Delayed Recovery of the Irminger Interior From Cooling in 2015 Due To Widespread Buoyancy Loss and Suppressed Restratification

Watermass transformation in the Irminger Sea, a key region for the Atlantic Meridional Overturning Circulation, is influenced by atmospheric and oceanic variability. Strong wintertime atmospheric forcing in 2015 resulted in enhanced convection and the densification of the Irminger Sea. Deep convection persisted until 2018, even though winters following 2015 were mild. We show that this behavior can be attributed to an initially slow convergence of buoyancy, followed by more rapid convergence of buoyancy. This two‐stage recovery, in turn, is consistent with restratification driven by baroclinic instability of the Irminger Current (IC), that flows around the basin. The initial, slow restratification resulted from the weak horizontal density gradients created by the widespread 2015 atmospheric heat loss. Faster restratification occurred once the IC recovered. This mechanism explains the delayed recovery of the Irminger Sea following a single extreme winter and has implications for the ventilation and overturning that occurs in the basin.

The recovery of the Irminger Sea is dictated by the restratification processes which converge warm, buoyant water on the interior, thickening the light water layer.Models and observations show that restratifying waters are advected from the boundary currents by eddies (Fan et al., 2013;Gelderloos et al., 2011;Katsman et al., 2004;Spall, 2004;Sterl & de Jong, 2022;Straneo, 2006aStraneo, , 2006b).In the Labrador Sea, these eddies are known to be produced by baroclinic instabilities in the boundary current (de Jong et al., 2016;Spall, 2004;Straneo, 2006a;Visbeck et al., 1996).In the Irminger Sea, Sterl and de Jong (2022) show that interannual variations in the seasonal restratification are strongly correlated with eddy kinetic energy (EKE) but do not explain what drives EKE variability.
The anomalous densification of the Irminger Sea in 2015 provided a unique opportunity to study restratification processes in the basin.We use repeat hydrography, an Argo product, and EKE from a satellite product to observe changes in both the Irminger interior and Irminger Current (IC) and investigate the dynamics that converge buoyancy on the interior.We show that while the 2015 wintertime buoyancy loss affected both the interior and the IC, the IC regains buoyancy much more quickly than the interior.The slow timescale for the recovery of the interior is primarily set by the time needed for the development of a large horizontal buoyancy gradient between the interior and IC.We show that the multi-year recovery processes and timescales in the Irminger Sea are a consistent extension of the seasonal restratification dynamics that have been described in the Labrador Sea.

Data
Conductivity, Temperature, Depth profiles collected along repeat occupations of the AR07-E hydrographic section (Figure 1a, bold black line) are used to observe full-depth changes in basin-wide potential density structure over the study period  Temperature and salinity from the Roemmich and Gilson Argo gridded climatology and anomalies (RG product) from January 2011 to December 2022 within the Irminger Sea are used to estimate the temporal variability in horizontal and vertical density structure.The RG product uses only adjusted, delay mode quality controlled Argo profiles with additional screening to exclude any floats with salinity drift to give monthly objective maps of temperature and salinity on 58 pressure levels between 0 and 2,000 dbar on a 1° × 1° grid (Roemmich & Gilson, 2009).
Eddy activity is quantified using the AVISO EKE product (AVISO+Altimetry, 2021).The product is derived from satellite altimetry and has a spatial resolution of 1/4° on a monthly time-step.

