Biogeosciences Observed acidification trends in North Atlantic water masses

The lack of observational pH data has made it difficult to assess recent rates of ocean acidification, particularly in the high latitudes. Here we present a time series that spans over 27 yr (1981–2008) of high-quality carbon system measurements in the North Atlantic, which comprises fourteen cruises and covers the important water mass formation areas of the Irminger and Iceland Basins. We provide direct quantification of acidification rates in upper and intermediate North Atlantic waters. The highest rates were associated with surface waters and with Labrador Sea Water (LSW). The Subarctic Intermediate and Subpolar Mode Waters (SAIW and SPMW) showed acidification rates of −0.0019±0.0001 and−0.0012± 0.0002 yr−1, respectively. The deep convection activity in the North Atlantic Subpolar Gyre injects surface waters loaded with anthropogenic CO 2 into lower layers, provoking the remarkable acidification rate observed for LSW in the Iceland Basin ( −0.0016± 0.0002 yr−1). An extrapolation of the observed linear acidification trends suggests that the pH of LSW could drop 0.45 units with respect to pre-industrial levels by the time atmospheric CO 2 concentrations reach ∼ 775 ppm. Under circulation conditions and evolution of CO2 emission rates similar to those of the last three decades, by the time atmospheric CO 2 reaches 550 ppm, an aragonite undersaturation state could be reached in the cLSW of the Iceland Basin, earlier than surface SPMW.


Introduction
The ocean acidification due to the increasing atmospheric CO 2 is a well-known fact (Bates et al., 2012;Doney et al., 2009;Raven, 2005) but the direct pH observations are sparse (Byrne et al., 2010;Tittensor et al., 2010;Wootton et al., 2008). Roughly 20-35 % of the anthropogenic CO 2 (C ant ) emissions are absorbed by the oceans (Khatiwala et al., 2009) and mitigate global warming. Since the beginning of the Industrial Revolution, the sea surface has seen a 30 % increase in hydrogen ion concentrations [H + ] (Caldeira and Wickett, 2005;Raven, 2005). The current acidification episode is occurring ∼ 100 times faster than any other acidity change in the last 50 million yr of Earth's history (Pelejero et al., 2010).
Ocean acidification is thought to be the onset for a number of cascading effects throughout marine ecosystems that may leave no time for many organisms to adapt, especially in the case of calcareous organisms (Feely et al., 2008;. It causes a combination of contrasted impacts on the marine environment, from reproductive larval survivorship and growth-related issues in several taxa  to the reduction of seawater's sound absorption coefficient (Ilyina et al., 2009). Cold-water scleractinian corals dwelling in intermediate and deep North Atlantic (NA) waters are particularly vulnerable to acidification (Guinotte et al., 2006;Raven, 2005).
The North Atlantic Subpolar Gyre (NASPG) is an important formation area of mode waters. These waters, formed in deep winter mixed layers, are identified by nearly uniform properties in the vertical, near the top of the permanent pycnocline (Thierry et al., 2008). The process of transformation of the warm, saline subtropical waters into intermediate and deep waters in the NASPG (McCartney and Talley, 1982;Read, 2001) results in several varieties of Subpolar Mode Water (SPMW) distributed around the gyre. The Labrador Sea Water (LSW), the densest variety of SPMW, is one of the thickest water masses in the NA and one of the main components of the lower limb of the Meridional Overturning

M. Vázquez-Rodríguez et al.: Observed acidification trends in North Atlantic water masses
Circulation (Thierry et al., 2008). The LSW has high contents of chlorofluorocarbons (CFCs) and anthropogenic carbon due to ventilation processes (Azetsu-Scott et al., 2003;Pérez et al., 2010). Thus, it is expected that those water masses will suffer changes in [H + ].
There are relatively few places where the carbon system has been surveyed thoroughly enough to generate a comprehensive database that can be used in the assessment of ocean acidification and its environmental impacts (Wootton et al., 2008). Several past and future pH projections have been proposed from Ocean General Circulation Models (GCMs) and model data (Orr et al., 2005), but in situ measurements documenting the evolution of ocean pH over time are limited (Wootton et al., 2008). The present work examines the temporal variability of pH in the main water masses of the North Atlantic using direct observations. Here we have gathered the available high-quality carbon system data that covered the NASPG between 1981 and 2008 ( Fig. 1a) to study the decadal acidification rates of the main North Atlantic water masses (Fig. 1b).

