Arctic climate and its interaction with lower latitudes under different levels of anthropogenic warming in a global coupled climate model

The model used in this study is the version 3.0.1 of the global coupled climate model EC-Earth (Hazeleger et al. 2010, 2012; Sterl et al. 2012), which is the successor of version 2.3 used for CMIP5, and which was also used by Batté and Doblas-Reyes (2015) and Davini et al. (2015). Compared to version 2.3, EC-Earth3.0.1 includes updated versions of its atmospheric and oceanic model components, as well as a higher horizontal and vertical resolution in the atmosphere.

The atmospheric component of EC-Earth is the Integrated Forecast System (IFS) of the European Centre for Medium Range Weather Forecasts (ECMWF). Based on cycle 36r4 of IFS, it is used at a T255 resolution, using a reduced Gauss-grid. The model has 91 vertical levels, thereof 50 above 200 hPa. The model top is at 0.01 hPa.

The ocean component is the Nucleus for European Modelling of the Ocean (NEMO, Madec 2008). It uses a tri-polar grid with poles over northern North America, Siberia and Antarctica with a resolution of about 1 degree (the so-called ORCA1-configuration) and 46 vertical levels (compared to 42 levels in the CMIP5 model version). The upper model level is at about 3 m and 10 levels are in the upper 100 m. The ocean model is based on NEMO version 3.3.1 and includes the Louvain la Neuve sea-ice model version 3 (LIM3, Vancoppenolle et al. 2012), which is a dynamic-thermodynamic sea-ice model. EC-Earth3.0.1 uses LIM3 with only one sea ice category. The atmosphere and ocean/sea ice parts are coupled through the OASIS (Ocean, Atmosphere, Sea Ice, Soil) coupler (Valcke 2006) every three hours.

Three 100-year long simulations with EC-Earth were performed, forced with three different constant greenhouse gas concentrations corresponding to year 2000 (EXP2000), 2015 (EXP2015) and 2030 (EXP2030), respectively. Concentration levels for year 2000 are based on observations; for years 2015 and 2030, the respective levels from the RCP4.5 emission scenario have been used. The CO₂ concentration used in EXP2000 is the observed value from year 2000 of 368.87 ppm; in EXP2015 and EXP2030, 399.97 ppm and 435.05 ppm are used, respectively. The CO₂ increase between EXP2030 and EXP2015 is thus slightly larger than the increase between EXP2015 and EXP2000. However, because of the logarithmical increase of the radiative forcing with increased CO₂ concentration, the increase in radiative forcing is nearly linear across our simulations. All three simulations were started from the same initial conditions obtained from the end of a 200-year long present day (using constant year 2000 forcing) control simulation. EXP2000 is a continuation of this present day control simulation using exactly the same model version and configurations.

For most of the analysis, year 21–100 of the simulations are used; the first 20 years are disregarded to allow the model to adjust to the modified external forcing. Results from existing studies (Döscher et al. 2010, Tietsche et al. 2011) indicated that near surface variables, as e.g. sea ice, recover from initial shocks within about one decade. Our EXP2000 simulation is in total 300 year long and even the deep ocean is in good equilibrium (see Fig. 1). This indicates that trends at the surface, and probably even in the AMOC, are mainly caused by natural variations and are not due to model drift in EXP2000. EXP2015 shows very small trends in Arctic T2m (−0.07 K/80 years) and SST (0.045 K/80 years), AMOC (0.24 Sv/80 years) and Arctic sea ice parameters (see Sect. 3.2; September ice extent trend: −1.6 × 10¹⁰ m²/80 years). The trends in EXP2015 are smaller than the trends in EXP2000, except for the deep ocean where we find a small drift of 0.03 K in 100 years. Given the fact, that EXP2000 is in equilibrium in the deep ocean, the trends in EXP2000 are probably mainly due to internal variability. Since the trends are smaller in EXP2015 than in EXP2000, we conclude that the drift in EXP2015 due to the initialization is small, and is no problem for the analysis in this study. In EXP2030, the trends are larger and the trend of the global mean SST reaches 0.24 K/80 years. Arctic temperature and the AMOC show a positive trend and the Arctic sea ice extent a negative trend. Most of these trends are related to a sudden warming in both the Arctic and the global mean after year 60 (Fig. 1a, b). It remains unclear if this warming is due to natural variability or a late response to the changes in the external forcing. In order to estimate the effect of the trends on the correlation between different variables in our simulations, we also performed all correlation analyses with detrended data. However, this had only a small effect on the results. Only in EXP2030, we see slightly reduced values for most of the correlations.

