Future changes in atmosphere, ocean and sea ice are closely linked, and it is often difficult to distinguish between causes and effects. In this section, we will—along with some evaluation of the simulated twentieth century Arctic climate—focus on the mean changes of Arctic key variables in the different EC-Earth future projections. Although we will briefly discuss the relation between key variables and indicate important processes that are involved in Arctic climate change we will leave detailed process analyses for future studies.
Arctic sea ice
Observations of the northern hemispheric ice extent during the last 30 years show a strong and accelerated negative trend, particularly during late summer. EC-Earth underestimates this trend in September but not in March and overestimates the ice extent both in September and March at the end of the twentieth century and beginning of the twenty-first century (Fig. 2). It is not until 2030–2040 that ice extents and trends in EC-Earth are comparable to today’s observations. Also, the modelled Arctic sea ice volume (not shown) is overestimated in the twentieth century compared to estimates of today’s observed ice volume (e.g. Belchansky et al. 2008; Rothrock et al. 2003). As for the ice extent, the simulated ice volume around 2030–2040 is comparable to today’s estimates. Thus, the relation between ice volume and extent in EC-Earth seems to be realistic.
The reduction of sea ice extent until 2050 is similar in all three projections. Thereafter, the decrease of sea ice extent differs strongly between the three different RCPs. The simulated September sea ice extent in RCP2.6 stabilizes after an abrupt increase around 2070 at about 1.5 million km2 below the modelled extent at the beginning of the twenty-first century. In RCP4.5, the reduction continues until around year 2070 and seems to stabilize on a very low level thereafter. The differences between the three RCP4.5 ensemble members is more pronounced after 2070 and several reduction events occur with partial recoveries thereafter. In RCP8.5, an abrupt sea ice reduction, which leads to almost total sea ice loss in September, takes place in all three members around 2060 without any recovery thereafter. The rapid sea ice reductions in RCP8.5 happen at a CO2 concentration of about 600 ppm. This is an increase of 115 ppm compared to year 2040, where EC-Earth simulates ice conditions similar to the observed conditions around 2010. Adding this 115 ppm to today’s CO2 concentration, the real world threshold, if existing, for an ice-free September would be around 500 ppm. This level is reached around 2040 and 2055 in RCP4.5 and RCP8.5, respectively. In RCP2.6, CO2 concentrations stay below 500 ppm through the entire twenty-first century indicating a low likelihood for an ice-free September.
The sea ice change in EC-Earth is strong compared to most CMIP3 models (Stroeve et al. 2007; Wang and Overland 2009); only a few CMIP3 models simulate an Arctic sea ice loss before 2060. However, due to different emission scenarios, the results are not completely comparable. RCP8.5 simulations with CCSM4 (Vavrus et al. 2011) show an ice free Arctic in September around 2070. This is slightly later than in our model even though CCSM4 starts with a lower initial ice extent in the end of the twentieth century.
Figures 3 and 4 show the spatial distribution of the annually averaged sea ice thickness and concentration at the end of the twentieth century, and changes by the end of the twenty-first century. Sea ice thickness reaches 3–4.5 m in most of the Central Arctic. Compared to existing observations and estimates of the ice thickness (Rothrock et al. 2003; Belchansky et al. 2008), this is an overestimation by about 0.5–1 m. At the Siberian coast, the overestimation is even larger, which is a typical problem of many coupled climate models and at least partly due to too weak offshore winds at the Siberian coast (Bitz et al. 2002; DeWeaver and Bitz 2006). The thick ice at the Siberian coast prevents the complete melting of ice during summer in this area. Along most ice edges, particularly in the Greenland Sea, sea ice extends slightly further south than in observations (Fig. 4b).
Ice thickness reductions until 2080–2100 are relatively uniform in the Central Arctic and reach 1–1.5 m in RCP2.6, 2–3 m in RCP4.5 and up to 4 m in RCP8.5. The largest sea ice concentration changes take place in the Barents Sea (Fig. 4c–e). Ice concentration is also substantially reduced along the ice edges of the Labrador and Greenland Seas. Here, the overestimation of sea ice in the twentieth century might contribute to the large reduction rates. Ice concentration changes are small in the Central Arctic, but still significant at the 95 %-level in all scenarios due to low interannual variability of ice concentration in the interior of the Arctic. The sea ice reduction in the Central Arctic is growing with increasing emissions and in RCP8.5, sea ice concentration is strongly reduced almost everywhere. The seasonal mean changes (Fig. 4f–i, only shown for RCP4.5) indicate the largest reduction in autumn, while in winter and spring the reduction is limited to the ice edges. The Barents Sea region has the largest ice concentration decrease in the Arctic throughout all seasons. The annual cycle of sea ice concentration changes have a similar pattern in RCP8.5 and RCP2.6 but the amplitudes differ.
