downward-pointing triangle, LSW). The rates of decrease in Ω aragonite of the cLSW layers are double the rate reported for waters deeper than 1500 m at the Iceland Sea time-series station (Fig. 5c, navy star), but similar to those at 1000-2000 m depth in the Iceland Sea (Fig. 5c, yellow square, IcS). However, the rate of decrease in Ω aragonite of cLSW is half that of the Arctic Deep Waters (1000-2000 m depth) of the Norwegian Basin (Fig. 5c, yellow square,  NB).
The DIC changes of the cLSW layers are the main contributors to the observed Ω aragonite decrease ( Fig. 4; Table S2), with changes in C nat reinforcing the decrease in Ω aragonite caused by C ant uptake in the Irminger Basin. The higher influence of C nat in the Ω aragonite decline of the cLSW layers compared to other layers is likely the result of the aging of this water mass (accumulation of DIC from remineralized organic matter) after its last strong formation event in the mid-1990s (e.g., Lazier et al., 2002;Azetsu-Scott et al., 2003;Kieke et al., 2007;Yashayaev, 2007). The formation history of cLSW can also explain the attenuation in the Ω aragonite decrease through changes in temperature and salinity. Enhanced ventilation towards the mid-to-late 1980s fostered the rapid formation of a large volume of cLSW in the Labrador Sea and transported cold waters from the surface into intermediate waters. The shutdown of the cLSW production in the mid-1990s (Lazier et al., 2002;Kieke et al., 2006;Rhein et al., 2007; caused the warming and salinization of this layer due to lateral mixing with surrounding waters, such as SPMWs, which have experienced an increase in salinity . These effects were more remarkable for the cLSW layer in the Irminger Basin than in the Iceland Basin due to the proximity of the former to the Labrador Sea. Therefore, the OA signal of the cLSW layers is reinforced by a reduction in ventilation (Chen et al., 2017).
Since CWCs mainly inhabit the LSW layers, we estimated the upward migration of the Ω aragonite isolines that may ultimately affect the metabolic status of aragonitic CWCs (e.g., Thresher et al., 2011;Hennige et al., 2014). The decrease in Ω aragonite of the uLSW layers led to a shoaling of the Ω aragonite isolines at a rate of 34 m yr − 1 in the Irminger Basin and 23 m yr − 1 in the Iceland Basin. For the cLSW layers, the Ω aragonite isolines shoaled at a rate of 10 m yr − 1 in the Irminger Basin and 6 m yr − 1 in the Iceland Basin. The 10-15 m yr − 1 shoaling rate of the ASH estimated for the center of the Irminger Sea for 1991-2016 by Pérez et al. (2018) are in agreement with ours for the cLSW in the Irminger and  (Table S3). Red dots represent ∆Ω aragonite for 1991-2018 in the water-mass layers (Table 1)  Iceland Basins. In comparison, the 6 m yr − 1 shoaling of the Ω aragonite isolines of the cLSW layer of the Iceland Basin is slightly higher than the 4 m yr − 1 shoaling trend reported for the Iceland Sea for 1985-2008 (Olafsson et al., 2009), and the 5.2 m yr − 1 shoaling trend for the subtropical South Pacific Ocean for 1994-2009 at 400 m depth (Murata et al., 2015). The shoaling of the Ω aragonite isolines of the uLSW layers is 12-17 times higher than the 2.0 ± 0.7 m yr − 1 estimated for the Ω aragonite isoline of 1.3 (contour within the uLSW; Fig. 2) in the high latitudes of the Southern Hemisphere for 1992-2011 (Williams et al., 2015). The cause of the higher shoaling rates of Ω aragonite in our analysis compared to those reported by Williams et al. (2015) is three-fold. First, Williams et al. (2015) evaluated the changes in Ω aragonite between only two years (1992 versus 2011 or 2005 versus 2011), being, therefore, more sensitive to inter-annual variability than our trend analysis. Second, Williams et al. (2015) only observed an anthropogenic signal in the upper 100 m, while the changes in Ω aragonite in the waters below the mixed layer result from either a slowdown in circulation or decreased ventilation. Therefore, the Pacific sector of the Southern Ocean was less affected by the anthropogenic signal than our study region, the Subpolar North Atlantic, causing the shoaling rates in the Pacific sector of the Southern Ocean to be significantly smaller than those in the Subpolar North Atlantic. Finally, when comparing shoaling rates of Ω aragonite one should note that they are a function of both the OA rates and the vertical gradient in the Ω aragonite ; therefore, similar OA signals could generate different shoaling rates depending on the Ω aragonite gradient.

