Future Atlantification of the European Arctic limited under sustained global warming

Modern Atlantification

The clustering algorithm applied to ORAS5 yields the current Atlantic domain to mainly encompass the southern side of the European Arctic, the West Spitsbergen area, the area close to the Barents Sea Opening and the southern Barents Sea (Fig. 1a,c,e,g, red lines), where the Norwegian Atlantic and West Spitsbergen currents flow (supplementary Figure S1). The position of the Atlantic-Arctic boundary is robust against different considerations of the vertical structure of the water column (see Methods). The multi-model ensemble captures the gross spatial features of the modern (1981–2010) Atlantic-Arctic boundary, especially in its eastern branch (Fig. 1a,c,e,g, blue shading). There, the S-shaped profile in MPI-ESM1-2-LR and IPSL-CM6A-LR reflects a realistic representation of the flow of the Norwegian Atlantic and Spitsbergen Polar currents, whereas the more linear profile in MRI-ESM-2-0 and MIROC-ES2L suggests a too weak circulation in the area. Models diverge more regarding the typical position of the Atlantic-Arctic boundary in the Fram Strait, reflecting uncertainties in the simulation of the West Spitsbergen current.

Mean state biases in the Atlantic-Arctic boundary are reflected in a mean state bias in the areal extent of the Atlantic domain in the European Arctic that is ubiquitous across models (Fig. 2a). Models overestimate ATLi from reanalysis especially in the Fram Strait, whereas distributions overlap in the Barents Sea region -most relevant for Atlantification- within the 1.5 interquartile range, except for IPSL-CM6A-LR (supplementary Figure S2). Notwithstanding mean state biases, simulated ATLi anomalies encompass reanalyses during the observational period (Fig. 3b), illustrating model-data consistency in the temporal evolution of Atlantification. ATLi trends (period 1958–2014) consistently quantifies the progress of Atlantification in ORAS5 (0.7–1.4%/decade) and in the ensemble mean (0.9–1.2%/decade), supporting that the forced component is predominant since the second half of the twentieth century.

Boxplots of single-model ensemble and reanalysis distributions of ATLi (a) and PAT (b) in various experiments. Shown are median (thick black line), interquartile range (IQR, boxes), 1.5*IQR (whiskers) and outliers (circles). Models are vertically aligned in panels a and b.

Evolution and trends of Atlantification and Pan-Arctic warming in the historical and future scenario (ssp585) periods. Trend analysis of ATLi (a) and PAT (c) ensemble means. Trend coefficients with p-value > 0.05 are masked out with black dots. The black cone encompasses trends estimated for the historical period alone. Time series of ATLi (b) and PAT (d) for all considered models and realizations, as well as the multi-model ensemble mean (black line). The green line in panel (b) illustrates the ATLi time series from ORAS5. Data are anomalies referred to the 1981–2010 average.

As expressed by pan-Arctic temperature, PAT (see Methods), the Arctic Ocean warms consistently across models and realizations during the historical period (Figs. 2, 3d). Ocean thermal inertia dampens high-frequency variability while highlighting multidecadal-to-centennial trends in PAT, which remain below 0.05 °C/decade in the ensemble mean during the historical period. Coherent decadal trends in ATLi and PAT indicate that Atlantification events historically occur during phases of pan-Arctic warming, while Arctification events occur during phases of pan-Arctic cooling (Fig. 3a,c). Multivariate regression of upper-ocean (0–200 m depth) temperature and salinity on ATLi and PAT supports that Atlantification and pan-Arctic warming constructively superposed in driving historical upper-ocean warming and salinification of the European Arctic, with comparable relative impacts (Fig. 4a,c, spatial fingerprints in supplementary Figures S3 and S4). The historical period thus features a tight coupling between local dynamics in the European Arctic and pan-Arctic properties.

Dependency of upper-ocean properties in the European Arctic on Atlantification and pan-Arctic warming. Box plots (line: median, box: interquartile range—IQR, whiskers: 1.5*IQR, + : outliers) are for local (grid-point) regression coefficients in the European Arctic domain: (a,b) upper-ocean potential temperature and (c,d) upper-ocean salinity on ATLi and PAT calculated for the (a,c) historical and (b,d) ssp585 experiments. Statistical significance, tested by means of a Mann–Whitney U-test, is reported in the stars above the plot as follows: *0.01 < p ≤ 0.05; **0.001 < p ≤ 0.01; ***10⁻⁴ < p ≤ 0.001; ****p ≤ 10⁻⁴.