Turbulent heat fluxes
Figure 5 shows modeled annual-averaged surface sensible and latent heat fluxes for the present climate, differences to ERA-Interim and changes for the two higher emission scenarios, RCP4.5 and RCP8.5. Averaged over the polar ice cap the fluxes are close to zero (Fig. 5a, b). The open-ocean is dominated by the upward fluxes in the northern North Atlantic, which are up to 100 ± 35 W/m2 in winter. Lack of direct observations makes model results difficult to evaluate. For winter, the only available, substantial data set is from the Surface Heat Budget of the Arctic Ocean (SHEBA) experiment (Uttal et al. 2002) while for summer, observations from several research cruises are obtained; for example, in addition to SHEBA, the Arctic Ocean Experiment 2001 (AOE-2001) and the Arctic Summer Cloud Ocean Study (ASCOS); see Tjernström et al. (2012) for a summary. Long data records are only available from the ice while evaluations over open water are difficult.
A comparison to ERA-Interim shows that the difference is near zero over large parts of the Arctic, as expected for the sea-ice dominated central parts of the Arctic. Larger differences, up to ±~50 W/m2, appear in large organized structures over the northern North-Atlantic and the Nordic Seas, where they reflect differences in ice conditions and SST. In particular the large north–south elongated difference in the Greenland Sea is a direct reflection of a shift in the ice edge between the observations and EC-Earth (compare Fig. 4).
The modeled winter sensible heat flux is reasonable compared to the SHEBA observations. The probability density function (PDF) of modeled fluxes (not shown) is skewed, with a peak at −10 W/m2, a negative tail to ~−50 W/m2 and a sharper cut-off around 10 W/m2. The observed PDF is wider, from about −30 to 20 W/m2 with a flat peak around −5 W/m2. While the modeled median summer sensible heat flux is also reasonable, the PDF-width is a factor of 2–3 too large; ±20–30 W/m2 in the model and ±~10 W/m2 in the observations. The modeled latent heat fluxes are too large in both seasons. In winter, modeled values are relatively normally distributed (~±10 W/m2) while in summer the distribution is skewed, from −10 to 20–30 W/m2 with a peak around zero. Observations indicate zero fluxes in winter and a skewed distribution in summer, also with a peak at zero but only from minus a few to ~5 W/m2. Both heat fluxes over open water, as expected, are large in winter (up to 100–200 W/m2) while in summer they are smaller, similar to the fluxes over the ice. Largely, the turbulent heat fluxes in EC-Earth are in good agreement with the CMIP3 model ensemble (Sorteberg et al. 2007).
The most extreme positive and negative local changes in the sensible heat fluxes under the lower emission scenarios are similar and reach ±35 W/m2 (Fig. 5e). In the highest emission scenario, the annually averaged changes do not grow proportionally (Fig. 5g). The primary reason for the changes seems to be related to the retreating northern North Atlantic sea-ice edge during winter. The changes are small in other parts of the Arctic Ocean. The change pattern has a large upward flux along the ice edge with a corresponding negative change further to the south. The high upward fluxes are due to winter cold-air outbreaks over the relatively warm open ocean that is shifting northward with the ice edge. In the RCP8.5 scenario (Fig. 5g) there is an additional small (5–15 W/m2) widespread increase over the sea-ice area, which is dominated by winter conditions and is likely due to the ice thinning or more leads and lower sea ice concentrations.
Changes in the latent heat flux exhibit a small maximum east of Iceland and a larger area with increased fluxes in the Barents Sea, extending into the Kara Sea (Fig. 5f, h). Maximum changes reach 35–40 W/m2, approximately constant across emission scenarios. Instead the area of the maxima increases with emissions. A slight decrease in the upward flux over the Greenland Sea remains essentially unchanged. The latent heat fluxes also exhibit a widespread increase in the annual average upward flux over the central Arctic Ocean, dominated by winter conditions. Unlike the sensible heat flux, the summer latent heat flux over the central Arctic Ocean shows a small but widespread decrease (~5–10 W/m2).
Figure 6 shows annual cycles of the median values of area averaged 3 hourly sensible and latent heat fluxes, for the present climate and two emission scenarios, analyzed separately for sea-ice and open-water grid points using a threshold at an ice fraction of 50 %. Note that for the RCP8.5 scenario there is little ice left in summer. However, as long as there is ice left in a grid box, even at appreciably <50 %, it will still affect the thermodynamics near the surface.
