Role of air-sea fluxes and ocean surface density on the production of deep waters in the eastern subpolar gyre of the North Atlantic

Wintertime convection in the North Atlantic Ocean is a key component of the global climate 10 as it produces dense waters at high latitudes that flow equatorward as part of the Atlantic Meridional Overturning Circulation (AMOC). Recent work has highlighted the dominant role of the Irminger and Iceland basins in the production of North Atlantic Deep Water. Dense water formation in these basins is mainly explained by buoyancy forcing that transforms surface waters to the deep waters of the AMOC lower limb. Air-sea fluxes and the ocean surface density field are both key determinants of the buoyancy15 driven transformation. We analyze these contributions to the transformation in order to better understand the connection between atmospheric forcing and the AMOC. More precisely, we study the impact of airsea fluxes and the ocean surface density field on the transformation of subpolar mode water (SPMW) in the Iceland Basin, a water mass that ‘pre-conditions’ dense water formation downstream. Analyses using 40 years of observations (1980–2019) reveal that the variance in SPMW transformation is mainly 20 influenced by the variance in density at the ocean surface. This surface density is set by a combination of advection, wind-driven upwelling and surface fluxes, the latter explaining ~30% of the variance in outcrop area as expressed by the surface area between the outcropped SPMW isopycnals. The key role of the surface density on SPMW transformation partly explains the unusually large SPMW transformation in winter 2014–2015 over the Iceland Basin. 25 https://doi.org/10.5194/os-2021-48 Preprint. Discussion started: 1 June 2021 c © Author(s) 2021. CC BY 4.0 License.


Introduction and Background
Recent observational studies have identified two main source regions for the dense waters that constitute the lower limb of the Atlantic Meridional Overturning Circulation (AMOC): the Nordic Seas, and the Irminger and Iceland basins (Bringedal et al., 2018;Chafik & Rossby, 2019;Lozier et al., 2019;Petit et 30 al., 2020). While these two regions produce approximately the same amount of dense waters, the interannual variability of the AMOC in the subpolar North Atlantic can be largely attributed to variability in the dense waters produced in the eastern subpolar region, rather than those imported from the Nordic Seas. A recent study based on observations shows that the production of dense waters in the Irminger and Iceland basins is mainly due to local buoyancy forcing (Petit et al., 2020). Specifically, between the 35 Greenland-Scotland Ridge and the OSNAP East section (Fig. 1), 7.0 ± 2.5 Sv of light water was transformed into dense water of the AMOC lower limb in 2014-2016. This transformation is consistent with an average overturning of 6.6-7.6 ± 3.8 Sv between the two sections during those same years.
Because the transformation of dense waters mainly drives AMOC variability in this region, we seek to 40 understand what factors drive variability in the transformation, which was twice as large in winter 2014-2015 as it was in winter 2015-2016 (Petit et al., 2020). Two variables are key to transformation estimates: the air-sea fluxes and the ocean surface densities. The air-sea fluxes are known to be highly variable in time over the subpolar gyre. For instance, Josey et al. (2019) showed strong seasonal and interanual variability of the net heat flux over the Irminger Sea from a high-resolution surface flux mooring. Changes 45 in net heat flux-as opposed to changes in the surface salinity field or in the surface freshwater flux-are considered to be mainly responsible for the amount of transformed water (Grist et al., 2014). Another recent study, however, has suggested that ocean surface properties play a role in the transformation of surface waters. Oltmanns et al. (2018) revealed that warm and fresh summers in the subpolar gyre are associated with a reduced buoyancy loss the following winter. Hydrographic property changes are thus 50 expected to affect the transformation of surface water by modifying the surface density fields.
Observational studies in the last few years have highlighted large temperature and salinity changes over the eastern subpolar gyre (Holliday et al., 2020;Josey et al., 2018). The density anomalies associated with these property changes are expected to influence the overturning over the Irminger and Iceland Basin,55 and thus the amount of dense water transported southward to the subtropical gyre (Jackson et al., 2016;Zou et al., 2020). At odds with this expectation, however, is a recent study (Fu et al., 2020) that showed a relatively stable AMOC state since the 1990s, in spite of large hydrographic property changes over the subpolar gyre during the same time period. This stability suggests either that hydrographic property changes have only a small influence on the transformation or that a lagged response of the overturning to 60 property changes applies over longer time scales than their 30-years period of observations. Motivated by these past studies, we assess the role of hydrographic property changes on the formation of dense water in the eastern subpolar North Atlantic.
In this study, we focus on the factors that determine the transformation of subpolar mode water (SPMW) 65 in the Iceland Basin. SPMW is identified by uniform hydrographic properties in the winter mixed layer of the subpolar gyre (McCartney & Talley, 1982) and is often defined by potential vorticity < 4x10 -11 m -1 s -1 over the eastern subpolar gyre (Talley, 1999). In contact with the atmosphere, its hydrographic properties vary along the cyclonic pathway of the subpolar gyre, increasing in density westward across the Iceland Basin from 27.3 to 27.5 kg m -3 Thierry et al., 2008). The SPMW, 70 considered the 'pre-conditioned' water mass that forms North Atlantic Deep Water (NADW), is further densified in the Irminger and Labrador Seas (de Boisséson et al., 2012;. The westward freshening and cooling of SPMW, evident in the OSNAP East hydrographic section from 2014-2016 (Fig. 1b), transforms light SPMW in the North Atlantic Current (NAC) branches to the 75 relatively dense NADW in the East Greenland Current. As observed by the spread of temperatures and salinities along the 27.74 kg m -3 in the Irminger Gyre (Fig. 1b), the NADW imported from the Labrador Sea to the Irminger Sea has a similar density to (Faure & Speer, 2005;Fischer et al., 2018;Jakobsen et al., 2003;Petit et al., 2019), but is slightly fresher and colder than, the NADW formed by SPMW densification within the eastern subpolar basin and exported via the East Greenland Current. The 80 https://doi.org/10.5194/os-2021-48 Preprint. transformation of SPMW in the layer 27.3-27.5 kg m -3 over the Iceland Basin (line segments 1-4 in Figure 1a) is the focus of our study.
As explained in Sect. 2, we use observational datasets and sensitivity experiments to explore the dependence of SPMW transformation on the air-sea fluxes and surface densities over a 40-year period. 85 We examine the linkage between buoyancy forcing and the surface density field in Sect. 3, while the sensitivity experiments are discussed in Sect. 4. Section 5 is focused on the unusually large SPMW transformation observed in winter 2014-15 and we summarize our results in Sect. 6.