Open image in new window

Note: our 100-year time series are too short to analyze multi-decadal variations in the ocean in detail. However, they allow to estimate the mean Arctic climate change caused by changes in the greenhouse gas forcing, and they allow to investigate the possible linkages between the Arctic and lower latitudes, and how these linkages might change under different greenhouse gas forcings.

Variations of the sea ice edge affect the atmosphere locally and possibly remotely. In a transient climate, the sea ice edge moves to the north with superimposed variations, which makes it almost impossible to extract the atmospheric response to the sea ice change from the sea ice variability. Note: our simulations differ only in the greenhouse-gas forcing. All other external forcings are the same in all three simulations. Thus, in contrast to observations and transient model simulations, our work really focuses on the differences caused by changes in the greenhouse-gas forcing.

The spatial distribution of sea surface temperature (SST) and its changes in EXP2015 and EXP2030 compared to EXP2000 in mid and high northern latitudes are presented in Fig. 2a–c. The main discrepancy from observed present day SST is a cold bias of several K in the sub-polar gyre. This cold bias is a typical problem of coarse resolution ocean models (Large and Danabasoglu 2006; Eden et al. 2004) and is due to a North Atlantic Current that is too zonal and misplaced to the south (see discussion in Sterl et al. 2012). The SST change in EXP2015 is dominated by a strong cooling in an area south of Greenland. In the Labrador Sea and the northeastern North Atlantic, surface temperature rises significantly, with up to 2 K northeast of Iceland. The North Pacific SST increases with up to 0.5 K. In EXP2030, the amplitude of the anomalies grows but the pattern remains similar. Meanwhile, the cooling south of Greenland gets even more pronounced. The changes of SST strongly affect the turbulent surface heat fluxes between the ocean and the atmosphere (sum of latent and sensible heat fluxes, Q_TLA, positive means heat flux into the ocean, Fig. 3). For instance, over the cooling area in the North Atlantic, significantly positive Q_TLA anomalies occur, which means that less heat is released to the atmosphere. East of this area, negative Q_TLA anomalies occur, reflecting increased SST. In addition to the effects of changes in SST, changes in the atmospheric temperature might contribute to modified vertical temperature gradients and changes in Q_TLA.

