Atmospheric impacts of an Arctic sea ice minimum as seen in the Community Atmosphere Model

Authors

  • Elizabeth N. Cassano,

    Corresponding author
    1. Cooperative Institute for Research in Environmental Sciences, University of Colorado, Boulder, CO, USA
    • Correspondence to: E. N. Cassano, Cooperative Institute for Research in Environmental Sciences, University of Colorado, UCB 216, Boulder, CO 80309. E-mail: ecassano@cires.colorado.edu

    Search for more papers by this author
  • John J. Cassano,

    1. Cooperative Institute for Research in Environmental Sciences, University of Colorado, Boulder, CO, USA
    2. Department of Atmospheric and Oceanic Sciences, University of Colorado, Boulder, CO, USA
    Search for more papers by this author
  • Matthew E. Higgins,

    1. The National Center for Atmospheric Research, Climate & Global Dynamics—Terrestrial Sciences Section, Boulder, CO, USA
    Search for more papers by this author
  • Mark C. Serreze

    1. Cooperative Institute for Research in Environmental Sciences, National Snow and Ice Data Center, University of Colorado, Boulder, CO, USA
    Search for more papers by this author

Abstract

There is growing recognition that reductions in Arctic sea ice extent will influence patterns of atmospheric circulation both within and beyond the Arctic. We explore the impact of 2007 ice conditions (the second lowest Arctic sea ice extent in the satellite era) on atmospheric circulation and surface temperatures and fluxes through a series of model experiments with the NCAR Community Atmospheric Model version 3 (CAM3). Two 30-year simulations were performed; one using climatological sea ice extent for the end of the 20th century and other using observed sea ice extent from 2007. Circulation differences over the Northern Hemisphere were most prominent during autumn and winter with lower sea level pressure (SLP) and tropospheric pressure simulated over much of the Arctic for the 2007 sea ice experiment. The atmospheric response to 2007 ice conditions was much weaker during summer, with negative SLP anomalies simulated from Alaska across the Arctic to Greenland. Higher temperatures and larger surface fluxes to the atmosphere in areas of anomalous open water were also simulated. CAM3 experiment results were compared to observed SLP anomalies from the National Center for Environmental Prediction/National Center for Atmospheric Research (NCEP/NCAR) Reanalysis data. The observed SLP anomalies during spring are nearly opposite to those simulated. In summer, large differences were shown between the observed and simulated SLP also, suggesting that the sea ice conditions in the months preceding and during the summer of 2007 were not responsible for creating an atmospheric circulation pattern which favoured the large observed sea ice loss. The simulated and observed atmospheric circulation anomalies during autumn and winter were more similar than spring and summer, with the exception of a strong high pressure system in the Beaufort Sea which was not simulated, suggesting that the forced atmospheric response to reduced sea ice was in part responsible for the observed atmospheric circulation anomalies during autumn and winter.

1. Introduction

As assessed from the satellite passive microwave data record from 1979 to present, Arctic sea ice extent exhibits downward linear trends in all months, largest for September, the end of the melt season (Serreze et al., 2007; Cavalieri and Parkinson, 2012; Stroeve et al., 2012). The increasing amount of open water has numerous implications for the energy balance in the Arctic. More open water means a lower surface albedo leading to an increase in summer heat storage in the ocean mixed layer (Manabe and Stouffer, 1980; Curry et al., 1995). In autumn and winter, a relatively cold atmosphere overlying open water then leads to large energy fluxes from the ocean to the atmosphere (Serreze et al., 2009; Screen and Simmonds, 2010; Serreze et al., 2011). Arctic sea ice extent for September 2007 was at the time the lowest in the satellite era (Comiso et al., 2008; Stroeve et al., 2008). Sea ice extent was especially low on the Pacific side of the Arctic as well as the Barents and Kara Seas (Figure 1). September 2007 now stands as the second lowest extent behind September 2012 (http://nsidc.org/arcticseaicenews/). Did the anomalous open water observed in 2007 influence the atmospheric circulation and if so how? This paper seeks to answer this question through a series of modelling experiments and comparisons with the observed atmospheric state. Our study is motivated by (1) observational evidence that through altering the Arctic energy budget, ongoing ice loss is already having impacts on atmospheric circulation (Francis et al., 2009; Francis and Vavrus, 2012), (2) projections from global climate models that sea ice extent will continue to decline through the 21st century both in summer (Holland et al., 2006; Wang and Overland, 2009) and during the cold season (Overland and Wang, 2007), and (3) whether the atmospheric circulation during the summer of 2007 was forced by existing ice anomalies in the months prior to that summer season.