For the present climate sensible heat flux over sea ice (Fig. 6a), median values are negative in winter, at about −6 W/m2, and near zero in May through September, while the latent heat fluxes (Fig. 6c) are small and positive in winter and peak at 10 W/m2 in June. The annual cycles exhibit significant changes for the different emission scenarios. For the highest emission scenario, the sensible heat-flux cycle even reverses sign, to a weak upward flux in winter and a comparable downward flux in summer. Changes in the annual cycle of the latent heat flux over sea ice is more complex with dual peaks for the RCP8.5 in December and May, at almost 10 and ~8 W/m2, respectively; the near-zero minimum occurs in August. The amplitudes of the annual cycles over open water are reduced for both fluxes. The present-day maxima in December at ~80 W/m2 are reduced to about 45 and 60 W/m2 in sensible and latent heat flux, respectively while summer values are unchanged.
We offer the following interpretation of these results. In the present climate, conditions in winter are dominated by sea ice leading to dry and cold air in the lower troposphere and consequently small latent and predominately downward sensible heat fluxes. As this air is advected over warm open water, either over leads or open ocean, the contrast is large and the resulting fluxes are large and upward. As ice fractions gradually diminish, cold and dry over-ice conditions gradually become less dominant and the air-mass transforms to open-ocean maritime conditions. For the highest emission scenarios in summer, large open-water areas will dominate; air advected over what little ice is left becomes moist and warm and the fluxes are therefore small or even downward.
Again, note that the annual cycles in Fig. 6 are derived over sea-ice covered ocean and open ocean separately. We do this in an attempt to better understand the physical processes responsible for the change. If we instead consider the annual cycles in turbulent surface fluxes for the whole Arctic Ocean region (i.e. disregarding surface type) the future scenarios show unchanged fluxes during summer and increasing fluxes during the remaining part of the year (not shown). This increase is explained by the retreat of sea-ice in the future scenarios and that sea-ice is associated with substantially lower surface fluxes than open ocean (Fig. 6).
Atmospheric temperature
In the twentieth century, the annually averaged 2-m air temperature (T2m, Fig. 7a) is about 2–3 K colder in EC-Earth compared to ERA-Interim-reanalysis (Fig. 7b) in most Arctic Ocean regions. In regions with to much ice, particularly over the East Greenland Current (EGC), and over Alaska it is up to 6 K colder. Over north-western Canada, the model is 1–4 K warmer than the reanalysis. Chapman and Walsh (2007) found a similar cold bias in the Arctic for the ensemble mean of 14 AR4-models but with a much larger cold bias in the Barents Sea. A recent study by de Boer et al. (2012) shows an Arctic wide cold bias of slightly below 2 K in the AR5-model CCSM4.
The simulated T2m changes in EC-Earth in the future (Fig. 7c–e) are strongly related to the changes in sea ice and heat fluxes, which agrees well to observations of the last decades (Screen and Simmonds 2010a). In RCP2.6, the warming in the twenty-first century is mainly concentrated on the Barents Sea region and surroundings. Here, annually averaged T2m increases by up to 10 K until the end of the twenty-first century. Over most land masses, the warming is not exceeding 2 K. In RCP4.5, the warming reaches 2–5 K over land and 4–10 K over the Arctic Ocean, over the Barents Sea up to 15 K. In RCP8.5, the warming north of 60°N is almost everywhere exceeding 6 K and the temperature increase over the Arctic Ocean is more than 10 K; up to 17 K over the Barents Sea. In all scenarios, the temperature change is significant at the 95 % significance level and the inter-ensemble variations are small compared to the change signal.
The temperature increase depends strongly on the season (Fig. 7f–i) and is largest in winter and autumn but with a different change pattern: the warming is more uniformly distributed in autumn while in winter it is particularly pronounced over the Barents Sea. The warming is smallest in summer because the surface stays near 0 °C until almost all sea ice has been melted. Only in RCP8.5, some Arctic Ocean regions warm up in summer due to the earlier onset of the sea ice melt period. The simulated temperature change in EC-Earth over the Arctic Ocean is at the upper end compared to the AR4-model ensemble but agrees rather well over land (Chapman and Walsh 2007). Note that the AR4-simulations were based on different emission scenarios than our AR5-simulations.
