Ventilation variability of Labrador Sea Water and its impact on oxygen and anthropogenic carbon: a review

1. Introduction

One of the few regions that ventilate the intermediate and deep North Atlantic is located in the Labrador Sea (figure 1). Here, the lightest component of the North Atlantic Deep Water (NADW), the Labrador Sea Water (LSW), is formed by wintertime deep convection [1–4]. Within and around the Labrador Sea, this water mass can be found from near-surface to 2500 m depth [5] and in more remote regions, the LSW occupies up to 1000 m of the water column [6]. The LSW is characterized by low salinities, low potential vorticity (PV; e.g. [7]) and high concentrations of dissolved oxygen [5] and transient tracers as, for instance, chlorofluorocarbons (CFCs) [8], sulfur hexafluoride (SF₆) [9], and anthropogenic carbon [10]. Both PV and CFCs have been used to infer the main pathways of LSW (figure 1) in the North Atlantic [7,8,11–13], and the time information inherent in the transient tracers provided time scales of spreading from the source region into the Southern Hemisphere [13–15].

• Download figure
• Open in new tab
• Download PowerPoint

The formation and ventilation of LSW receives a lot of attention because of its connection to the large-scale circulation. In state-of-the-art climate models [16], in high-resolution ocean–ice models [17] and in ocean reanalysis [18], the variability in the LSW formation is mainly responsible for the multi-annual to decadal variability of the climate relevant Atlantic Meridional Overturning Circulation (AMOC). On multi-decadal and longer time scales, the AMOC is projected to weaken under global warming scenarios, and the weaker AMOC co-occurs with less ventilated LSW in the future [16].

The main factor governing the depth and spatial extent of deep convection is the air–sea heat flux, which is linked to large-scale atmospheric modes like the North Atlantic Oscillation (NAO; e.g. [19]). During positive NAO phases, winter air temperatures over the Labrador Sea tend to be lower than normal and winds stronger, and the ocean responds with increased heat losses to the atmosphere as shown in the time series of the cumulative heat flux (figure 2). The smoothed time series of the NAO and the heat flux correlate well: high NAO phases have higher heat fluxes from the ocean to the atmosphere. The cooling of the surface layer increases the density and leads to homogenization and ventilation of the water column through vertical mixing. Depending on the atmospheric state and the initial vertical density stratification in the Labrador Sea in the previous autumn, convection can reach down to 2500 m depth [2,16].

• Download figure
• Open in new tab
• Download PowerPoint

Convection transfers ocean surface signals into the interior of the Labrador Sea and subsequently through horizontal advection and mixing into the intermediate to deep layers of the Atlantic Ocean. For instance, high surface concentrations of oxygen, anthropogenic carbon (C_ant) or CFCs are incorporated into the LSW [10,20,21] and spread laterally along the LSW pathways. Under weaker convection conditions, the input of surface signals is reduced, and export and mixing with less ventilated water reduce oxygen, CFC and SF₆ concentrations in the LSW.

In the recent decade, special attention has been drawn to the ongoing warming and salinification trend observed in the Labrador Sea and in downstream regions that ended in the three subsequent winters 2013/2014 to 2015/2016. To put these events into a long-term context and tracing their impact along the LSW spreading pathways towards 47° N at the southern boundary of the subpolar gyre, we first review the time series of LSW parameters in the main formation area of LSW, the central Labrador Sea. After discussing the circulation of LSW and the convectively formed anomalies in oxygen and C_ant, the link between LSW formation and the AMOC variability and its connection to cold surface anomalies on a multitude of time scales is highlighted.

Here, we use the LSW definition of Stramma et al. [22] for consistency with earlier works [4,23]. An alternative definition based on various properties, e.g. density, of newly formed LSW can be found in other studies [5]. A more detailed discussion on the characteristics of LSW vintages and their evolution with time can be found, for instance, in [3]. To reflect the different convection histories [25,26], the LSW bound by the isopycnals σ_Θ = [27.68, 27.80] kg m⁻³ is separated in two modes: the lighter shallower mode is dubbed upper LSW (uLSW) with densities σ_Θ = (27.68–27.74) kg m⁻³, and the denser mode is deep LSW (dLSW) with σ_Θ = (27.74–27.80) kg m⁻³. South of 45° N, a mixture of LSW vintages formed during weak and strong convection periods is present. Following Rhein et al. [13], LSW south of 45° N is bound by σ_1.5 = (34.60, 34.75) kg m⁻³. By this definition, the core of the ventilated LSW is always included in that region.

