On average, the Arctic Ocean has a strong halocline because it receives freshwater from river runoff, net precipitation minus evaporation and Pacific Water which has relatively low salinity (Serreze et al., 2006; Haine et al., 2015; Carmack et al., 2016). The surface water in the Arctic Ocean is much fresher than that in the North Atlantic, where it is exported to, so it is traditionally called Arctic freshwater, and the amount of freshwater is often quantified with the so-called freshwater content (FWC). The major upper ocean circulation in the Arctic basin, basically the surface geostrophic flow, is mainly determined by the halosteric contribution to sea surface height because the temperature-induced ocean density change is relatively small (Carmack, 2007). Understanding the changes in the large-scale circulation in the upper Arctic Ocean is thus by large equivalent to understanding the changes in the spatial distribution and magnitude of the FWC in the Arctic Ocean (see Figure S1 in the Supplementary Material ). The FWC at each location can be computed as

where S is ocean salinity, S_ref is a reference salinity and D is the isohaline depth of S = S_ref . FWC (in meter) represents the amount of freshwater contained in the ocean column. It is equivalent to the amount of pure (zero-salinity) water required to be taken out from an ocean column so that the mean salinity in the ocean column is changed to the reference salinity. In this paper S_ref = 34.8 psu if not otherwise stated. Integrating FWC laterally in a region we can get the volumetric FWC (with a unit km³).

When understanding the wind-driven circulation in the Arctic Ocean, the impact of sea ice should be considered. Sea ice mediates momentum transfer from the atmosphere to the ocean due to its seasonally strong internal stress (Hibler, 1979; Martin et al., 2014). As sea ice declines, the response of the upper ocean circulation to winds can be strengthened due to enhancement in ocean surface stress (Davis et al., 2014; Wang et al., 2021c). Through dedicated numerical simulations it was found that the sea ice decline in the 2000s dramatically modified the ocean surface stress so that more surface freshwater was shifted toward the converging Beaufort Gyre region, thus contemporarily reducing the amount of freshwater in the Eurasian and Makarov basins (Wang et al., 2019b). The rapid freshwater accumulation in the Beaufort Gyre region in the 2000s can be attributed to both sea ice meltwater and other freshwater components that were made more available to the Beaufort Gyre by sea ice decline (Wang et al., 2018a), including Mackenzie River runoff and Pacific Water (Proshutinsky et al., 2019). That is, Arctic sea ice decline has amplified the effect of wind forcing on freshwater distribution and upper ocean circulation, which is clearly demonstrated by using numerical simulations in Wang (2021). Wang (2021) shows that, in comparison to the sea ice condition in the 1980s, the sea ice condition in the 2010s increases the response of the Arctic Ocean to Arctic Oscillation and Beaufort High wind perturbations by up to 50% for integrated FWC and up to about 100% for upper ocean velocity regionally. The nature that the Arctic FWC and upper ocean circulation are dynamically driven by winds implies their predictability (Marshall et al., 2017; Johnson et al., 2018; Cornish et al., 2020). However, because Arctic sea ice is in a rapid decline, a quantitative prediction of Arctic FWC will very possibly fail without considering the changes in sea ice (Wang et al., 2018a).

