Variability of sea ice in the Barents Sea
In this section, sea ice variability in the Barents Sea and the formation of particularly strong and weak sea ice conditions are analyzed by using the control integration of the model. The Barents Sea is a transition zone with no ice at all in the southwest and permanent ice cover in the northernmost part (Fig. 2a). The first empirical orthogonal function (EOF) of annual mean sea ice concentration in the control integration explains 26.4% of the variance of northern hemispheric sea ice concentration (Fig. 2b). The first EOF is dominated by a dipole with one center over the Barents, Kara and Greenland Seas and the other over the Labrador Sea. The pattern is similar to EOF 1 of 40-year winter data from the National Snow and Ice Data Center (Deser et al. 2000). Compared to our study, Deser et al. (2000) found a slightly higher explained variance (35%) and largest variability in the Greenland Sea while our simulations show largest variations in the Barents Sea. This might be due to the fact that Deser et al. (2000) analyzed winter means while we use annual means. The time series of our EOF 1 is highly correlated with both sea ice extent (r = 0.94) and sea ice volume in the Barents Sea (IVB, r = 0.96). The correlation with the total northern hemispheric sea ice extent is 0.81 indicating the importance of sea ice variations in the Barents Sea. This agrees with results by Bengtsson et al. (2004) who performed sensitivity simulations with both AGCM and AOGCM to analyze the impact of sea ice change on Arctic surface air temperature (SAT, mean of 70–90°N). They found that reduced sea ice concentration particularly in the Barents Sea is the main reason for increased Arctic temperature. Also, Goosse and Holland (2005) showed that Arctic SAT variability is highly related to sea ice concentration and surface air temperature (SAT) in the Barents and Kara Seas by analyzing a 650-year simulation with an AOGCM.
Our simulations of the annual mean sea ice volume of the Barents Sea (Fig. 3) show a high variability at interannual to multi-decadal timescales with peaks exceeding 95% significance at about 2.5, 5.5, 9 and 20 years (Fig. 4). Annual mean values of IVB in our control integration vary between 0.5 × 1011 and 7 × 1011 m3. Many other studies analyzing Arctic climate and sea ice variability found peaks at roughly 10 years. Mysak and Venegas (1998) suggested a 10-year climate cycle in the Arctic, which should be characterized by a clockwise propagation of sea ice anomalies through the Arctic and a coexisting standing oscillation in sea level pressure (SLP) anomalies. A similar mechanism is responsible for decadal variations in the Fram Strait sea ice export (Koenigk et al. 2006, 2008). Polyakov and Johnson (2000) analyzed NCEP/NCAR reanalysis data and related decadal sea ice variations to decadal variations in the AO. A similar variation has been found by Hilmer and Lemke (2000). However, our model control integration does not show a significant decadal variability of the AO.
Other studies found significant variability on longer time scales. Goosse et al. (2002) used a global coupled climate model and found a dominant peak at 15–20 years in Arctic sea ice volume. Divine and Dick (2006) found a 20–30-year-oscillation in historical data of Greenland and Barents Seas sea ice edge. These authors (2006) also showed a strong seasonal cycle of the ice edge with only little ice left in the Barents Sea in August. Our simulations show similar results. IVB is at maximum in late spring with up to 1 × 1012 m3 and at minimum in late summer/early autumn where sea ice disappears in several years of the control integration. However, standard deviations of seasonal IVB are large in all seasons and vary between 0.98 × 1011 m3 in autumn and 1.57 × 1011 m3 in spring. The correlation between IVB in one season and the following three seasons reaches about 0.75, 0.5 and 0.3 independent of the starting season. Hence, both summer and winter IVB depend partly on the amount of IVB of the preceding seasons. The autocorrelation of annual mean IVB is 0.51 and 0.22 for a lag of 1 and 2 years, respectively.
