3.1 Upper ocean freshening in response to sea ice melting
Using the low-res model, we conduct four experiments that differ in either the type of radiative perturbation applied or the strength of the modified radiative fluxes. We analyze in more detail the shortwave (SW) experiment, in which sea ice surface albedo is reduced due to increased shortwave absorption.
In response to the imposed perturbations, annual mean sea ice and volume decrease rapidly in all experiments within the first 5–10 years and then remain fairly steady for the rest of the simulation (Fig. 1a, b). Sea ice volume decreases in all seasons, with a larger reduction in late summer (e.g., 70% in summer and 40% in winter in the SW experiment, Fig. 1e). The amount of total sea ice volume loss is generally consistent with the strength of the imposed radiative perturbation, although SW undergoes a greater sea ice volume loss than LW, despite similar changes in Arctic sea ice area (Fig. 1d), which suggests a slightly higher radiative imbalance in the SW experiment than in the LW.
We first focus on the SW experiment and compare its climatological sea ice concentration to the observations (Fig. 2) and the control (supplementary Fig. S2). September sea ice extent decreases all over the Arctic and the Nordic Seas, with the largest sea ice melting occurring around sea ice margins, especially over the shelf area and along the path of the East Greenland Current (Fig. 2d). In March, the area-integrated total sea ice area reduces by less than 10% (Fig. 1d), but there are important spatial changes in the North Atlantic. In particular, the model shows generally reduced sea ice extent in the basins south of Greenland (Fig. 2c), suggesting a weakening of local sea ice formation and sea ice export from the Arctic ocean (supplementary Fig. S3). The reduction of sea ice extent to the northeast of Iceland, which is on the margins of winter sea ice cover in the control climate, indicates a retreat of winter sea ice edge in this region. Meanwhile, a slightly increased sea ice extent is found between the Barents Sea and Norwegian Sea, and within the Irminger Sea. These locations trace the North Atlantic Current, suggesting anomalous cooling along its path. This is in line with the weakened AMOC transport in these experiments (supplementary Fig. S4).
Unquestionably, the initial melting of sea ice is expected to freshen the Arctic. Assuming that the total freshwater from the initial melting in SW is spread evenly over the upper 300 m of the ocean area north of 60˚N with a mean salinity of 33 psu, this would give a transient salinity reduction of about 0.2 psu. This effect should be relatively short-lived, probably a few decades long, as the released freshwater will spread laterally and mix with waters below. The actual salinity reduction in the model, however, tells a different story (Fig. 1c). We observe a fast salinity decrease in the first 10 years followed by a persistent gradual freshening over a century before the system reaches a new equilibrium. While the first 10 years of freshening in SW (ΔS = 0.2 psu) can be largely attributed to the rapid sea ice melting, the final salinity anomaly of about 0.6 psu is greater than the initial sea ice melting could have induced; nor does this anomaly decay with time. What causes this enhanced Arctic freshening?
3.2 Changes in salinity distribution
Figure 3 shows the area-weighted vertical profiles of time-mean salinity and temperature in the Arctic region (north of 60˚N). At the end of the simulation, the entire ocean column becomes fresher, even though the freshening is most pronounced in the upper ocean. The temperature profile reveals an increase of maximum temperature in the layer of Atlantic inflow water.
We next examine the spatial patterns of salinity anomalies in the upper ocean (0-300 m) and at mid-depth (300–700 m) during the fast sea ice contraction (the first 15 years), the slow adjusting phase (years 16–80), and the final stage (years 151–200), respectively (Fig. 4 and supplementary Fig. S5). Over this timeframe, the Arctic ocean freshwater content must increase in the upper ocean to reduce its salinity. Indeed, in the first 15 years (Fig. 4a), the freshening mainly occurs in the upper Arctic ocean, with the most significant salinity decrease along the Siberian shelf. Meanwhile, the Nordic Seas and the North Atlantic exhibit increased salinity, particularly in the western part of the Norwegian Sea and south of Greenland, corresponding to the most significant wintertime sea ice retreat (Fig. 2c). Along the mid-layer of 300–700 m, the North Atlantic, the Nordic Seas, and a part of the central Arctic become saltier than the control, despite surface sea ice melting. In the years to follow, the upper layer gets fresher, and freshwater anomalies spread out into the subarctic and mid-latitude North Atlantic (Fig. 4c). Salinity at mid-depths is also generally reduced, but areas south of the Greenland and Norwegian Sea remain saltier both in the upper and mid-depth layers (Fig. 4d). By the end of the simulation, both ocean layers become significantly fresher (Fig. 4e, f).