Consistently across simulations and reanalyses, the Atlantic-Arctic boundary approximates the winter sea-ice edge but remains more locally confined than the latter through the year (Fig. 1b,d,f,h and supplementary Figures S5 and S6) maintained by buoyancy fluxes. The Atlantic-Arctic boundary position is therefore not a mere expression of sea-ice dynamics, although the latter contribute to explain the diversity of simulated behaviors.

Projected future evolution of Atlantification

The ensemble coherence extends beyond the observational period, as the models yield a strongly consistent temporal evolution of the Atlantification across models and realizations over the whole historical and ssp585/ssp245 experiments, independent of the future scenario considered. Following the initial stationary phase with lack of interdecadal trends from 1850 to the first quarter of the twentieth century, the ensemble-mean evolution of ATLi exhibits a progressive acceleration culminating around the mid-twenty-first century (Fig. 3b, ssp245 in supplementary Figure S7). The Atlantic maximizes its areal gain around 2060, followed by a stationary phase until 2100 characterized by a northward displacement of the eastern branch of the Atlantic-Arctic boundary (Fig. 1b,d,f,h). This sustained Atlantification occurs across all models, realizations and scenarios (Fig. 2a), thus representing a forced response under global warming. Under ssp585 conditions, MIROC-ES2L and MRI-ESM-2–0 yield similar peak ATLi values (0.45–0.50) but achieved through a small and homogeneous progression along the modern position in MIROC-ES2L, and a stronger progression along the northern branch of the Atlantic flow in MRI-ESM-2–0 (Fig. 1d,f). In IPSL-CM6A-LR the Atlantic domain exhibits a geographically incoherent progression yielding an areal gain of only about 5% (Fig. 1h). MPI-ESM1-2-LR yields the strongest future Atlantification (ATLi exceeding 0.50), associated with an advancement of the Atlantic similar to MRI-ESM-2-0 but extending further north (Fig. 1b). The simulated Atlantic-Arctic boundary evolution is even more diverse across models in the Fram Strait, further contributing to the ensemble spread of projected future Atlantification.

The strong projected (ssp585) sea-ice loss results in the European Arctic facing absence of sea ice in late summer in the second half of the twenty-first century and a substantial retreat of the sea-ice edge in winter at the end of the century, when in the Barents Sea it is confined to above around 80°N, that is, much north of the projected Atlantic-Arctic boundary position (Fig. 1b,d,f,h and supplementary Figure S8). This spatial detachment suggests that in a warmer world different mechanisms govern Arctic sea-ice dynamics and the Atlantic-Arctic boundary. The strong sea-ice melting can be interpreted within the framework of pan-Arctic warming. As expressed by PAT, Arctic warming accelerates around 2010 up to exceeding 0.1 °C/decade by 2030 and 0.2 °C/decade by the end of the twenty-first century. Despite the diversity of biases in the modern PAT climatology, the ensemble agrees on a future pan-Arctic warming of about 1 °C under ssp585 and 0.7 °C under ssp245 (Fig. 2b), suggesting that the destabilizing feedbacks that determine Arctic amplification are consistently still operative at the end of the twenty-first century under both scenarios. The rate of Atlantification increases since 2010 as pan-Arctic warming accelerates, until Atlantification detaches from pan-Arctic warming around 2060 (Fig. 3b,d). Pan-Arctic warming becomes predominant in driving an overall upper-ocean warming and freshening of the European Arctic (Fig. 4b,d), while Atlantification remains a significant contributor to temperature changes only in the eastern and north-eastern region (supplementary Figures S9 and S10). Therefore, Atlantification/Arctification aligns with pan-Arctic warming/cooling in the historical period, but both phenomena decouple under sustained global warming.

Natural roots of Atlantification

Preindustrial experiments under transient (past1000/past2k) and unperturbed (piControl) conditions allow framing modern and projected Atlantification, and its relation with pan-Arctic warming, in the context of natural climate variability. All models agree that Atlantification is already distinguishable from natural variability: ATLi values for the modern period are consistently significantly higher (p_ranksum < 10^–5) than transient late preindustrial (1851–1880) and unperturbed preindustrial values (Fig. 2a). However, modern Atlantification does not fully exceed the range of natural variability, as early-21^st-century ATLi values remain occasionally below the 95^th percentile of piControl values depicting intrinsic variability (supplementary Figure S11). This is expected to change under continued global warming, where future distributions for both ssp585 and ssp245 only overlap with intrinsic variability above its 95^th percentile, except for IPSL-CM6A-LR. Accordingly, the exceptional character of modern Atlantification will be fully revealed in the next decades. Further pointing to the exceptional contribution of external forcing to modern Atlantification, non-smoothed ensemble-mean values since 2007 exceed previous maxima during the whole preindustrial millennium and ensemble-mean trends in the historical period exceed preindustrial trends on all time scales (Fig. 5a,b).