The annual rate of change of IVB is determined by transport of sea ice across the boundaries of the Barents Sea and melting and freezing of sea ice within the Barents Sea (Table 1) within one year. The mean ice transports in the Barents Sea and vicinity in our control integration are shown in Fig. 5. The sea ice transport across the boundaries of the Barents Sea with Kara Sea and Central Arctic is mainly directed towards Barents Sea (in the following we call this ice import although it can get negative). The mean annual sea ice import into the Barents Sea amounts to 0.87 × 1012 m3/year in the control integration. Kwok et al. (2005) used satellite measurements to estimate the winter ice volume transport across the line Svalbard–Franz Josef Land between 1994–2003. They found a mean ice volume transport of 0.04 × 1012 m3/winter. However, the difference to our study is probably not as large as the numbers suggest because Kwok et al. (2005) did not consider the entire year and analyzed only about half of the border to Barents Sea. In our model, about 60% of the ice import into the Barents Sea takes place between Svalbard and Franz Josef Land. The rest is mainly imported between Franz Josef Land and Novaya Zemlya and a small part through the Kara Strait into the Barents Sea. Additionally, Kwok et al. used a rather short time period, which was dominated by strongly positive NAO. Our simulations show a reduction of the ice import into Barents Sea by about 30–50% in high NAO-years since anomalously southerly winds occur between Svalbard and Franz Josef Land (Fig. 6). The variability of annual export in our simulation is very high with a standard deviation of 0.33 × 1012 m3/year. Maximum and minimum ice imports are 2.06 × 1012 and −0.25 × 1012 m3/year. Nevertheless, our model overestimates the ice import into the Barents Sea. The ice transport across the southern and western boundaries is normally directed towards Greenland and Norwegian Seas and takes mainly place south of Svalbard (we call this ice export). Hence, the convergence of sea ice transport in the Barents Sea (import minus export) is governed by the import (correlation between import and convergence is 0.90).
Table 1 Means, maximum and minimum values, and standard deviations of rate of sea ice volume change (rate of ice vol. change), ice transport over northern and eastern border of Barents Sea (ice import), ice transport over southern and western border (ice export) and freezing minus melting in the control integration in 1012 m3/year
The annual rate of change of IVB depends strongly on the convergence of sea ice transport into the Barents Sea (r = 0.70) and therefore on the import (r = 0.57). Thermodynamics play only a minor role for interannual variability of IVB. The correlation of the thermodynamic sea ice volume change (calculated as residuum from the convergence of sea ice transport in the Barents Sea and rate of IVB change within 1 year) with rate of change of IVB is −0.31 and with the import −0.84. Hence, large sea ice imports into the Barents Sea are related to strong melting of sea ice in the Barents Sea. The reason is that more sea ice reaches areas where ocean temperatures are normally above the freezing level and hence a larger amount of sea ice can be melted. This also explains why the year-to-year rate of change of IVB is rather small compared to changes in the ice import and thermodynamic ice volume changes. Correlations between annual mean IVB itself and ice transports are similar to the values above-mentioned.
The ice transport into the Barents Sea is mainly dominated by the local wind stress. It is highly correlated with the SLP-gradient between northern Svalbard and northern Novaya Zemlya (r = 0.79). Anomalously high pressure over Svalbard and below normal pressure over Novaya Zemlya lead to anomalously northerly winds, which transport much ice into the Barents Sea. A cross-correlation analysis between the SLP-gradient between Svalbard and Novaya Zemlya and ice transport into the Barents Sea indicates that the SLP-gradient governs the ice transport at high frequencies below 5 years (not shown).