The winter (DJF) and summer (JJA) Arctic mean near-surface temperatures (spatial mean for 70–90°N) are shown in Fig. 8. The mean Arctic T2m at the end of the twentieth century is about 3 and 2 K colder in EC-Earth than ERA-Interim in winter and summer, respectively. The warming in the twenty-first century is much more pronounced in winter and reaches 5, 8 and 18 K in RCP2.6, RCP4.5 and RCP8.5, respectively. In summer, the temperature increase is typically below a few degrees, reaching as much as 5 K only for the RCP8.5 scenario. The annual mean Arctic T2m change reaches up to 12 K in RCP8.5, which is slightly above the values found in CCSM4 for the same region in RCP8.5 (Vavrus et al. 2011). The interannual variability is much larger in winter than in the summer, as expected, in both model and ERA-Interim. The natural variability is also substantial in winter, when the difference between the ensemble members can reach several degrees and persists over decadal time scales. Although these differences are small compared to the change-signal, a larger ensemble size would reduce the uncertainties due to internal variations. The decadal scale variations also indicate enhanced decadal predictability for winter temperatures.
The zonally mean vertical temperature distribution is dominated by a strong winter near surface inversion in the high Arctic in the twentieth century (Fig. 9a). The warmest average temperatures typically occur at about 850–900 hPa. At the surface, the temperature is up to 6 K colder than the air aloft, primarily due to the effect of the negative net radiation at the surface. Over the Arctic Ocean, the insulating properties of the sea ice and snow also play an important role. In summer (Fig. 9b), on average, no temperature inversion is simulated by the model but the temperature decrease with height is relatively small compared to lower latitudes. The vertical temperature distribution compares well to the ERA-Interim reanalysis (Fig. 9c, d). The cold bias in the Arctic in EC-Earth is slightly reduced with increased height. Also the spatial pattern of the inversion strength in EC-Earth during winter compares well to the ERA-Interim reanalysis (not shown) and is smaller and thus more realistic than in most CMIP3 models (Medeiros et al. 2011).
The simulated future changes differ distinctively between summer and winter (Fig. 9e–j). In summer, the atmosphere is relatively uniformly warmed by about 0–2 K in RCP2.6, 1–3 K in RCP4.5 and 3–6 K in RCP8.5. The warming near the surface is slightly reduced north of 80°N because of the melting sea ice and the cold ocean. In winter, the warming near the surface is strongly amplified compared to lower latitudes. The temperature amplification decreases with increasing height and above 600 hPa, no amplification can be found. Thus, the Arctic atmosphere becomes less stable during winter; in RCP8.5 the winter temperature inversion totally disappears at the end of the twenty-first century. This vertical warming distribution agrees well to the trend in ERA-Interim reanalysis data (Screen and Simmonds 2010b).
Sea level pressure
The sea level pressure (SLP) in the twentieth century (Fig. 10a) is well simulated in the Arctic compared to ERA-Interim reanalysis (Fig. 10b) and biases are small compared to most other global climate models (Chapman and Walsh 2007; de Boer et al. 2012). In an area from Alaska across the Bering Strait towards Siberia, the annually averaged SLP is underestimated by 1–3 hPa. Over the European Arctic, the SLP is overestimated by 1–2 hPa.
The change of SLP in the twenty-first century (Fig. 10c–e) is small in RCP2.6; mostly within the range of ±1 hPa and only in parts of the Barents Sea and Bering Sea significant at the 95 %-significance level. Here, where the ice margins move northward, the decrease reaches 1.5 and 2 hPa, respectively. In RCP4.5, the largest SLP reductions occur in the Barents Sea, and north of Greenland with up to 3 hPa. SLP is slightly decreased over most of the Arctic Ocean and the American Arctic land masses and slightly increased over the northeastern North Atlantic and northwestern Europe. In RCP8.5, the change pattern is similar but the amplitude is substantially larger than in RCP4.5. The anomalous pressure gradient across the sub-polar North Atlantic causes anomalously southwesterly winds here. The ensemble mean changes in RCP4.5 and RCP8.5 are significant in most of the Arctic but compared to most other Arctic parameters, the SLP change signal varies relatively strong between members. While the spatial change pattern over the Arctic Ocean is relatively robust, the SLP increase over the North Atlantic varies substantially across model members. Also Deser et al. (2012) showed that the SLP-change signal shows a high spread between model members. Most AR4-models show a similar decrease in SLP over the Arctic in the twenty-first century and many of them also show maximum decreases in the Bering Strait and/or the Barents Sea regions. However, in contrast to EC-Earth, most AR4 models also show a SLP reduction over western and middle Siberia (Chapman and Walsh 2007). SLP change patterns in RCP8.5 simulations with CCSM4 (Vavrus et al. 2011) are similar to our RCP8.5 pattern but their amplitude is about 50 % smaller.