To determine the origin of the mid-depth anomalous salinity, we examine a latitude-depth transect of zonally-averaged salinity (Fig. 5a-c) and temperature (Fig. 5d-f) anomalies within the 10–15˚W sector (marked by the red dashed line in Fig. 4c). This sector is chosen because it coincides with the path of the East Greenland Current, includes the Greenland-Iceland sill, and is close to the positive upper ocean salinity anomaly northeast of Iceland. During the first 15 years (Fig. 5a, d), significantly saltier and warmer surface water within the Norwegian sea penetrates all the way down to the ocean sill. As discussed above, this surface anomaly is due to the retreat of sea ice cover, which lead to reduced freshwater input and greater ocean exposure to direct solar radiation. Here, salinity effects dominate density anomalies. That reduced sea ice cover can cause surface anomalies to pervade through the water column, which emphasizes the sensitivity of convection to surface buoyancy in this region. This anomalously salty and warm water continues to travel northward into the Greenland Sea and the Arctic ocean after reaching the depth of neutral buoyancy at around 200-500m. A similar effect has been found in recent Arctic observations (Timmermans et al. 2018).
Another segment of salty and warm water comes from the southern boundary at 200-500m depth (Fig. 5b-c, e-f), which represents water inflow from the North Atlantic. This shows a strengthened Arctic circulation that coexists with a weakened AMOC (also found in Bitz et al., 2006). The strengthened salt advection is likely caused by the estuarine circulation response (Stigebrandt 1981; Nummelin et al. 2016; Pemberton and Nilsson 2016; Lambert et al. 2019) as well as changes in gyre circulations (supplementary Fig. S6). This discussion reveals that the mid-layer salty water comes from both surface flux-induced convective overturning and horizontal advection from the North Atlantic.
3.3 Changes in freshwater content, surface freshwater fluxes and ocean circulation
To understand the persistent upper-ocean Arctic freshening, next we calculate the simulated changes in the freshwater content (FWC) of the Arctic region. The liquid FWC with respect to a reference state is defined via ocean salinity as:
where S(x, y, z) is salinity, Sref is the reference salinity of 34.8 psu, and D is the depth of integration. We consider FWC integrated over the total ocean depth, as well as FWC for the upper 300 m.
Figure 6a shows the time series of the anomalous FWC (in km3) integrated over the Arctic region (north of 60˚N). The total-depth anomalous FWC gradually increases over the first ~ 130 years then stabilizes at 8x104 km3. In the first 30–40 years, the FWC increase primarily occurs within the upper 300 m. In the new quasi-equilibrated state, the anomalous FWC of the upper ocean accounts for nearly 70% of the total FWC change. As part of this FWC increase in the upper ocean, the central Arctic FWC (which includes the Canadian and Eurasian Basins) increases by 5x104 km3 (equivalent to a salinity decrease of 0.7 psu), accounting for about 90% of the total upper ocean FWC increase north of 60˚N.
The change of the Arctic ocean FWC can come from surface freshwater fluxes (SFC), convergence of horizontal and vertical advection (ADV), and the diffusion term (DF): =SFC+ADV+DF. As the diffusion term is small, we mainly consider changes of surface freshwater fluxes and convergence of freshwater by advection (Fig. 6b). Immediately after the perturbation initiation, the rapid sea ice melting leads to an increase of surface freshwater flux. This anomalous surface flux slowly abates as the sea ice adjusts towards stabilization. Around year 40, the surface freshwater flux reaches to a balanced state, with an anomalous ~ 0.03 Sv freshwater input to the ocean compared to the control. It was found that surface buoyancy anomalies are dominated by net freshwater/salt fluxes induced by sea ice melting and brine rejection (e.g., Fig. 6 in Liu et al., 2018). The anomalous surface freshwater input of 0.03 Sv is therefore the result of the strengthened sea ice seasonal cycle with enhanced summer melting (Fig. 1).
In the first 40 years, the trend of FWC( )is almost entirely attributed to surface freshwater fluxes associated with sea ice melting, while the effect of total-depth freshwater transport is small (Fig. 6b). Subsequently, after year 40, the freshwater export by ocean circulation increases and eventually balances the surface freshwater fluxes, leading to a quasi-equilibrium.
To better understand the key processes involved in the FWC increase, we analyze the spatial patterns of surface freshwater flux anomalies (Fig. 7) and ocean circulation (Fig. 8). Annual mean anomalous freshwater fluxes are consistent with the patterns of anomalous upper-ocean salinity (Fig. 4a and 7b). The central Arctic receives a large amount of freshwater from sea ice melting, while the Nordic Seas and regions along the East Greenland Current experience generally reduced surface freshwater input owing to reduced sea ice cover and export. The annual mean pattern is largely dominated by anomalies in summer (Fig. 7c) when sea ice melting is the strongest. In winter (Fig. 7d), the central Arctic shows a reduced freshwater flux. The amplitude of the surface salinity seasonal cycle is consistent with that of the freshwater flux, which scales with the imposed perturbation intensity (Fig. 1f and Fig. 7a).