A lead-lag correlation analysis between annual mean oceanic heat transport into the Barents Sea (OHT) and IVB shows that the highest correlation occurs at lag 0 with 0.56. Performing the same correlation analyses but for high pass filtered data from 1 to 10 and 1 to 5 years provides correlation coefficients of 0.35 and 0.09, respectively. Furthermore, the correlation of OHT with the rate of change of IVB is only 0.16. This and the fact that thermodynamic sea ice volume change is negative in years with a positive total ice volume change indicate that OHT into the Barents Sea plays only a minor role for interannual sea ice variability in the Barents Sea. However, the importance of OHT increases with increasing time scales. This becomes evident in a cross-spectrum analysis between normalized values of IVB and the OHT (Fig. 4). The power spectrums agree nicely for time scales exceeding 5 years. A high coherence between IVB and OHT can be seen at long time scales and the phase difference is continuously near 0. In contrast, at short time scales no substantial relation occurs between the two time series. A lag-lead correlation analysis between OHT and IVB using a low pass filter eliminating all periods below 10 years shows similar results. The highest correlation coefficient occurs at lag 0 with 0.74 (0.72 if heat flux leads 1 year, 0.70 if IVB leads one year). In contrast to the atmosphere with its strong interannual variability, OHT shows pronounced decadal and longer scale variations and hence governs the IVB at long time scales.
We calculated a lag regression analysis between annual mean heat content of the upper 200 m and IVB (not shown). During heavy ice conditions in the Barents Sea, the heat content is reduced in the Barents and in the entire Nordic Seas. The heat content of the upper 200 m spatially integrated over the Barents Sea and IVB are correlated with 0.33. Eleven-year running means of IVB and heat content are correlated with 0.58. Similar to the heat transport into the Barents Sea, also the heat content shows strongest variability at longer time scales than those considered in this study. However, the correlation between Barents Sea ocean heat content of the upper 200 m and OHT is only moderate with 0.37 for annual means and 0.46 for 11-year running means. Decadal variations of the atmosphere–ocean heat flux will be analyzed in detail in an upcoming study.
Figure 7 shows a lag regression analysis between annual mean IVB and SLP. Four years before IVB is at maximum, a dipole with positive center near Iceland and negative anomalies between Newfoundland and Spain is formed. This pattern intensifies in the following years and is similar to the negative NAO pattern. However, the pattern is slightly shifted to the north and the center of the positive pole is situated over Svalbard, which leads to a strong SLP-gradient between Svalbard and Severnaya Zemlya. Hence, ice transport into the Barents Sea is anomalously large in these years and a positive anomaly of IVB is formed. It has to be noted that the autocorrelation of this regression pattern is very low. The pattern should therefore not be understood as a standing pattern, which exists for several years. Analyses of the 15 largest IVB events show that such a pattern normally occurs in one or two of the years before or during the IVB event and transports a huge amount of sea ice into the Barents Sea, where the sea ice volume stays above normal for the next 2 or 3 years. Goosse et al. (2003) performed a very long integration with a coarse resolution OAGCM and analyzed one extreme Arctic ice volume event. This event was mainly characterized by sea ice covering the entire Barents Sea and extending as far south as the Lofot Islands. The atmospheric pattern leading to this ice event is dominated by a positive pressure anomaly over the Norwegian Sea and a negative anomaly over Greenland. This pattern strongly reduces Fram Strait sea ice export and keeps the ice in the Arctic. Additionally, ice transports into the Barents Sea are increased. Although this pressure pattern differs from our regression pattern between SLP and IVB it also includes an enhanced pressure gradient across Barents Sea.
The correlation between annual mean NAO-index and annual mean IVB in our control integration is −0.29 for lag 0 and −0.25 when NAO leads by 1 and 2 years. The correlation for winter values is slightly smaller. The first principal component (PC) of EOF 1 of annual mean sea ice concentration and NAO-index is correlated with −0.33. Vinje (2001) found that the correlation between winter NAO-index and April sea ice extent in the Barents Sea is time dependent. In the periods 1900–1935 and 1966–1996, the correlation was between −0.5 and −0.6 but much weaker during 1864–1900 (r = −0.3) and 1935–1966 (r = −0.36). Holland (2003) used a global coupled model and found a correlation of 0.59 between NAO-index and EOF 1 of sea ice concentration. Sorteberg and Kvingedal (2006) stated that the wintertime link between NAO and sea ice extent in the Barents Sea is only moderate and not the dominating factor. They showed that cyclones moving from the Nordic Seas to the Arctic and Siberian coast dominate sea ice conditions in the Barents Sea. Obviously, the impact of the NAO on the Barents Sea depends strongly on the exact position of the Iceland Low extension into the Arctic because this extension determines the wind forcing across the boundaries of the Barents Sea.