Methods
The study period, January 2011 to December 2022, is split into three periods: pre-strong convection (January 2011 to March 2015), during strong convection (April 2015 to March 2018), and post-strong convection (April 2018 to December 2022).Temperature and salinity profiles from each AR07-E occupation are interpolated onto a regular 2D grid and averaged over each period to create composite summertime potential density sections across the Irminger Sea (Figures 1b-1d).
Within the Irminger Sea the study focuses on two regions: the interior and the IC on the eastern boundary (Figure 1a, red and magenta crosses respectively).While exchange of waters is also known to occur on the western boundary, through eddies (Fan et al., 2013;Le Bras et al., 2020) and slantwise convection (Le Bras et al., 2022), the western boundary is notably less buoyant than the eastern boundary (Figure 1) (Våge et al., 2011) with less available potential energy for the development of baroclinic instabilities (Pedlosky, 1979).As such, it is expected that exchange with the eastern boundary dominates the convergence of buoyant water on the interior.This is consistent with idealized and observational studies in both the Labrador and Irminger Seas (Spall, 2004;Sterl & de Jong, 2022;Straneo, 2006b;Våge et al., 2011); with Fan et al. (2013) showing that the most buoyant eddies found in the Irminger interior are from the eastern boundary.Thus, we create spatially-averaged time-series of 10.1029/2023GL106501 3 of 9 potential density for only the interior and IC on the eastern boundary using the RG product (Figures 2a and 2b).Potential density referenced to the surface is calculated at each grid-point then an area-averaged potential density profile is calculated for the interior and IC.
Similarly, a time-series of the depth-integrated buoyancy content in the 100-1,000 m layer is calculated for the interior and IC using the RG product (Figure 2c).First, area-averaged in situ density profiles are calculated for the interior and the IC as above.Second, the buoyancy content within the interior, B int , and IC, B ic , is calculated as where ρ(z) is the average in situ density profile in each region, ρ 0 is defined to be 1,037 kg m −3 representing the densest water in the upper 2,000 m, and g is the gravitational constant.
The rates of recovery of the interior and IC are defined as the rate at which buoyancy is gained in each region.The rates are estimated by using a Least Squares fit of a linear trend with a seasonal cycle to the time-series of B int and B ic : where coefficients, a, m, c, and d are optimized for the fit.m is the linear trend, c and d determine the amplitude of the seasonal cycle, and ω is the period of the seasonal cycle (Wunsch, 1996).
The seasonal restratification of the interior is quantified by the buoyancy gained in the interior over the restratification period.The restratification period is defined as the time from minimum B int to maximum B int each year such that the buoyancy gain during restratification is     = max() − min() (Figures 3b and 3c).Lastly, the horizontal density gradient across the IC is approximated by differencing the buoyancy content between an inner point (x1) and an outer point (x2) of the IC (Figure 1a) using Equation 1 at each time step, ΔB x = B x2 − B x1 (Figure 3a).Following Spall (2004) and Straneo (2006a), we square ΔB x at the start of the restratification period each year (Figure 3b).
Δ 2  is used as the horizontal buoyancy difference sets both the number of eddies and the additional buoyancy content the eddies can transport to the interior (Spall, 2004).Additionally, EKE is quantified by spatially averaging all AVISO grid-points in the IC region (EKE ic , Figure 1a) and taking the time-mean over the restratification period each year (EKE R , Figure 3c).We then calculate linear correlations between