Figure 1.

The (a) annual mean, (b) JF, (c) MAM, (d) JJA, (e) SOND, and (f) September difference in sea ice concentration between the sea ice in 2007 and climatology. Cool (warm) colours have more (less) sea ice in 2007 compared with climatological sea ice. Negative values are further indicated by crosshatching.

A number of past studies have examined atmospheric responses to altered sea ice conditions. Bhatt et al. (2008) used the Community Climate System Model version 3 (CCSM3) to study the response to 1995 sea ice conditions (before 2007, the year 1995 had the lowest June–September ice extent in the satellite record). The results, focusing on April–October, included larger surface fluxes and higher surface air temperatures in the area of the open water as compared with climatological sea ice. The large scale circulation response to reduced sea ice in the western Arctic was higher sea level pressure (SLP) over the North Pacific and reduced precipitation in southern Alaska suggesting a weakened North Pacific storm track. Alexander et al. (2004) used the same years as Bhatt et al. (2008) but focused on the cold season, November through March. Over areas of anomalous open water, they found an increase in surface heat fluxes, near-surface warming, enhanced precipitation and evaporation, and lower SLP. Petoukhov and Semenov (2010) used the atmospheric general circulation model (AGCM) ECHAM5 to investigate the relationship between cold Eurasian winters and anomalous open water in the Barents and Kara Seas region. They found that lower-tropospheric heating over the Barents and Kara Seas could lead to anomalous cold easterly advection over northern Eurasia. Honda et al. (2009) ran an AGCM to investigate the relationship between anomalously cold winter air temperature anomalies in Europe extending to the Far East and anomalous open water in the Barents and Kara Seas region in late autumn. Anomalous turbulent heat fluxes from the additional open water tended to induce an amplification of the Siberian High which intensified cold northerlies in the Far East.

Regional climate models have also been used to examine the atmospheric response to altered sea ice conditions. Rinke et al. (2006) forced a regional climate model over an Arctic domain with two different sea ice and sea surface temperature (SST) boundary conditions but exactly the same lateral boundary conditions. Areas of higher SSTs and reduced sea ice thickness and concentration were associated with stronger upward heat fluxes and higher 2 m air temperatures. They did not find a simple relationship between anomalies in SST, sea ice, and change in storm tracks which they argue may result from a dominance of the lateral boundary forcing. Semmler et al. (2004) studied atmospheric impacts using two regional climate model experiments focused over the Fram Strait region. The experiments differed in the treatment of sea ice—one experiment had sea ice prescribed by satellite data and therefore grid cells could have partial sea ice, and in the other the sea ice was either 0 or 100% depending on the SST. For the experiment with more realistic sea ice, turbulent heat fluxes were often directed upwards due to the presence of leads and polynas leading to an increase in cloud cover and precipitation. The experiment with more realistic sea ice also compared more favourably to observations.

Several previous studies have looked at responses to 2007 sea ice conditions. Balmaseda et al. (2010) ran several experiments using the operational European Centre for Medium-Range Weather Forecasts (ECMWF) global seasonal forecasting system S3. For their 2007/2008 experiment using observed SSTs they found a significant impact on atmospheric circulation with anomalous high pressure over the Arctic and low pressure centres over western Europe and northwest North America. For their experiments with 2007/2008 sea ice but predicted SSTs the response to a given ice anomaly was found to be conditioned by the background mean state of the ocean–atmosphere. Kay et al. (2011) used the Community Atmosphere Model version 4 (CAM4) to evaluate the boundary layer response to the 2007 sea ice extent. These included short-term observationally constrained forecasts and longer-term experiments in which the atmosphere freely evolves. They found that for July near-surface temperature and humidity were little affected by sea ice loss in that month. For September, they found near-surface decreases in stability and humidity increases aloft over the areas of anomalous open water. Blüthgen et al. (2012) used 2007 sea ice conditions to force ECHAM5 and found an increase in oceanic heat uptake in summer, increased oceanic heat loss in fall, and a pronounced negative SLP anomaly over the Eastern Arctic in late summer.