Dataset
A total of fourteen cruises with high-quality carbon system measurements were selected to follow the temporal evolution of pH in the North Atlantic (Fig. 1a, Table 1). The used cruise data can be accessed at CARINA site http://store. pangaea.de/Projects/CARBOOCEAN/carina/index.htm. We also used the climatological WOA05 data available at http: //www.nodc.noaa.gov/OC5/WOA05/pr woa05.html.
Over time different analytical procedures were used to measure pH, so different adjustments and corrections were applied to the raw data (Table 1) to create the pH dataset used in this study. The pH measurements in the database were determined either potentiometrically (using pH electrodes; Dickson, 1993) or, more commonly, with a spectrophotometric method that used m-cresol purple as a pH indicator in either scanning or diode array spectrophotometers (Clayton and Byrne, 1993). The spectrophotometric pH at sea has a reported analytical precision of approximately 0.0004 pH units (Clayton and Byrne, 1993). Periodic checks of the pH measurement precision with Certified Reference Material (CRM) during the FOUREX and OVIDE cruises indicated a precision better than 0.002 pH units. However, the uncertainties inherent to the use of equimolal tris buffers to obtain the constants used by Clayton and Byrne (1993) suggest a final uncertainty for spectrophotometric pH measurements of ∼ 0.004, as reported in Table 1. This value is consistent with our comparisons between the measured and the calculated pH using the acid dissociation constants in Dickson and Millero (1987) and reported in Clayton and Byrne (1993). All pH measurements that had not been originally reported  1. (a) shows the study area and selected cruises. The thick black lines delimit the Irminger, Iceland and Eastern North Atlantic (ENA) Basins; (b) shows the main NASPG water masses considered for this study over the salinity distribution of the OVIDE 2004 section, which gives representative coverage of the NASPG. The vertical thick white lines delimit the Irminger, Iceland and ENA Basins. The isopycnals (horizontal white lines; σ θ , in kg m −3 ) are the ones listed in Sect. 2.2 and Table 2. The water mass acronyms stand for: SAIW = Subarctic Intermediate Water; LSW = Labrador Seawater; NADW = North Atlantic Deep Water; SPMW = Subpolar Mode Water; NACW = North Atlantic Central Water; MW = Mediterranean Water. The lowercase letters "c", "u" and "l" before the acronyms denote, respectively, "classical", "upper" and "lower" water mass types/branches. in the seawater scale (pH SWS25 ; Millero, 2007) were converted to it from either the total or the NBS pH scale using the corresponding acid dissociation constants (Dickson and Millero, 1987), following the CARINA database second quality control recommendations for pH data scale unification and cruise adjustments . The acid dissociation constants of HF or HSO − 4 (Millero, 2007) were used to convert pH values originally reported in the total scale (those measured spectrophotometrically; Table 1) to the SWS scale. The pHs measured potentiometrically were all reported on the NBS scale and were converted to the SWS Table 1. List of selected North Atlantic cruises (Fig. 1a). Acronyms denote: P.I. = principal investigator; S = variable measured with spectrophotometric techniques; P = variable measured with potentiometric techniques; Calc = pH calculated from C T and A T using the thermodynamic equations of the carbon system (Dickson et al., 2007) and a set of carbon dioxide dissociation constants (Dickson and Millero, 1987). Uncert. = analytical uncertainties of spectrophotometric, potentiometric, and calculated pH. Adjustments of C T , A T (in µmol kg −1 ) and pH suggested from several a posteriori crossover analysis (Velo et al., 2009Pierrot et al., 2010)  and c below. b C T analysed with SOMMA (Johnson et al., 1993) and calibrated with CRMs given analytical accuracy ±2 µmol kg −1 , except in 1981 TTO cruise that was determined potentiometrically (Bradshaw et al., 1981) without CRMs with accuracy ±4 µ mol kg −1 . c A T analysed with potentiometric titration and determined by developing either a full titration curve (Millero et al., 1993;Dickson et al., 2007) or from single point titration (Pérez and Fraga, 1987;Mintrop et al., 2002). Analytical accuracy ±4 µ mol kg −1 .
