Introduction

The Arctic Ocean, a critical component of Earth’s climate system, is undergoing rapid changes1,2,3,4. Among various Arctic regions, the large, clockwise-rotating Beaufort Gyre in the Arctic Ocean’s Canada Basin stands out as one of the most prominent features5. The Beaufort Gyre serves as a vast reservoir of freshwater, accumulating input from river runoff, precipitation, melting sea ice, and low-salinity water from the Pacific6,7,8,9. In the central Beaufort Gyre, surface salinity fell below 27 during the summer months in recent years10,11, and the isohaline corresponding to the Arctic mean salinity of 34.812 is located at approximately 400 meters depth.

Arctic freshwater storage is crucial for upper ocean stratification, influencing ocean stability, vertical heat flux and thus sea ice melting13,14,15. Sea ice retreat can amplify Arctic warming and climate change16. Furthermore, the release of freshwater from the Beaufort Gyre into the North Atlantic can affect upper ocean stratification and wintertime deep convection in regions of dense water formation, thereby impacting the Atlantic Meridional Overturning Circulation (AMOC)17,18,19,20, a key player in regulating global climate. Moreover, the gyre’s ability to retain and release freshwater has strong implications for the Arctic Ocean’s salinity and density gradients, influencing the broad circulation pattern in the Arctic Ocean21. Importantly, changes in the Beaufort Gyre circulation and stratification can also modify local concentration of dissolved oxygen and acidified water, with profound impacts on marine ecosystems22,23.

During the past two decades, the Beaufort Gyre has experienced notable changes. These changes are marked by increases in the gyre’s sea level and circulation strength24, the volume of freshwater it retains25,26,27, and mesoscale eddy activity28, induced by factors such as sea ice decline and atmospheric circulation changes29,30,31,32. In particular, the amount of freshwater stored in the Beaufort Gyre reached an unprecedented high in the late 2010s3,11,33,34. Part of the anomalous freshwater from the Beaufort Gyre, released over the past few years, has accumulated in a buffer zone north of Greenland and the Canadian Arctic Archipelago35. As the Arctic continues to lose sea ice, the state of the Beaufort Gyre is expected to shift, potentially leading to substantial changes in the Arctic Ocean’s physical and biogeochemical environment.

The Beaufort Gyre’s importance for climate and marine ecosystems points to the need for a better understanding of how it will evolve in a warming climate. Predicting future ocean changes relies on coupled climate model projections, but simulating the Arctic Ocean and the Beaufort Gyre faces challenges. Current climate models often struggle with adequately representing salinity, temperature and stratification in the Arctic Ocean36,37,38,39. In particular, most Coupled Model Intercomparison Projects (CMIP) models exhibit a fresh bias in the Arctic halocline37 and simulate an oversized Beaufort Gyre38. Ocean-sea ice model simulations forced by prescribed atmospheric reanalysis fields tend to have similar model biases40,41. It was found that increasing horizontal resolution can help reduce model biases in simulated Arctic hydrography42,43. The recent 6th phase of CMIP (CMIP6) does not demonstrate clear improvements in simulating Arctic hydrography compared with previous CMIP phases37,41, likely due to the continued use of coarse resolutions in most CMIP6 simulations. To improve the fidelity of future projections for the Arctic Ocean, efforts are needed to reduce model biases.

This paper utilizes CMIP6 dataset and dedicated high-resolution ocean-sea ice model simulations to elucidate future changes in Arctic Ocean freshwater and circulation (see Methods). We find that the Beaufort Gyre will increasingly accumulate freshwater in a warming climate, playing a dominant role in shaping the Arctic Ocean’s circulation pattern. High-resolution simulations predict a much stronger spin-up of the Arctic Ocean circulation compared to CMIP6 models. Correcting a severe, common model bias in atmospheric circulation considerably intensifies projected increases in freshwater content, sea surface height, ocean current speeds, and eddy activity in the Beaufort Gyre region, further underscoring the dominance of the Beaufort Gyre inflation in future Arctic Ocean changes.

Results

Dominance of the Beaufort Gyre inflation in a warming climate

The CMIP6 historical simulations show a spatial concentration of freshwater in the Canada Basin (Fig. 1a), which is associated with a dome in sea surface height (Fig. 1d), indicating an anticyclonic gyre circulation. However, in comparison with the freshwater content (see Methods for definition) based on observed salinity climatology44 and sea surface height and surface geostrophic currents obtained from satellite observations21,24, the simulated Beaufort Gyre exhibits an overly extensive area that encroaches into the Eurasian Basin (Fig. 2a, b and Supplementary Figs. S1, S2). The simulated salinity in individual CMIP6 models tends to show large biases relative to observations and there is a considerable spread among the models (Supplementary Fig. S3). As a result, the spread in simulated Beaufort Gyre freshwater content is also large (Fig. 3a).

