Temporal variability of the anthropogenic CO2 storage in the Irminger Sea

The anthropogenic CO 2 (C ant ) estimates from cruises spanning more than two decades (1981–2006) in the Irminger Sea area reveal a large variability of the C ant storage rates in the North Atlantic Subpolar Gyre. During the early 1990’s, the C ant uptake rates doubled the average rate for 1981–2006, whilst a remarkable drop to almost half 5 that average followed from 1997 onwards. The C ant storage evolution runs parallel to CFC12 inventories and is in good agreement with C ant uptake rates of increase calculated from sea surface pCO 2 measurements. The North Atlantic Oscillation shift from a positive to a negative phase in 1996 led to a reduction of the air-sea heat loss in the Labrador Sea. The consequent convection weakening accompanied by an increase 10 in stratiﬁcation lowered the e ﬃ ciency of the northern North Atlantic CO 2 sink.


Introduction
The ocean is a CO 2 sink that during the 1990s has removed 2.2±0.4 Pg C·yr −1 from the atmosphere out of the total 8.0±0.5 Pg yr −1 of anthropogenic carbon (C ant ) emitted to the atmosphere directly from human activities (Canadell et al., 2007). The 15 North Atlantic Subpolar Gyre (NASPG) has the largest C ant storage per unit area (∼80 mol C·m −2 on average) of all oceans, holding 38% of the oceanic C ant storage (Sabine et al., 2004). The key mechanism responsible for this large CO 2 uptake is the Meridional Overturning Circulation (MOC). The MOC transports warm surface waters with high C ant loads from low latitudes to the northern North Atlantic (Watson et  found consequences for global climate due to the associated decrease in oceanic C ant uptake (Sarmiento and Le Quéré, 1996) and heat transport (Drijfhout et al., 2006). Several Ocean General Circulation Models (OGCMs) have suggested that the decadal variability of the MOC is closely related with the variability of Labrador Seawater (LSW) formation rates (Böning et al., 2006). On the other hand, the long-term evolution of the 5 MOC such as the possible weakening during the 21st century might be related to a decrease in the density of the Denmark Straight Overflow Water (DSOW) and the Iceland-Scotland Overflow Water (ISOW) (Böning et al., 2006). These water masses meet in the Irminger Sea, where the Deep Western Boundary Current originates (Yashayaev et al., 2008). 10 The Irminger basin has been proposed as a LSW formation region (Pickart et al., 2003), in addition to the Labrador Sea. Independently of the formation region, two modes of LSW are typically defined: the classical LSW (cLSW, sometimes referred to as deep LSW) and the less dense upper LSW (uLSW) (Kieke et al., 2006). The LSW is formed in winter, when deep convection caused by intense air-sea heat loss results in 15 the formation of homogeneous layers of up to 2000 m. The ambient stratification and wind forcing intensity are determinant factors in this convective process (Dickson et al., 1996;Curry et al., 1998;Lazier et al., 2002). During the early 1990's, the strongly positive North Atlantic Oscillation (NAO) index forced an impressive convection activity down to more than 2000 m (Lazier et al., 2002). This resulted in the formation of the 20 thickest layer of cLSW observed in the past 60 years (Curry et al., 1998). This energetic convection period abruptly ended in 1996 with the shift of the NAO index to a negative phase. Nonetheless, weaker convection events continued to take place in the central Labrador Sea and formed the less dense uLSW. It was first detected in the western side of the Labrador Sea during the second half of the 1990's (Azetsu-Scott et al., 2003;25 Stramma et al., 2004). Decadal time series of layer thicknesses of both LSW types corroborate that, far from exceptional, uLSW is an important product of the convection activity in the Labrador Sea (Kieke et al., 2007). These time series show that the strong formation processes of cLSW in the early 1990's are actually the exceptional Interactive Discussion events. The temporal evolution of CFCs in the Labrador Sea indicates that after a strong increase in the early 1990's, CFC12 concentrations started to decline towards the end of the decade in the cLSW body (Azetsu-Scott et al., 2003). This CFC12 decrease was also observed during the early 2000's in the Labrador and Irminger Seas (Kieke et al., 2007). The fluctuations of convection in the NASPG can modify the 5 expected oceanic C ant uptake rates in a likewise and parallel manner to CFCs. In this study we have gathered hydrographical measurements and results from seven cruises conducted in the Irminger Sea. The aim is twofold: a) study the evolution of C ant concentrations in subsurface waters, LSW, North East Atlantic Deep Water (NEADW) and DSOW in order to b) evaluate the variability of the oceanic uptake rates of C ant 10 linked to the fluctuations of the convective processes in the NASPG, and the reduction of the formation of cLSW.