scale as specified in Pérez and Fraga (1987). Some of the cruises listed in Table 1 did not perform direct pH measurements, but obtained total alkalinity (A T ) and dissolved inorganic carbon (C T ) data instead. In such cases the pH values were calculated in the SWS scale from A T and C T data using the thermodynamic equations of the carbon system (Dickson et al., 2007) and a set of CO 2 dissociation constants (Dickson and Millero, 1987). An uncertainty of ±0.006 was estimated for these calculated pHs by random propagation of the reported A T and C T uncertainties (±4 and ±2 µmol kg −1 , respectively). The exception to the latter is the TTO cruise, which had a reported C T uncertainty of ±4 µmol kg −1 that caused the estimated uncertainty for calculated pH to be slightly higher (±0.008; Table 1). During the A16N cruise pH was determined spectrophotometrically, but the spatial resolution was not as good as in the case of C T and A T , so we used pH values calculated from C T and A T for this cruise instead. The AR7E and A01E cruises (Fig. 1a) had a comprehensive amount of C T measurements, yet very little potentiometric A T data. Given the scarcity of A T data, the equation A T = S/35 · (2294.7+1.37 [Si(OH) 4 ]) (R 2 = 0.97; [Si(OH) 4 ] refers to silicate concentration) given by Pérez et al. (2010) was applied to the AR7E and A01E datasets to generate A T values at the sampling depths of measured C T . The pH was then calculated from C T and A T data as mentioned above.

pH data analysis
The dataset spans 27 yr ) and a wide spatial coverage of the study area ( Fig. 1a; Table 1) that was divided into three basins: the Irminger, Iceland and East North Atlantic (ENA). These three basins and their geographical boundaries have also been defined and used in Pérez et al. (2010). The Irminger Basin boundaries are defined by the main axis of the Reykjanes Ridge and the east coast of Greenland. The Iceland Basin is defined as the region bounded between the main axis of the Reykjanes Ridge and the line joining the Eriador Seamount and the Faroe Islands. The ENA Basin extends south from the Eriador-Faroe Line over the Rockall Trough, the Porcupine Bank, and the Biscay and Iberian Basins (Fig. 1).
In order to evaluate the temporal variability of pH in the water masses of the North Atlantic, the water column was divided into five layers delimited by potential density (σ θ isopycnals) for each region (Fig. 1b). To determine the isopycnal boundaries of the North Atlantic Deep Water (NADW) we followed Lherminier et al. (2010), who established different layers by potential density intervals on the basis of the hydrographic properties and circulation of the different water masses along the OVIDE section. They discriminate between the two components of NADW: the lower NADW (lNADW), spreading from the bottom to σ θ = 45.84, and the upper NADW (uNADW) that spreads in 5220 M. Vázquez-Rodríguez et al.: Observed acidification trends in North Atlantic water masses the density range 36.94 < σ θ < 45.84. We took the density range 37 < σ θ < 45.84, which is almost identical, because the isopycnal σ θ = 37 seemed to delimit better the deepest boundary of the cLSW core (coincident with the uNADW upper density limit) in the Iceland and ENA Basins. For the spreading of LSW in the ENA Basin, the density range selected (32.35 < σ θ < 37) is very close to the one in Lherminier et al. (2010). Following Ríos et al. (1992) the Mediterranean Water (MW) layer is delimited by 27.2 < σ θ < 32.35 and the North Atlantic Central Water (NACW) layer is established from surface to σ θ < 32.35, according to the spreading of these water masses in the zone. For the Irminger and Iceland Basins, the potential density limits were established following Kieke et al. (2007) and Yashayaev et al. (2008). So for the Iceland Basin the layer of SPMW is found between 100 m and σ θ = 27.6. The upper and classical LSW (uLSW and cLSW) spread in the density ranges of 27.68 < σ θ < 27.76 and 27.76 < σ θ < 27.81, respectively. In the Irminger Basin, the Sub Arctic Intermediate Water (SAIW) spreads between 100 m and σ θ = 27.68. The uLSW and cLSW are found between 27.68 < σ θ < 27.76 and 27.76 < σ θ < 27.81, respectively. The North Atlantic Deep Water (NADW, which includes contributions from ISOW) is delimited by 27.81 < σ θ < 27.88, and the Denmark Strait Overflow Water (DSOW) by σ θ > 27.88 (Fig. 1b).

Basin normalization of average pH SWS25
The average pH SWS25 was obtained for each layer and year in the three basins, following the averaging and "basinreferencing" methodology that Pérez et al. (2008Pérez et al. ( , 2010 and Ríos et al. (2012) used for C ant . The spatial coverage of cruises over years is variable and this can cause significant biases in the observed average layer properties in each year. These small differences can potentially introduce spatial biases in the average pH SWS25 due to different ventilation stages and rates of each water mass. Therefore, for each basin the pH SWS25 was normalized to better represent the pH SWS25 in each considered layer of the basin (Fig. 1) by adding a new term named pH SWS25-BA (where "BA" stands for basin average). This term represents the deviation of pH SWS25 (average from cruise data) from the pH SWS25 basin average (pH SWS25-BA ).