Results
The anomalous water mass transformation that occurred in 2015 notably changed the density structure of the whole Irminger Sea (Figures 1b-1d).Following the 2014/2015 winter there is thinning of the light water layer (27.64-27.72 kg.m −3 ) and a thickening of intermediate water layer (27.72-27.78kg.m −3 ) across the Irminger interior, consistent with de Jong and de Steur (2016) and de Jong et al. (2018).Thickening of the intermediate water layer is also seen on both the eastern (between x1 and x2) and western (between −44°E and −42°E) boundaries, resulting in a steepening of the horizontal density gradients (Figure 1c).Between 1,000 and 1,500 m, in the interior isopycnals deepen due to the convergence of anomalously fresh intermediate water recirculating from the Labrador Sea (Fox et al., 2022;Zunino et al., 2020); whereas there is little change at depth on the boundaries.In the post-strong convection period the density structure is returning to the pre-strong convection state, with a deepening of the 27.64 and 27.72 kg.m −3 isopycnals in the interior and a flattening of the isopycnal slope across the boundaries (Figure 1d).As discussed in Section 3 we now focus on the interior and the IC on the eastern boundary and find that the two regions recover from the density anomaly differently (Figure 2, Table 1).In the interior, the density anomaly persists from 2015 through 2018, after which isopycnals above 1,000 m begin to deepen (Figure 2a).This is reflected in the slow rate of buoyancy gain in the interior in the first 3 years after 2015, followed by more rapid gain in the subsequent 5 years (Figure 2c, red).In contrast, in the IC, the density anomaly is quickly exported, with the 27.64 kg.m −3 isopycnal deepening rapidly in 2015 and 2016, then stabilizing after 2018 (Figure 2b).This is seen in the rapid gain of buoyancy in the IC in the first 3 years after 2015, followed by slower gain in the subsequent 5 years (Figure 2c, pink).Estimates of the rates of recovery for the interior and IC are shown in Table 1.
The difference in the evolution of the buoyancy content in the interior and the IC results in a change in the horizontal buoyancy gradient across the IC, ΔB x (Figure 3a).Before the onset of strong convection ΔB x oscillates around 1 m 2 s −2 .In 2015 there is little change in the mean of ΔB x as both the buoyancy content on the inner and outer edge of the IC decrease similarly.However, from 2016 to 2018, there is a steady increase in ΔB x due to the more rapid increase in buoyancy on the outer edge of the IC (at x2) than on the inner edge (at x1).In the years after 2018 ΔB x appears to decrease slightly, consistent with eddy flux from the IC to the interior working to flatten the isopycnal gradient.
Lastly, the buoyancy gain during restratification is compared to the horizontal buoyancy gradient and the EKE.Changes in the horizontal buoyancy difference are correlated with changes in the buoyancy gain in the interior over the restratification period, with r Bx = 0.57 (p = 0.06) (Figure 3b).The correlation between EKE and buoyancy gain is even higher, with r eke = 0.82 (p = 0.002) (Figure 3c).While these correlations are calculated at zero lag, each data point is an average over the 8-month restratification period such that any lag on shorter timescales, for example, due to the duration of eddy advection, is lost.This high correlation at zero lag is consistent with Fan et al. (2013) who show that eddies with lifetimes of around 4 months travel from the IC to the interior.The improved correlation between EKE and buoyancy gain during restratification indicates that while baroclinic instabilities are likely a major source of buoyancy they are not the only eddy mechanism transporting buoyancy into the interior.