In correspondence to the atmospheric circulation, the regression analysis between annual mean sea ice thickness in the Arctic and IVB (Fig. 8) shows a small positive ice thickness anomaly in the Laptev Sea and below normal ice thicknesses at the Canadian coast four years before the maximum. In the following years, the wind anomalies lead to increased ice transports from the North American coast across the Arctic towards Barents Sea, Kara Sea and the European Arctic. Sea ice volume in the Barents Sea is accumulated over two to three years until it reaches its maximum. At the same time, sea ice thickness is strongly decreased at the North American coast. Obviously, a several-year-long redistribution of ice in the Arctic due to anomalous atmospheric circulation leads to the formation of the IVB maximum. Again, it has to be noted that this should not be understood as a several year-long continuously anomalous ice transport into the Barents Sea but as increased ice transport in parts of the period before and during the maximum IVB. Sea ice thickness stays above normal for two to three years in the Barents and Kara Sea (Fig. 8e, f) after the maximum while the negative anomaly at the North American coast disappears rather fast.
Interannual climate response to sea ice anomalies in the Barents Sea
To analyze the response of climate conditions to large and small IVB, composite analyses are calculated. All years from the control integration are considered with IVB-anomalies exceeding ±1 standard deviation. It has to be noted that it is difficult to discriminate between the climate response to maximum IVB and the climate conditions causing maximum IVB. This difficulty occurs particularly at zero lag. Thereafter, a large part of the anomalies can be associated with the ice anomaly in the Barents Sea.
In the year of maximum IVB, albedo is strongly enhanced in the Barents and western Kara Sea. The center of the anomaly is situated between northwestern Novaya Zemlya and Franz Josef Land), where albedo is increased by up to 20% due to enhanced sea ice cover. In the following 2 years, albedo increases by up to 15% and up to 10%, respectively (Fig. 9a, d, h). Hence, absorption of short wave radiation is substantially reduced in the Barents Sea. Laine (2004) analyzed the summer albedo in the Arctic from radiometer data of the period 1982–1998 and showed considerable annual variability in the Barents/Kara Seas. Interestingly, he found a rather small correlation (r = 0.3) between summer albedo and sea ice concentration in this area but much higher correlations in most other Arctic areas. In contrast, the correlations between sea ice concentration and albedo, both for summer and annual means averaged over the Barents Sea, exceed 0.9 in our control integration. Although Laine used rather short time series and observations of summer sea ice concentration are difficult due to melt-ponds, it seems that our model overestimates the relationship between ice concentration and albedo due to the imposed dependence of surface albedo on sea ice concentration and surface temperature.
The enhanced ice cover and thickness during large IVB strongly reduces the heat flux from ocean to atmosphere (sensible + latent heat flux, not shown). Annual anomalies are largest with up to −30 W/m2 between northwestern Novaya Zemlya and Franz Josef Land. During winter, when heat fluxes are particularly large due to large temperature differences between the cold atmosphere and the comparatively warm ocean surface, the average heat flux anomalies reach −100 W/m2. The sum of sensible and latent heat flux accounts for about two-third of the total net heat flux anomalies in the Barents Sea while roughly one-third is due to short and long wave radiation flux anomalies.
The strongly reduced ocean heat release in the Barents Sea leads to a local high pressure anomaly (Fig. 7e) one year after the maximum IVB. However, the impact on the large-scale atmospheric circulation is limited. SLP only responds significantly in the Barents Sea.