Another important aspect of the dynamic impact of sea ice on the Arctic Ocean is its stabilization effect on the Beaufort Gyre. Previously transport by mesoscale eddies was thought to be the main process that stabilizes the Beaufort Gyre FWC and circulation (Manucharyan and Spall, 2016; Yang et al., 2016; Armitage et al., 2020). Recent studies suggested that the status of the Beaufort Gyre FWC and circulation should be viewed as a result of three-way balance: wind stress, eddy flux and the so-called ice-ocean governor (Meneghello et al., 2018a; Doddridge et al., 2019; Wang et al., 2019a; Meneghello et al., 2020). The ocean surface stress that drives the Ekman convergence of surface freshwater in the Beaufort Gyre region depends on the velocity difference between sea ice and surface ocean. A decrease (an increase) in FWC in the Beaufort Gyre results in a decrease (an increase) in the anticyclonic ocean surface geostrophic current, which induces an anticyclonic (a cyclonic) anomaly in the ocean surface stress; The latter functions to stabilize the Beaufort Gyre, as its effect is to oppose the further decrease (increase) of the Beaufort Gyre FWC. This negative feedback relies on large sea ice internal stress in wintertime (Kwok and Morison, 2017) and is dubbed ice-ocean governor (Meneghello et al., 2018a). Its stabilization effect on the Beaufort Gyre was clearly demonstrated using observations (Dewey et al., 2018; Meneghello et al., 2018b; Zhong et al., 2018) and realistic Arctic Ocean modeling (Wang et al., 2019a).

Below the Arctic halocline the modified Atlantic Water circulates cyclonically in the intermediate depth range along topography slopes (Woodgate et al., 2001; Karcher and Oberhuber, 2002; Schauer et al., 2002; Gerdes et al., 2003). The focus of this paper is on the upper Arctic Ocean, but the halocline changes in the eastern Eurasian Basin are closely linked to the changes in the underlying Atlantic Water layer in the process of Arctic Atlantification. Therefore, we briefly allude to Atlantic Water inflow related to our topic here. See Timmermans and Marshall (2020) for a recent comprehensive review on Arctic Ocean circulation dynamics including the Atlantic Water layer.

Warm, saline Atlantic Water enters the Arctic Ocean through both the Barents Sea Opening and Fram Strait ( Figure 1 ). The ocean heat transport through the Barents Sea Opening is higher, but most of the ocean heat in this branch is lost to the atmosphere within the Barents Sea in the climate condition of the 20th century (Smedsrud et al., 2013), while this situation will change with time in a future warming climate (Årthun and Eldevik, 2016; Shu et al., 2021). In the 2000s and 2010s, a warming trend together with an increase in volume transport in the West Spitzbergen Current (WSC) was found in observations (Beszczynska-Moeller et al., 2012; von Appen et al., 2016; Wang et al., 2020, also see Figure S2 ) and model simulations (Wang et al., 2020). Different mechanisms could drive the Atlantic Water inflow, including wind-driven mixing interior to the Arctic that draws water in and winds exterior to the Arctic that drive water into the Arctic following bathymetry contours (Timmermans and Marshall, 2020). Ascertaining what drives the Atlantic Water inflow remains to be a subject of active research, but model simulations have proved that Arctic sea ice decline has caused an enhanced ocean heat transport through Fram Strait and considerably contributed to the warming of the Atlantic Water layer in the eastern Eurasian Basin (Wang et al., 2020).

The Atlantic Water heat can relatively easily reach sea ice and produce ice-free areas around Svalbard in winter (Ivanov et al., 2016; Athanase et al., 2020). However, along the cyclonic circulation pathway of Atlantic Water towards the eastern Eurasian Basin and Amerasian Basin, the halocline thickens and strengthens, thus insulating the mixed layer and sea ice from the warm Atlantic Water layer (Rudels et al., 1996). Observations reveal that the halocline in the eastern Eurasian Basin has experienced remarkable changes in the early 21st century, becoming weaker along with an uplift of the underlying Atlantic Water layer (Polyakov et al., 2018; Polyakov et al., 2020b). It is unclear if this phenomenon of Atlantification in the eastern Eurasian Basin represents a typical trend under climate change, or a decadal variability that was possibly enhanced by climate change. At least the analysis of paleoceanographic records has indicated that the aforementioned ocean warming in the Fram Strait in the early 21st century is unprecedented for the past 800 years (Tesi et al., 2021).