Fig. 1: Mean states and future changes in freshwater content (FWC), sea surface height (SSH) and surface geostrophic current (SGC) speeds in CMIP6 models.
figure 1

a–c Multi-model mean of FWC in 1981–2000 (a), 2081–2100 (b), and the difference between the two periods (c). d–f, the same as (a–c), but for SSH. g–i, the same as (a–c), but for SGC speeds.

Full size image
Fig. 2: Freshwater content (FWC) in the historical period.
figure 2

a Observational estimates based on PHC344. b CMIP6 multi-model-mean (MMM). c FIO-ESM2.1 simulation. d FESOM2 high-resolution simulation without wind correction. e FESOM2 high-resolution simulation with wind correction. Model results are averaged over the period of 1981–2000.

Full size image
Fig. 3: Dominance of the state change of the Beaufort Gyre in CMIP6 simulations.
figure 3

a Time series of freshwater content (FWC) in the Beaufort Gyre region. The thick solid line denotes the multi-model mean (MMM), the thin gray lines denote individual models, and the shading indicates model spreads ( ± 0.5 standard deviation). b, MMM changes of FWC between 1981–2000 and 2081–2100 in the Beaufort Gyre region in comparison with other Arctic regions. BG: Beaufort Gyre region (Canada Basin), MB: Makarov Basin, EB: Eurasian Basin, ESS: East Siberian Sea, LS: Laptev Sea, BKS: Barents-Kara seas. The inset depicts the Arctic regions analyzed in this study. The error bars indicate model spreads. c–d the same as (a–b), but for sea surface height (SSH). e-f, the same as (a–b), bur for surface geostrophic current (SGC) speeds.

Full size image

The freshwater content is projected to increase in most areas of the Arctic deep basin (the Arctic area excluding continental shelf) in future warming climate, with the largest increase occurring in the Canada Basin (Fig. 1b, c). Compared to the period of 1981–2000, the freshwater content in the Beaufort Gyre region increases by an average of 12 m in the period of 2081–2100 according to the CMIP6 multi-model mean (MMM) (Fig. 3a, b), which corresponds to an increase of about 75%. This accumulation of freshwater in the Beaufort Gyre region clearly surpasses that of other Arctic regions, including the Eurasian and Makarov basins, as well as the shelf seas (Figs. 1c and 3b). As a result, the spatial asymmetry in Arctic freshwater distribution, with a concentration in the Canada Basin, will grow with climate warming.

In the future warming climate, the sea surface height in the Arctic deep basin is projected to increase, consistent with the increase in freshwater content (Fig. 1e, f and Supplementary Fig. S4a). The sea surface height in this study refers to the dynamic sea level, which is relative to the global mean sea level and does not contain contributions from global total water mass changes. The Beaufort Gyre region is expected to experience the largest rise in sea surface height, with an average increase of about 24 cm during the period of 2081–2100 compared to the period of 1981–2000 (Figs. 1f, 3c, and d). The shallow shelf seas will also experience an increase in sea surface height, but this is not mainly attributable to ocean freshening (Fig. 1c, f). Instead, it is likely a result of increased water mass over the continental shelf due to deep-basin steric expansion, a phenomenon particularly pronounced in the Arctic45,46. The CMIP6 models predict strong freshening (Fig. 1c) and warming4,37 trends in the Arctic deep basin, indicating the occurrence of deep-basin steric expansion, which can increase the water mass over surrounding continental shelf as suggested in previous studies45,46. Among the Siberian shelf seas, the East Siberian Sea shows the largest increase in mean sea surface height, with a magnitude slightly less than the mean increase in the Beaufort Gyre region (Fig. 3d).

Associated with the doming of the Beaufort Gyre, the surface geostrophic current in the gyre is expected to intensify in the future (Fig. 1h, i). The CMIP6 MMM indicates an increase of slightly more than 1 cm s−1 in the mean current speed between the two considered periods, larger than in other Arctic basins and shelf seas (Fig. 3e, f). The future changes in surface geostrophic current in the Beaufort Gyre region are not significantly correlated with changes in sea surface height across the CMIP6 models (Supplementary Fig. S4b), likely because the models simulate an oversized Beaufort Gyre, meaning the changes in the spatial gradients of sea surface height in the Beaufort Gyre region are not well represented. The mean surface geostrophic current speeds in shelf seas do not increase proportionally with the rise in sea surface height (Fig. 3d, f). This is because the relatively large spatial gradients of sea surface height — and thus the associated surface geostrophic currents—are confined to the continental slope areas (Fig. 1d–i).