2. Long-term variability ventilation and properties of Labrador Sea Water in the central Labrador Sea

Density changes in the surface layers of the Labrador Sea are caused by changes in temperature and salinity. On the interannual to decadal time scales, the main driver of deep convection in the Labrador Sea is the winter air–sea heat flux. Surface salinity anomalies, mainly associated with freshening events such as the Great Salinity Anomaly in the late 1960s to early 1970s [27], are of lesser significance to convection. During that time period, the combination of reduced winter cooling due to an anomalously warm atmosphere and increased buoyancy of the freshened surface waters led to the shutdown of convective renewal of the deeper waters, accompanied by their warming and salinification [27]. A third factor is the convection history: several subsequent years with no or very weak convection strengthen the density stratification, increasing the threshold of the buoyancy flux needed for deeper convection to develop. On the other hand, deep convection repeated over several/multiple winters may ‘help’ convection in the next winter, which is known as convective preconditioning [28].

Depending on the density stratification, in winters with moderately to extremely cold air masses over the Labrador Sea, the surface heat loss is sufficient to cool the mixed layer to the density range of LSW. Vigorous mixing that can exceed depths of 2000 m homogenizes the water column from the surface to the convection depth [29], thus transferring the cold, fresh, oxygen and CFC-rich signals characterizing the mixed layer into the interior of the ocean. In short, intensified convection tends to replenish the LSW reservoir, also making it colder and fresher. In years without or only weak convection, the LSW warms and becomes more saline by lateral mixing with the warmer and saltier water in the environment. Occasionally, ventilation of mostly shallow uLSW density layers also occurs in the Irminger Sea [25,30,31].

The variability of the convective activity in the Labrador Sea is reflected in the time series of temperature, salinity, oxygen and anthropogenic carbon inferred from measurements in the central Labrador Sea (figures 2 and 3). Time periods with intense convection are revealed by cool and fresh anomalies in 1972–1976 [1], and more strikingly in 1987–1994 [24]. From 1994 to 2007, convection did not reach densities σ_Θ > 27.75 kg m⁻³ [23,25,26]. The shallower uLSW modes kept their fresh signal, while the denser dLSW subsequently warmed and became more saline by mixing with ambient water. LSW formation rates have been calculated from CFC inventory changes since 1997, i.e. shortly after the cessation of the intense convection of the early 1990s [25,26]. Between 1997 and 2005, only the uLSW was ventilated, and the formation rate decreased continually from 8 Sv in 1997–1999 to less than 1 Sv in 2003–2005 [23]. The latter time period was also studied with hydrographic data with sufficient spatial coverage in the Labrador Sea. Although between 2003 and 2005 the LSW warmed and became more saline, the increased vertical homogeneity and lower PV pointed to ventilation reaching the uLSW density layers in accord with the results from the change in the CFC inventories [32]

• Download figure
• Open in new tab
• Download PowerPoint

In winter 2007/2008, however, the anomalous heat loss led to convective overturning extending to 1600 m depth and reached densities that had not been ventilated since 1994 [24]. This event interrupted the warming trend that persisted since the mid-1990s (figure 2). The high heat flux in January–February 2008 was caused by the presence of anomalous cold continental air masses over the Labrador Sea. The atmospheric anomaly was a local and short-lived episode [5].