The pH SWS25-BA term was computed from cruise data and expressed as individual correction elements for each cruise and layer in the three basins (Table 2) as follows: where c stands for "cruise" and subscript i denotes "property" (1 = Si(OH) 4 ; 2 = AOU; 3 = θ; 4 = S). The X c i and X WOA05 i terms are, respectively, the average magnitudes of the ith properties calculated from direct observations along the cruise track and from WOA05 data in the respective basins ( Table 2). The a i factors are the regression coefficients that were calculated in each basin and layer from a multiple linear regression (MLR) fit (Eq. 3) of the pH SWS25 averages vs. the averaged i properties using data from the fourteen cruises (Table 2). The obtained a i regression coefficients are listed in Table 3.
All terms and scripts in Eq.
(3) have the same meaning as in Eq.
(2). Also, the X c i terms for i = 1 through 4 are the same as in Eq. (2). The same is true for the a i coefficients in equation (2). Actually, the purpose of Eq. (3) is obtaining those a i values to be used in Eq. (2). k are the independent terms. The X 5 = xCO atm 2 values used as input parameters in Eq. (3) are the averages for the year of the corresponding cruise c. The xCO atm 2 records were obtained from time series from meteorological stations in the NASPG (Storhofdi, Iceland; CIBA, Spain; Mace Head, Ireland; Ocean Station C, US; Pico-Azores, Portugal; and Terceira Island-Azores, Portugal), that are part of the global cooperative air-sampling network managed and operated by the National Oceanic and Atmospheric Administration (NOAA) Carbon Cycle Greenhouse Gas group (http://www.esrl.noaa.gov/gmd/ccgg/flask. html). The a 5 term associated with the xCO atm 2 variable (Table 3) in Eq. (3) is not used in Eq. (2) since the pH SWS25-BA term should only include the effect of variables with spatial variation. Such xCO atm 2 terms are required when calculating the a i coefficients (Eq. 3, Table 3), since xCO atm 2 co-varies with pH 25SWS . By including a 5 in Eq. (3) we remove from the rest of a i factors the transient influences that co-vary with pH SWS25 . Considering that pH varies over time because of the xCO atm 2 change, the inclusion of this variable in Eq.
(3) insures that coefficients of the other properties that change mostly spatially are more consistent than if the xCO atm 2 is not included.

Results
The vertical distributions of pH SWS25 along a section between the Iberian Peninsula and Greenland are shown in Fig. 2, providing a first look at the evolution of pH over the last two decades. The general pattern of pH SWS25 follows the natural distribution expected, with higher pH values at the surface and lower pH in deep waters: the high values of pH SWS25 above the seasonal thermocline, in the photic layer (uppermost ∼ 400 m), respond to the photosynthetic activity of primary producers that withdraw dissolved CO 2 from seawater. The deep and less ventilated NADW has low pH SWS25 . The NADW is located generally below 2500 dbar (σ 2 > 37.00; Fig. 1b) mainly in the deep ENA Basin and shows weak signs of acidification over the last two decades, Table 2a. Temporal evolution (1981Temporal evolution ( -2008 of the values (average ± standard error of the mean) of salinity, potential temperature, AOU, silicate concentrations, measured pH (pH SWS25 ), pH basin corrections ( pH SWS25-BA ) and basin-normalized pH (pH SWS25-BA = pH SWS25 + pH SWS25-BA ) for the water masses considered in the: (a) Irminger; (b) Iceland; and (c) ENA Basins. The WOA05 lines give the climatological data used as reference values (see Eq. 2). Irminger Basin.  although there exist slight differences between the upper and lower NADW branches (uNADW and lNADW). The branch of uNADW that spreads westward into the Iceland Basin mixes with LSW (Yashayaev etl al., 2008) forming a pH gradient that shows decreasing pH values over time. The influence of LSW in the uNADW is also revealed by the AOU and Si(OH) 4 values of the uNADW, which are lower than those in the lNADW layer (Table 2c). In the Irminger, Basin the decreasing trends of pH SWS25 are clearly visible in the most recently ventilated waters like the uLSW and DSOW (Fig. 2). The latter shows low pH SWS25 in 2004 and 2008 and higher values in 2006 due to the different NAO conditions . The most evident sign of acidification is detected between 1000 and 2000 m depth, where the volume of water with pH values below 7.725 thickens over time.