Open image in new window

Open image in new window

Many future climate projections showed the smallest warming south of Greenland (Stocker et al. 2013) but only very few models simulated a significant cooling as found in our study. Also transient future climate projections with EC-Earth2.3 showed reduced warming but no cooling in this area (Koenigk et al. 2013). Observational based data, however, show a significantly negative temperature trend in this area south of Greenland (Rahmstorf et al. 2015), widely debated as the “Atlantic cold blob”. Rahmstorf et al. (2015) related this cooling to a reduction in the AMOC, especially after 1970. The strong cooling in our model might partly be due to the constant forcing as opposed to the transient forcing in CMIP5. The AMOC (Fig. 1b) and the associated oceanic northward heat transports vary on multi-decadal time scales. Thus, the transient year 2030 climate is affected by ocean water masses that have been formed decades before, in a climate with still high AMOC activity. In our experiments, the use of a constant forcing might increase the cooling effect from changes in the northward oceanic heat fluxes compared to the increased greenhouse gas forcing in transient climate simulations. All three experiments show large variations and a tendency to a weakening of the AMOC in the first two decades, particularly in EXP2030. Thereafter, the AMOC increases again but its average stays about 3 Sv smaller in EXP2030 than in EXP2000; also in EXP2015 a significant reduction occurs (compare Table 1). This reduction in our quasi-equilibrium simulations is almost three times as large as the AMOC reduction between year 2030 and year 2000 in transient CMIP5 projections with EC-Earth, and as large as the change until the second half of the twenty-first century in the CMIP5 simulations (Brodeau and Koenigk 2015). Towards the end of our 100-year simulations, the differences between the three simulations decrease. While part of this reduction is due to relatively low AMOC values in the last 20 years of EXP2000, we cannot rule out that a partial recovery of the AMOC in EXP2015 and EXP2030 contributes as well. Results by Blackport and Kushner (2016) indicated a recovery of an initial AMOC-reduction to Arctic sea ice loss after a few 100 years. Our experiment setup differs substantially from the experiments done by Blackport and Kushner (2016) but our time series are too short to see if a similar AMOC-recovery would take place. A partial recovery of the AMOC would likely lead to a less pronounced cold temperature blob.

The sea surface salinity (SSS) is strongly reduced in the same area of the North Atlantic where the ocean surface is getting cooler (Fig. 2d–f). Again, this is likely the consequence of the reduced transport of warm and salty water masses into this region. Also in the Central Arctic, the surface gets significantly fresher, likely due to increased freshwater input from rivers and increased precipitation (Koenigk et al. 2007, 2013). The ocean circulation stores most of the additional freshwater in the Beaufort Gyre or transports it in the Transpolar Drift Stream towards Fram Strait. An increased freshwater storage in the Beaufort Gyre is in agreement with observations (Giles et al. 2012). The increase of the salinity along some coastlines is likely caused by enhanced mixing due to longer periods with open water. Particularly along the North American coast, the mixed layer depth is increased (not shown). Enhanced sea ice melt in the Arctic might locally affect the surface salinity as well.

A somewhat surprising increase of salinity occurs in the Labrador Sea. This is surprising since the CMIP5 future simulations with EC-Earth (Brodeau and Koenigk 2015) indicated strongly reduced deep water convection in the Labrador Sea. This prevents that the relatively fresh surface layer is mixed with the underlying saltier layers, which would lead to a further reduction of SSS and convective activity. Both EXP2015 and EXP2030 show an increase of up to 300 m of the mixed layer depth (MXLD) in the area in the Labrador Sea where SSS increases. However, Fig. 2g–i show that this increase of MXLD occurs east of the main convection area in the Labrador Sea. In the main convection area, MXLD is reduced as expected.

The increase of MXLD by 300 m is likely not sufficient to substantially contribute to the North Atlantic Deep Water formation. The depth of the maximum AMOC is around 900 m in EC-Earth (not shown) and thus convection depth exceeding 900 m would be needed to feed the deep water path to the south. To further investigate this hypothesis, we calculate the “deep mixed volume” (DMV) for the Labrador Sea. The DMV is an index for the deep convection. It integrates the mixed water masses below a specific depth over a specific region (Brodeau and Koenigk 2015). Here, we use the same depth criterion of 1000 m as done by Brodeau and Koenigk (2015) but we have moved the box for the convective region of the Labrador Sea region used in Brodeau and Koenigk (2015) slightly to the northwest to capture the more realistic placement of the deep convection in EC-Earth3.0.1 compared to EC-Earth2.3. Figure 4a shows less frequent deep convective events in the Labrador Sea in EXP2015 compared to EXP2000 but the amplitude of the events is still similar. In EXP2030, however, both frequency and amplitude are substantially reduced. EXP2030 shows almost no deep convective activity between year 20 and 50 but a resurgence occurs from year 65 to 85. The latter period falls together with the warming period in EXP2030. Brodeau and Koenigk (2015) found a highly significant correlation between decadal variations of the DMV in the Labrador Sea and the AMOC a few years thereafter. Also in our simulations, the DMV is clearly related to the AMOC. Indeed, almost every peak in the AMOC can be explained by deep convection in the Labrador Sea (compare Figs. 1b and 4a, c, e). The correlation between 11-year running mean values of the DMV and the AMOC reaches 0.68, 0.80 and 0.73 in EXP2000, EXP2015 and EXP2030, respectively (Table 2). The correlation is largest when the DMV leads the AMOC by 4 years and the significance exceeds the 95 % significance level in all three simulations (using a student t test and taking the smoothing of the time series into account). A detailed description of the linkages between the DMV and the AMOC and the causes for the DMV variability is given in Brodeau and Koenigk (2015).