In the following years, the Labrador Sea resumed a moderate state of convection, with convection depths around 1000 m [4,5,28]. In winter 2013/2014, the mid and high latitudes of the North Atlantic were subject to exceptionally strong latent and sensible heat loss, primarily driven by anomalously strong northerly airflows that originated in the Nordic Seas [33,34]. The Labrador Sea responded with a cooling, freshening and oxygenation of the LSW down to 1700 m depth and produced the most ventilated LSW since the beginning of the twenty-first century. In January–February 2014, the maximum average daily heat loss (56°–60° N, 59°–54° W) significantly exceeded the mean for the preceding winters [33]. The cold surface temperatures persisted, and in summer 2015, it was up to 2°C colder than normal over much of the subpolar gyre [35]. The anomaly was caused by a continuing anomalous air–sea heat loss from late 2014 to spring 2015 and the re-emergence of colder anomalies that developed in preceding years [35]. Through the extended phase with strong air–sea heat fluxes, the deep convection in the Labrador Sea persisted also in winters 2014/2015 and 2015/2016 and extended beyond 2100 m in the latter winter [5,28]. This most recent convection intensity resulted in a shift in the LSW properties back towards conditions observed in the early 1990s [28].

Previous changes of the oxygen distribution in the Labrador Sea have been described by van Aken et al. [36] (years 1950–2007), by Stendardo & Gruber [37] (years 1962–2009) and over recent years by Yashayaev & Loder [5,28] (years 2008, 2011, 2015–2016). Figure 3a completes and extends these time series and highlights the oxygen evolution in the central Labrador Sea from 1992 to 2016. The previously mentioned authors remarked on the switch between high oxygen and low oxygen periods in the LSW, caused by ventilation events as observed in other parameters like temperature and salinity (figure 2). Historic oxygen data from the pre-1990s are sparse with a considerable data gap in the 1980s. Despite this shortcoming, it is still possible to observe the gradual increase in oxygen between the 1960s and 1990s in agreement with the overall cooling of the entire water column in the Labrador Sea from 1966 to 1994 [37]. In 2008, convection reached to 1600 m depth, freshening and cooling the LSW somewhat (figure 2). The volume of the convected water was smaller than in 2014/2015 [27], and the averaging over the convection region obscured the signal in 2008. The low oxygen values in 2011 are connected to an anomalous warm, saline and less dense LSW (figures 2 and 3). Furthermore, the water column of the Labrador Sea was the warmest and most saline observed throughout the past 40 years, and the year 2011 was the second year in a row with convection depths less than or equal to 800 m. After 2012, the oxygenation of LSW re-emerged, and in spring 2016, oxygen concentrations exceeding 295 µmol kg⁻¹ extended down to 1500 m depth (figure 3a). Although the oxygen increase since 2012 is impressive, the concentrations are still less than in the years with intense deep convection in the early 1990s, where oxygen concentrations larger than 295 µmol kg⁻¹ covered the water column down to 2000 m depth. One reason for that is the temperature dependence of the oxygen solubility. In 2015, the LSW layer was still warmer by 0.5°C than 20 years ago (figure 2). This temperature difference is equivalent to a difference in the oxygen solubility equilibrium of about 4 µmol kg⁻¹ with colder temperatures favouring a higher uptake of oxygen.

The influence of the convective activity in the Labrador Sea on the storage of C_ant is shown in figure 3b. C_ant is calculated from CFC observations, using the transit time distribution (TTD) method [10,38]. The reduction in the uptake of C_ant by LSW between the NAO high period related with intense convection in the 1990s and the period of weaker convection after 1997 has already been reported [10,39]. Figure 3b shows that this reduction mainly occurs in the deeper LSW layer, whereas C_ant uptake in the uLSW follows the increase in the surface layer caused by rising atmospheric CO₂. Owing to the reduced ventilation of the dense LSW, C_ant increased only slowly in layers with σ_θ > 27.74 kg m⁻³ between 1996 and 2013, lagging the faster increase rate close to the surface. The reoccurrence of deep reaching convection in winter 2013/2014 has a direct impact on the uptake of C_ant in LSW and leads to a rapid increase also in the denser LSW mode.

The dramatic shift from the dominance of the denser mode dLSW over uLSW after the cessation of intense convection in 1994 is highlighted by a comparison of the layer thickness of the two modes (figure 4). In the early to mid-1990s, dLSW covered a depth range of more than 2000 m, while uLSW was an almost negligible entity. After 1997, only uLSW was ventilated and produced, and in 2004, both layers had a thickness of about 1000 m. Since that time, uLSW was the dominant mode of LSW occupying the Labrador Sea, with the thickness in 2014 exceeding 1300 m. Owing to intensifying convection in 2015, the thickness of the dLSW layer increased again.