The OSNAP observations
We used gridded data across the OSNAP East section to observe the evolution of hydrographic properties in the Iceland Basin and Irminger Sea ( Fig. 1; Lozier et al. (2017)). Based on thirty-two moorings, the 100 gridded section has a nominal horizontal resolution of ~25 km from the Scottish shelf to the southeast tip of Greenland, and a vertical resolution of ~20 m (Lozier et al., 2019). The data are composed of continuous measurements of salinity, temperature and velocity between August 2014 to May 2018. The gridded data set was constructed by combining OSNAP mooring data with other observations in the region following an Objective Analysis (Li et al., 2017). The other observations include Argo Profiling 105 Floats, OSNAP gliders, shipboard Conductivity, Temperature and Depth stations carried out during the 2014 and 2016 OSNAP cruises, and the WORLD Ocean Atlas 2018 climatology.

Transformation estimated from Atmospheric Reanalysis
To estimate the transformation of surface water by air-sea fluxes, we use heat and freshwater fluxes of the European Centre for Medium Range Weather Forecasts Reanalysis 5 (ERA5) atmospheric reanalysis 110 (Poli et al., 2016). These monthly fields are combined with monthly surface temperature and salinity fields over the eastern subpolar gyre between 1980-2019. Surface temperature is derived from ERA5, while salinity, at 5-m depth, is derived from quality-controlled profiles in EN4.2.1 (Good et al., 2013).
With a horizontal resolution of 1°, the gridded salinity field is subsampled onto the finer ERA5 horizontal grid of 30 km. We also use EN4.2.1 to calculate potential vorticity ( is the Coriolis parameter and ρ is the potential density. The evaporation (E), precipitation (P), and net heat flux into the ocean (Q) are used to estimate the surfaceforced transformation across an isopycnal, s, according to (Speer & Tziperman, 1992;Tziperman, 1986;Walin, 1982): 120 where, a is the thermal expansion coefficient, b is the haline contraction coefficient, CP is the specific heat, and S is the 5-m salinity. Temperature and salinity are used to calculate surface density (Gill, 1982). For each 125 month and each isopycnal, s, the local buoyancy flux (term in square brackets) is integrated over the surface area of the associated density bin, Ds. Thus, SFT has a non-zero value for those months when the specified isopycnals outcrop, otherwise it is set to zero. To focus on the transformation within the SPMW layer (27.3-27.5 kg m -3 ), we estimate the transformation across the isopycnal 27.4 kg m -3 toward higher or lower densities using a density bin size of Ds = 0.2 kg m -3 . From this equation, we note that air-sea 130 fluxes influence the transformation both directly, through the buoyancy flux term, and indirectly by partially setting the area covered by a particular surface density range, which is also influenced by the ocean circulation. We thus expect these two factors (i.e. surface density and buoyancy flux) to share some dependence. We explore this dependence in Sect. 3.