Precipitation and evaporation
The precipitation in EC-Earth in the twentieth century (Fig. 11a) is about 10–30 % smaller over the Arctic Ocean compared to the ERA-Interim reanalysis (Fig. 11b); over Alaska, parts of Siberia and the north eastern Atlantic and northern Europe, there is 10–30 % more precipitation in EC-Earth. Serreze and Hurst (2000) compared ERA-reanalysis to existing Arctic precipitation data sets and concluded that ERA realistically simulates precipitation over most of the Arctic except for the Atlantic sector where precipitation is somewhat underestimated. Note, that also data sets of precipitation over the Arctic Ocean have large uncertainties due to coarse resolution of observations in space and time and gauge undercatchment issues.
All scenario simulations show a general increase of precipitation in the entire Arctic except for small areas in the Nordic Seas (Fig. 11c–e). The increase is largest in the Barents Sea, Greenland’s southeast coast and the south coast of Alaska; up to more than 300 mm/year in all RCPs. In the Central Arctic, precipitation changes strongly differ with the scenario; the increase is about 20 mm/year in RCP2.6 and up to 100–150 mm/year in RCP8.5. Over a number of land regions and in parts of the Beaufort Sea, the precipitation change is not significant at the 95 %-level in RCP2.6. Vavrus et al. (2011) found a comparable but spatially more uniform increase over the Arctic in CCSM4.
Evaporation over the Arctic Ocean is small in the twentieth century and reaches 20–50 mm/year (not shown). Over the Arctic land areas, evaporation reaches up to 400 mm/year and over the sea near the ice edge almost 1,000 mm/year. Also for evaporation, observational data are uncertain. However, over land along 65°N, observations suggest an annual mean evaporation between 200 and 350 mm (Serreze and Hurst 2000), which fits relatively well to EC-Earth. Precipitation in EC-Earth exceeds evaporation (P − E > 0) in most Arctic areas except for some smaller regions in the Nordic Seas. In the twenty-first century, evaporation is strongly enhanced. The change pattern resembles the one of precipitation. Except for near the ice margins, where P − E is near zero or even slightly negative, the change in P − E is positive and slightly increasing with growing RCP.
Cloudiness and radiative forcing
EC-Earth twentieth century simulated cloud variables are compared to APP-x satellite estimates (Wang and Key 2005; Karlsson and Svensson 2011) and ERA-Interim analysis in Fig. 12 (left column). The three EC-Earth simulations and ERA-Interim annual cycles of cloud amount over the Arctic (70–90°N) are very similar. In summer, EC-Earth and ERA-Interim are close to the APP-x cloud fraction of 84 %. In winter, EC-Earth overestimates cloud fraction by 15 % and ERA-Interim by 10 % compared to the observations. However, the cloud fractions are within the range of the substantial across-model spread found for CMIP3 models in Arctic winter (Vavrus et al. 2009; Karlsson and Svensson 2011). It should be emphasized that cloud fraction is not a very well defined variable in general and even more so in the Arctic where a substantial part of the clouds are optically very thin. The EC-Earth total cloud water path (TWP), the sum of the liquid water path (LWP) and ice water path (IWP), is at minimum in January at 40 g/m2 (90 % IWP) and peaks in August at 100 g/m2 (60 % LWP).
The longwave (LW) and shortwave (SW) surface cloud radiative forcings (CRFs) are calculated from the difference in all sky and clear sky net LW and SW fluxes at the surface. The LW warming by EC-Earth and ERA-Interim clouds varies from 20 W/m2 in winter to 55 W/m2 in summer. The slightly larger cloud fraction and less cloud condensate in EC-Earth compared to ERA-Interim result in very similar values of the LW CRFs (Fig. 12e). In winter, EC-Earth and ERA-Interim LW CRFs are 10 W/m2 lower than the APP-x estimate. The modeled SW cloud cooling effect is strongest in July with −70 W/m2, about 5 W/m2 less than for ERA-Interim due to the smaller amount of cloud condensate in EC-Earth. EC-Earth total cloud forcing is positive from September to May with maximum warming in October, ~45 W/m2, and cooling in July, −20 W/m2, which is in good agreement to the APP-x data (not shown).
At the end of the twenty-first century, the changes in the cloud variables increase with increasing emission scenario (Fig. 12, right column). The mean total cloud fractions increase a few percent in winter and spring due to increased cloudiness over the sea-ice while cloudiness is decreased over the Barents Sea region where the warming is the highest and the sea ice is strongly reduced. Vavrus et al. (2009) analyzed the cloud fraction changes for 20 CMIP3 models for SRES A1B emission scenario and found an increase in the ensemble mean monthly cloud fractions all year (4–5 % in winter and 1–2 % in summer). The ensemble mean change was dominated by models with low winter cloud fraction in the twentieth century while models with initially high winter cloud fraction, as is the case for EC-Earth, showed very small changes.