The interannual variability of the freshwater content, sea surface height and surface geostrophic current in the Beaufort Gyre region is projected to increase in future warming climate (Fig. 4a–c), consistent with previous studies showing increased ocean response to winds when ocean surface stress and air-sea momentum transfer is enhanced by sea ice decline47. In contrast, the magnitudes of their seasonal cycles are projected to decrease (Fig. 4d–f), which can be explained by the weakening of the seasonal cycle of surface freshwater fluxes associated with diminishing sea ice seasonal melting and freezing.

Fig. 4: Interannual and seasonal variability in CMIP6 simulations.
figure 4

a–c The standard deviation (SD) of annual mean (a) freshwater content (FWC), (b) sea surface height (SSH) and (c) surface geostrophic current (SGC) speeds in the Beaufort Gyre. The SD was calculated after detrending. Red marks in the boxes indicate the median and the bottom and top edges of the boxes indicate the 25th percentile and 75th percentile, respectively. The black vertical lines extend to the most extreme data points not considered outliers. The black circle denotes outliers. Individual models are indicated by red dots. d–f, Seasonal variability of (a) FWC, (b) SSH and (c) SGC speeds in the Beaufort Gyre. Annual means are subtracted to obtain seasonal cycles.

Full size image

Overall, the CMIP6 models demonstrate that the inflation of the Beaufort Gyre, including the increases in freshwater storage, sea surface height and geostrophic current speeds, will surpass those in other Arctic basins and shelf seas. The accumulation of freshwater in the Arctic Ocean is consistent with the projected rise in Arctic surface freshwater fluxes in a warming climate (Supplementary Fig. S5). Across the CMIP6 models, increases in Beaufort Gyre freshwater content are correlated with both the rise in surface freshwater fluxes and the decrease in surface stress curl (corresponding to increased Ekman convergence and downwelling) (Fig. 5a, b), two of the main factors contributing to Beaufort Gyre freshwater accumulation3. Transport by mesoscale eddies acts to counteract freshwater accumulation in the Beaufort Gyre48,49. The correlation coefficients between changes in freshwater content and changes in surface freshwater fluxes and surface stress curl are not very high (Fig. 5a, b), suggesting that the implementation of eddy parameterization, which varies across CMIP6 models, has considerable impacts on simulated freshwater content.

Fig. 5: Correlation relationship across CMIP6 models.
figure 5

a Correlation between changes in Beaufort Gyre freshwater content (FWC) and Arctic surface freshwater flux (SFF). b Correlation between changes in Beaufort Gyre FWC and surface stress curl (SSC). c Correlation between changes in Beaufort Gyre surface stress and near-surface wind speeds. d Correlation between changes in Beaufort Gyre surface stress and sea ice extent (SIE). The changes are the differences between 2081–2100 and 1981–2000.

Full size image

The increase in ocean surface stress in the Beaufort Gyre region correlates more strongly with rising wind speeds than with decreasing sea ice cover across the CMIP6 models (Fig. 5c, d). However, wind speeds in the Beaufort Gyre region are projected to increase by approximately 12% from 1981–2000 to 2081–2100 in the CMIP6 MMM, while ocean surface stress is projected to increase by about 48% over the same period, suggesting that sea ice decline plays an important role in amplifying ocean surface stress. This will be further demonstrated through the following sensitivity simulations.

Stronger ocean spin-up in high-resolution simulations

We performed a high-resolution FESOM2 simulation forced by atmospheric fields derived from a coupled climate model, which covers both the historical period and future projection (see Methods). This high-resolution simulation demonstrates a dominance of Beaufort Gyre inflation in the Arctic Ocean, supporting the main conclusions based on CMIP6 models (Figs. 6 and 7). Employing high resolution considerably improves the representation of the salinity vertical profile (Supplementary Fig. S3) and the spatial pattern of the Beaufort Gyre (Fig. 2). Our results indicate that, despite larger biases in the simulated mean state of salinity in individual CMIP6 models, their MMM changes in freshwater content and sea surface height averaged in the Canada Basin are similar to our high-resolution simulation (Fig. 6a, b).

Fig. 6: Dominance of the state change of the Beaufort Gyre in high-resolution and CMIP6 simulations.
figure 6

a Changes of freshwater content (FWC) between 1981–2000 and 2081–2100 in the Beaufort Gyre region in comparison with other Arctic regions. BG: Beaufort Gyre region (Canada Basin), MB: Makarov Basin, EB: Eurasian Basin, ESS: East Siberian Sea, LS: Laptev Sea, BKS: Barents-Kara seas. These regions are depicted in Fig. 3b. b the same as (a), but for sea surface height (SSH). c the same as (a), but for surface geostrophic current (SGC) speeds.