Open image in new window

Table 2

Correlations calculated from 11-year running mean values

Correlations, 11-year running means	EXP2000	EXP2015	EXP2030
DMV Labrador Sea → AMOC DMV leads 4 years	0.68 (0.64)	0.80 (0.82)	0.73 (0.75)
DMV Labrador Sea → T2m Arctic DMV leads 4 years	0.72 (0.70)	0.85 (0.87)	0.85(0.72)
AMOC → BSO lag 0	0.76 (0.71)	0.81 (0.81)	0.81(0.72)
AMOC → Sep Arctic ice extent lag 0	−0.78(−0.74)	−0.52 (−0.55)	−0.67 (−0.73)
AMOC → T2m Arctic lag 0	0.77 (0.73)	0.57 (0.58)	0.67 (0.78)
Heat transport BSO → T2m Arctic lag 0	0.78(0.74)	0.85 (0.85)	0.92 (0.79)
Heat transport BSO → Sep Arctic ice extent lag 0	−0.86 (−0.84)	−0.81 (−0.81)	−0.87(−0.70)
Heat transport Bering Strait → Sep Arctic ice extent, lag 0	−0.72 (−0.68)	−0.47 (−0.50)	−0.92 (−0.91)

Cursive and bold numbers denote a correlation exceeding the 95 and 99 % significance level, respectively. In brackets, the correlations of detrended time-series are shown

Similar to the Labrador Sea, the MXLD in the Greenland Sea and in the Irminger Sea are strongly reduced. The DMV for the Greenland-Iceland-Norwegian Seas (GIN-Sea, Fig. 4b, d, f) shows a strong reduction in amplitude, particularly in EXP2030, compared to EXP2000 and might contribute to a reduced AMOC as well. However, no clear relation between the variations of the DMV in the GIN-Sea and the AMOC could be found (Figs. 1b, 4) and correlations are not significant. This might be explained by the less direct effect of the GIN-Sea bottom waters on the AMOC compared to the deep water that is formed in the Labrador Sea. During the complicated travel across the overflows towards the North Atlantic, the GIN-Sea bottom waters are more mixed with other water masses, and contribute thus less distinctly to the AMOC than the deep water that is formed in the Labrador Sea.