135
Previous studies have shown the utility of transformation maps to highlight the migration of intense transformation for different density bins over the subpolar and subtropical gyres Maze et al., 2009). Although the equation above can be used to integrate the transformation for any density bin, the integrand is calculated only over the surface area for the specified isopycnals that bracket the density of the source water for SPMW. Thus, the integration yields the magnitude of the 140 transformation over the domain. In Sect. 4, we visualize the spatial distribution of the SPMW transformation variability by mapping the standard deviation of SFT and explore its sensitivity to two factors: the surface density and the air-sea fluxes. As further described below, annual SPMW transformations are estimated with time variable and/or monthly climatological surface density ( , which includes the variability of surface salinity and temperature) and air-sea fluxes (ℱ, which includes the 145 variability of Q, E and P).

Surface area for the source water
The spatial pattern of the climatological mean surface density shows that the SPMW layer outcrops over a large part of the Iceland Basin during winter (Fig. 1a). To quantify the interannual variability of this outcrop between winters, we estimate the area between the two outcropped isopycnals of the SPMW layer 150 (27.3-27.5 kg m -3 ) over our study domain, which is bounded by the OSNAP East section (1-4), the top of the Reykjanes Ridge and the Iceland-Faroe-Scotland Ridge, as indicated in Fig. 1a. This area is hereafter referred to as the surface area of the source water, or simply the surface area.
The mean surface area of the source water over the Iceland Basin is 3.67x10 11 m 2 and is, surprisingly, 155 highly variable over the period 1980-2019, with a standard deviation of 1.53x10 11 m 2 (Fig. 1c).

Gyre Boundary estimated from AVISO 160
To estimate the subpolar gyre boundary, we use monthly absolute dynamic topography fields from the gridded ¼° AVISO (Archiving, Validation and Interpretation of Satellite Oceanographic data center) altimeter products distributed by CMEMS (Copernicus Marine Environment Monitoring Service). For ease, absolute dynamic topography is hereafter referred to as sea surface height (SSH). Following previous work, the gyre boundary is defined as the largest closed contour of the monthly SSH field with 165 1-cm contour intervals (Foukal & Lozier, 2017).