In autumn, in contrast to what a majority of the CMIP3 models showed (Vavrus et al. 2009), there is a widespread Arctic reduction in cloudiness of about 10 % for RCP8.5, while for RCP4.5 and 2.6 the mean reductions of 5 and 2 % are concentrated to the Barents Sea region where the sea-ice is reduced. The warming near the surface and at low atmospheric levels dominates over the humidity changes leading to decreased relative humidity at low levels and thereby decreased low level and total cloudiness. However, the total cloud water paths increase fairly evenly all year by 5, 10 and 25 g/m2 in the RCP2.6, 4.5 and 8.5, respectively. The summer and autumn changes are due to increases in LWP over the whole region while IWP decreases over Barents Sea (not shown). For winter and spring the TWP changes are dominated by increased IWP over the remaining sea-ice, moderated by reductions of IWP over the Barents Sea.
The increases in cloud fraction and large increases in TWP in winter and spring over the remaining sea-ice lead to increases in the LW CRF’s by 5–10 W/m2, implying a larger cloud induced warming. The low amounts of liquid water in twentieth century EC-Earth mixed-phase clouds makes the model prone to large changes in cloud emissivity and in the amount of LW emitted to the surface (Willén et al., manuscript in preparation). In summer, despite of the increase in TWP over most of the Arctic, the changes in LW CRF’s are close to zero for RCP2.6 and RCP4.5 since the present day cloud emissivities are already close to unity and the changes in surface LW fluxes are smaller (not shown). For RCP8.5 the LW CRF is reduced in summer and autumn due to the reduction in cloud fraction centered over the Barents Sea, the decrease in low-level clouds and increase in high-level clouds reduce the surface LW CRF’s but increase the top of atmosphere LW CRF’s (not shown).
The EC-Earth SW cloud cooling increases as a result of the increases in LWP in summer and autumn which leads to optically thicker cloud and less SW flux to the surface. The SW CRFs become more negative with the largest decrease in June of −15, −30 and −55 W/m2, for the RCP 2.6, 4.5 and 8.5 simulations, respectively. The increase in SW cloud radiative cooling is both attributed to the increases in cloud albedo and to the decreases in surface albedo, due to the retreat of the sea-ice (Fig. 4). The cloud albedo changes are responsible for about half of SW CRF reductions seen in Fig. 12h, i.e. EC-Earth clouds reduce the sea-ice albedo feedback.
At the end of the twenty-first century, the total cloud forcing in the EC-Earth simulations is positive from September to May for RCP2.6 and 4.5 and from September to April for RCP8.5 with maximum warming in October of about 50 W/m2 for all emission scenarios. The cloud forcing is negative from June to August for RCP2.6 and 4.5 and from May to August for RCP8.5, with maximum cooling in July varying from −30, −50 to −70 W/m2 for RCP2.6, 4.5 and 8.5 scenarios, respectively.
Atmospheric meridional energy transport
The meridional, atmospheric energy transport reaches 4PW in the Northern Hemisphere (NH) mid-latitudes and decreases towards the North Pole. The largest part of the transport is accomplished by the dry-static component, whereas the latent transport stands for a minor contribution. The total transport at a given latitude Φ
0
is defined as:
$$ \int\limits_{\Upphi = \Upphi_0} {\int\limits_{0}^{{p_{s} }} {(k + c_pT + gz + Lq)} } v\frac{dp}{g}dx $$
where the first three terms in the integral constitute the dry-static transport, and the last term the latent transport. Here Φ is latitude, p
s
surface pressure, k kinetic energy, c
p
specific heat capacity at constant pressure, T temperature, g gravity, z geopotential height, L latent heat of condensation, q specific humidity, v meridional wind component, p pressure, and x is the zonal coordinate. Estimations based on EC-Earth and ERA-Interim are shown in Figs. 13a and 14a. The transports are estimated with a 6-h resolution from fields at model hybrid levels. For ERA-Interim a correction is applied to take into account transports associated with erroneous mass fluxes (Trenberth 1997; Graversen 2006; Graversen et al. 2007).
EC-Earth and ERA-Interim are in a fairly good agreement at most high, northern latitudes. In the NH mid-latitudes the disagreement is around 10 % where EC-Earth underestimates the total transport relative to ERA-Interim. The annual EC-Earth transports are also roughly similar to estimates from the NCEP-NCAR reanalysis reported by Trenberth and Stepaniak (2003), but are somewhat larger than the estimation based on rawinsonde measurements over the period 1963–1973 documented by Oort and Peixoto (1983).
The twenty-first century transport changes are shown in Fig. 13b and the split into seasons in Fig. 14b. All seasons except summer show an increase of the latent, but a decrease of the dry-static component north of 60°N. In summer the total changes are positive but small.