The ocean volume transports into the Arctic show pronounced decadal variations (Fig. 5, left). Both the inflow into the Arctic through the Barents Sea Opening (BSO) and the outflow through the Fram Strait are slightly increased in EXP2015 and EXP2030. However, the increase is small compared to the amplitude of the decadal variations. The heat transports through BSO and Fram Strait (Fig. 5, right) are substantially enhanced, particularly in EXP2030. The increase in the heat transport through BSO is partly due to the larger volume flux but is predominantly the consequence of warmer temperatures, which also explains the amplified increase in EXP2030. In contrast, the variations of the heat transport through BSO are mainly governed by variations in the volume transport. After year 60, the heat transport through BSO in EXP2030 increases and stays at a higher level with reduced decadal variations. The volume transport through BSO is also high in this period and its variability is small. In EXP2030, this increase in ocean heat transport into the Arctic coincides with the warming of the Arctic after year 60 (Fig. 1). In this period, the increased convective activity in the Labrador Sea and the resulting increase in the AMOC might contribute to the warmer temperatures and enhanced ocean transports through BSO as well. We find significant correlations between 11-year running means of the DMV in the Labrador Sea and the BSO heat transport and T2m in the Arctic (Table 2). Correlations between 11-year running means of the AMOC and the BSO heat transport also reach about 0.8 in all three simulations (Table 2) and the heat transport through BSO is highly correlated with the mean Arctic T2m. This underlines the importance of the ocean heat transport into the Arctic for both Arctic climate variation and change. However, beside the increasing heat transport through BSO after year 60, the heat transport through the Bering Strait is also very high between year 70 and 90 in EXP2030 and might contribute to the Arctic warming after year 60.

Open image in new window

Table 2 shows a significantly positive correlation between the AMOC and mean Arctic temperature. A possible explanation is that the AMOC affects the heat transport into the Arctic and thus consequently the Arctic sea ice and air temperature. As such, the reduction of the AMOC should have a dampening effect on the Arctic temperature in EXP2015 and EXP2030. By regressing the 11-year running mean values of the AMOC on the Arctic temperature in our three simulations, and taking the standard deviation of the AMOC into account, the reduction of the AMOC between EXP2030 and EXP2000 would lead to a reduction of the Arctic temperature increase by 0.6–1.1 K. However, this assumes that the processes that link the variability of the AMOC with Arctic T2m are the same as the processes that relate the mean changes in the AMOC with the changes of Arctic T2m.

EC-Earth tends to underestimate the observed seasonal cycle of the sea ice extent (Fig. 6). The ice extent is lower than in NSIDC-data (average over 1980–2013; Cavalieri et al. 1996) in March but, except for EXP2030, higher in September. The ice extent difference between EXP2015 and EXP2000 is substantially smaller than the difference between EXP2030 and EXP2015. The simulated mean reduction between September ice extent in EXP2030 and EXP2000 reaches about 1.3 million km² in our model simulations and thus accounts for less than half of the observed reduction since 1980.

Open image in new window

The simulated ice volume loss is quite large (from 15 to 9.9 million km³) but also smaller than estimates for the real ice volume reduction. However, sea ice extent and volume in the Arctic (Fig. 6) show pronounced decadal-scale variations in our simulations. These variations could mask or enhance human-induced trends at interannual to decadal scales. This finding is in agreement with a recent study by Swart et al. (2015) analyzing the internal sea ice variability in CMIP5 models.

Not only the ice reduction but also the simulated T2m increase in the Arctic in EXP2030 (1.78 K) is smaller than in ERA-interim since 1980 (about 2 K). Still, it seems that Arctic sea ice in our model might respond less sensitively to global warming than in reality.

Sea ice extent in EXP2030 is reduced after year 60, which is consistent with the warm Arctic temperature after year 60. As discussed in Sect. 3.1, the ocean heat transports into the Arctic through both the Barents Sea Opening and Bering Strait are high after year 60. Increased ocean heat fluxes through these sections are likely to reduce the Arctic sea ice extent as shown by previous observational (Schlichtholz 2011, Woodgate et al. 2010) and modelling studies (Koenigk and Brodeau 2014). We find a strongly negative correlation, exceeding −0.8 in all three experiments (Table 2), between the low-pass filtered (11-year running means) heat transport through BSO and the Arctic sea ice extent in September. The relation between the heat transport through the Bering Strait and sea ice extent seems to be more variable, the correlations vary substantially between the three simulations (Table 2).