• Download figure
• Open in new tab
• Download PowerPoint

3. Main pathways of Labrador Sea Water and C_ant inventories in the subpolar North Atlantic

Convection variability induces anomalies in hydrographic and other properties of LSW, and these signals are exported along the main LSW pathways into the Irminger Sea, the eastern Atlantic and into the subtropics (figure 1; e.g. [6–8,13,15,21]). By mixing and lateral exchange with surrounding water masses along the pathways, these signals or properties are gradually changed, and extrema imprinted in the Labrador Sea are smoothed on the way south. Decadal observations of anthropogenic tracers such as CFCs allow the calculation of basin-wide distributions of water mass fractions and thus highlight spreading pathways. To do so, Rhein et al. [13] applied the TTD approach to CFC data from the subpolar North Atlantic. They estimated the fraction of the two LSW modes (uLSW and dLSW) younger than 20 years within the respective density ranges for two different periods (1996–1999 versus 2007–2010) representing different states of convection intensity. Following this approach, figure 5a,b presents maps showing the spatial distribution of water mass fractions for the entire LSW layer (uLSW + dLSW). The years 1996–1999 represent the period after the cessation of convection intense and deep enough to form and renew dLSW, while the lighter uLSW was still produced. Over the years 2007–2010, convection intensity was generally lower. In both cases, the highest fractions of young, recently formed LSW are observed in the Labrador and Irminger Seas, with substantially smaller fractions in the second period 2007–2010. This reflects the weakened formation of LSW over these years. In the eastern subpolar Atlantic, the fraction of young LSW increased, due to the ongoing import of ventilated LSW formed in the 1990s. As a consequence, LSW in the western basins became less ventilated and LSW in the eastern basins more ventilated in 2007–2010.

• Download figure
• Open in new tab
• Download PowerPoint

Weakening ventilation reduces the supply of oxygen and transient tracers into the LSW layer and therefore has also consequences for the oceanic uptake and storage of C_ant [10]. The oceans absorb and store a significant portion of the C_ant emissions, but large uncertainties remain in the quantification and regional distribution of this carbon sink [16]. Using the same time periods as in figure 5a,b, also the C_ant inventories were calculated for the subpolar North Atlantic. To highlight the change in the C_ant inventories caused by the ventilation changes between the two time periods, the C_ant inventory in 1996–1999 is projected to the year 2010, assuming that the ocean was in steady state (figure 5c). If that would be indeed the case, the observed C_ant distribution in 2007–2010 (figure 5d) should closely resemble the projected inventory (figure 5c). The general features are similar: higher inventories are found in the western basins, especially the Labrador Sea, while the eastern Atlantic shows enhanced inventories along the known spreading pathways of LSW. A closer look reveals that the observed inventories (figure 5d) are smaller than the projected ones, mainly in the Labrador and Irminger Seas, indicating that the rate of input of C_ant decreased. This is explained by a decrease in ventilation rate, supported by the decrease in young LSW between the two time periods. In the eastern Atlantic, C_ant rose faster than expected from the projection. As already mentioned, the eastern basin became more ventilated due to the ongoing import of young LSW from the 1990s.

4. Labrador Sea Water export into the South Atlantic

Based on 25 years of CFC observations in the Atlantic, the Deep Western Boundary Current (DWBC) shows up as the main and fastest conduit transporting LSW from its formation region into the tropical Atlantic. It represents the most direct link between these regions and carries the least diluted LSW properties [11,13,40]. Interior pathways also exist, especially in the transition zone between the subpolar and the subtropical gyre [13,41,42]. Here, the 47° N section [42–45] crosses this regime from the Canadian Shelf in the west towards the European Shelf in the east delivering annually repeated hydrography and tracer measurements for the western basin since 2007 (less frequent in the eastern basin and for the whole section in earlier years).