Full size image
Fig. 7: Future changes in freshwater content (FWC), sea surface height (SSH) and surface geostrophic current (SGC) speeds in high-resolution simulations.
figure 7

a–c Changes in (a) FWC, (b) SSH and (c) SGC between 1981–2000 and 2081–2100 in the simulation without wind corrections. d–f, The same as (a–c), but for the simulation with wind corrections. In the latter simulation, wind biases associated with the sea level pressure biases depicted in Fig. 8g are subtracted from the applied wind forcing.

Full size image

The future increase in surface geostrophic current speeds is notably greater in the high-resolution simulation compared to the CMIP6 models (Fig. 6c), which is apparent in most areas of the Arctic Ocean (Figs. 1i and 7c). The discrepancy in the Beaufort Gyre region is linked to the larger spatial gradients in freshwater content and sea surface height in the Canada Basin captured by the high-resolution simulation (Supplementary Fig. S6). Furthermore, the high-resolution simulation is eddy-permitting. The eddy kinetic energy (EKE) in the Beaufort Gyre region is relatively low in the historical period due to sea ice dissipation, while it is projected to increase with sea ice decline under a warming climate (Supplementary Fig. S7a,c)50. Thus, the varying surface geostrophic currents associated with enhanced eddy activity also contribute to the increased current speeds in the high-resolution simulation. The current speeds in the Beaufort Gyre region, calculated from monthly mean sea surface height, are 5.3 cm s−1 larger in 2081–2100 than in 1981–2000 (Fig. 6c). This increase is much larger than the simulated increase of about 1 cm s−1 in the CMIP6 MMM.

In addition to the basin-wide ocean spin-up under a warming climate (Fig. 6c), the high-resolution simulation projects a future strengthening of Arctic boundary currents along the continental slopes (Fig. 7c). This trend is substantially weaker in the CMIP6 models (Fig. 1i).

Greater-than-projected Beaufort Gyre inflation

The Beaufort Gyre is driven by the Arctic high-pressure system centered over the Canada Basin, known as the Beaufort High (Fig. 8a). In the CMIP6 models, the simulated Beaufort High during the historical period has a magnitude similar to observations, but its center is displaced toward the Eurasian sector (Fig. 8b), with a negative bias in sea level pressure in the western Arctic and a positive bias in the eastern Arctic (Fig. 8c). In the future warming climate, the Arctic sea level pressure is expected to decrease (Fig. 8b, d). In most CMIP6 models, the locations of their Beaufort High center in both the historical period and future scenarios remain closer to the central Arctic than observed (Supplementary Figs. S8 and S9). Considering different atmospheric reanalysis products and different historical periods, the sea level pressure biases in CMIP6 models remain similar (Supplementary Fig. S10). Assuming the models maintain the same biases in the future, we subtract the biases from the sea level pressure over time. This correction results in a stronger Beaufort High with its center repositioned over the Canada Basin, rather than the central Arctic (Fig. 8e). The coupled climate model FIO-ESM2.1, from which we derived atmospheric fields to drive FESOM2 simulations, shows a bias pattern similar to CMIP6 models (Fig. 8f–i).

Fig. 8: Atmospheric circulation bias in climate model simulations.
figure 8

a Sea level pressure (SLP) in atmosphere reanalysis JRA55-do averaged over the period of 1960–2009. b CMIP6 multi-model mean (MMM) SLP averaged over the period of 1960–2009. c, SLP bias of CMIP6 MMM averaged over the period of 1960–2009. d, CMIP6 MMM SLP averaged over the period of 2081–2100. e Corrected (CRD) CMIP6 MMM SLP averaged over the period of 2081–2100. f–i, the same as (b–e), but for the SLP from the FIO-ESM2.1 simulation.

Full size image

We corrected the wind forcing for the entire period by subtracting the wind biases averaged over the historical period and used both the uncorrected (FESOM2 NFC) and corrected (FESOM2 FC) winds to drive the high-resolution FESOM2 simulations (see Methods). By comparing the two simulations we can understand how the wind biases influence the simulated changes in the Beaufort Gyre.

The corrected winds considerably increase the accumulation of freshwater in the Beaufort Gyre region (Fig. 7a, d), which can be explained by enhanced ocean surface stress and thus Ekman transport toward the Canada Basin (Supplementary Figs. S11,S12). The increase of the freshwater content in the Beaufort Gyre from 1981–2000 to 2081–2100 is 32% larger with the wind correction (Fig. 9a and Supplementary Fig. S13). Correspondingly, the future increase in sea surface height in the Beaufort Gyre is 45% larger with the wind correction (Figs. 7b, e and 9b). The surface geostrophic current speeds in the Beaufort Gyre region are also intensified with the wind correction (Fig. 7c, f), reaching a future increase of about 7 cm s−1, which is 27% greater than the case without wind correction (Fig. 9c). Comparing the two high-resolution simulations, it is evident that the wind biases lead to an underestimation of the dominance of the Beaufort Gyre inflation in shaping the Arctic Ocean changes (Figs. 6 and 7). Since the same wind correction was applied throughout the simulation, the fact that the simulation with wind correction shows a larger inflation of the Beaufort Gyre in the future warming climate reveals the role of sea ice decline in increasing ocean surface stress, and consequently, Ekman convergence toward the Beaufort Gyre region (Supplementary Figs. S11,S12).