Influence of buoyancy loss on the structure of the upper ocean density
We begin our study of the influence of air-sea fluxes and ocean surface densities on the SPMW transformation in the Iceland Basin by evaluating the independence of these variables. We first examine the influence of the local buoyancy flux on the vertical density profile for each winter over the OSNAP 170 period from 2014 to 2018, a period during which Petit et al. (2020) identified large interannual variability in the buoyancy forcing. These profiles reveal the ventilation of SPMW (again, identified by potential vorticity < 4x10 -11 m -1 s -1 ) at the beginning of winter and its restratification at the onset of spring (Fig. 2a).
The 2014-2015 winter stands out among these profiles, as it is marked by both the strongest buoyancy flux (8x10 -6 W m -2 in December; Fig. 2a) and the deepest SPMW (600 m in March; Fig. 2b). Conversely, 175 the 2016-2017 winter is associated with a weak buoyancy flux (4x10 -6 W m -2 in December) and a shallow SPMW (250 m in March). Despite these agreements, the linkage between the SPMW thickness and the buoyancy flux over the period 1999-2019 is weak, with a correlation of only 0.52 (Fig. 2c). It is possible that this lower than expected correlation can be explained by year to year variability in the surface density field. Even if the buoyancy forcing is relatively strong over the Iceland Basin for a given winter, the 180 surface area needs to be large enough to have a sizeable impact on the SPMW layer. This possibility raises the question as to the influence of the buoyancy forcing on the surface area.  (2)), and the difference in SPMW thickness March-April between the end of the first (dzma(1)) and second (dzma(2)) winters over 1999-2019. The SPMW thickness was estimated by the depth of the potential vorticity isoline 4x10 -11 m -1 s -1 over the source water area (as shown in Fig.   2b). EN4.2.1 data before 1999 are not used because of large uncertainties due to scarce observations at depth. (d) Correlation between the 190 buoyancy flux averaged over December-January-February in the second winter (Buoydjf(2)), and the difference in surface area between the end of the first (areama(1)) and second winters (areama (2)). To answer this question, we subtract the surface area at the end of each winter from the surface area at the end of the following winter, for each year. We compare this area difference to the buoyancy forcing 195 over the second winter as we want to understand whether the buoyancy forcing is setting the surface area observed at the end of that second winter (Fig. 2d). Though a strengthening of the buoyancy forcing generally leads to an expansion of the surface area, the buoyancy flux in a given winter can explain only 30% of the surface density change in the Iceland Basin. The remaining variance can be attributed to ocean advection, mixing and wind-driven upwelling, although no single process is likely dominant. To 200 conclude, we assume that the dependence of the surface density field on the air-sea fluxes is sufficiently small so that we can separately explore the contributions of the air-sea fluxes and ocean surface densities on the interannual variability of SPMW transformation.

225
However, the spatial pattern of the buoyancy flux variance differs considerably from the spatial pattern of the SPMW transformation variance. Over the subpolar gyre, the largest variance in SPMW transformation is localized over the Reykjanes Ridge, Hatton-Rockall Plateau and Rockall Trough (Fig. 3d), while the largest variance in buoyancy flux is localized over the Labrador Sea and the NAC 230 (Fig. 3c). Over the area defined by our black box, large variance in buoyancy flux is noted over the interior of the Iceland Basin (Fig. 3e), while the variance in SPMW transformation is small (Fig. 3f). These large spatial differences suggest that buoyancy fluxes are unlikely to be the only driver of variability in SPMW transformation. The same analysis has been performed over the two periods differentiated by their standard deviation in surface areas in Fig. 1c ( 1980-1999and 2000Fig. 5). The interannual variability of SPMW 250 transformation is driven primarily by the variance in surface density during both periods. This comparison also shows a relatively stronger variance of the transformation in 1980-1999 compared to 2000-2019, although their maxima remain localized over the Reykjanes Ridge and Rockall Trough. This highly variable transformation during the first period is consistent with a relatively large variation of surface area during those years (Fig. 1c). 255

Unusually large SPMW transformation in winter 2014-15
We next examine the conditions leading to intense transformation of SPMW over the Iceland Basin. The monthly time series of SPMW transformation in Fig. 6a clearly shows that the transformation of surface water to a density greater than 27.4 kg m -3 reached its largest value since 1980 during the 2014-15 winter, 270 with a net transformation of 23.9 Sv in January 2015. Though the buoyancy anomaly is relatively strong during this winter, it is not the largest anomaly observed over the 40-year record (Fig. 6b), suggesting that additional conditions are required for large transformation. An examination of winter 2010-11 reveals that in spite of sizeable buoyancy forcing, the transformation appears to be limited by a relatively small surface area. Winter 2013-14 shows the opposite: though the surface area is sizeable the relatively 275 weak buoyancy forcing appears to limit the transformation. In contrast to these two winters, the large SPMW transformation in winter 2014-15 is associated with relatively strong buoyancy fluxes and a relatively large surface area that covers most of the Iceland Basin as defined by the black box in Fig. 6c.
By comparing the surface areas of the three winters (Fig. 6c), we note that their contraction or expansion 280 is mainly related to the position of the isopycnal 27.3 kg m -3 over the southeastern part of the Iceland Basin, south of our box. To characterize the spatial pattern of the surface area during winters of 'small' and 'large' areas, we select the five winters associated with the smallest, and five associated with the largest, surface areas of the source water (Fig. 7a). These areas were estimated over a larger box (shown in Fig. 7c) so as to include the shift of the isopycnal 27.3 kg m -3 south of the OSNAP line. The source 285 area is localized over the Reykjanes Ridge and the Rockall Plateau during 'small' winters ( Fig. 7c), while it is localized over the interior of the Iceland Basin and the Rockall Trough during 'large' winters ( Fig. 7d). The position of the isopycnal 27.3 kg m -3 over the Reykjanes Ridge shifts north of the Charlie-Gibbs Fracture Zone (CGFZ) during 'small' winters, and south of the CGFZ during 'large' winters.