To estimate the acidification rates of the water masses, we normalised the discrete in situ pH SW S25 data to basin-average conditions (pH SWS25-BA ), as described in Sect. 2.2. The average correction ( pH SWS25-BA ) applied to the studied region is 0.003±0.009 (Table 2). On average, the largest corrections correspond to the Irminger Basin (0.007 ± 0.009), while in the Iceland and ENA Basins they are smaller (0.003 ± 0.009 and 0.002 ± 0.010, respectively). In the Irminger Basin no correction was applied to the uNADW and DSOW layers (Table 2a). The highest average corrections in this basin were applied to the uLSW (0.014±0.008) and cLSW (0.012±0.005) layers, and the highest individual correction (0.027 ± 0.003) corresponds to the SAIW in 1997. The smallest average pH SWS25-BA corrections in the Iceland Basin correspond to the uLSW (0.000±0.003) and the largest to the SPMW layer (0.008 ± 0.014), to which also the highest individual correction was applied (0.003 ± 0.005), corresponding to the 1991 A01E cruise. In the ENA Basin the smallest average corrections correspond to the LSW and NACW layers (0.0012 ± 0.004 and 0.0045 ± 0.004, respectively) and the largest to the MW (0.014 ± 0.002), where the highest individual corrections were also applied (0.023 ± 0.002) in 1998 and 2003, both cruises conducted along meridian 20 • W.
In general, we can see a trend of decreasing pH over time for both pH SWS25 and pH SWS25-BA in all basins and layers (Table 2). These decreasing pH SWS25 trends tend to be more pronounced in the Irminger and Iceland Basins and less marked in the ENA Basin ( Table 2). The SAIW and uLSW layers in the Irminger Basin show strong decreasing pH SWS25 trends in the period 1981 to 1997 (positive NAO index) and less pronounced ones from 2002 to 2008. In the deepest layers (cLSW, uNADW and DSOW) the pH SWS25 trends are lower and there is also a minimum value in 1997, when the NAO phase changed from positive to neutral/negative. Similar pH SWS25 trends are observed in the Iceland Basin, with a noticeable decrease from 1981 to 1997 during the high NAO, followed by a slower rate of pH SWS25 decrease. Differently, the lowering of pH SWS25 in the ENA Basin shows a more continuous and steady trend, with a maximum during 1981 and a minimum in 2006, in the NACW and LSW layers. Also Table 3. List of a i coefficients obtained for Eq. (2) using the expression in Eq. (3) in each water mass and basin. Between brackets are the properties associated with each a i coefficient and the corresponding units. All a i coefficients have been scaled up by a factor of 10 3 , except for the salinity ones (a 4 ). The n.s. ("not significant") variables explained very little of the pH variability and weakened the overall MLR fit. They were therefore rejected according to a stepwise method of MLR solving. Estimated errors refer to the error of the MLR fits. at the ENA Basin, the uNADW and lNADW show rather constant pH SWS25 values over time, with no clear trends. The pH SWS25 signal of the MW layer is noisier due to the important variations in salinity caused by the mixing between MW and other water masses, and as a consequence of changing cruise tracks during the considered time period. The evolution of the average pH SWS25-BA between 1981 and 2008 in each layer and basin is plotted in Fig. 3. The error bars on the graph represent the error of the mean and the uncertainty due to the normalization of the data. The general pattern seems to be a decrease of the acidification rates over depth, in all basins. The lowest pH SWS25-BA vs. time slopes are found in the ENA Basin, while the fastest acidification rates correspond to recently ventilated waters like the SAIW (−0.0019 ± 0.0001 yr −1 ), the uLSW (−0.0017 ± 0.00004 yr −1 ) -both of them in the Irminger Basin -, and the SPMW (−0.0012 ± 0.0002 yr −1 ), in the Iceland Basin. The pH SWS25-BA of cLSW in the Iceland Basin presents a remarkable average decrease of −0.0016 ± 0.0002 yr −1 , unlike in the Irminger and ENA Basins (−0.00089 ± 0.00004 and −0.0008 ± 0.0001 yr −1 , respectively). The layer of uN-ADW shows negative pH SWS25-BA vs. time slopes from the Irminger (−0.0010 ± 0.0001 yr −1 ) to the Iceland Basin (−0.0008 ± 0.0002 yr −1 ) due to the influence of ISOW and to the mixing with LSW. Overall, the lNADW and uNADW in the ENA Basin are the least acidified water masses over time, with low pH SWS25-BA vs. time slopes. Their regression fits are, in addition, statistically non-significant (both pvalues > 0.2) and their pH SWS25-BA vs. time slopes are small, namely: 0.0002 ± 0.0002 yr −1 (R 2 = 0.15; p-value = 0.57) and −0.0003 ± 0.0001 yr −1 (R 2 = 0.28; p-value = 0.47) for lNADW and uNADW, respectively. The MW in the ENA Basin showed a moderate acidification rate (−0.0006 ± 0.0001 yr −1 ) due to its known capacity for C ant drawdown by entrainment from surface layers (Ríos et al., 2001;Álvarez et al., 2005).