Fig. 9: Underestimated state change of the Beaufort Gyre due to bias in atmospheric circulation simulations.
figure 9

a Anomaly of Beaufort Gyre freshwater content (FWC) relative to the 1981-2000 mean. b-f, the same as (a), but for (b) sea surface height (SSH), (c) surface geostrophic current (SGC) speeds, (d) upper 300 m available potential energy (APE), (e) upper 300 m mean kinetic energy (MKE), and (f) upper 300 m eddy kinetic energy (EKE) in the Beaufort Gyre region. The blue lines denote the results of the high-resolution simulation without wind corrections, and the red lines denote those with wind corrections. The percentages of the differences between the two simulations averaged over 2081–2100 are shown in each panel. The EKE data are only saved from the simulations for some years.

Full size image

Due to greater freshwater accumulation, the increase of available potential energy (APE) in the upper 300 m of the Beaufort Gyre from 1981–2000 to 2081–2100 is 34% larger with the wind correction (Fig. 9d). With the wind correction, the mean kinetic energy (MKE) in the upper 300 m of the Beaufort Gyre is projected to increase by 70% more (Fig. 9e). Owing to the greater increase in freshwater content and APE, the increase in Beaufort Gyre eddy kinetic energy (EKE) is 61% larger (Fig. 9f). While correcting winds does not obviously affect the EKE in the historical period, it has a strong impact in the future warming climate (Supplementary Fig. S7), highlighting the role of sea ice decline in amplifying the effects of wind anomalies. The wind correction also strengthens the projected increase in freshwater content, sea surface height and surface geostrophic current speeds in the Makarov Basin, although the strengthening is less pronounced than in the Beaufort Gyre region (Fig. 6).

Discussion

The Beaufort Gyre regulates the Arctic Ocean’s physical and biogeochemical environment and influences deep water formation in the subpolar North Atlantic3,11,22. Understanding and predicting its changes in a warming climate is therefore crucial. Using results from CMIP6 coupled climate models and dedicated high-resolution simulations, we find that the Beaufort Gyre will accumulate more freshwater, experience a rise in sea level, and undergo intensified circulation as the climate continues to warm. This inflation is more pronounced than in other Arctic basins and shelf seas, causing the Beaufort Gyre to become a more dominant feature of the Arctic Ocean. The growing asymmetry in the spatial distribution of Arctic Ocean freshwater, with an increasing concentration in the Canada Basin, implies rising importance of the Beaufort Gyre. While the Beaufort Gyre will experience dominant inflation, our high-resolution simulations also indicate a widespread spin-up of ocean circulation throughout the Arctic.

Atmospheric cyclone (low-pressure system) activity in the Arctic has shown an increasing trend over the last few decades51 and is projected to continue increasing in the future52. An observed increase in the intrusion of cyclones from lower latitudes into the Arctic has tended to weaken the atmospheric Beaufort High53,54. In addition, the Arctic is also warming two to three times faster than the global average16,55. Consistent with the effects of cyclones and the generation of thermal lows, there will be a reduction in Arctic sea level pressure in a warming climate (Fig. 8 and Supplementary Fig. S14). However, the Arctic winds will rather strengthen (Fig. 5c) due to weakened atmospheric stability56,57. The resulting increase in ocean surface stress and the projected increase in surface freshwater fluxes together will lead to Beaufort Gyre inflation.

Across the analyzed CMIP6 models, the projected increase in ocean surface stress in the Beaufort Gyre region is not correlated with the decrease in sea ice extent (Fig. 5d). However, this does not imply that the decline in sea ice does not contribute to the simulated increases in ocean surface stress and freshwater accumulation. First, ocean surface stress is influenced by ocean surface currents in the presence of seasonal sea ice. Consequently, uncertainties in the simulated freshwater content associated with factors such as eddy and mixing parameterizations, may weaken the correlation between changes in ocean surface stress and sea ice extent across the models. Second, dedicated modeling studies, in which winds remain the same while Arctic sea ice cover is reduced by increasing Arctic air temperature, have shown that sea ice decline considerably enhances ocean surface stress and Ekman transport58. Similarly, the decline of sea ice over the past two decades has been found to play a key role in the spatial distribution of freshwater in the Arctic Ocean by increasing ocean surface stress33. Third, as shown by our high-resolution simulations, although the same wind anomaly is added to the wind forcing throughout the simulations, the differences in the ocean are much more pronounced in the future than in the current climate (Fig. 9). This clearly demonstrates the effects of sea ice decline in amplifying ocean surface stress.