Discussion
These results show that restratification dynamics govern the differing rates of recovery of the interior and IC from an anomalously dense and weakly stratified state.Due to the widespread nature of the density anomaly (de Jong et al., 2020;Josey et al., 2019), the IC becomes noticeably denser in 2015 such that there is initially very little change in the horizontal buoyancy gradient between the interior and IC (Figures 3a and 3b).As the IC is a fast-flowing current (Våge et al., 2011), the anomalously cold waters are flushed from the region relatively rapidly and replenished by lighter, warmer waters from upstream.The recovery of the IC results in an increase in the horizontal buoyancy gradient between the interior and IC from 2016 to 2018, driving faster flow in the IC from 2018 to 2020 (Fried & Jong, 2022) coincident with a peak in the seasonal buoyancy gain in the interior.This is consistent with baroclinic instabilities resulting in an eddy flux of buoyancy into the interior, as proposed by de Jong et al. (2016), Spall (2004), andStraneo (2006a).
The recovery of each region can be roughly quantified by comparing the deficit in buoyancy content at a given time, relative to the buoyancy content in March 2015 and with the seasonal cycle removed, to the deficit in April 2015 (Table 1).While it is expected that the recovery of the interior will take longer than that of the IC given its isolated nature, the interior only recovers 12% of the lost buoyancy in the first 3 years-recovering far slower Irminger interior 0.9 ± 0.5 2.7 ± 0.6 12 79 Irminger Current 9.3 ± 1.0 2.2 ± 0.3 70 94 than the typical restratification timescale of a single season (Straneo, 2006a).We propose this is due to buoyancy loss in the IC resulting in, initially, little change in the horizontal buoyancy gradient, limiting convergence of buoyancy by baroclinic instabilities and delaying the recovery.Notably, the 8 years it takes the interior to recover 79% of the buoyancy deficit is much longer than the single season the buoyancy was lost over and is similar to the 10 years that typically separate strong convective events in the Irminger and Labrador Seas (de Jong & de Steur, 2016;van Aken et al., 2011).
While many processes drive exchange between the boundary current and the interior, only a few processes converge buoyancy (i.e., light waters) into the interior.In addition to baroclinic instabilities, baroclinic Rossby waves and other eddy generating mechanisms (e.g., barotropic instabilities) on the eastern boundary may provide an additional buoyancy flux to the interior (Våge et al., 2011).These processes could explain the stronger correlation of buoyancy gain at the interior with EKE than with the horizontal buoyancy gradient.Additional processes may contribute to exchange of intermediate and dense water, such as export on the western boundary (Le Bras et al., 2020Bras et al., , 2022) ) and recirculation in the Irminger gyre (Våge et al., 2011).Here we assume that the effect of these processes on the rate of convergence of light water on the interior is small compared to the baroclinic instability process described above.
The delayed recovery of the interior has implications for wintertime convection.Zunino et al. (2020) found that deep convection was able to persist for 3 years following the strong winter of 2015 due to pre-conditioning of the water column.We posit that the pre-conditioning lasted multiple years due to the lack of increased seasonal eddyflux of buoyancy from the IC.Additionally, Biló et al. (2022) show that a surface salinity anomaly propagated around the Irminger Sea during the late 2010s, suggesting that freshwater accumulation in the interior caused strong convection to cease after 2018.Our results suggest that the freshening of the boundary current may have contributed to the buoyancy gain in the IC that led to the recovery of the interior and ended pre-conditioning.
There are a couple limitations of the analysis presented here.First, the AVISO EKE product only resolves large eddies, with a resolution of ∼40 km (Chelton et al., 2011), so smaller-scale turbulence and mixing resulting in exchange between the IC and the interior are not captured here.Second, the parameterization developed by Spall (2004) to describe the eddy flux of buoyancy is based on a model of the Labrador basin, but the Irminger basin has notably different bathymetry: on its eastern boundary the Irminger Sea is bounded by the relatively shallowly sloping Reykjanes Ridge (in a large-scale sense when averaging over very rough small-scale bathymetry).
The parameterization given by Spall (2004) is expected to hold when the bottom slope is less than 1.5 times the isopycnal slope, which is not the case at any point in the time-period studied here.However, the relatively high correlation between the buoyancy gain during restratification and the square of the horizontal buoyancy gradient suggests that the parameterization holds moderately well.Additionally, the steeper bottom slope on the western boundary still supports the assumption that the eastern boundary is the main source of waters that converge buoyancy to restratify the interior.
In addition to clarifying the restratification process, these results also have important implications for ocean biogeochemistry.Feucher et al. (2022) found that the main drivers of the oxygen variability in the region were changes in vertical mixing and horizontal advection, but do not specifically explore eddy-advection from the IC.Given that oxygen concentrations are known to be low over the Reykjanes Ridge (Petit et al., 2019) we suggest that changes in the eddy-flux of water from the IC may contribute to variability in oxygen concentrations in the interior, in addition to the delayed recovery allowing for reventilation of waters not yet exported from the interior (Feucher et al., 2022).Further work is needed to determine whether variability in eddy-flux from the IC significantly affects the oxygen budget of the Irminger interior.