Discussion
The acidification of waters in the upper layer of the NASPG here assessed from in situ pH measurements spanning the last three decades (1981 to 2008) shows very similar tendencies of pH decline to those observed in the time series stations ES-TOC (29 • 10 N, 15 • 30 W) and BATS (31 • 43 N, 64 • 10 W), in the Subtropical Atlantic. In the Irminger Basin, the observed pH SWS25-BA decrease rates for SAIW and uLSW are −0.0019 ± 0.0002 and −0.0017 ± 0.0001 yr −1 , respectively, similar to those obtained by Olafsson et al. (2009) for surface waters during the winter (0.0024 yr −1 ). The slight difference with the values reported by Olafsson et al. (2009) likely comes from the fact that the surface isopycnals here considered include thick layers of mode waters with lower interannual variations. The acidification rates here obtained for SAIW and uLSW in the Irminger Basin  Basins (c). The inset boxes give the acidification rates ± standard error of the estimate (in 10 −3 pH units yr −1 ) and correlation coefficients (R 2 ). Each of the points in the scatter plots represents the average pH of a particular water mass at the time of the measurement (cruise) (pH SWS25-BA ; Table 2). Considering the ample time interval of this study , these pH averages represent well the annual means. The error bars represent the error of the mean plus the uncertainty due to the basin normalization of the data (Sect. 2.2.1).
are also comparable to those reported for the subtropical North Atlantic, at the ESTOC site, in surface waters and in the mixed layer (−0.0017 yr −1 ) during the decade 1995(Santana-Casiano et al., 2007González-Dávila et al., 2010), and also at the BATS site, in surface waters (−0.0016 yr −1 ), from 1983 to 2011(Bates et al., 2012. In the ENA Basin, the decreasing pH SWS25-BA rate in the NACW (−0.0009 ± 0.0001 yr −1 ) is similar to those computed at the ESTOC site at 300 and 600 m (−0.0010±0.0004 and −0.0008 ± 0.0003 yr −1 , respectively) for the decade 1995(González-Dávila et al., 2010. At 3500 m, the decrease rate of pH SWS25-BA here obtained for lNADW (0.0002 ± 0.0002 yr −1 ) has a very low pH SWS25-BA vs. time correlation coefficient (r 2 = 0.15; Fig. 3c) and is therefore not significant, yet similar to that given by González-Dávila et al. (2010) (−0.0002 ± 0.0002 yr −1 ) for the same water mass between 1995 and 2004. On the contrary, at the layer where the MW spreads, around 1000 m, González Dávila et al. (2010) reported a pH decrease rate of −0.0008 ± 0.0003 yr −1 , which is slightly higher (considering the associated uncertainties) than our pH change rate (−0.0006 ± 0.0001 yr −1 ) for the same water mass. This difference could be due to the way MW is defined in our work compared to González-Dávila et al. (2010): they consider MW as a mix of at least three different water types (including MW, Antarctic Intermediate Water and NACW) at the east North Atlantic (González-Dávila et al., 2010).
Ocean uptake and chemical equilibration of C ant with seawater results in a gradual reduction of seawater pH and saturation states ( ) for calcium carbonate (CaCO 3 ) minerals. However, other contributions to these pH reductions such as ventilation of the water masses or remineralization of organic matter exist. We have checked whether these obtained pH SWS25-BA decrease rates would follow the acidification trends expected mainly from C ant uptake contribution, using the C ant rates given by Pérez et al. (2010). The necessary pH values to obtain such rates were calculated using the expression (∂pH/∂t) ANT = (∂C ANT /∂t) (∂pH/∂C T ) (S,A T ) , where (∂pH/∂t) ANT is the expected variation over time of anthropogenic pH (i.e., due to C ant ); (∂C ANT /∂t) is the corresponding C ant storage rate (from Pérez et al., 2010); and (∂pH/∂C T ) (S,A T ) is the variation with respect to C T of a pH calculated from the thermodynamic equations of the marine inorganic carbon system (as described in Sect. 2), using the available A T data and salinity measurements.