Oxygen measurements in the western basin along this section reveal the highest concentrations to be found in the DWBC at the Canadian continental slope with maximum values in the years before 2008 (figure 6). Owing to the reduced convection activity, oxygen decreased not only in the DWBC but also farther east, as less oxygenated LSW invaded the interior of the Newfoundland Basin. Any impact resulting from the 2008 episode of more intense convection is—as in the central Labrador Sea (figure 3)—not visible here. The increased convection starting in 2013/2014 has led to an increase in the oxygen concentrations in the most recent years in the DWBC (figure 6). Both the temporal and spatial features of CFCs and oxygen point to the DWBC as the main pathway for recently ventilated LSW.

• Download figure
• Open in new tab
• Download PowerPoint

uLSW anomalies are also transported through the Flemish Pass, a shallow passage (sill depth 1200 m) located to the west of Flemish Cap. The pass cuts almost meridionally through the Canadian continental shelf at about 47° W, providing a shortcut for uLSW to the south [44]. The major part of uLSW however follows the DWBC around Flemish Cap. The interannual variability of uLSW properties at both locations at 47° N (Flemish Pass and DWBC) follows the general time evolution of uLSW in the Labrador Sea (figure 7). Main signals like the warming and salinification in 1995 and in 2003, and the decrease in 2008 are also found downstream, compatible with a spreading time of less than 1 year [22]. The 20-year trends for temperature and salinity of the central Labrador Sea and the two sites at 47° N agree with each other within the uncertainties. Temperature and salinities at 47° N however are higher than in the formation region. Schneider et al. [44] found that the uLSW in the boundary current was already warmer and saltier at 52°–53° N, speculating that most of the transition between the cooler, fresher LSW in the convection region and the more warmer and saltier conditions found on both downstream locations (figure 7) already took place in the Labrador Sea between the interior and the boundary current.

• Download figure
• Open in new tab
• Download PowerPoint

Other sites with continuous measurements reporting the advent of LSW anomalies in the DWBC originating in the Labrador Sea are at ‘Line W’ (39° N, 69° W) [40] and at 26° N [46], supporting the continuous spreading of LSW anomalies along the DWBC pathway. However, not all characteristics of LSW are maintained throughout the pathway to the Southern Hemisphere. Further away from its source, LSW has lost the salinity minimum signal by mixing with salty water from the Mediterranean Sea and by encountering the fresher Antarctic Intermediate Water. The CFC maximum of LSW, however, remains [47,48]. Rhein et al. [13] presented the most recent analysis on the LSW pathways from the Labrador Sea to 20° S. They calculated age and fraction of young LSW and compared the distributions with the spreading of the Denmark Strait Overflow water, the densest contribution to NADW. In the subpolar gyre, the pathways north into the Irminger Sea and east into the northeast Atlantic are clearly visible by the high fractions. South of the subpolar gyre, the elevated fraction of young LSW (figure 8) emphasizes the role of the DWBC as the fastest export pathway carrying the highest fraction of young water [13].

• Download figure
• Open in new tab
• Download PowerPoint

In the transition zone between the subpolar and the subtropical gyre, a relatively strong downstream decrease in young LSW in the boundary current was observed [13]. Here, the North Atlantic Current flows closely to the DWBC and carries older deep water towards the north. This old water mixes with the young water in the southward flowing DWBC leading to an apparent slower propagation in the tracer ages. Interior pathways are also visible in the tracer age distributions and in other hydrographic properties [41,42].

In the tropical Atlantic, the young LSW follows the well-known zonal flow of deep water that leaves the boundary [13,47,48]. Despite these zonal excursions, a notable export of young LSW with the DWBC into the Southern Hemisphere is still obvious (figure 8).

5. Labrador Sea Water anomalies, the Atlantic Meridional Overturning Circulation and the ‘cold blob’

Under global warming, in most of the climate models and the climate model mean, the surface air temperatures in years 2080–2100 show a conspicuous cool anomaly in the North Atlantic at roughly 40°–60° N [49]. This cooling is related to a reduction in the AMOC in the models and thus to a reduced northward meridional heat transport and also to a weaker deep-water production [49,50]. A cold trend over the North Atlantic is also found in the overall global warming pattern in the historical surface temperatures from 1901 to 2010 (dubbed ‘cold blob’), although this cooling trend is not statistically different from zero (e.g. fig. 2.21 in [51]). Ignoring the latter, Rahmstorf et al. [52] considered this so-called ‘cold blob’ as an indication to argue that the AMOC had already slowed down during the twentieth century.