Atlantification and Pan-Arctic warming in the climatic context of the last millennium. Trend analysis of ATLi (a) and PAT (c) ensemble means. Trend coefficients with p-value < 0.05 are masked out with a black dot. The black line cone encompasses trends estimated for the historical period alone. Time series of ATLi (b) and PAT (d) for all considered models, as well as ensemble mean (black line), for the historical and parent past1000/past2k experiments. All series are anomalies with respect to the period 850–2014, smoothed with a 15-year moving average. For IPSL-CM6A-LR, spurious trends of 0.002 1/century for ATLi and 0.04 °C/century for PAT are preliminarily removed from the data.

In both individual simulations and the ensemble mean, preindustrial variability includes recurrent potent Atlantification and Arctification events, with duration and intensity comparable to modern Atlantification, revealing an almost symmetrical natural pattern of variability of the Atlantic-Arctic boundary around its natural climatological position (Fig. 5a,b and supplementary Figure S11). In all models, the ratio of ATLi standard deviations calculated for the past1000/past2k and piControl experiments only slightly exceeds the value of one (MPI-ESM1-2-LR: 1.12; MRI-ESM-2-0: 1.04; MIROC-ES-2L: 1.07; IPSL-CM6A-LR: 1.24). Arctification and Atlantification therefore pertain to the intrinsic features of climate variability, and this intrinsic component remains predominant for transient natural variability of the Atlantic-Arctic boundary. External forcing thus appears to primarily exert its role by setting the phase of variability generated by intrinsic dynamics.

A major Arctification episode is coherently simulated across models in the early nineteenth century (Fig. 5b), arguably as a forced response to the 1809-Tambora volcanic cluster³⁷. The beginning of modern Atlantification is rooted in the first half of the nineteenth century, if it also encompasses the recovery from an earlier Arctification episode. Spatial patterns further support a natural origin of modern Atlantification: the ensemble yields a coherent progression of the Atlantic-Arctic boundary across the Barents Sea from late preindustrial to modern times, whereas models largely disagree on changes in the Fram Strait (supplementary Figure S12), similar to what is obtained for projected future changes (Fig. 1b,d,f,h).

Consistent with indications from the historical experiment, the preindustrial period features a recurrent coexistence between Arctic warming/cooling episodes and Atlantification/Arctification throughout the past1000/past2k experiment (Fig. 5c,d). Global warming scenarios therefore project a disruption of a coherent natural behavior coupling Atlantification/Arctification and pan-Arctic warming/cooling that lasted on a millennial time scale.

Dynamics of limited future Atlantification

Consistently across ensemble members, upper-ocean warming accelerates in the European Arctic during the late twentieth century while salinity decreases starting from the early twenty-first century, both changes contributing to a decrease in upper-ocean density (Fig. 6a,b,c). This general tendency masks a differential upper-ocean warming in the Atlantic and Arctic domains: except for IPSL-CM6A-LR, the average seawater temperature gradient (Atlantic minus Arctic) increases substantially from the 1970s until peaking in the 2050s, indicating that the Atlantic domain warms at a faster rate compared to the Arctic domain (Fig. 6d). The co-occurrence of this peak with sea-ice free late summers in the European Arctic (supplementary Figure S8) points toward a potential central role of sea-ice dynamics. For instance, sea-ice albedo and ocean latent heat loss by the ocean due to sea-ice melting contribute to slower warming of the Arctic domain. From the 2050s onwards, when sea ice retreats north of the European Arctic (supplementary Figure S8), the Atlantic-Arctic temperature gradient weakens in all models, indicating that the Arctic domain warms faster than the Atlantic domain. The Atlantic-Arctic gradient in salinity slightly increases in the historical experiment, but trajectories diverge in ssp585, yielding near-zero ensemble-mean projected future trends (Fig. 6e). Consistently across ensemble members, the Atlantic-Arctic gradient in density remains stable until around 2020 and then becomes more negative –it strengthens– until around 2050, when it stabilizes again (Fig. 6f).