Changes in Arctic sea surface height and the associated barotropic geostrophic circulation can influence the Atlantic Water layer circulation located beneath the halocline on different time scales as illustrated by model simulations (Lique et al., 2015; Hinrichs et al., 2021). Observations of chemical tracers indicate that both the strong cyclonic surface circulation in the mid-1990s associated with the strongly positive Arctic Oscillation and the strengthened anticyclonic Beaufort Gyre circulation in the past two decades left an imprint on the Atlantic Water layer circulation (e.g., Karcher et al., 2012; Smith et al., 2021). Therefore, observing and understanding the upper ocean circulation are also crucial for interpreting the ocean circulation changes at depth.

Previous studies based on observations and model simulations have significantly improved our understanding of the ocean dynamics in the Arctic Ocean as described above. However, considering the major changes in the large-scale circulation of the upper Arctic Ocean during 2000-2019, the previous studies mostly focused on part of the period or part of the Arctic basin. In this paper we will provide a synthesized view of these changes based on numerical simulations with support of satellite-derived dynamic ocean topography and observed FWC estimates. We will consider interannual, pentadal and decadal time scales with the cyclonic and anticyclonic ocean circulations on an equal footing. The focal aspect will be quantifying how strongly Arctic sea ice decline influences the wind-driven large-scale circulation in the Arctic and the Arctic Atlantification in the eastern Eurasian Basin.

We note that this paper does not cover all aspects of the Arctic Ocean dynamics. Recently some other Arctic Ocean topics have also been reviewed, including ocean circulation dynamics in general (Timmermans and Marshall, 2020), freshwater in the 2010s (Solomon et al., 2021), ocean heat transports in relation to Arctic sea ice (Docquier and Koenigk, 2021; Smedsrud et al., 2022), and Arctic Atlantification and Borealization (Polyakov et al., 2020a; Ingvaldsen et al., 2021).

In Section 2, we present a few idealized wind-perturbation numerical simulations to demonstrate how the ocean responds to Arctic Oscillation and Beaufort High forcing. The results underpin the explanation of the wind-driven circulation changes presented in Section 3. In Section 3 we disentangle the individual role of winds and sea ice decline in driving the upper Arctic Ocean changes. Discussions and summaries are provided in Section 4 and 5, respectively.

To underpin the discussion in the next section, results from three wind-perturbation numerical simulations will be presented here to demonstrate how the Arctic Ocean circulation responds to different wind forcing. The employed model is the global Finite Element Sea-ice Ocean Model (FESOM, Wang et al., 2014; Danilov et al., 2015; Wang et al., 2018b) with 4.5 km horizontal resolution in the Arctic Ocean. First, the model was spun up for 20 years starting from the PHC3 hydrography climatology (Steele et al., 2001). The wind forcing used to drive the model is the normal year (1-year repeating) version of the Coordinated Ocean-ice Reference Experiments Phase II (CORE2) atmospheric forcing (Large and Yeager, 2009). Then four simulations were conducted for 5 years each, using either the same wind forcing as during the spin-up (as a control run) or modified winds with some idealized perturbations added to this forcing (wind-perturbation experiments). All these experiments started from the same ocean and sea ice conditions obtained from the spinup simulation.

The winds in the three wind-perturbation experiments were perturbed with wind anomalies representing the cases of the positive Beaufort High anomaly, positive phase of the Arctic Oscillation and their combination, respectively ( Table 1 ). Figures 4A–C show the corresponding wind anomalies and associated sea level pressure anomalies. The Beaufort High perturbation ( Figure 4A ) is the same as that used by Marshall et al. (2017) and Wang et al. (2019a), and its magnitude is similar to the anomaly of the Beaufort High in 2007, when it was abnormally high. The magnitude of the Arctic Oscillation perturbation ( Figure 4B ) is similar to that of the first EOF of the Northern Hemisphere sea level pressure, the Arctic Oscillation pattern. Note that for the purpose to explore the effect of the wind perturbations, other atmospheric forcing fields and the ocean-ice model configurations were kept the same in all the simulations.