Eddies act to counteract increases in freshwater content in the Beaufort Gyre28,48,49. CMIP6 models, which have very coarse resolution and parameterized eddy effects, exhibit a large spread in the simulated Arctic Ocean hydrography and freshwater content as shown in this work (Figs. 3 and 5) and previous studies36,37,38,59. The misrepresentation of the spatial extent of the Beaufort Gyre in these coarse-resolution models leads to an underestimation of the projected increase in Beaufort Gyre circulation, potentially causing biases in ocean-ice stress. The spatial extent of the Beaufort Gyre is more realistically simulated in the high-resolution simulations. They predict much greater increases in surface geostrophic current speeds in the Beaufort Gyre and along Arctic continental slopes than CMIP6 models. Although we cannot exclude possible uncertainties in projected ocean changes associated with uncertainties in the global atmospheric fields used to drive our high-resolution simulations, our study addresses the benefit of using high-resolution ocean models. It remains to see how further increases in model resolution, reaching a scale of eddy-rich ocean, might influence the quantitative representation of future freshwater content in the Beaufort Gyre region.

Present climate models commonly exhibit a bias in simulating the atmospheric Beaufort High, extending it too far into the Eurasian Arctic (Fig. 8). This bias tends to lower the freshwater content in the Canada Basin and the strength of the Beaufort Gyre. Notably, the impact of this bias on the ocean increases with sea ice decline. This also highlights the role of sea ice decline in enhancing ocean surface stress and air-sea momentum transfer under a warming climate. Our results suggest that the dominance of the Beaufort Gyre inflation is likely underestimated in CMIP6 models. Our high-resolution simulations indicate that correcting the wind bias also intensifies the projected increase in Beaufort Gyre eddy activity in a warming climate.

The Beaufort Gyre is a prominent feature of the Arctic Ocean in the present climate. Its strong inflation in a warming climate implies an increasing role in modulating the distribution of freshwater and chemical tracers in both the Arctic Ocean and the subpolar North Atlantic, with greater impacts on climate and marine ecosystems. Sustained observations and improved climate models are important for better understanding and predicting the evolving role of the Beaufort Gyre.

Methods

CMIP6 dataset

We used historical and projection simulations from 39 CMIP6 coupled climate models (Table S1). The employed CMIP6 simulations cover the periods 1950–2014 for historical experiments and 2015–2100 for projection experiments. The projections used in this study are based on the SSP585 scenario, which represents a pathway of rapid, fossil-fueled development with an effective radiative forcing of 8.5 Wm−2 in 210060. By using this scenario, the time series of the analyzed variables illustrate their evolution in line with the trajectory of climate warming, from the current climate to a world with 4 °C of global temperature increase61. We analyzed the following diagnostics from these CMIP6 models: freshwater content (FWC, equation 1), sea surface height, surface geostrophic current speeds, salinity profiles, surface freshwater fluxes (including runoff, precipitation, evaporation and fluxes associated with sea ice thermodynamic processes), sea level pressure, sea ice extent, ocean surface stress, surface stress curl and near-surface wind speeds. Not all the required variables are available in every CMIP6 model for analyzing these diagnostics (Table S1). In particular, the number of models providing sea surface height (dynamic sea level), ocean surface stress and grid information allowing to calculate surface stress curl is limited. For consistency, we used models that provide both three-dimensional salinity and sea surface height data when these two variables are linked or compared, including Figs. 1, 2, 3. Due to the relatively small size of the ensemble, the fingerprint of individual models’ internal variability is not fully eliminated in the MMM, as shown in Fig. 3a, c, e.

Our evaluation of CMIP6 MMM results reveals that the simulated Arctic precipitation minus evaporation in the historical period closely matches reanalysis and synthesized climatology, while river runoff is likely underestimated (Supplementary Fig. S5a). The simulated changes in Arctic sea ice extent are in good agreement with satellite observations (Supplementary Fig. S5b). The models tend to have fresh biases in the halocline and saline biases near the surface (Supplementary Fig. S3). The spatial extent of the Beaufort Gyre is considerably overestimated (Fig. 2 and Supplementary Figs. S1,S2). The large ocean biases in CMIP6 models highlight the need for high-resolution simulations, which are used in this study. Additionally, CMIP6 models show a common bias in Arctic sea level pressure, with the center of the Beaufort High displaced toward the Eurasian Basin (Fig. 8 and Supplementary Figs. S8,S9). This motivates the correction of the wind bias in our high-resolution simulations. Details are provided in the sections below.