The pH SWS25-BA decrease in the layers of cLSW, uNADW and DSOW (Irminger Basin), and of SPMW and uNADW (Iceland Basin) do follow such expected acidification trend due mainly to C ant entry. However, the rest of the considered water masses actually show some deviations from these calculated human-induced acidification patterns. In the layers of uLSW (Irminger and Iceland Basins) and cLSW (Iceland Basin) there is a component (∼ 50 %) of the observed acidification trends that is not explained by the uptake of C ant , and is attributed to organic matter remineralization. The SAIW layer in the Irminger Basin presents an intermediate case compared to the previous ones: ∼ 75 % of the pH SWS25-BA decrease comes from the influence of C ant . In contrast with what was observed in the Irminger Basin, in the upper layer of the ENA Basin the C ant -induced acidification is partially compensated by the increase in ventilation (higher CO 2 removal via enhanced photosynthetic activity) of the eastern NACW (ENACW) that, overall, produces lower acidification rates than expected.
From our set of pH SWS25-BA observations we have made projections of future pH levels (Fig. 4). The Iceland Basin is particularly suitable for extrapolating the pH trends from Fig. 3b into the future, given the good coverage of measurements available in this region. This characteristic will confer added robustness to the projected acidification trends. The SPMW and cLSW were selected for such projections because out of the considered water bodies, they are some of the most susceptible ones to anthropogenic acidification, and also because they have strong pH SWS25-BA vs. time fits (Fig. 3b). The projections were calculated under the assumption that the acidification trends in Fig. 3 and the ocean's general circulation behave similarly to what has been observed during the last three decades, for the rest of the 21st century.
It is known that the strength and the phase of the NAO index affect water mass ventilation and C ant uptake rates . However, the fact that the NAO phase was close to neutral both during the 1980s and the 2000s should minimise potential biases in the proposed linear projections of pH, which are based on observations from the results here obtained (Fig. 3). Although such linear extrapolation is not constrained, several works have demonstrated that the decline of carbon system parameters, like [CO 2− 3 ], is almost linear for predictions made between 2000 and 2050 (Zeebe and Wolf-Gladrow, 2001;Hauck et al., 2010). The buffering effect of carbonate minerals and biogenic CaCO 3 dissolution can be disregarded since these processes tend to occur in deep waters over timescales that are at least one order of magnitude larger than the one here considered. We can therefore assume that on decadal timescales (our observational time span) pH will evolve in the future analogously to what we have observed in surface (SPMW) and intermediate (cLSW) waters. This timescale is also within the time frame in which the xCO atm 2 range considered for the pH projections ( Fig. 4) will be reached, under a business-as-usual CO 2 emission scenario.
Concerning the assumption of the ocean's general circulation, there is the caveat that the increased stratification of surface layers expected in the future (Friedlingstein and Prentice, 2010) could possibly hamper water mass ventilation processes and potentially bring about a decrease of pH (acidification), because C ant would not be as effectively transported towards the ocean interior via deep convection and water mass formation processes . Therefore, if such increased stratification prediction holds true in the future, assuming a steady state for the general circulation can potentially cause overestimates in the pH values of the linear projections for surface and intermediate waters in Fig. 4. Nevertheless, this acidification slowdown process due to the decrease in C ant entry could be counterbalanced by the enhanced remineralization rates of organic matter in the upper and intermediate ocean layers that would develop in such a scenario of increased stratification.
According to the obtained pH projections in Fig. 4, the pH of surface waters in the Iceland Basin could drop ∼ 0.35 Fig. 4. Extrapolation of the observed linear trends of acidification for the SPMW and cLSW in the Iceland Basin. On the x-axis, the projections range from the pre-industrial 280 ppm to future 800 ppm of atmospheric xCO 2 (molar fraction of CO 2 ). The prediction bands give the 95 % confidence intervals for the projected linear trends. The aragonite saturation states ( arag ; as percentages) for present xCO 2 (∼ 380 ppm) and for the horizons of 500 and 750 ppm are shown in the top boxes.
units with respect to the pre-industrial era by the time atmospheric CO 2 reaches 800 ppm, which is consistent with outputs from coupled climate/carbon-cycle models (Caldeira and Wickett, 2005;Orr et al., 2005). In the case of cLSW, the linear projection predicts a pH decrease of more than 0.45 units with respect to pre-industrial pH values by the time atmospheric xCO 2 reaches ∼ 775 ppm (about twice the present atmospheric concentration of CO 2 ). This result is 0.25 pH units lower than the values predicted by the wellknown climate-carbon coupled model in Caldeira and Wickett (2003) for the same xCO atm 2 and ocean region, meaning that according to our results, the cLSW would acidify at a faster rate than expected from numerical simulations. The difference between our observation-based prediction and the latter model (Caldeira and Wickett, 2003) could be due to the fact that our data is largely extrapolated and also that it is still difficult for General Circulation Models (GCMs) to model accurately the Meridional Overturning Circulation (MOC), its NAO-related variability (Danabasoglu et al., 2012) and the deep winter convection of the NASPG. The NAO-related MOC variability has a strong influence on C ant storage in the NASPG  and it can be therefore expected that this will affect the long-term variability of pH too, in a way models cannot quite account for yet. In this sense, our results are a good complement and reference for models outputs. On the other hand, different authors have reported that ocean acidification might in fact be proceeding more rapidly than models had predicted (Wootton et al., 2008), as the contemporary CO 2 emissions are actually exceeding future scenarios based on business-as-usual emission rates . Such reports are consistent with the lower pH predictions we obtained compared to Caldeira and Wickett's (2003) results.