The difference of the model results between the wind-perturbation experiments and the control run reveals the response of the ocean to the wind perturbations. The positive Beaufort High forcing causes the Beaufort Gyre to accumulate freshwater, with some minor reduction in the FWC around the Beaufort Gyre ( Figure 4D ). Therefore, the total FWC integrated over the Arctic Ocean is increased. The doming of the sea surface height resembles the spatial pattern of the FWC anomaly, resulting in a spin-up of the anticyclonic Beaufort Gyre ( Figure 4G ).

With the positive-phase Arctic Oscillation forcing, the Makarov Basin and part of the Eurasian Basin are drained, but the surrounding regions, including the Beaufort Sea, have some freshwater gain ( Figure 4E ). The magnitudes of the FWC increase in these surrounding regions are relatively small, so the total FWC integrated over the Arctic Ocean is reduced under positive Arctic Oscillation (not shown). In accordance with the drop of the FWC in the Makarov and Eurasian basins, the sea surface height depression leads to a cyclonic circulation anomaly centered over the Makarov Basin ( Figure 4H ). The effect of the positive Arctic Oscillation demonstrated here is consistent with established understanding based on observations and models (Maslowski et al., 2000; Zhang et al., 2003; Steele et al., 2004; Condron et al., 2009; Morison et al., 2021). We note that the increase of the sea surface height over the Eurasian continental shelf is unproportionally large when considering the FWC anomaly (cf. Figures 4E, H ). This indicates that the change of the sea surface height over the shallow shelf is more determined by mass change (reflected in bottom pressure change), rather than the halosteric effect which plays a major role in the deep basin area. In this paper our focus is on the ocean dynamics in the deep basin area.

In the third experiment both the positive Beaufort High and positive-phase Arctic Oscillation perturbations are enforced, which is motivated by the atmospheric conditions in the second half of the 2000s and 2010s ( Figure 3A ). In these periods, the Arctic Oscillation was positive in some of the years, though on average not as strong as in the period from 1989 to the mid 1990s. In this experiment, the changes in FWC and sea surface height are roughly the sum of the two cases when only one of the forcing is applied ( Figures 4F, I ). In particular, the inflation of the Beaufort Gyre is slightly stronger, but spatially confined on its northwest side, compared with the case when positive Arctic Oscillation forcing is absent ( Figures 4D, G ). On the other hand, it is also clear that the ocean convergence driven by the positive Beaufort High forcing has a stronger impact on the Beaufort Gyre. That is, the positive Arctic Oscillation alone could facilitate the accumulation of freshwater in the Beaufort Gyre by discharging freshwater from the Makarov Basin, but an efficient spin-up of the Beaufort Gyre needs ocean convergence in the Canada Basin.

The impact of the wind forcing on the upper ocean circulation is illustrated with a dye tracer injected in the Bering Strait representing Pacific Water and the velocity averaged over the upper 150 m ( Figures 4J–L ). In comparison with the control simulation ( Figure S3 ), the positive Beaufort High forcing mainly strengthens the anticyclonic Beaufort Gyre circulation without significant effect on the location of the Transpolar Drift Stream ( Figure 4J ). The positive Arctic Oscillation forcing can shift the Transpolar Drift Stream and the front between Pacific- and Atlantic-derived waters from the Lomonosov Ridge to the Alpha-Mendeleev Ridge, thus expanding the cyclonic circulation to the Makarov Basin and reducing the spatial extent of the anticyclonic circulation ( Figure 4K ). Although the exact location of the Transpolar Drift Stream depends on the magnitude and duration of the Arctic Oscillation forcing, the ocean response revealed here clearly represents the ocean changes under positive Arctic Oscillation as observed in the 1990s (Carmack et al., 1995; Morison et al., 1998; Steele and Boyd, 1998; McLaughlin et al., 2002). The responses of the Pacific Water circulation and spatial distribution to the Beaufort High and Arctic Oscillation forcing presented here are also consistent with other model simulations (Aksenov et al., 2016; Hu et al., 2019). In the presence of both the positive Beaufort High and Arctic Oscillation, the location of the Transpolar Drift Stream is the result of the competition between the two atmospheric modes ( Figure 4L ).