The surface temperature time series in the North Atlantic reveals that the appearance of the ‘cold blob’ region since 1901 is governed by large decadal variability and not so much by a long-term trend (e.g. fig. 2.22 in Hartmann et al. [51]). The surface warmed until the 1940s, followed by substantial cooling in the 1950s–1980s, and beginning in the mid-late 1990s, the subpolar North Atlantic warmed. As discussed above, the history of LSW ventilation is known for the last 70 years or so by hydrographic measurements [3]. It is evident that cold atmospheric anomalies are linked to more intense convection activities in the Labrador Sea. How this variability is related to the strength of the AMOC cannot be studied with observations, because the appropriate time series are lacking or are not long enough [53]. Danabasoglu et al. [17] analysed the AMOC variability in 20 global ocean–sea ice coupled models covering the time period 1958–2007 and related this to variations of the mixed layer depth in the LSW formation area and the changes in the NAO (figure 9). The modelled AMOC decadal variability in this time period is of the order of 1–2 Sv, i.e. about 10% of the mean transport. Until the mid-1970s, the AMOC was relatively steady and slightly weaker than the average of 1958–2007. This was followed by a stronger AMOC with maximum transports in the mid-1990s. Afterwards, the AMOC weakened until the end of the integration in 2007. The strengthening of the AMOC from the mid-1970s to the mid-1990s is connected to enhanced deep water formation and associated mixed layer deepening (convection) in the subpolar North Atlantic and especially in the Labrador Sea (figure 9). The enhanced convection activity is driven by high surface buoyancy fluxes and stronger wind stress related to the positive phase of the NAO, which is linked to a colder than average atmosphere over the Labrador Sea (figure 9). In short, cold atmospheric surface anomalies (a cold blob) are linked on decadal to multi-annual time scales to large air–sea heat fluxes, intense ventilation and a stronger than normal AMOC.

• Download figure
• Open in new tab
• Download PowerPoint

Only if we go to time scales longer than decadal does variability of ocean dynamics exert an important influence on the surface heat fluxes. Then, cold sea-surface temperatures are correlated with weaker air–sea heat fluxes [54], linking cold anomalies with reduced ventilation and formation of deep water and a potentially weaker AMOC as found in the climate models [49]. So far, it seems that the atmospheric modes/anomalies drive the interannual to decadal variability of the ventilation in the Labrador Sea since 1900 and thus the AMOC. Rahmstorf et al. [52] suggested that melting of the Greenland ice sheet has already contributed to the slowdown of the AMOC. Additional freshwater might reduce the density in the surface layer and suppress ventilation and formation of deep water. Until now, the air–sea heat fluxes set the convection scenario, but over the past 20 years, the mass loss from Greenland had almost quadrupled [55]. The crucial question is how much of the melt leaves the boundary current around Greenland and is imported into the convection region and how long is it accumulated there. Böning et al. [56] analysed a global ocean circulation model with a resolution capable to capture the relevant small-scale processes in the subpolar North Atlantic (grid size 3 km). The authors concluded that only a small fraction of the melt entered the convection area and did not yet impact convection in the Labrador Sea, but might do so in the coming years.

6. Summary

Ventilation of about 30% of the subpolar North Atlantic occurs in a relatively small region in the central Labrador Sea, and the LSW formed there by deep convection is a main contributor to the deep water masses of the North Atlantic and thus to the AMOC. Owing to hydrographic measurements, the history of deep convection of the past 70 years is known fairly well with the convection depth and temperature/salinity anomalies as proxies: periods of intense convection were 1972–1976, and even more so in 1987–1994. Anomalous deep convection in winter 2007/2008 interrupted the long period with weaker ventilation. In winter 2013/2014, anomalous cold air masses were present, and a more intense convection activity was resumed. The variability in the formation of LSW has consequences for the ventilation of the interior ocean and thus for the spatial distribution of oxygen, anthropogenic carbon and trace gases. With suppressed convection, less oxygen and C_ant enters the mid-depth ocean, and these changes can be followed along the LSW pathways.