Dataset and method
Six cruises spanning through 25 years (1981-2006) of high-quality carbon measurements in the Irminger Sea area have been selected for this study (Fig. 1, Table 1).

15
Unlike in the most recent cruises from FOUREX or the OVIDE project, the TTO-NAS analytics did not include certified reference materials for their total inorganic carbon (C T ) measurements. For the TTO-NAS, Tanhua et al. (2005) performed a crossover analysis with an overlapping more recent cruise. Based on a comparison with modern Certified Reference Material-referenced data, they suggest a correction for 20 TTO-NAS C T measurements of approximately −3.0 µmol·kg −1 , which has been applied to our dataset. In order to evaluate and interpret the variations of C ant uptake rates we have focused on six water masses delimited by the density (σ θ ) intervals established following Kieke et al. (2007) and Yashayaev et al. (2008) , namely: from the upper 100 m to σ θ =27.68 kg·m −3 we find the subsurface layer; The uLSW is found be-  (Fig. 1b). To estimate the anthropogenic CO 2 the ϕC o T method from Vázquez-Rodríguez et al. (2008) 1 is applied. It is a data-based, back-calculation method that constitutes an improved version of the classical ∆C* approach (Gruber et al., 1996). The ϕC • T method uses 100-200 m data as the only reference to build the parameterizations needed. This 5 sub-surface layer avoids the seasonal variability of surface properties, thus making the derived parameterizations more representative of water mass formation conditions. The method does not rely on CFC measurements and takes into account the temporal variation of the CO 2 air-sea disequilibrium (∆C dis ). The overall uncertainty of the method has been estimated in 5.2 µmol·kg −1 by means of random error propagation 10 over the precision limits of the parameters involved in the calculation of C ant . Regarding the specific inventory estimates, errors were estimated using a perturbation procedure for each layer and the total water column. They were calculated by means of random propagation with depth of a 5.2 µmol·kg −1 standard error of the C ant estimate over 100 perturbation iterations, and are plotted on Fig. 3.

Results
The salinity, potential temperature (θ), apparent oxygen utilisation (AOU=O eq 2 -O meas 2 , which represents the sum of the biological activity undergone by the sample since it was last in equilibrium with the atmosphere) and C ant in the Irminger Basin from 1981 to 2006 are shown in Fig. 2. The TTO-NAS shows a strong vertical stratification at the 20 LSW level between the first 1000 m and the relative salinity maximum of the NEADW (>34.95), related to the low convection activity in the Labrador Sea in the late 1970's (Kieke et al., 2006). It must be reminded that it takes two years for the LSW to spread BGD 5,[1587][1588][1589][1590][1591][1592][1593][1594][1595][1596][1597][1598][1599][1600][1601][1602][1603][1604][1605][1606]2008 Temporal variability of the anthropogenic CO 2 storage in NNA  (Yashayaev et al., 2008). The temperature minimum (θ<1.5 • C) in the DSOW is also remarkable within the considered time-span. Compared to TTO-NAS, the AR7E (1991) cruise shows cooler LSW. The strong vertical homogenisation down to 1800 m in 1991 suggests local LSW formation at the Irminger Sea, which caused the saltier NEADW signal to shrivel. The same year, the LSW layer also thick-5 ened substantially in the Labrador Sea (Kieke et al., 2006). In 1997, the low salinity LSW invaded lower layers, beyond the 2000 m depth, while surface stratification slightly increased. The temperature of the cLSW reaches its minimum values for the period of observation, in agreement with Yashayaev et al. (2008). The DSOW temperature signature (θ<2 • C) practically disappeared during the FOUREX cruise, but recovered during the OVIDE cruises history, most importantly in 2004. It is likely that this variability is linked to that of the entrainment downhill of the sills. Alternatively, it must be noted that the FOUREX cruise intersects the DWBC to the south of the OVIDE line, and this may partially account for this difference. During this period of biennial sampling (2002, 2004 and 2006) an increase in salinity (especially at the LSW and NEADW layers) and 15 stratification is observed, coinciding with a period of weak winter convection (Kieke et al., 2006). The described thermohaline evolution concurs with the θ/S results shown by Yashayaev et al. (2008) for the LSW core: A salinity and temperature minimum is recorded in 1996 at the Irminger Sea (which is two years behind the θ/S minimum at the Labrador Sea) followed by a progressive salinization and warming due to lateral 20 mixing that can be observed along the σ 1 =36.93 isopycnal and extends to the rest of LSW density range. In 1981, the AOU profiles nicely separate close to