Simulated evolution of oceanic conditions in the European Arctic in the historical and future scenario (ssp585) periods. Anomaly of the average potential temperature (a), salinity (b) and potential density (c) in the European Arctic. Anomaly of the differences in potential temperature (d), salinity (e) and potential density (f) between the Atlantic and the Arctic domain (Atlantic minus Arctic). Anomalies are calculated by subtracting the average value in the 1981–2010 period from the time series. Color code: MPI-ESM1-2-LR (orange), MRI-ESM-2-0 (red), MIROC-ES2L (blue), IPSL-CM6A-LR (purple), ensemble mean (black).

Therefore, the density decrease of Arctic waters is consistently slower than Atlantic waters in the ensemble, despite model specificities in the relative contributions of salinity and temperature changes. The spatial patterns of end-of-21^st-century upper-ocean density anomalies further reveal ensemble consistent features and model specificities (Fig. 7a,b,c,d, shading). All models agree on seawater density decrease near eastern Svalbard, where the Spitsbergen Polar Current typically flows southward. Models disagree instead on the density decrease in the Central Bank and surrounding areas, which traces back on the diversity of simulated changes in the Atlantic-Arctic boundary. MPI-ESM1-2-LR and MIROC-ES2L yield the most consistent anomalous density gradients across the Atlantic-Arctic boundary in the Barents Sea, seen as alignment between the boundary position and the anomalous isopycnals. In IPSL-CM6A-LR Atlantic-Arctic gradients are weakest in the Central Bank, whereas they even have an opposite sign, i.e., negative Atlantic-Arctic anomalous gradient, in the southeastern basin.

Projected ocean circulation changes contrasting Atlantification. Upper-ocean density (color maps) and baroclinic pressure gradient (arrows) differences between the averages calculated over the future (2071-2100, ssp585 experiment) and modern (1981–2010) periods for (a) MPI-ESM1-2-LR, (b) MRI-ESM-2-0, (c) MIROC-ES2L and (d) IPSL-CM6A-LR. For each model, the spatial average of the density difference is removed before mapping, amounting to (a) −0.52 kg/m³, (b) −1.32 kg/m³, (c) −0.78 kg/m³ and (d) −0.71 kg/m³, hence an upper-ocean density decrease prevails over the European Arctic. Thick black lines correspond to the Atlantic-Arctic boundary position calculated for the future period.

These changes reflect altered ocean–atmosphere interactions in the region: the anomalous density pattern superposes in the Barents Sea on winter wind-stress-curl anomalies, especially in MPI-ESM1-2-LR and MRI-ESM2-0 (supplementary Figure S13); changes in the winter net surface freshwater fluxes also contribute to the density anomaly, consistently following across models a strengthening of their historical climatological pattern of precipitation minus evaporation (supplementary Figure S14). The retreat of the sea ice edge further contributes to enhanced surface ocean turbulent heat losses in the Arctic portion of the Barents Sea (supplementary Figure S15). The overall picture is that local atmosphere–ocean interactions and local wind forcing substantially contribute to the density gradient between Atlantic and Arctic waters in the European Arctic, where the effect is mediated by sea ice changes.

The density anomalies are associated with a baroclinic pressure gradient force (Fig. 7 a,b,c,d, arrows). Besides model specificities –e.g., the southeastern basin in IPSL-CM6A-LR as mentioned above–, the ensemble consistently yields anomalous baroclinic pressure gradient force vectors in the area near the Atlantic-Arctic boundary that point south-westward, that is towards the Atlantic domain, which describes a countercurrent that contrasts Atlantification. Models lose coherence in strength and direction of the baroclinic pressure gradient force away from the Atlantic-Arctic boundary, which supports the interpretation that the dynamics described above pertain to a regional oceanic mechanism.

Large-scale oceanic precursors of the density anomalies are unclear. The ensemble yields a consistent 21^st-century progressive weakening of the Atlantic meridional overturning circulation in its historical climatological core at tropical northern latitudes, but also stationary conditions or even temporary strengthening in its shallow polar component until around 2050 (supplementary Figure S16). In MPI-ESM1-2-LR these changes are reflected in latitudinal divergent trends in the meridional ocean heat transport driven by the overturning circulation, with reduction at tropical/mid latitudes and increase at subpolar/polar latitudes (supplementary Figure S17). These changes dominate the total meridional ocean heat transport and are associated with a shrinking and north-eastward intensification of the subpolar gyre with respect to its historical climatology (supplementary Figure S18a). The subpolar gyre also strengthens in MRI-ESM-2-0, but with different magnitude and spatial pattern compared to MPI-ESM1-2-LR, whereas IPSL-CM6A-LR yields a slowdown of the subpolar gyre (supplementary Figure S18b,c).