The interannual to decadal variability in convection intensity so far is driven by the variability of atmospheric anomalies: cold atmospheric temperatures cause strong ocean heat losses in winter, weakening or erasing the density stratification in the water column and initiating deep convection. The cold air could be locally constrained over the Labrador Sea as in 2008 or widely spread over most of the North Atlantic between 40° N and 60° N as, for instance, in 2014–2015. The long intense convection period between 1987 and 1994 was linked to a strong positive phase of the NAO, with colder than average air temperatures over the Labrador Sea. Whether the variability of the convection modifies the transport of the AMOC cannot be discerned from observations, because of the lack of long-term relevant time series. In state-of-the-art coupled ocean–sea ice models, air–sea heat flux anomalies associated with LSW convection are the main driver of AMOC variability on annual to decadal time scales, while shorter periods are driven by local wind stress variability. On multi-decadal time scales, the link between AMOC and ocean heat loss still holds, but on these long periods, the ocean is the driver, linking cold air temperature anomalies with reduced LSW formation and a weaker AMOC.

It is obvious that long-term (several decades) continuous observations of the AMOC transport variability in the subpolar North Atlantic or at least of the main contributors are urgently needed. Long-term observations studying the state of convection in the Labrador Sea have to be carried out to document changes that are occurring and provide data to evaluate models used for future predictions. The availability of hydrographic data has been significantly augmented by the Argo Program [57], allowing a subseasonal coverage and a much more detailed vision on convection (figure 2). However, the shipboard measurements are also needed in the future for calibration issues and also for the measurement of transient tracers, nutrients and other carbon cycle-related parameters.

In order to analyse the role of increasing melt from the Greenland ice sheets on the LSW formation, the analysis of helium and neon isotopes in the Labrador Sea will be added to the tracer suite in the near future. These noble gases quantify the fraction of glacial melting in the convection area in an unambiguous way. Many processes may change the salinity in the Labrador Sea, but elevated neon concentrations can only originate from glacial melting [58]. For now, the import of glacial melt into the central Labrador Sea and its consequences for LSW convection was studied with models [56]. Recently, a time series of freshwater fluxes (1978–2013) into the Labrador Sea was reconstructed from observations. It shows an increase in freshwater in the last two decades, and part of it might enter the central Labrador Sea [59]. However, since 2013/2014, LSW convection increased, driven by the elevated air sea buoyancy fluxes, overriding the potential influence of the freshwater increase.

Data accessibility

The data reported here are taken from the Glodapv2 repository (http://cdiac.ornl.gov/oceans/GLODAPv2/). Argo data were collected and made freely available by the International Argo Program and the national programmes that contribute to it (http://www.argo.ucsd.edu, http://argo.jcommops.org). The Argo Program is part of the Global Ocean Observing System. The time series of the LSW layer thickness, oxygen and C_ant concentrations have been augmented by recent own data from Bedford Institute and from IUP-MARUM. The IUP-MARUM and BIO data are available upon request. The IUP-MARUM temperature, salinity, oxygen and CFC data as well as the gridded data on which the oxygen figures in this paper are based have been submitted to the PANGAEA data repository.

Authors' contributions

M.R. prepared the first design of the manuscript and wrote the first draft. R.S., D.K., I.S. and I.Y. cooperated in producing the figures. All authors contributed equally to improve the design and to revise the manuscript critically for important intellectual content. For the revised version, M.R. provided the first draft, and R.S., D.K. and I.Y. provided improved figures, and all authors contributed to the revision of the text and contributed to the response to the reviewer's comments. They contributed to the analysis, interpretation, writing and approved of the submitted revised manuscript.

Competing interests

We declare we have no competing interests.

Funding

M.R. received funding from the German Ministry for Education and Research (BMBF) in the RACE programmes (grants 3F0651C, 03F0729A), from the DFG (grants RH25/36, RH25/40) and from the DFG funded Cluster of Excellence ‘The Oceans in the Earth-System—MARUM’, subproject OC1 and D.K. from DFG-FLEPVAR, grant KI1655-1/1.