We note that the applied wind perturbations also lead to changes in sea ice thickness (that is, solid freshwater), sea ice concentration and drift ( Figure S4 ). Both the ocean circulation and sea ice state respond to the wind perturbations as a coupled system with interaction between them, which is investigated in Wang et al. (2021a) and not the focus of the current study. The results shown in this section will serve as a reference when understanding the role of winds in impacting the upper Arctic Ocean changes during 2000–2019 in the next section.

In this section a historical (hindcast) model simulation and two sensitivity experiments ( Table 2 ), synthesized with some related observations, will be used to explain the upper Arctic Ocean changes in the 2000s and 2010s. FESOM on the same model grid as used in the last section is employed here. The historical simulation was driven by the JRA55-do v.1.4 atmospheric data sets (Tsujino et al., 2018) from 1958 to 2019. To disentangle the role of Arctic sea ice decline, we conducted a sensitivity experiment in which the historical thermal forcing fields (near-surface air temperature, shortwave and longwave radiation fluxes) over the Arctic Ocean were replaced with climatological forcing. The one-year-long climatological thermal forcing was obtained by averaging the JRA55-do data from 1970 to 1999 for each time step (3-hourly). This experiment used the same historical winds as in the historical simulation, and is called wind_vari run hereafter. In the other sensitivity experiment, the thermal forcing was the same as in the historical simulation, but winds were kept the same every year. We followed the recommendation of Stewart et al. (2020) to use the one-year-long winds compiled from 1st May 1990 to 30th April 1991 of JRA55-do, which represent a neutral state of the atmospheric circulation from a global perspective. This experiment is called thermal_vari simulation. Both the sensitivity simulations were conducted from 2001 to 2019 starting from the end state of the historical simulation in 2000. The Arctic domain over which the atmospheric forcing was replaced in the sensitivity simulations is indicated by the magenta lines in Figure 1 . These simulations are similar to those used in Wang et al. (2018a; 2019b; 2020), but with the simulations extended to 2019 to cover the full length of the past two decades.

The historical simulation has a reasonable performance in reproducing the observed mean sea ice drift, concentration and thickness ( Figure 2C and Figures S5, 6 ), mean upper ocean circulation ( Figure 2D ), the pentadal changes in sea surface height ( Figures 6D–I ), the variation of the liquid FWC ( Figure 7 ), the trend and variability of Arctic sea ice volume ( Figure 8A ) and the trend and variability of Atlantic Water temperature in Fram Strait ( Figure S2 ). Despite some noticeable model biases, the simulation can serve as a basis for discussing the Arctic Ocean changes in the past two decades. We will first use the historical simulation to illustrate the upper ocean circulation changes in relation to winds, and then use the sensitivity simulations to demonstrate the extremely important role of Arctic sea ice decline in setting the changes in the upper Arctic Ocean circulation and in enhancing the Arctic Atlantification of the eastern Eurasian Basin.

We will compare the sensitivity experiments with the historical simulation to demonstrate the individual impact of sea ice decline and winds on the upper Arctic Ocean changes in the early 21st century.