air-sea equilibration subsurface waters from the old NEADW. The relative AOU minimum at the bottom West of the Irminger Basin indicates the marked presence of DSOW. The C ant concentrations fol-25 low a similar pattern to AOU: high values, close to saturation (32 µmol·kg −1 ), near the surface and lower values (∼45% of saturation) towards the bottom. During the first deep convection events in the early 1990's there is a significant and parallel increase in the C ant and oxygen loads in the upper 1500 m. Nevertheless, the NEADW region BGD 5,[1587][1588][1589][1590][1591][1592][1593][1594][1595][1596][1597][1598][1599][1600][1601][1602][1603][1604][1605][1606]2008 Temporal variability of the anthropogenic CO 2 storage in NNA shows higher AOU along with slightly lower (around 15%) C ant values than in 1981, denoting the presence of older waters. The high ventilation of the water column caused by the strong deep convection between 1991 and 1997 (Kieke et al., 2006) resulted in smaller AOU and higher C ant values during FOUREX than previously recorded for the LSW in the Irminger Basin. In 1997, the C ant concentrations (30±0.7 µmol·kg −1 ≈80% 5 saturated) at the LSW core is at a maximum and AOU at a minimum. The NEADW layer is richer in oxygen, less saline and it contains about 50% more C ant than in 1991, suggesting that more intense mixing processes occurred with the upper bounding LSW layer. In the CGFZ, where LSW and NEADW flow in opposite direction, mixing processes are enhanced, and in 1997 the LSW transport in this area was particularly 10 intense (Lherminier et al., 2007). During the OVIDE period, convection was weak in the Labrador Sea (Kieke et al., 2006) and no deep convection was recorded in the Irminger Sea (Yashayaev et al., 2008) resulting in a re-stratification of the water column and an aging of the deep-water masses. This is particularly evident in the cLSW, whose AOU increased by 10 µmol·kg  Table 2. The average thickness of the layers and their percentage contribution to C ant specific inventories are also given in Table 2. The thickness was calculated as the average distance between layers weighed by the separation between stations. The averages for the rest of properties were computed integrating vertically and horizon- tally, and then dividing by the area of the corresponding layer. The C sat ant is estimated from the average temperature and salinity of the layer, assuming full equilibrium of surface waters with the average atmospheric molar fraction of CO 2 (xCO 2 ) at the year of each cruise. A quantitative evaluation of the previously described interdependences between the variability of AOU, C ant and ventilation has been attempted by plotting the 5 percentage of C sat ant vs. AOU from Table 2. The term %C sat ant is independent of the atmospheric CO 2 increase since it is referred to the C sat ant concentration in the corresponding sampling year. In this sense, %C sat ant is comparable to oxygen, whose atmospheric concentration is stationary. Hence, it is expected that recently equilibrated (young) waters will have low AOU and high %C sat ant values, while the opposite is expected in older wa-10 ters that have undergone large remineralization of organic matter. We found that the largest temporal variability of AOU and %C sat ant is in the cLSW layer, where both variables are highly correlated (R 2 =0.94). The uLSW shows also a significant %C sat ant vs. AOU correlation (R 2 =0.74). When all Irminger Sea water masses in Table 2 (except for the DSOW) are considered altogether a correlation of R 2 =0.91 is obtained. This suggests 15 that in the Irminger Sea, %C sat ant for the sampling year could be estimated fairly accurately from AOU in the main water masses. Hence, for a given AOU value the %C sat ant would be invariable, independently from the sampling year. Knowing the %C sat ant and the atmospheric pCO 2 for the sampling year, C ant could thereby be estimated. This AOU vs %C sat ant dependence establishes an empirical quantitative relationship based on 20 a simplified mixing model with subsurface waters and NEADW as end-members. The former would represent the highly ventilated young waters from the winter mixed layer (WML) whereas the latter would stand for the older components. The deviations from this hypothetical mixing line can be due to lateral advection, to interannual or decadal variability of water mass formation or to differential biological activity rates. 