Changes in Ω aragonite in deep waters (ISOW and DSOW)
This study provides the first published trends of Ω aragonite in the deep waters of the Irminger and Iceland Basins. During 1991-2018, Ω aragonite decreased by − 0.0014 ± 0.0002 yr − 1 in the ISOW layer in the Iceland Basin, and by − 0.0020 ± 0.0002 yr − 1 and − 0.0021 ± 0.0003 yr − 1 in the ISOW and DSOW layers of the Irminger Basin, respectively ( Fig. 3; Table S2). The entrainment of intermediate waters into the overflow layers (van Aken, 2000;Dickson et al., 2002;Sarafanov et al., 2010) is likely to transfer the OA signal from intermediate to deep waters, hence, making the Ω aragonite trends of the ISOW and DSOW layers similar to those of the cLSW layers (Fig. 5c,d). The entrainment of intermediate waters into the overflow waters may explain why we observe a statistically significant rate of Ω aragonite decrease in the deep waters of the Irminger and Iceland Basins while no significant rate of Ω aragonite decrease was found in the deep waters of the Norwegian Basin (Fransner et al., 2020).
Changes in DIC and TA constitute the primary drivers of the temporal evolution of Ω aragonite in the deep waters ( Fig. 4; Table S2), with changes in sTA significantly offsetting the reduction in Ω aragonite caused by the increase in sDIC for the Irminger Basin. Changes in temperature and salinity in the overflow waters did not significantly affect ΔΩ aragonite . The substantial influence of C ant on the Ω aragonite decrease in the overflow layers (DSOW and ISOW) stands out, with significant parallel pH decreases in these water-mass layers (García-Ibáñez et al., 2016). The Ω aragonite decrease due to C nat for ISOW and DSOW could derive from the increase in TA from excess carbonate mineral dissolution and/or from the mixing with cLSW and relatively old North Atlantic Deep Water, with high DIC accumulation from organic matter remineralization (Fig. S1e). The increase in sTA in the overflow waters, counteracting the decrease in Ω aragonite , may have been transmitted from SPMW and LSW through entrainment Sarafanov et al., 2010;Hansen et al., 2016).

Projected future changes in Ω aragonite
Our projections of future Ω aragonite relative to atmospheric xCO 2 indicate that the entire water column of the Irminger and Iceland Basins will likely be undersaturated for aragonite by the time xCO 2 reaches 876 ± 83 ppmv (approximately twice the present atmospheric xCO 2 ; Fig. 6; Table S4). Atmospheric xCO 2 will likely reach 876 ppmv before the end of the century according to SSP5 (Riahi et al., 2017), which is the SSP with the highest emissions (the only SSP that results in a radiative forcing pathway close to the highest Representative Concentration Pathway RCP8.5; Kriegler et al., 2017) and the one we have followed most closely (e.g., Sanford et al., 2014). Numerical models based on RCP8.5 also project aragonite undersaturation in the North Atlantic north of 50 • N by 2100 (Feely et al., 2009;Matear and Lenton, 2014). The intermediate waters of the Irminger and Iceland Basins will become undersaturated for aragonite much sooner. The uLSW layers will likely become undersaturated for aragonite when in equilibrium with atmospheric xCO 2 of 757-808 ppmv ( Fig. 6b; Table S4). The cLSW layers will be close to undersaturation for aragonite when in equilibrium with an atmospheric xCO 2 of 558 ppmv (an xCO 2 level corresponding to 2 • C global warming for the SSP5 Baseline Marker; Riahi et al., 2017), reaching aragonite undersaturation when atmospheric xCO 2 reaches 628-669 ppmv ( Fig. 6c; Table S4). The ISOW layer of the Irminger Basin will be undersaturated for aragonite when in equilibrium with an atmospheric xCO 2 of 463 ppmv ( Fig. 6d; Table S4), 52 ppmv above the current atmospheric xCO 2 level.
Aragonite undersaturation of the cLSW and ISOW layers will create a chemically unfavorable state for CWCs that inhabit these depths ( Fig. 1; e.g., Maier et al., 2009;Ragazzola et al., 2012). Physiological acclimation to OA (higher levels of CO 2 ) was observed in Lophelia pertusa during long-term incubations (e.g., Form and Riebesell, 2012;Hennige et al., 2015). But the exposed bases of the reefs become more susceptible to bio-erosion and mechanical damage when exposed to higher levels of CO 2 , with a corresponding reduction in the structural integrity of the reef (e.g., Hennige et al., 2015) that could ultimately cause reef collapse. Therefore, the potential increase in the dissolution of the dead bases of the reefs could lead to a net loss of CWC reefs, rather than the reduction in calcification rates of the live coral (Hennige et al., 2015). The systemic weakening and ultimate loss of the foundation framework from OA would prevent future reef and mound growth and lead to an ecosystemscale habitat loss for existing CWC reefs (Hennige et al., 2020).