The large freshwater flux input within the central Arctic co-occurs with the strengthening of the Beaufort Gyre (Fig. 8a, c, e), which increases the storage of the upper ocean freshwater made available by the strong summer sea ice melting (in this case, reducing freshwater export). Meanwhile, the Atlantic subpolar gyre weakens, even though the East Greenland Current and West Greenland Current strengthen (Fig. 8a, c, e). The upper 300 m of the ocean maintain a net freshwater export across 60˚N of about 0.01 Sv. Note that the freshwater transport in the North Atlantic sector is small throughout the simulation (Fig. 6c), suggesting that the subpolar Atlantic has a small impact on the Arctic upper-ocean freshwater content, and that the feedback from a weakened AMOC on the Arctic upper-ocean freshening is small (more on this in Sect. 4).
In summary, Arctic sea ice decline generates an imbalance between the surface freshwater flux (due to enhanced seasonal melting of sea ice) and the export of freshwater (by ocean circulation). In the process of adjusting to the new state, freshwater accumulates in the Arctic, reducing the upper-ocean salinity. The sea ice thermodynamic adjustment with seasonal melting and re-freezing, and its interactions with the ocean circulation, is key to the upper-ocean freshening.
We see a general agreement between different numerical experiment; however, we find that the LW experiment has a smaller total sea ice volume reduction but a fresher Arctic at the end of the simulation than in the SW experiment (Fig. 1a-c). It is found that the LW experiment generates stronger net surface freshwater fluxes into the ocean than the SW during the adjustment stage, resulting in a higher freshwater accumulation before a new quasi-equilibrium is reached.
3.4 Arctic ocean freshening and salinity changes in the North Atlantic
In the previous discussion in Sect. 3.2, we found that the enhanced convection in a part of the Norwegian Sea and the region south of Greenland may allow surface salinity anomalies to spread down the water column. We now examine the relevant processes to understand how they impact the upper-ocean freshening in the Arctic and, more generally, the salinity distribution in the Arctic and North Atlantic.
Figure 9a shows changes in the upper-ocean stratification averaged over the last 50 years in the SW experiment, which closely follows changes in surface freshwater fluxes (Fig. 7b). The southwestern corner of the Norwegian Sea and the Labrador Sea south of Greenland show reduced stratification and therefore may be prone to convection. Elsewhere in the Arctic and subarctic North Atlantic stratification increases. In particular, stronger stratification in the eastern North Atlantic is related to the reduction in northward salt transport from the subtropics by the weakened AMOC. The anomalous March mixed layer depth shows generally similar features as the changes in upper-ocean stratification (supplementary Fig. S7). The area to the south of Iceland, which is a major part of the original deep convection zone in the control simulation, experiences a significant shoaling of the mixed layer depth, which reflects the weakening of deep convection, and the AMOC slowdown, caused by fresh and warm anomalies at the ocean surface (Liu et al. 2018). The mixed layer deepens in the Beaufort Gyre region, which is paralleled by increased freshwater storage (Fig. 8).
To understand the salinity changes and related processes, we analyze salinity changes over time in two critical regions with the largest buoyancy loss – the region to the south of Greenland and the southwestern corner of the Norwegian Sea (marked in Fig. 9a). For the south of Greenland (box A, Fig. 9b), the anomalously saline water at the surface is transported down via strong convective events to approximately 1000 m depth, where it combines with the upstream dense water and continues to travel south. Year 10 marks the arrival of meltwater from the Arctic, as the upper 300 m begins to freshen up. The convective activity continues throughout the next ~ 200 years, as can be suggested from the correlation between the mixed layer deepening and high salinity signals in the deeper ocean (Fig. 9b and supplementary Fig. S8). Similar features are also present in the Norwegian Sea (box B, Fig. 9c), where anomalous high salinity is present from 50 m to ~ 1500 m during the first century. Sources of this salinity anomaly include convection, sinking of saltier shelf water due to the stronger winter brine rejection, and advection of the Atlantic water (Fig. 5). This saline water then mainly sinks to the deep ocean while the upper ocean gets fresher.
Note that the Labrador Sea was never a deep-water formation site in the control simulation of this model. Likewise, despite the reduced stratification and deepened mixed layer that allow for episodic convective events during the first decades of the sea ice perturbation experiments, these experiments do not develop sustained deep convection in the Labrador Sea. This contrasts the sea ice perturbation experiments with the higher-resolution model which undergoes activation of deep convection in this region (Sect. 3.5 and Li et al. 2021).
We further examine the temporal evolution of the FWC and freshwater transport in the layer of 700–1100 m north of 60˚N (Fig. 9d, e). We find that the FWC has a negative anomaly in the first 30 years, consistent with the increased salt content (see Figs. 5b and 9b-c). This FWC decrease is concurrent with freshwater import by horizontal advection, suggesting that it is vertical processes that cause the increase of salt content at this depth.