25 The temporal evolution of the average C ant in each of the five layers is plotted in Fig. 3. The subsurface layer (Fig. 1b)  tions. The uLSW trend from 1981 to 1991 follows its upper bound subsurface layer, keeping up with the atmospheric CO 2 increase and maintaining its %C sat ant . The maximum thinning of the uLSW layer from 1981 to 1997 coincides with the end of the maximum convection period in the Irminger Sea (Kieke et al., 2006). The cLSW almost doubled its thickness and average C ant content during this period of time at a 5 rate even superior to the atmospheric one. All of it derives from the increased convection processes that occurred in the NASPG between 1991 and 1997 (Kieke et al., 2006). The noteworthy decrease in C ant and layer thickness during the OVIDE cruises caused by the hindered ventilation entails the increase in the AOU levels due to isopycnal mixing. In the NEADW layer, C ant shows a large increment from 1991 to 1997 parallel to the salinity and AOU drop. This suggests the possibility of important diapycnal mixing with the upper re-ventilated cLSW. The DSOW layer will represent a very small proportion in terms of total storage in the Irminger basin given its short thickness. It shows an analogous behaviour to NEADW, i.e., there is an increase in average C ant from 1991 to 1997, matched with a drop in AOU caused by the incorporation of 15 young water by entrainment downhill of the Iceland-Scotland sills. Regarding the total storage, its temporal evolution shows a global increase at different paces. From 1981 to 2006 the average rate of increase in the specific C ant storage for the Irminger Sea has been 1.1±0.1 mol C·m −2 ·yr −1 . During the early 1990's, this rate was more than twice the mean (2.3±0.6 mol C·m −2 ·yr −1 ). Compared with this period of exacer-20 bated C ant storage rate in the Irminger Sea, the average C ant uptake during the following decade (1997)(1998)(1999)(2000)(2001)(2002)(2003)(2004)(2005)(2006) underwent an important fall (Fig. 3)

Discussion
As Doney and Jenkins (1988) pointed out, ocean regions affected by strong convection processes tend to acquire large air-sea disequilibria. This applies not only to CO 2 , but also to atmospheric gases with higher air-sea transfer velocities such as oxygen or CFCs. The same processes affecting oceanic C ant uptake will be determining the 5 distribution and uptake rates of CFCs. Several works have focused on the CFCs in the NASPG after the 1970's, and the patterns they described support the C ant trends obtained in the present study. The specific inventories of CFC12 in the core of the cLSW found in both the Labrador and Irminger Seas grew until approximately the first half of the 1990's. As Lazier et al. (2002) have illustrated, the formation of cLSW ceased after 1997. The lack of supply of low-salinity/CFC-rich surface waters led to an annual increase in cLSW salinities and a decrease in CFCs because of isopycnal mixing. The CFC12 inventories started to decline strongly from 1997 to 1999 and kept decreasing at a slower rate until 2003 (Azetsu-Scott et al., 2003;Kieke et al., 2007). These evidences seem to correspond with the same pattern here observed for the 15 average C ant in the cLSW, which decreases from 1997 to 2002 (Fig. 3). According to Azetsu-Scott et al. (2003), the marked increase in CFC12 concentrations in the Irminger basin prior to 1995 in the uLSW core reduced to almost standstill from that point until 2001. This result is also in good agreement with the patterns of average C ant obtained for the same years and region in this study (Fig. 3). Kieke et al. (2007) have 20 pointed that this significant increase in CFC12 inventories in the uLSW for the NASPG best describes the situation in the Labrador Sea, rather than in the Irminger, where the CFC12 inventory has a more subtle increase during that period. In spite of this remark, our C ant results for the uLSW in the Irminger follow the expected trend, within the associated error margins. With respect to the NEADW and DSOW, Azetsu-Scott et 25 al. (2003) have shown that from 1991 to 2000 the CFC12 storage in the Irminger Sea increased up to 80% in both the NEADW and DSOW layers. Whilst this happened at a steady rate in the case of NEADW, the interannual variability was larger for DSOW. Over the same period, the C ant storages here obtained increased by 50% and 40% in the NEADW and DSOW layers, respectively. There are a few differences in the environmental behavior of CFCs and CO 2 that may account for the dissimilarities in magnitude of their inventories. The former is not affected by the Revelle factor, it has a greater solubility in cold waters and its atmospheric rate of increase is different to that 5 of CO 2 Although the NASPG has a net gain of C ant by horizontal advection (Mikaloff-Fletcher et al., 2006;Álvarez et al., 2003), estimating how much it is stored and how much of that comes from direct exchange with the atmosphere entails certain difficulty. Indirect estimates of the air-sea C ant fluxes can be obtained by combining C ant storage results 10 with horizontal transports into carbon budget balances from closed box studies (as in Mikaloff-Fletcher et al., 2006). The C ant storage rate for the