Similar to the ice extent, the amplitude of the annual cycle of ice volume seems to be slightly underestimated in our EC-Earth simulations. However, note that ice thickness observations are still extremely uncertain and that the reference data from the Pan-Arctic Ice Ocean Modeling and Assimilation System (PIOMAS, Zhang and Rothrock 2003) used here should not be considered as the truth. In contrast to sea ice extent and air temperature, sea ice volume (Fig. 6c, d) is already considerably reduced between EXP2015 and EXP2000, and no clear acceleration until 2030 can be seen. This might indicate that the sea ice first has to be thinned down to a critical thickness before the ice extent is substantially reduced. This is partly in contrast to results from Holland et al. (2006), who linked the rate of summer ice retreat mainly to the interplay of simulated natural variability and forced changes. In agreement to our study, Holland et al. (2006) found an important influence of the ocean heat transport into the Arctic on sea ice variability and decline.

The ice export out of the Arctic constitutes an important linkage between the Arctic and lower latitudes by its influence on the deep water formation in the North Atlantic, particularly in the Labrador Sea, and consequently on the global ocean circulation (Jungclaus et al. 2005; Holland et al. 2001). Observationally based estimates suggested exports of about 0.7–1 Sv (Vinje et al. 1998, Kwok and Rothrock 1999, Schmith and Hansen 2003), which fits well to the simulated values of EC-Earth: in EXP2000, the ice export through Fram Strait reaches about 0.8 Sv in average, 0.7 Sv in EXP2015 and drops below 0.5 Sv in the last 30 years of EXP2030 (Fig. 6e, f). This reduced ice export in EXP2030 goes along with a thinning of the ice (not shown), which overcomes a potential increase in velocity, and is in line with future projections (Koenigk et al. 2007, 2013; Vavrus et al. 2012). Together with the drop in the mean ice export, the interannual to decadal variations are strongly decreased in the last 30 years of EXP2030. Figure 7 shows the spatial distribution of sea ice concentration in EXP2000 and the changes in EXP2015 and EXP2030. EXP2000 simulates too much ice in the Greenland Sea, and slightly too little ice along the rest of the ice edges in March compared to the OSISAF satellite data product (Eastwood et al. 2010). Generally, the ice concentration is too high along the ice edges in September. In EXP2015, in March, the ice concentration is mainly decreased in the Barents Sea, north of Iceland and in the Sea of Okhotsk. A slight increase can be seen in the Greenland Sea south of Svalbard in both EXP2015 and EXP2030. This increase might be related to northerly wind anomalies in this area (Fig. 8a, c). The changes in sea ice concentration strongly affect Q_TLA with a dipole-like pattern (Fig. 3); the strongest turbulent heat loss to the atmosphere moves northward together with the ice edge in EXP2015 and EXP2030.

Open image in new window

Open image in new window

In September, sea ice decreases mainly in the Barents Sea and in the Beaufort Sea along the North America coast in EXP2015. In EXP2030, we see a more general retreat of sea ice along all ice edges with largest reductions in the Barents and Kara Seas. Compared to the sea ice change in transient CMIP5 climate projections with EC-Earth 2.3 (Koenigk et al. 2013), the response is somewhat less focused on the Barents Sea region.

A number of studies suggested a possible link between the observed sea ice reduction and the large scale atmospheric circulation and mid-latitude air temperature (Jaiser et al. 2013; Inoue et al. 2012; Petoukhov and Semenov 2010; Hopsch et al. 2012; Overland et al. 2011; Peings and Magnusdottir 2014; Rinke et al. 2013; Koenigk et al. 2016). Most of these studies used either the difference between the last decade (with little ice) and the previous two (with much ice), or they used detrended time series to assess the atmospheric response to sea ice variability and trend. One common problem for all these studies is the fact that observational time series are very short. Moreover, they are derived from a climate in transition, which means that both external and internal forcing have likely changed during the three decades of observations.