The aragonite saturation state is defined as arag = [Ca 2+ ][CO 2− 3 ]/K sp , where square brackets indicate seawater ion concentrations and K sp is the apparent solubility product of aragonite (Mucci, 1983). Because [Ca 2+ ] is highly and positively correlated with salinity, arag is largely determined by variations in  ]. This characteristic makes arag an optimum indicator of the environmental availability of dissolved carbonate ions.
From the measured pH data and our pH projections (Fig. 4) we calculated the arag of the SPMW and cLSW in the Iceland Basin for atmospheric xCO 2 values of 380 (∼ present day), 500 and 750 ppm (see insets in Fig. 4). The results suggest that cLSW would actually reach aragonite undersaturation ( arag < 1) by the time atmospheric CO 2 reaches ∼ 550 ppm and not 900 ppm, as suggested by the numerical model predictions in Orr et al. (2005). The high-NAO enhanced ventilation that occurred towards the mid-late 1980s fostered the fast formation of a massive vintage of cLSW (Kieke et al., 2007;Yashayaev et al., 2008). The rapid subduction of this newly formed cLSW injected C ant from the surface into intermediate waters, transporting C ant much faster than via downward diffusion alone and thus causing a faster acidification rate in the cLSW (where organic matter remineralization also contributes significantly to the pH lowering) than in the SPMW, where C ant influence is the main contributor to acidification. Depending on the future CO 2 emission rates the 550 ppm threshold at which, according to our projections, cLSW would face aragonite undersaturation, could be passed by 2050 or before (Nakicenovic et al., 2000;Caldeira and Wickett, 2005;Feely et al., 2009). Guinotte et al. (2006) have in fact pointed out that some deep-sea coldwater corals may experience undersaturated waters as early as 2020 under an IPCC "business-as-usual" CO 2 emission pathway, which is in good agreement with our observationbased results for the Iceland and Irminger Basins.
The data analysis also showed that the aragonite saturation depth (lysocline = isopleth where arag = 1) has shoaled at a rate of 7 and 4 m yr −1 between 1981 and 2008 in the Irminger and Iceland Basins, respectively. The latter is in agreement with previous local studies (Olafsson et al., 2009). The fast rate of lysocline shoaling in the Irminger Basin is promoted by the intense NAO-enhanced deep convection that injects ventilated, CO 2 -rich waters into deeper ocean layers (Messias et al., 2008), as mentioned previously. For comparison's sake, the shoaling rates of the lysocline were estimated to be ∼ 0.2 m yr −1 during the Paleocene-Eocene Thermal Maximum (55 million yr ago), when a massive natural release of CO 2 into the atmosphere caused global temperatures to raise more than 5 • C in less than 10 000 yr (Pelejero et al., 2010).

Conclusions
The progressive acidification of North Atlantic waters has been assessed from direct pH observations spanning the last three decades. The increasing atmospheric CO 2 concentrations have largely affected the pH of surface and intermediate waters in the three studied North Atlantic regions, at varying extents. Most importantly, the LSW has shown very high acidification rates that are amongst the highest found in the NASPG. In the Irminger Basin, the acidification rate of cLSW responds to that expected from the influence of C ant , while in the Iceland Basin only about 50 % of the observed pH change in the cLSW is anthropic. The SAIW has the fastest acidification rate observed (−0.0019 ± 0.0002 yr −1 ), and 75 % of this pH decrease is anthropogenic. In contrast, the C ant contribution to the acidification rates in the ENACW is partially compensated by the ventilation of this water mass, thus explaining the moderate acidification rates observed in the upper layers of the ENA Basin (compared to the Iceland and Irminger Basins). Predictions from an observation-based extrapolation of the current acidification trends and rates are in agreement with model results (Caldeira and Wickett, 2005;Orr et al., 2005) in surface layers. However, our results indicate that the intermediate waters of the North Atlantic (LSW in particular) are getting acidified more rapidly than models predicted.