The change in the atmospheric energy transport will likely affect the Arctic climate (Graversen 2006). The energy-divergence change over the Arctic will directly cause cooling or warming. In addition, changes in the humidity divergence may modify the greenhouse effect over the Arctic, due to changes in both humidity and cloudiness, hereby altering the surface-energy budget. Hence, the reduction at the Arctic boundary of the atmospheric meridional energy transport over the twenty-first century in the three darker seasons, as simulated by EC-Earth, will likely in itself reduce the Arctic temperature amplification. However, at the same time, the increase of the latent component may enhance the greenhouse effect and contribute to an enhancement of the amplification. Also note that zonal variations of the transport change and non-linear feedbacks may be important: If the transport increases over areas with thin sea ice, the warming effect may be larger than the cooling effect from a corresponding reduction of the transport elsewhere. A warm anomaly over thin sea ice may melt the ice and expose the ocean surface whereby feedbacks such as that of surface albedo are invoked. In contrast, a cold anomaly over an ice area where the ice is normally retained by the end of the summer would reduce the ice melt but cause little albedo change.
Arctic Ocean circulation
The Arctic ocean circulation in layers near the surface is characterized by a pronounced Beaufort Gyre, transpolar currents from the Siberian coast towards Fram Strait and southerly flows through Fram Strait as well as inflowing waters through the Barents Sea opening and further along the Siberian coast. Limited observations make an evaluation difficult. However, results from e.g. Morison et al. (2012) indicate a similar circulation pattern of the real Arctic Ocean mixed layer but the Beaufort Gyre seems to be slightly too extended in our model. Figure 15 compares the ensemble mean currents below the mixed layer in the Arctic at 109 m for the end of the twentieth century and at the end of the twenty-first century in RCPs 2.6, 4.5 and 8.5. In RCP2.6 and RCP4.5, the currents are generally strengthened compared to the twentieth century but the patterns stay similar. The reduction in sea ice leads to increased wind stress on the ocean and thus to enhanced currents. A comparison to the SLP changes (Fig. 10c–e) indicates that enhanced south-westerly wind stress is responsible for enhanced inflow into the Barents Sea. In RCP8.5, the inflow through the Barents Sea is further strengthened due to additional increase in the south-westerlies over the Nordic Seas. Also the circulation pattern in the Arctic Basin changes substantially. The size of the Beaufort Gyre is reduced and is displaced towards Chukchi Sea. This leads to strongly reduced velocities in the transpolar drift. Anomalously on-shore winds at the Siberian coast (Fig. 10e) are responsible for at least parts of the simulated changes.
Arctic Ocean temperature and salinity
The twentieth century Arctic Ocean is characterized by a cold and fresh upper layer and warmer and more saline water masses in deeper layers (Fig. 16). Compared to observations (World Ocean Atlas 2009: Antonov et al. 2010 and Locarnini et al. 2010), the upper layer in EC-Earth is slightly too cold and too saline. Furthermore, the observed warm layer at around 500 m, which is caused by the warm and saline inflowing Atlantic water, is not sufficiently pronounced and does not appear at all in the twentieth century. The CCSM4 model (Jahn et al. 2012) show similar to our model a too saline upper layer and no pronounced intermediate Atlantic water masses. The inflowing warm Atlantic water masses are rapidly mixed into larger depth after entering the Arctic Ocean in EC-Earth. Within the twenty-first century, a more pronounced warm water layer is formed. In RCP2.6, the cold surface layer persists during the entire twenty-first century while in RCP8.5 the surface layer is warmed most, by up to 3 K.
Salinity in the twenty-first century decreases by up to −1.5 psu in the upper 100 m. This agrees well to model results by Koenigk et al. (2007) who found a maximum decrease of −1.8 psu near the surface in the A1B scenario. The salinity differences between the RCPs in our simulations are not as pronounced as for temperature but we see a clear tendency to more freshening in the higher emission scenarios.
Arctic Ocean freshwater budget
The Arctic Ocean freshwater budget has been calculated with a reference salinity of 34.9 psu, taken from the twentieth century inflowing Atlantic water in our model.
In the twentieth century, the simulated freshwater transports (Fig. 17a) agree relatively well to observational based estimates (Serreze et al. 2006). However, the liquid freshwater exports through Fram Strait and the Canadian Archipelago are underestimated by about 30 %. For Fram Strait, this is partly compensated by too large sea ice export. The river runoff into the Arctic is too small and might explain the high surface salinity and low liquid freshwater export biases. Similar to EC-Earth, the CCSM4-model is overestimating Fram Strait ice export and underestimating the liquid freshwater export (Jahn et al. 2012). In contrast to EC-Earth, CCSM4 is overestimating the river runoff but shows a more realistic export through the Canadian Archipelago.