Thus, we find that the strengthened sea ice seasonality and the associated changes in sea ice transport freshen the Arctic upper ocean while making ocean deep layers in the Arctic ocean and subarctic North Atlantic more saline (during the first several decades of the perturbation experiments). This resembles a distillation process that would remove salt from the upper ocean and increase salinity at depth.
The salt accumulated in deeper layers of the subarctic North Atlantic is then advected southward by the Deep Western Boundary Current (DWBC) as seen in a Hovmoller diagram of zonally-averaged anomalous salinity estimated at the typical depths of DWBC (1700–2000 m) in the Atlantic Ocean (Fig. 10a). The most saline water mass, first seen at 55–60˚N at year 10, propagates southward into mid-latitudes over the following 40 years. Analyzing area-averaged salinity anomalies at three latitudinal bands (averaged over 40–55˚W, 1700–2000 m) along the meridional transport pathway. With time, the subarctic North Atlantic Ocean becomes fresher, while the ocean at lower latitudes becomes more saline. The high to low latitudes salinity contrast is further amplified by the AMOC slowdown and hence the reduced upper-ocean northward transport of saline subtropical waters.
The evolution of depth-integrated ocean salt content further illustrates the process by which salt is taken out of the subarctic ocean (Fig. 10b-e). Positive salt content anomalies evident in the Nordic Seas and in the Labrador Sea during the first decades of the perturbation experiment are gradually transported out of the high-latitudes over approximately 80 years. The pathway of the export is mainly along DWBC (Fig. 10e and supplementary animation). Eventually, a strong negative salinity anomaly develops in the eastern North Atlantic at the latitude of 45˚N, resulting largely from the reduced northward transport by the AMOC. These changes together lead to a pronounced freshening in the mid-to-high latitude North Atlantic but increased salinity in the tropical and subtropical ocean (supplementary Fig. S9).
3.5 Results from the higher-resolution model
A persistent freshening of the Arctic is also evident in the SW sea ice decline experiment (Fig. 11) using a higher-resolution CESM configuration (f19_gx1v6) with a nominal 1˚ ocean horizontal grid spacing; for further model details see Sect. 2. In this model the upper-ocean salinity falls to its minimum around years 20–30 then increases slightly, reaching a new balance with an average salinity reduction of 0.2 psu in the second half of the experiments relative to the Control. The total adjustment timescale is also about 100 years.
The vertical profiles of the Arctic average salinity and temperature in the high-res perturbation experiment share similar features as the low-res model (Fig. 12). However, in the high-res model, the upper ocean freshening is primarily confined to the upper-ocean of the central Arctic, while the Barents Sea, the Nordic Seas and the rest of subarctic Atlantic become more saline by the end of the perturbation simulations (Fig. 13). By comparing the temporal evolution of the Arctic FWC (Fig. 14), we find that the differences in the responses between the two models are largely attributed to the interplay between freshwater advection and changes in ocean circulation. In particular, the subpolar gyre circulation in the high-res model strengthens and begins to export the anomalous meltwater (Fig. 14b and Fig. 8). By year 30, the freshwater export entirely offsets (and will overwhelm later) the positive local freshwater flux. Around year 30, the AMOC also starts to recover from the initial weakening (supplementary Fig. S4), joining force with the strengthened subpolar (Fig. 8) to import more saline Atlantic water into the Arctic. As a result, the total depth-integrated freshwater content decreases. The upper-ocean salinity increases over the subpolar Atlantic and the Nordic Seas (Fig. 13), due to both decreased surface flux and the increased transport of saline Atlantic water (supplementary Fig. S10). Note that despite the enhanced salinity import, the central Arctic upper ocean still shows persistent freshening.
It is critical that he two models have very different responses in ocean circulation. While the advection from the Atlantic has very little effect on the Arctic upper ocean FWC in the low-res model (Fig. 6c), it is a dominant factor in the high-res model (Fig. 14b-c). In the control climate, the North Atlantic subpolar gyre, the North Atlantic Current and the Norwegian Current are stronger in the high-res model than in the low-res model (Fig. 8b). Salt advection by the North Atlantic Current can therefore have a stronger effect on the Nordic Seas and even the Barents Sea.
In fact, changes in the North Atlantic are an essential component of changes in the AMOC. Accordingly, the AMOC responses are also different between the two models: in the low-res perturbation experiments, the AMOC weakens continuously during the first ~ 100 years, approaching a new quasi-equilibrium; in the high-res model, the AMOC weakens by only 10% during the first 20–30 years, which is followed by a full recovery and a slight overshoot (supplementary Fig. S4). Li et al. (2021) find that the diverging AMOC behaviors in the two models are related to differences in the AMOC stability properties, which are controlled by the model mean states, including the basin-wide mean surface freshwater fluxes.