Here, we investigate the relationship between sea ice anomalies in autumn and SLP and T2m in the following winter, in our three quasi equilibrium simulations, using correlation analysis. This analysis is performed based on variations of sea ice area in eight different Arctic regions as defined in Koenigk et al. (2016) (Northern Hemisphere, Barents-Kara Seas, Greenland Sea, Labrador Sea-Baffin Bay, Laptev-East Siberian Seas, Chukchi-Bering Seas, Beaufort Sea, Central Arctic). The general result is that the correlation between autumn sea ice and winter SLP is weak in all three simulations, and that air temperature shows the strongest response in the location of the ice anomaly and its surroundings. In the following, we will therefore only shortly discuss the response to November ice anomalies in the Barents and Kara Seas area (BAKA, Fig. 12) since previous studies found that the atmospheric response is largest to November sea ice anomalies in the BAKA region. The correlation between November ice area in the BAKA region and SLP in the following winter is below 0.3 everywhere in EXP2000. A small local response in the Kara Sea region can be seen, which is significant at the 95 % level. The T2m response in EXP2000 shows somewhat higher correlations: a significantly negative correlation from the Nordic Seas across the BAKA area (here correlation is maximal and exceeds −0.6) and Siberia to the North Pacific. This means that less ice in the BAKA area leads to a warming. The explanation for this local response is relatively simple: little sea ice in November in the BAKA area leads to reduced sea ice extent in winter—due to the persistence of the ice anomaly—and consequently to warmer temperatures in the Barents Sea. Significantly negative T2m anomalies occur also over southern North America, and positive anomalies over the western North Pacific. These anomalies seem to be related to SLP anomalies, which are not significant but have the potential to advect cold and warm air masses in the regions of the T2m anomalies. Very small positive correlations can also be seen over Central Asia.

Open image in new window

In EXP2015, the correlation between November ice and winter SLP is still weak, although some slightly larger areas with significant correlations are found over eastern Siberia and central Europe. The strongest negative correlations between ice and T2m occur in the Barents Sea and its surroundings. Furthermore, we see more wide-spread negative correlations extending from Florida across the North Atlantic towards southern Europe, and negative correlations over the northeastern North Pacific. In EXP2030, the SLP response is similar as in EXP2015 over eastern Asia, but in addition, a significantly negative correlation occurs over the subtropical North Atlantic and over the western North Pacific. Significantly negative temperature correlations extend now all the way from the Caribbean across the North Atlantic, following the North Atlantic Current into the Arctic and further into the North Pacific and eastern Asia. Over a small area of the North Atlantic that spreads from eastern Canada to the south of Greenland, a positive correlation is found.

We performed the same correlation analysis with detrended data (not shown). The results are generally similar to the results from the analysis using the raw data. However, significantly positive correlations between sea ice and T2m occur over the North Atlantic subpolar gyre (up to r = 0.32) and weaker negative correlations along the North Atlantic Current compared to the correlations of the raw data.

None of our three simulations reproduces the suggested relation between reduced sea ice in the Barents Sea area and a trend towards more negative NAO winter atmospheric conditions. However, if we, instead of using the entire 80-year period, perform the same correlation analysis for different 30-year periods, we find large variations in the relation between sea ice and SLP (Fig. 13). Note, that correlations exceeding ∓ 0.38 are significant at the 95 % significance level, based on a two-sided t test and assuming 30 degrees of freedom and a normal distribution. For EXP2000 and EXP2030, 30-year periods exist with a NAO-like response pattern to reduced sea ice area in the BAKA area in November. The correlation coefficients reach similar amplitudes as the observed correlations (compare Koenigk et al. 2016, Fig. 4). Thus, EC-Earth is able to simulate a SLP response, which is similar to the observed response. Interestingly, neighboring 30-year periods show a NAO + like response of the winter SLP to ice reductions in the BAKA area. This shows that the response in our model is not robust over time and it indicates that 30-year periods are very short to make robust statements about a possible relationship between sea ice variations and mid-latitude climate variations.

Open image in new window