295
We explore whether the difference in these spatial patterns for the SPMW surface area can be related to an extension or contraction of the subpolar gyre. For this exploration, the subpolar gyre boundary was estimated with the largest closed contour of SSH in January of those winters (Fig. 7b). Any difference in the subpolar gyre boundary between these years is localized over the northern part of the Reykjanes Ridge 300 and the northwest corner of the NAC. Overall, however, we cannot attribute small or large density areas Other possible mechanisms responsible for changes in the surface area include variability in ocean advection and the winter re-emergence of relatively dense SPMW. South of OSNAP East, an exceptionally strong heat loss (Josey et al., 2018) formed an unusually large volume of dense SPMW (Grist et al., 2016) in the winter 2013-14. Grist et al. (2016) analyzed the long-term impact of anomalous SPMW formation on regional climate and found that some of this water mass emerges the following 310 autumn/winter. Thus, we suggest that the large formation of SPMW in winter 2013-14 over the southern part of the Iceland Basin contributed to the large source area of SPMW found over the northern part of the Iceland Basin the following winter 2014-15.

Discussion and Conclusion
Using 40 years of observations , we show that the production of SPMW is correlated to the 320 surface area of the source water and to buoyancy forcing over the Iceland Basin, at a similar level for each variable (R=0.7 and R=0.67 respectively). Our analysis reveals that these two variables are weakly dependent in this region: variance in the air-sea fluxes can explain ~30% of changes in the wintertime surface density field over the Iceland Basin. We thus infer that ocean advection, mixing and wind-driven upwelling account for the remaining variability in the surface area of the source water. 325 The direct and indirect (via surface area changes) influences of buoyancy forcing combine to account for about half of the SPMW transformation variability. However, observation-based sensitivity experiments reveal that the spatial distribution of the SPMW transformation is most sensitive to surface density changes, which sets the surface area for the source waters, as opposed to the direct influence of the air-330 sea fluxes, which is weak at these densities.
A combination of strong buoyancy forcing and a large source area during the 2014-2015 winter produced the largest SPMW transformation in 40 years. Possible causes for the large source area observed over the northeastern part of the Iceland Basin during that winter include a large local surface heat loss during this 335 winter, the re-emergence of a large volume of SPMW formed upstream the previous winter (Grist et al., 2016), and/or the advection of anomalously dense waters into the region.
Further exploration is needed before a specific attribution for the large transformation in winter 2014-2015 is made, however we do note that a number of recent papers have documented strong advective 340 changes in the Iceland Basin over the past few years. In one study (Ortega et al., 2020) the NAC is shown to drive large salinity and temperature property changes at the surface and subsurface, while in another study (Holliday et al., 2020), strong property changes are attributed to changes in the fraction of water from the Labrador Sea that reaches the Iceland Basin. These studies, highlighting the importance of ocean advection, are consistent with a previous paper that showed large hydrographic property changes for 345 SPMW localized over the Reykjanes Ridge that cannot only be explained by the variability of the local air-sea fluxes (Thierry et al., 2008).
Finally, we now understand that the preconditioning of the surface waters in the Iceland Basin is a major contributor of the total waters carried within the lower limb of the AMOC (Petit et al. 2020), and thus a 350 key determinant of AMOC variability within the subpolar North Atlantic. Hence, our study highlights the importance of understanding the factors that influence the surface density field in the Iceland Basin-whether by advection from the subtropics or the western subpolar gyre, or by the influence of local winds that bring cold water to the surface via mixing and/or upwelling, or through surface buoyancy loss-since that factor is of prime importance in determining the transformation of SPMW.  Competing interests. The authors declare no competing interests.