As for regional modelling studies focusing on 2007, Strey et al. (2010) used Polar WRF (Skamarock et al., 2008; Cassano et al., 2011) in an ensemble of simulations with a North Pole centred domain extending south to approximately 30°N. The simulations used 2007 lateral atmospheric boundary conditions and SSTs for September–December. For the set of ensemble members testing the impact of decreased sea ice, 2007 sea ice concentration was used. For control ensemble members, 1984 sea ice concentration and extent were employed, with everything else the same as in the 2007 sea ice experiment. Focusing their results on October–November, they found increased latent heat fluxes and large temperature increases over the area of anomalous open water (focused in the western Arctic) and also over the Gulf Stream area which they attributed to a decrease in SLP over eastern North America and an associated increase in cold air advection in this area. Difference maps showed a ‘trough-ridge-trough’ pattern from the area of anomalous open water (a large decrease in SLP) roughly to the North Atlantic where positive SLP anomalies were modelled. In general, the simulations showed circulation changes throughout the atmosphere with higher tropospheric heights over western North America and lower constant pressure heights over eastern North America in the 2007 sea ice case, with these features broadening with height. The subsequent WRF-based study by Porter et al. (2012) used observed sea ice and SSTs from a low (2007) and high (1996) ice year, in addition to an experiment using a mixed SST field between 2007 and 1996, for three 15-member ensembles to sample a large range of climatic variability. They found the largest local response in October and November with increased turbulent heat fluxes which heated and moistened a vertically deep layer of the atmosphere. They also found an increase in cloud cover affecting the surface and atmospheric energy budgets.

Other studies have investigated the atmospheric response to projected future sea ice conditions. Higgins and Cassano (2009) used the NCAR CAM3 to ascertain winter impacts. They compared experiments using climatological sea ice from 1980 to 1999 to those with climatological sea ice extent from 2080 to 2099 from an ensemble of CCSM3 A1B scenario runs. The later period sees a higher frequency of strong cyclones over the central Arctic Ocean and increased precipitation across the Arctic. Deser et al. (2010) also used CAM3 to ascertain the impacts of projected sea ice conditions for 2080–2099 from the A1B scenario using an eight-member ensemble mean of CCSM3 simulations. The largest impact on the Arctic surface energy budget was found for winter where the largest temperature differences between the open water and atmosphere are located. The circulation response was greatest during the cold season with an insignificant response during the summer. Singarayer et al. (2006) used the Hadley Centre Atmospheric Model 3 (HadAM3) to investigate the impacts of projected 21st century changes in sea ice extent. The experiments performed in this paper are similar to those of Singarayer et al. in that the sea ice extent is altered but climatological SSTs are used. Singarayer et al. (2006) found significant increases in surface temperature primarily in winter, in association with an increase in upward sensible surface heat fluxes over areas of open water. Precipitation increased over areas with lower ice extent due to an increase of evaporation, most prominent over the marginal ice zone. They also found a reduction in SLP over much of the Arctic and a southward shift of the North Pacific storm track.

Regarding observational studies, Agnew (1993) found that weaker (stronger) Aleutian and Icelandic lows attend heavy (light) ice winters suggesting a role of differences in local and regional heating patterns. In addition, the weakening of both lows in heavy ice conditions would reduce the meridional atmospheric circulation and atmospheric poleward heat transport. Overland and Pease (1982) found a westward shift in cyclone tracks in the Bering Sea in light ice years. Francis et al. (2009) found that summers with anomalously low sea ice extent tend to be followed by temperature and circulation anomalies resembling the negative mode of the North Atlantic Oscillation that persist in to the winter. Francis and Vavrus (2012) compared circulation patterns for the decade of 2000–2010 to those for the previous 30 years. The period 2000–2010 with lower summer ice extent is characterized by a weaker 1000–500 hPa thickness gradient (leading to weaker zonal winds) between the high and mid-latitudes as well as elongated ridging (further slowing eastward Rossby wave propagation). They argue that this could lead to more persistent weather patterns in the mid-latitudes, increasing the probability of extreme weather events. Jaiser et al. (2012) also found that low sea ice years are associated with greater atmospheric heating from open water in autumn and winter that reduces vertical static stability, and suggests that sea ice changes exert a remote impact on atmospheric circulation anomalies. Overland and Wang (2010) showed an increase in surface temperatures and those at 850 hPa since 2002 which contributed to an increase in the 1000–500 hPa thicknesses and an anomalous tropospheric easterly wind component in the western Arctic.