It is important to note that climate models do not necessarily reproduce observed variability in the historical period. For example, if the strengthening of the anticyclonic winds over the Canada Basin during the last two decades3 is absent in a coupled climate simulation, we would not expect the simulated freshwater content in the Beaufort Gyre to increase as much as observed (Supplementary Fig. S13). This discrepancy also helps explain why the simulated sea surface height and surface geostrophic currents are weaker than those from recent satellite observations (Supplementary Figs. S1,S2).

High-resolution simulations

We performed high-resolution simulations using the Finite volumE Sea ice-Ocean Model (FESOM2)62,63. FESOM2 is a global general circulation ocean model that employs unstructured meshes, allowing for variable horizontal resolutions and thus enabling efficient simulations.

The model configuration used in this study has a horizontal resolution of 4.5 km in the Arctic Ocean, decreasing to about 24 km in the subpolar region and 1∘ in most other parts of the global ocean. The resolution is also slightly increased to 1/3∘ in the equatorial belt and along coastlines. The mesh consists of 47 z-levels, with 5 m vertical spacing at the surface, which coarsens with depth. Eddy diffusivity is scaled with local horizontal resolution. In the Arctic region, the resolution is eddy-permitting, so parameterization for mesoscale eddies is switched off. This model setup was described and evaluated within the framework of Ocean Model Intercomparison Project phase 2 (OMIP2)43. When forced by atmospheric reanalysis fields, it demonstrates good performance in reproducing the observed mean state and temporal changes in the Arctic Ocean, including key parameters such as salinity, temperature, mixed layer depth, halocline depth, freshwater content and Arctic gateways transports43.

The surface forcing used to drive FESOM2 is derived from the output of the coupled climate model FIO-ESM v2.164, an improved version of FIO-ESM v2.0, which participated in CMIP665. FIO-ESM v2.0 demonstrated good skills in simulating atmospheric fields, ranking in the top two across 14 CORDEX (COordinated Regional Downscaling EXperiment) domains, including the Arctic region66. The update to FIO-ESM v2.1 aims to further improve sea ice simulations by upgrading its sea ice component model and improving ice-ocean heat exchange parameterization64. The atmospheric fields and river runoff derived from FIO-ESM v2.1 cover the historical period (1900–2014) and the SSP585 future scenario (2015–2100). These fields include near-surface winds, air temperature, specific humidity, downward shortwave radiation, downward longwave radiation, rainfall, and snowfall, with a spatial resolution of 1.25∘ × 0. 9∘ and a temporal resolution of three hours.

The FESOM2 simulations are initialized using observed temperature and salinity from Polar Science Center Hydrographic Climatology version 3 (PHC3)44. The simulations cover the entire period from 1900 to 2100. Compared to observations, FESOM2 historical simulations provide a more accurate representation of the salinity vertical profile (Supplementary Fig. S3) and the spatial pattern of the Beaufort Gyre than CMIP6 models and FIO-ESM v2.1 (Fig. 2 and Supplementary Figs. S1,S2). Therefore, we expect that the FESOM2 simulations offer more reliable predictions of future Arctic Ocean changes upon the reliability of the applied surface forcing.

Forcing correction

We found that nearly all CMIP6 models exhibit a common bias in simulated sea level pressure in the historical period, with the Beaufort High excessively extending into the Eurasian Arctic compared to JRA55-do67 reanalysis data (Supplementary Fig. S8). The CMIP6 MMM sea level pressure shows a positive bias in the eastern Arctic and a negative bias in the western Arctic (Fig. 8b, c). FIO-ESM2.1 displays a similar bias (Fig. 8f, g). Using different reanalysis products for different periods, the CMIP6 models’ bias is found to be similar (Supplementary Fig. S10). Although the exact cause of this common model bias is unknown, it is reasonable to assume that climate models tend to have similar biases in their projection simulations. Indeed, the projected Beaufort High in future warming climate shows a spatial displacement similar to the historical simulations despite an overall reduction in Arctic sea level pressure (Fig. 8b, d and Supplementary Fig. S9). If we subtract the sea level pressure bias of the historical simulations from the sea level pressure of the future projections, we recover the canonical pattern of sea level pressure, with the center of the Beaufort High located over the Canada Basin (Fig. 8e, i). If we apply this correction to the sea level pressure of the historical simulations (Fig. 8b, f), by definition, the resultant sea level pressure is exactly the same as the reanalysis data used to calculate the bias.