Near surface air temperature anomalies are particularly large over the Barents Sea. The temperature is reduced by more than 2 K in years with maximum IVB and is 2 and 1.5 K colder after one and two years, respectively (Fig. 9b, e, i). In contrast to SLP, the 2 m air temperature anomaly spreads over large areas of Siberia, northeastern and eastern Europe. Although the response of the atmospheric circulation is rather localized, this anomaly and the mean atmospheric circulation advect the cold air masses over large regions. Wu et al. (2004) analyzed SAT during winters with heavy and light ice conditions in the Greenland–Barents Seas from NCAR/NCEP reanalysis data (Kalnay et al. 1995). They found a large-scale SAT anomaly pattern with sharp local-scale anomalies along the ice edge. They argued that the atmospheric circulation, which is dominant during several heavy and light winters, respectively, is responsible for the large-scale anomalies while the local-scale anomalies are a feedback to the displaced sea ice edge.
Over the northwestern North Atlantic, the simulated temperature is below normal as well. This is connected with a southward displacement of the North Atlantic Current (NAC) and a weakening of the NAC due to reduced westerlies over the North Atlantic during the formation process of the IVB maximum (Fig. 7). Eden and Willebrand (2001) showed that a negative NAO anomaly is associated with a reduction of the strength of the subpolar and subtropical gyre, which goes along with a reduced NAC. The atmospheric forcing over the North Atlantic during the formation of high IVB in our simulations is similar to the negative NAO case analyzed by Eden and Jung (2001). Similar to our results, these authors found negative surface temperature anomalies in the NAC in periods with negative NAO conditions indicating a common cause for the cold anomaly over the NAC and heavy ice conditions in the Barents Sea.
The temperature response at 850 hPa is much weaker compared to the surface and the center is not as sharply localized over the Barents Sea (Fig. 9c, f, j). The large 2 m air temperature anomaly in the Barents Sea is due to exchange processes at the surface. The vertical atmospheric temperature gradient within the first 1,500 m is reduced by 1–2 K. Hence, stability of the boundary layer is increased (decreased) during heavy (light) winter ice conditions in the Barents Sea. This agrees with results of Wu et al. (2004) who found a more stable winter atmosphere over sea ice than over open water.
Precipitation in the Nordic Seas is significantly smaller in years with large sea ice volume in the Barents Sea (Fig. 9d, g, h). In the Barents Sea itself, precipitation is reduced by up to 20%. In the following 2 years, precipitation stays below normal in the Barents Sea. The reduced oceanic heat release and the more stable atmosphere lead to a decrease of convective precipitation events. Large-scale precipitation is also reduced since fewer cyclones propagate into Barents Sea during years with heavy ice conditions. Note that precipitation shows almost no response over land areas, which is in contrast to air temperature.
The effect of minimum IVB on climate (not shown) is nearly symmetric. The reduced sea ice cover and thickness lead to decreased albedo, above normal oceanic heat release, negative SLP anomaly, strongly increased temperatures and enhanced precipitation in the Barents Sea in years during and after the minimum. Both distribution and amplitude of the response are comparable to the large IVB case.
Anomalies of ocean surface temperature (SST, mean of 12 m thick uppermost grid cell in our model) and surface salinity (SSS) are subject to rather strong multi-decadal variations. Hence, we used a 11-year Hanning high-pass filter for the composite analysis of SST and SSS to extract the interannual response to the sea ice anomalies (Fig. 10). The SST pattern is characterized by large cold anomalies in the Barents Sea and in the North Atlantic and a smaller warm anomaly in the Labrador Sea during years with large IVB. The SST anomaly in the Barents Sea is due to cooling from above by enhanced sea ice and a cold atmosphere. A displacement and weakening of the North Atlantic Current (NAC) is responsible for the cooling in the North Atlantic. Anomalous advection of warm surface waters and easterly wind anomalies reduce sea ice cover in the Labrador Sea, which leads to a warming in the Labrador Sea along the ice edge. The positive anomaly in the Labrador Sea disappears rather fast while the negative anomalies stay significant for one to two years. The SST pattern is similar to that of 2 m air temperature (Fig. 9b, e, i) but the response is more localized and weakens faster.