The historical simulation reasonably reproduces the sea ice decline as compared with the PIOMAS sea ice volume ( Figure 8A , Schweiger et al., 2011). In the wind_vari simulation, the trend of the Arctic sea ice is eliminated, albeit wind-driven variability remains ( Figure 8A ). The historical simulation has thinner, less compact and more mobile sea ice than the wind_vari simulation ( Figure S7 ). As a result, the ocean surface stress is larger in the historical simulation nearly everywhere in the Arctic basin ( Figure 8B ). The ocean-ice stress depends on both the sea ice drift and ocean surface velocity, while the ocean surface velocity changes with sea surface height which can be influenced by the ocean surface stress. With such a feedback, it is not surprising that the direction of the ocean surface stress difference between the two simulations does not follow their sea ice drift difference ( Figure 8B vs. Figure S7 ). Averaged over the period 2001–2019, the associated Ekman transport difference between the two simulations (historical minus wind_vari) is directed from the Eurasian continental shelf towards the Eurasian Basin and from the Lomonosov Ridge towards the Makarov Basin ( Figure 8C ). We will see below that this dynamic impact of the sea ice decline is crucial for determining the FWC spatial distribution, salinity and depth of the halocline, and the upper ocean circulation.

The Arctic surface freshwater flux had a significant upward trend in the past two decades, with sea ice melting accounting for about 40% of the trend ( Figure 9A ). Integrated over the last 20 years, sea ice meltwater accounts for about 70% of the total anomalous surface freshwater ( Figure 9B ). In the wind_vari simulation, the anomalous sea ice meltwater is eliminated, while other surface freshwater budget terms remain nearly the same as in the historical simulation (Evaporation could be influenced by sea ice coverage, but we found that its change is much smaller than the overall trend of other freshwater fluxes). Therefore, the difference in the surface freshwater flux between the historical and wind_vari simulations is mainly due to the difference in the sea ice thermodynamic growth rate, which is shown in Figure 9C . Positive values indicate that the historical simulation has larger surface freshwater flux into the ocean over most of the area inside the Arctic Ocean.

Another important change in the upper Arctic Ocean in the early 21st century is the weakening of the halocline and the shoaling of the lower halocline boundary in the eastern Eurasian Basin found from the analysis of hydrography observations (Polyakov et al., 2017; Polyakov et al., 2018; Polyakov et al., 2020a; Polyakov et al., 2020b). The concern about this phenomenon, Arctic Atlantification, is due to its potential role in accelerating sea ice decline through reducing the insulation between the warm Atlantic Water layer and sea ice. The main driver of this phenomenon needs a better understanding.

In Figures 12A–I , the changes of FWC, mean salinity in the upper 150m and the depth of the 0°C isotherm (the lower halocline boundary) between 2000 and 2019 are shown for both the historical and wind_vari simulations and their differences. For the whole two decades from 2000 to 2019, the FWC in the eastern Eurasian Basin is slightly reduced ( Figure 12A ). In the period from 2000 to 2004, which is not investigated above, there is an increase in the FWC in the eastern Eurasian Basin and the impact of sea ice decline is relatively small ( Figure S8 ). Without sea ice decline, there would have been a strong increase in the FWC in the Eurasian Basin over the past two decades ( Figure 12B ). The sea ice decline reduces the FWC in the Eurasian Basin ( Figure 12C ) and this effect slightly overweighs the otherwise FWC increase in the eastern Eurasian Basin ( Figure 12B ), leading to the FWC reduction there ( Figure 12A ).

The impact of the sea ice decline on the upper ocean salinity in the eastern Eurasian Basin is consistent with its impact on the FWC. That is, the sea ice decline is responsible for the salinification in the halocline, the halocline weakening in the eastern Eurasian Basin ( Figure 12F ). The depth of the boundary between the halocline and the Atlantic Water layer in the eastern Eurasian Basin has little change in the wind_vari simulation ( Figure 12H ), but it shallows for about 30-40 m due to sea ice decline ( Figures 12G, I ). The shoaling of the boundary is similar to the observed changes in this region (Polyakov et al., 2020b). The sea ice decline enhances the Ekman transport directing from the Eurasian Basin toward the Amerasian Basin on average ( Figure 8C ), which flushes surface freshwater out of the Eurasian Basin, thus increasing the halocline salinity and causing the uplift of the lower halocline boundary in the Eurasian Basin ( Figures 12F, I ).