The change in the total freshwater transport (solid + liquid) out of the Arctic is relatively small despite a strong increase in the freshwater input by P − E and river runoff in the twenty-first century. After 2030, the total export is slightly increased and this trend is accelerated in RCP8.5 at the end of the twenty-first century. Investigating the individual straits, a strong increase of the total freshwater export through the Canadian Archipelago is evident, while the total exports through the Fram Strait and through the Barents Sea are slightly reduced. The transport through Bering Strait shows a slight increase. After 2060, we see accelerated changes in RCP8.5 compared to RCP4.5. The freshwater transports are dominated by the liquid transport except for Fram Strait (Fig. 17c) where the ice export dominates. In the twenty-first century, ice export is strongly reduced while the liquid export increases.
The additional freshwater input into the Arctic exceeds the increase in the exports in the twenty-first century simulations. As a consequence, the Arctic Ocean fresh water content grows (Fig. 17d). This growth is similar in all RCPs since the larger freshwater input in the higher RCPs is to a large degree compensated by larger exports.
Ocean heat transport
The ocean heat transport into the Arctic is shown in Fig. 18. We calculated the heat transport across 70°N as residuum of the integrated surface heat fluxes and the change in ocean heat content north of 70°N (Fig. 18a). The heat transports through the Arctic openings (Fig. 18b) have been estimated by calculating the product of velocity perpendicular to the opening, the difference of ocean temperature to a reference temperature, the density of the water mass and the specific heat capacity of the water. Similar to most other studies, we use 0 °C as the reference temperature. We tested the robustness of the results against changes in the reference temperature and found that the mean heat fluxes are different but that the twenty-first century changes are insensitive.
The twentieth century heat transport across 70°N sums up to 0.27PW. This is in good agreement to observational based estimate by Oliver and Heywood (2003) who found a transport of 0.2 ± 0.08PW across a section at approximately 70°N between Greenland and Norway. It also compares well to model results by Jungclaus and Koenigk (2010), who found a transport of 0.28PW across 70°N. In the twenty-first century, the heat transport is strongly growing and reaches 0.32, 0.41 and 0.6PW in RCP2.6, RCP4.5 and RCP8.5, respectively.
The total heat transport into the Arctic Ocean reaches 50TW; 20TW are transported through Barents Sea (across a line connecting Svalbard with the Kola Peninsula at 69°N, 37°E) and about 15TW each through Fram Strait and Canadian Archipelago. Measurements indicate heat fluxes of about 50TW through the Barents Sea Opening (Skagseth et al. 2008) and 30–40TW through Fram Strait (Schauer et al. 2008). Obviously, EC-Earth underestimates the heat flux through Fram Strait. The underestimation of heat and volume flux through Fram Strait into the Arctic is a common problem in global coupled models, probably due to insufficient resolution and is also seen in the CCSM4-model (Jahn et al. 2012). The Barents Sea section used by Skagseth et al. (2008) was further to the south and west than our section and loss of heat to the atmosphere within the Barents Sea can explain at least parts of the differences. At the end of the twenty-first century, the heat flux through the Barents Sea is strongly enhanced and the ensemble averages reach 100TW and 160TW in RCP4.5 and RCP8.5, respectively. The single RCP2.6 member reaches about 70TW at 2050 but decreases somewhat thereafter (not shown). The strong increase of heat transport in the Barents Sea is due both to the temperature increase of the transported water masses and increased northward volume transport. The volume transport through Barents Sea is enhanced from 2.7 Sv at the end of the twentieth century to 3.7 Sv and 4.2 Sv in RCP4.5 and RCP8.5, respectively. The increase is probably mainly caused by strengthened south-westerly winds in the Nordic Seas (compare Fig. 10).
The heat transport through the Barents Sea opening governs sea ice variations in the Barents and Kara Sea on decadal scale time periods. The correlation is −0.75 (−0.95) between 10-year running means of heat transport and the detrended (including trend) Barents Sea/Kara Sea ice area time series. We hypothesize that the increasing ocean heat transport strongly contributes to the reduced sea ice cover in the Barents and Kara Sea region and thus also contributes to the Arctic temperature amplification.
The Arctic Ocean heat content is strongly growing in the twenty-first century (Fig. 18c). However, the integrated heat flux anomaly into the Arctic in the twenty-first century is about twice the heat content anomaly. This means that about 50 % of the inflowing ocean heat anomaly in the twenty-first century is either used to melt sea ice or passed to the atmosphere.