Sea level pressure is used in Fig. 8 to illustrate the bias in the atmospheric circulation and the concept of forcing correction, but it is the winds, not sea level pressure, that are used to drive model simulations. To assess the impacts of this common bias on the simulated Arctic Ocean, we corrected the winds derived from FIO-ESM2.1 and performed another FESOM2 simulation. First, we calculated the temporal-mean biases of the zonal and meridional winds relative to the JRA55-do atmospheric reanalysis67 averaged over the period 1960–2009. The biases were calculated for each 3-hourly record separately. Then, these 3-hourly biases were subtracted from the 3-hourly winds for the entire period (1900–2100). This forcing correction does not change the interannual variability and trend of winds. Comparing the FESOM2 simulation with forcing correction (FC) and the one without forcing correction (NFC) reveals the impacts of the common bias in atmospheric circulation. As shown in the main text, we find that the inflation of the Beaufort Gyre is likely underestimated in CMIP6 simulations due to this atmospheric circulation bias.

Analysis of ocean model data

The FWC in a water column is defined as follows:

$${{{\rm{FWC}}}}=\int_{D}^{{{{\rm{surface}}}}}({S}_{{{{\rm{ref}}}}}-S)/{S}_{{{{\rm{ref}}}}}{{{\rm{d}}}}z,$$
(1)

where S denotes ocean salinity, Sref the reference salinity, D the depth where ocean salinity is equal to the reference salinity, and z the vertical coordinate. The reference salinity is set to 34.8, the mean salinity of the Arctic Ocean12. The FWC in a column refers to the amount of pure water that can be extracted so that the mean salinity of the column equals the reference salinity. As changes in steric height (that is, changes in sea level due to ocean density changes associated with temperature and salinity changes) in the Arctic deep basin are mainly determined by salinity changes, FWC is directly correlated with sea surface height. In practice, this relationship allows FWC changes to be derived using satellite-observed sea surface height changes3,24,25,26. To keep the consistency, we used CMIP6 models that provide both salinity and sea surface height data for both historical and projection simulations for our analysis of FWC and sea surface height.

The available potential energy (APE) in the upper 300 m is defined following previous studies68:

$${{{\rm{APE}}}}=\int_{{300{\rm {m}}}}^{{{{\rm{surface}}}}}g(\rho -{\rho }_{{{{\rm{ref}}}}})z{{{\rm{d}}}}z,$$
(2)

where g is the gravity acceleration, ρ is potential density, and ρref is the potential density at 300 m depth. An increase in FWC is expected to increase APE, thus increasing mesoscale eddy activity.

We use the Reynolds averaging method to split the total kinetic energy (TKE) into mean kinetic energy (MKE) and eddy kinetic energy (EKE) components. MKE represents kinetic energy associated with the mean flow, while EKE represents kinetic energy associated with turbulent and eddy fluctuations in the flow. We decompose ocean velocity to a time mean component (indicated by overbars) and an eddy component (indicated by primes), \({{{\bf{u}}}}=(u,v)=(\overline{u}+{u}^{{\prime} },\overline{v}+{v}^{{\prime} })\). MKE and EKE are calculated as follows:

$${{{\rm{MKE}}}}=\frac{1}{2}({\overline{u}}^{2}+{\overline{v}}^{2}),$$
(3)
$${{{\rm{EKE}}}}={{{\rm{TKE}}}}-{{{\rm{MKE}}}}=\frac{1}{2}(\overline{{u}^{2}}+\overline{{v}^{2}})-\frac{1}{2}({\overline{u}}^{2}+{\overline{v}}^{2})=\frac{1}{2}(\overline{{u}^{{\prime} 2}}+\overline{{v}^{{\prime} 2}}).$$
(4)

Here, \(\overline{u}\), \(\overline{v}\), \(\overline{{u}^{2}}\) and \(\overline{{v}^{2}}\) are monthly means from the model output. \(\overline{{u}^{2}}\), \(\overline{{v}^{2}}\) are saved only for the periods of 1981–2000 and 2081–2100 in the FESOM2 simulations, so EKE is available for these periods.

Observational and reanalysis datasets

The following datasets are used in this study to evaluate model results: the PHC3 gridded salinity data based on in-situ observations44, sea surface height and surface geostrophic currents from satellite measurements21,24, time series of freshwater content in the Beaufort Gyre derived from in-situ salinity observations3,11, synthesized surface freshwater fluxes8, river runoff estimates based on observations69, precipitation and evaporation data from the ERA-Interim reanalysis70, sea level pressure from the JRA55-do67, ERA571, and MERRA272 reanalysis products, and sea ice extent data from satellite observations73. The observational datasets used here are either the only available data of the type (spatial and temporal coverage) we require or known high-quality data. Because long-term Arctic-wide precipitation, evaporation, total sea surface freshwater flux, and sea level pressure are not available from observations, we used reanalysis data and synthesized values from previous studies.