Irminger Sea was first estimated byÁlvarez et al. (2003)  MPD is defined as the quotient between the C ant water column inventory and the C ant concentration in the mixed layer (C ml ant ), and it can be interpreted as an index of the convection activity in the considered region. For the calculation, they assumed a fixed C ant rate of increase of 0.85µmol·kg −1 ·yr −1 and calculated an average and constant MPD for the Irminger basin of 1739±381 m by approximating C ml ant ≈ C sat ant in the correspond-20 ing years. The average C ant storage rate for the 1981-2006 period in the Irminger Sea here obtained is 1.1±0.1 mol C·m −2 ·yr −1 . The calculated average MPD for the considered years is 1715±63 m (Fig. 3). This is quite in agreement withÁlvarez et al. (2003), although the MPD values are seen to vary, especially in the strong convection periods such as from 1991-1997. The 1.5±0.3 mol C·m −2 ·yr −1 estimate fromÁlvarez et 25 al. (2003) is larger than the average 1.1±0.1 mol C·m −2 ·yr −1 here obtained and, anyhow, lower than the estimated rate between 1991-1997 of 2.3±0.6 mol C·m −2 ·yr −1 . Some factors accounting for the temporal variability of C ant storage rates may explain some of these discrepancies, primarily: a) The time variability of the MPD can affect Interactive Discussion notoriously the C ant storage rates, especially between 1991 and 1997 (8% larger than the average). There can exist exceptional interannual stages where the storage rates can amount up to twice (1991)(1992)(1993)(1994)(1995)(1996)(1997) or almost half (1997)(1998)(1999)(2000)(2001)(2002)(2003)(2004)(2005)(2006) the long-term average; b) The C ant rate of increase must consider the dependence of the C sat ant with temperature, which is intimately connected with the Revelle factor (it describes how 5 the pCO 2 in seawater changes for a given change in C T , and vice versa). The capacity for ocean waters to take up C ant is inversely proportional to the Revelle factor, which depends on temperature. The C ant rates of storage can change ∼2% per • C. Our observations can also be compared with other works on the secular variation of sea surface pCO 2 . In the NASPG, Lefèvre et al. (2004) reported a mean increase 10 of 1.8 µatm·yr −1 between 1982 and 1998. This corresponds to a ∆C T =0.77 µmol·kg −1 per annum for an average sea surface temperature of 5 • C in the Irminger Sea. If an average MPD of 1715 m is considered for this region and time period, the above ∆pCO 2 translates into a C ant storage increase rate of 1.35 mol C·m −2 ·yr −1 . This is very close to the 1.22±0.03 mol C·m −2 ·yr −1 estimated here from Fig. 3 for the 1981-1997 period.

15
From sea surface pCO 2 measurements Schuster et al. (2007) showed that the sink of atmospheric CO 2 in the North Atlantic was subject to important interannual variability. They estimated an annual decrease of the North Atlantic uptake of −1.1 mol C·m −2 ·yr −1 between 1997 and the mid-2000's. They argued that the main causes for this change were the declining rates of wintertime mixing and ventilation between surface and sub-20 surface waters due to increasing stratification. In the present work we have estimated this same rate of decrease to be −1.6±0.4 mol C·m −2 ·yr −1 for the 1997-2006 decade in the Irminger Sea. Corbière et al. (2007) estimated an annual increase in wintertime sea surface pCO 2 of 3.0 µatm·yr −1 between 1993 and 2003 utilizing data from a shipping route from Iceland to Newfoundland. They attributed this finding principally to the 25 increasing sea surface temperatures linked with the shift of the NAO index into a negative phase after wintertime 1995. According to Schuster et al. (2007), this corresponds to an annual decrease of the C ant storage of −1.6 mol C·m −2 ·yr −1 , which is in excellent agreement with our estimates. In summary, the general decline of the NASPG CO 2 sink is supported by the data here obtained and it is corroborated by other results like those from Schuster et al. (2007) or Corbière et al. (2007). There is a documented variability of the Labrador Sea deep convection that reached its maximum activity during the early to mid 1990s, correlated with positive NAO phases. The maximum C ant storage rate occurred during 5 this maximum convection period. The observed convection decrease since 1997 in the NASPG is mainly tied to the enhanced stratification and the consequent reduced heat loss in the northern North Atlantic (Lazier et al., 2002;Azetsu-Scott et al., 2003;Kieke et al., 2007). All these fluctuations are embodied in the NAO index variability. The above factors have a direct influence on the physical pump of CO 2 and add to the BGD 5,[1587][1588][1589][1590][1591][1592][1593][1594][1595][1596][1597][1598][1599][1600][1601][1602][1603][1604][1605][1606]2008 Temporal variability of the anthropogenic CO 2 storage in NNA