Surface salinity is significantly reduced in the Barents Sea at lag 0. Since sea ice volume is above normal, melting is intensified and salinity reduced. This negative salinity anomaly weakens in the following year and disappears at a lag of 2 years. In the Kara Sea, where sea ice volume is also above normal, a significant negative salinity anomaly occurs after 2 years. Houssais et al. (2007) forced a coupled ocean–sea ice model with winds and air temperatures of the positive AO-phase, which normally leads to slightly enhanced sea ice concentration and thickness in the Barents Sea. They did not find any strong salinity response in the Barents Sea but showed a substantial positive salinity anomaly in the Kara Sea similar to the anomaly in our simulations. Sundby and Drinkwater (2007) analyzed salinity data since 1947 and found in contrast to Houssais et al. (2007) a clear relation between the 3-year mean NAO-index and salinity north of the Kola Peninsula. Furthermore, they showed that high salinity is related to a high propagation speed of the salinity anomaly and vice versa. Furevik (2001) analyzed hydrographic sections in the Barents Sea opening and showed that at least the temperature of the inflowing water is related to the phase of the NAO.
The weakening and displacement of the NAC reduces SSS in the NAC in our simulation. The weakening of the NAC leads to less transport of warm and salty water into this area and cold and fresh sub-polar waters penetrate further to the south. Along the Siberian coast, SSS anomalies are strong but statistically only marginally significant because salinity variations are very high at the Siberian coast due to a strong dependence of melting and freezing processes on the atmospheric circulation and variations of the river runoff.
Seasonal climate response to sea ice anomalies in the Barents Sea
In the following, the seasonal climate anomalies associated with anomalous IVB are discussed. Figures 11 and 12 show SLP and 2 m air temperature during large IVB (exceeding mean +1 standard deviation) in winter (DJF, a), spring (MAM, b), summer (JJA, c) and autumn (SON, d) and the following seasons. The season with maximum anomalous IVB (except for summer) is characterized by a SLP pattern similar to the lag 0—regression pattern of annual values. During winter, spring and autumn, pressure over the Nordic Seas is anomalously high with maximum over Svalbard and anomalously low further south over the North Atlantic. During summer, the pattern is dominated by anomalously low SLP in Kara and Laptev Sea. However, in all seasons, a rather strong (weakest in summer) SLP-gradient between Svalbard and Novaya Zemlya leads to strong sea ice transport into the Barents Sea. It is difficult to say what part of the SLP anomaly at lag 0 is a response to the ice anomaly in the Barents Sea. However, in all seasons—except for summer—the SLP anomalies are most pronounced around Spitsbergen and in the Barents Sea. This may lead to the conclusion that the atmosphere in this area reacts to the sea ice anomaly below at lag 0. Wu et al. (2004) used reanalysis data and performed composite analysis of SLP for light and heavy winter ice conditions in the region Greenland/Barents Sea. Similar to us, they found the strongest SLP anomaly in the area of large ice conditions and concluded that sea ice anomalies partly determine the local SLP anomalies. Independent of the season of maximum ice volume, the SLP anomalies in the following seasons are largest in winter and spring. Temperature differences between atmosphere and ocean and hence heat flux anomalies are particularly large in this time period. Main characteristics of the winter and spring pattern are positive SLP anomalies over the Barents Sea and over north and northeastern Europe. Furthermore, SLP tends to be above normal over the Arctic Ocean and below normal over the northwestern North Atlantic. A two-sided t test shows that at least parts of these anomalies are significant at the 95% level. The anomalous autumn SLP pattern is similar but with a smaller amplitude. The center of the positive response over the Barents Sea moves with the ice edge towards Kara Sea. In autumn after maximum IVB in spring, no significant values at all can be seen in mid and high northern latitudes. The smallest anomalies occur during summertime, when the differences between ocean and atmosphere temperature and hence heat fluxes are small. Changes in ice cover and thickness have therefore only a small impact on the atmospheric circulation in autumn. The anomalies in the second year after maximum IVB are generally small.