Summary
The gradual reduction over time of Ω aragonite in the Subpolar North Atlantic Ocean was assessed from observations in the Irminger and Iceland Basins spanning the last three decades . Upperocean and intermediate waters of the Iceland and Irminger Basins were supersaturated for the aragonite form of CaCO 3 , while waters below 2500 m depth were undersaturated. The entire water column of both basins showed statistically significant decreasing Ω aragonite trends, with the greatest decrease in upper-ocean waters. The Ω aragonite decrease in intermediate waters, which 44% of the CWC reefs of the study region inhabit, has caused the Ω aragonite isolines to shoal at a rate of 6-34 m yr − 1 in the Irminger and Iceland Basins. We attribute the Ω aragonite changes mainly to the oceanic C ant uptake, reinforced in the cLSW layers by increases in net remineralization as ventilation rates decreased and mean water age increased. The increase in salinity (and thus in [Ca 2+ ]) in the SPMW layers reduced the Ω aragonite decrease resulting from the increase in DIC. For the uLSW layers, its recent ventilation counteracted the decrease in Ω aragonite from C ant uptake. For cLSW and deep waters, the Ω aragonite decreases caused by increases in DIC were offset by increases in TA. Taking advantage of the observed Ω aragonite change for 1991-2018, we inferred its future change for increases in atmospheric CO 2 . Our projections suggest that the entire water column of both basins would be undersaturated for aragonite when in equilibrium with an atmospheric xCO 2 of ~880 ppmv (which will be reached by the end of the century according to the SSP5 climate scenario), which is in agreement with projections from numerical models. However, intermediate layers between 1000 and 2000 m depth would reach aragonite undersaturation more rapidly, when in equilibrium with an atmospheric xCO 2 exceeding ~630 ppmv, an xCO 2 level slightly above that corresponding to 2 • C global warming. The future aragonite undersaturation of the intermediate layers is likely to affect the distribution, calcification rates, and structural stability of the CWC reefs, important hotspots of biodiversity in the North Atlantic Ocean.

Author contributions
MIGI wrote the manuscript and prepared the figures. MIGI, MF, and AV contributed to the acquisition of data. MIGI, NRB, DCEB, MF, and AV discussed the results and reviewed the manuscript and supporting information.

Data availability statement
Data supporting the conclusions of this article are available through the authors and GLODAPv2.2020 .

Declaration of Competing Interest
The authors declare no competing financial interest.

Acknowledgements
We are grateful to the captains, crew, and researchers who contributed to the acquisition and processing of hydrographic and ocean carbon data. The research leading to these results was supported through the EU FP7 project CARBOCHANGE "Changes in carbon uptake and emissions by oceans in a changing climate", which received funding from the European Commission's Seventh Framework Programme under grant agreement No 264879. For this work, M.I. García-Ibáñez and A. Velo were supported by the BOCATS2 Project (PID2019-104279GB-C21) cofunded by the Spanish Government and the Fondo Europeo de Desarrollo Regional (FEDER). M.I. Garcia-Ibanez was also supported by the UK's Natural Environment Research Council CUSTARD grant [NE/ P021263/1], part of the RoSES programme. N.R. Bates was funded by National Science Foundation award (OCE-1258622). M. Fontela was funded by Portuguese national funds from the FCT-Foundation for Science and Technology through project UID/Multi/04326/2019 and CEECINST/00114/2018. This project has also received funding from the European Union's Horizon 2020 research and innovation programme under grant agreement No 820989 (project COMFORT, Our common future ocean in the Earth system -Quantifying coupled cycles of carbon, oxygen, and nutrients for determining and achieving safe operating spaces with respect to tipping points). The work reflects only the author's/authors' view; the European Commission and their executive agency are not responsible for any use that may be made of the information the work contains. We thank Fiz F. Pérez for his constructive and useful discussion that greatly helped to improve the paper. We thank the two anonymous reviewers whose insightful comments and suggestions helped improve and clarify this manuscript.

Appendix A. Supplementary data
Supplementary data to this article can be found online at https://doi. org/10.1016/j.gloplacha.2021.103480.