Conclusion
The densification of the Irminger Sea in 2015 provided an opportunity to study the dynamics of the subsequent restratification and recovery.We find that the recovery of the Irminger interior was two-stage: an initial slow buoyancy gain while the IC recovered, which set up the necessary conditions to drive a more rapid second-stage recovery of the interior.The restratification of the subsurface water column is shown to be due to eddy-shedding from the IC along the Reykjanes Ridge, primarily as a result of baroclinic instabilities.This result is consistent with findings in the Labrador Sea (de Jong et al., 2016;Spall, 2004;Straneo, 2006a) and speculations about the Irminger Sea (Fan et al., 2013;Sterl & de Jong, 2022).The delayed recovery of the interior is a direct outcome of was supported by Grants OCE-1756272 and OCE-1948482.S.G.P. was supported by NOAA Grant NA20OAR4320278.MF de Jong is financially supported by the Innovational Research Incentives Scheme of the Netherlands Organisation for Scientific Research (NWO) under Grant 016.Vidi.189.130.We thank all the field teams (scientists, technical staff and crew) that made the hydrographic data collection possible.We thank Tiago Biló for providing the streamlines for the average circulation in the Irminger Sea in from 2006 to 2020 (Figure 1a).We thank the reviewers for their rapid responses and comments which helped to improve the manuscript.

•
Widespread buoyancy loss across the Irminger interior and Irminger Current (IC) delayed the recovery of the interior from strong cooling in 2015 • Baroclinic instabilities shed from the IC are the dominant source of buoyancy restratifying the sub-surface Irminger interior • It is important to consider changes in the IC when considering drivers of variability in convection, ventilation, and the Atlantic Meridional Overturning CirculationSupporting Information:Supporting Information may be found in the online version of this article.
).Eight occupations between 2010 and 2022 are used: biennial occupations between 2010 and 2018 by the Observatoire de la variabilité interannuelle et décennale en Atlantique Nord (OVIDE) project, and occupations in 2015, 2020, and 2022 by the Overturning in the Subpolar North Atlantic Program (OSNAP).All sections were conducted in either June or July, except the 2014 section which was conducted in May and the 2022 section which was conducted in August.No additional data quality control was carried out.

Figure 1 .
Figure 1.(a) Potential density at 300 m in the Irminger Sea with 1,500, 2,000, and 2,500 m isobaths (thin black lines) and Absolute Dynamic Topography streamlines (red) overlain to illustrate the mean circulation.Locations of data used in this paper are also shown: RG grid-points used in the interior area average (red crosses); RG grid-points used in the Irminger Current (IC) area average (magenta crosses); AR07-E hydrographic section (bold black line); x1 (blue cross) and x2 (yellow cross) between which ΔB x is calculated; and the region over which the area-averaged EKE ic is calculated (magenta box).Mean potential density AR07-E hydrographic sections across the Irminger Sea averaged over the (b) pre-strong convection period (2010, 2012, 2014); (c) during strong convection period (2015 and 2016); and (d) post-strong convection period (2018, 2020, and 2022), with the 27.64, 27.72, and 27.78 kg.m −3 isopycnals (red) and the locations of x1 and x2 for the ΔB x calculation (black dashed) overlain.

Figure 2 .
Figure 2. Time-series of the area-averaged potential density in the (a) interior and (b) Irminger Current (IC) (see Figure 1a for locations), in the upper 1,500 m with the 27.64, 27.72, and 27.78 kg.m −3 isopycnals overlain (red).The mixed layer depth (black dashed) is shown in (a).(c) Depth-integrated buoyancy content over the 100-1,000 m layer for the interior (red) and IC (pink); thin lines show the raw monthly time-series, thick lines show the smoothed time-series after applying a 4-month moving mean.Thick black dashed lines show linear trend from Least Squares fit for each period.

Figure 3 .
Figure 3. (a) Monthly time-series of the horizontal difference in buoyancy content (integrated over the 100-1,000 m layer) across the Irminger Current (IC), from x1 to x2 (see Figure 1a), circles show the ΔB x value at the start of the restratification period.(b) Comparison of the interannual variability in the square of the horizontal difference in buoyancy content over the IC at the start of the restratification period,  Δ 2  (blue) and the buoyancy content gain in the interior over the restratification period in the 100-1,000 m layer,     (red).(c) Comparison of the interannual variability in the average eddy kinetic energy (EKE) in the IC region during the restratification period, EKE R (black) and

Table 1
Rates of Recovery in the Irminger Interior Compared to Those in the Irminger Current for the During-Strong Convection Period (April 2015 to March 2018) and thePost-Strong Convection Period (April 2018 to December 2022)