Budget analysis has shown that the eastward shift of the surface freshwater from the eastern Eurasian Basin in the last decades was accompanied by an increase in the concentration of saline Atlantic Water, being the main reason for the increase in halocline salinity in the Eurasian Basin (Wang et al., 2019b). It was also found that only about 20% of the increased Atlantic Water concentration above the 34.8 isohaline in the Eurasian Basin can be attributed to the Fram Strait branch (Wang et al., 2019b). Therefore, although sea ice decline can increase vertical mixing (e.g., Rainville and Woodgate, 2009; Polyakov et al., 2020b; Rippeth and Fine, 2022), thus increasing halocline salinity through mixing saline Atlantic Water upwards, this process is not the major contribution to the salinification of the halocline in the Eurasian Basin. In addition, the positive salinity anomaly in the Barents Sea induced by sea ice decline could contribute to the positive salinity anomaly in the eastern Eurasian Basin too, but the larger salinity anomaly associated with sea ice decline in the eastern Eurasian Basin than in the upstream Barents Sea implies that this contribution is small ( Figure 12F ).

There were upward trends in the temperature and poleward volume transport of the Atlantic Water in the Fram Strait in the past two decades ( Figure S2 ). It was found that the Arctic sea ice decline has contributed to these trends possibly through modifying the Nordic Seas circulation (Wang et al., 2020), thus considerably intensifying the Atlantic Water layer warming in the eastern Eurasian Basin (cf. Figures 12G, K ). The warming of the Atlantic Water layer with an uplifted upper boundary also has impacts on the downstream region, especially at the Chukchi Plateau ( Figure 12I ).

Our above analysis demonstrates that the Arctic Atlantification in the eastern Eurasian Basin in the past two decades is largely a consequence of Arctic sea ice decline. Using a set of numerical simulations with different atmospheric warming levels prescribed, it was shown that a further sea ice decline in a future warming climate can potentially cause even stronger salinification of the upper ocean and uplift of the Atlantic Water layer in the Eurasian Basin (Wang et al., 2021c). However, the future stratification in the Arctic Ocean also depends on how different freshwater sources will change.

The analysis in Figure 12 also provides insights for the Canada Basin in contrast to the Eurasian Basin. Without sea ice decline, the salinity in the upper 150 m increases in the Canada Basin and decreases in the Eurasian Basin ( Figure 12E ). This can be explained by the weak but positive sea level pressure anomaly in the central Arctic basin averaged over the past 20 years ( Figure S9 ). The sea level pressure anomaly roughly resembles a negative Arctic Oscillation pattern, which tends to shift the surface freshwater circulation pathway from the Amerasian Basin towards the Eurasian Basin (Wang et al., 2021b). The sea ice decline completely changes this tendency. It considerably reduces the upper ocean salinity and deepens the halocline in the Canada Basin, and has an opposite effect in the Eurasian Basin ( Figures 12F, I ).

At the end we need to comment on the thermal_vari experiment with respect to the Atlantification in the Eurasian Basin. In this experiment, saline anomaly was absent in most parts of the eastern Eurasian Basin as implied by the plot of the FWC anomaly ( Figure 11C ). The repeat-year wind forcing used in this experiment does not represent the climatology as indicated by the sea level pressure anomaly ( Figure 11D ). The thermal_vari result shows that the sea ice decline together with an arbitrary wind forcing may not reproduce the observed halocline changes in the eastern Eurasian Basin. On the other hand, although the sea level pressure anomaly over the Eurasian Basin averaged over the last two decades is positive ( Figure S9 ) and causes a strong freshening in the Eurasian Basin (in the wind_vari experiment, Figure 12E ), adding the sea ice decline in the model can reverse the trend, leading to salinification in the halocline of the eastern Eurasian Basin (in the historical run, Figure 12D ). This indicates that the sea ice decline is the main driver for the observed Atlantification in the eastern Eurasian Basin.