The SLP response after low IVB is almost symmetric to the response to large IVB. SLP is significantly reduced in the Barents Sea and surroundings (not shown) and again, the response is smallest in summer.
The 2 m air temperature shows a strong cooling of more than 2 K centered over the Barents Sea in all seasons with large positive IVB. This negative anomaly spreads out laterally and is significant over the entire northern and northeastern Europe and over western and middle Siberia. During summer, the anomaly has a slightly reduced extension. In the seasons after high IVB, air temperature stays much colder than normal in the Barents Sea mainly due to the strongly reduced ocean heat release. The anomaly reaches up to 4 K in winter, spring and autumn but stay below 2 K in summer. The cold temperatures extend to northern Russia, Kara Sea, Laptev Sea and the European Central Arctic in winter and spring. In autumn, the anomaly extends mainly eastward along the Siberian coast. The cold air is advected with the mean atmospheric circulation to the north and east. In summer, the cold region is more limited in its extension. After seasons with low IVB, the anomaly patterns of air temperature are almost symmetric and show much warmer temperatures than normal.
Bengtsson et al. (2004) performed simulations with an AGCM forced with the Global Sea Ice and Sea Surface Temperature Data Set (GISST) for the twentieth century. They used a discontinuity in the sea ice data set leading to a mean reduction of the sea ice area by 2 × 106 km2 after 1949 to analyze the atmospheric response. This leads to a decreased sea ice concentration by 5–40% in the Barents Sea. Although not only sea ice in the Barents Sea was decreased, the local response in the Barents Sea was similar to our results from the control integration. The reduced sea ice concentration in Bengtsson et al. led to an increased winter ocean heat release by up to 150 W/m2. The temperature increased by 6 and more Kelvin and SLP responded with a local reduction of about 1.5 hPa. The response is about 50% higher than in our composite analysis. However, this can mainly be explained by a stronger sea ice reduction in the Barents Sea in their study.
Rinke et al. (2006) analyzed the impact of lower-boundary forcing on the mean state of the atmosphere by forcing a regional atmosphere model with two different sets of sea ice and SST. They found that local air temperature is colder (warmer) and SLP higher (lower) where sea ice concentration is increased (reduced). The effect is most pronounced along the displaced ice edge and much weaker in the Arctic’s interior. Moreover, their results show a strong annual cycle of the air temperature differences with large changes in winter and small changes in summer. This agrees well with our findings for the Barents Sea. Interestingly, SLP anomalies do not show an annual cycle in their simulation, which is in contrast to our results. Keup-Thiel et al. (2006) analyzed the future climate development in the Barents Sea region with a regional atmosphere model. The anomaly patterns of summer and winter air temperature for the time slice 2011–2030 in their simulations compare well with the response to negative IVB in our control integration (not shown).
As noted above, it is difficult to separate between the effect of anomalous sea ice anomalies in the Barents Sea and the response to the forcing leading to this sea ice anomaly in a lag 0 regression or composite analyses of the control integration. Since NAO influences IVB, we subtract the NAO-signal from the lag 0 composite in the following. We determine the NAO-index during high and low (exceeding mean ± 1 standard deviation) IVB and calculate the SLP and 2 m air temperature pattern belonging to these NAO-indexes. The pattern gained by this simple procedure is subtracted from the lag 0 pattern during high and low IVB (Fig. 13). The by far largest SLP anomalies occur in the Barents Sea itself and the negative anomaly over the North Atlantic has almost disappeared. The air temperature pattern at lag 0 is now very similar to the lag 1 pattern and does not show the dipole over the Labrador Sea anymore. However, these patterns still include the non-NAO forcing, which is responsible for the formation of sea ice volume anomalies in the Barents Sea. Therefore, additional sensitivity studies are necessary to isolate the response of atmospheric climate conditions to sea ice variations in the Barents Sea.