Numerous studies have investigated causes of the 2007 sea ice loss. It is widely agreed that a primary driver was a summer atmospheric circulation pattern featuring anomalous high pressure north of the Beaufort Sea and unusually low pressure over eastern Siberia which has persisted through 2012, particularly in early summer (Overland et al., 2012). This promoted persistent southerly winds in the Laptev and East Siberian Seas favouring strong melt and ice transport away from the coast (Stroeve et al., 2008). L'Heureux et al. (2008) link the anomalous anticyclone to a strongly positive phase of the Pacific-North American (PNA) pattern. Other factors include warmer waters flowing through the Bering Strait triggering the onset of solar-driven melt (Woodgate et al., 2010), thinning of the ice pack which had preconditioned it for increased melt (Maslanik et al., 2007; Stroeve et al., 2008), anomalously high SSTs (Steele et al., 2008), and large ice bottom melting from solar heating of the upper ocean (Perovich et al., 2008). Zhang et al. (2008) performed modelling experiments of the 2007 sea ice and found that pre-conditioning, anomalous winds, and ice-albedo feedbacks were all important.

Our study of the impacts of 2007 ice conditions on the atmosphere use experiments with a global climate model designed to eliminate competing impacts of differing lateral boundary conditions and other changes in surface conditions. We focus on the response of the complete annual cycle in addition to the impacts on particular seasons. Section 'Data and methods' discusses the experimental design and describes the model used. Section 'Results' describes results from the experiments. Section 'Comparisons with observations' discusses the results in the context of the observed atmospheric state to investigate the contribution of the atmospheric circulation to the dramatic summer loss of ice in 2007. Section 'Summary and conclusions' summarizes the work and conclusions drawn.

2. Data and methods

2.1. CAM model description

The CAM3 used here is fully described by Collins et al. (2004) and Collins et al. (2006a). CAM3 is the atmospheric component of the CCSM (Collins et al., 2006b). In this paper CAM3 is run in stand-alone mode. In this mode it is integrated with the Community Land Model (CLM; Oleson et al., 2004), a thermodynamic sea ice model, and a data ocean model (Collins et al., 2006a). The resolution for the experiments is T42 with the standard 26 vertical levels, which have been used in previous similar studies (Higgins and Cassano, 2009; Deser et al. 2010).

2.2. Experimental design

We performed two experiments—a 2007 sea ice experiment and a control experiment. Sea ice concentrations, extents, and SSTs were updated monthly for each simulation. Two 30-year simulations were performed for both experiments. Thirty-year simulations were chosen so as to have a significant number of ensembles to overcome model noise and natural variability allowing us to capture the climatological signal from the experiments. The primary difference between the two experiments was the sea ice concentration and extent used for the lower boundary conditions with some changes to the climatological SSTs described below. For the control experiment, the annual cycle of climatological sea ice as provided by the CAM3 model distribution was used in a repeated 30-year simulation. The climatological (covering the time period of 1950–2001) sea ice dataset is a blend of the Hadley Centre Global Sea Ice and Sea Surface Temperature (HadISST; Rayner et al., 2003) used through 1981 and the Reynolds SST dataset for subsequent years (Reynolds et al., 2002).

For the 2007 sea ice experiment, the observed annual cycle of sea ice from 2007 was used in a repeated 30-year simulation. Sea ice conditions are based on the Advanced Microwave Scanning Radiometer—Earth Observing System (AMSR-E)/Aqua Daily L3 25 km Brightness Temperature and Sea Ice Concentration Polar Grids dataset (Cavalieri et al., 2004). SSTs from the climatological dataset provided with CAM3 model distribution were used for both experiments (with some changes as described next) so the primary difference between each experiment is the sea ice concentration and extent. The SSTs were altered in areas with differences between the 2007 and climatology sea ice as follows: if the sea ice concentrations were within 25% of each other, the climatological SST was used. If the grid point was open water (less than 15% coverage) in 2007 but ice covered in the climatological sea ice dataset, a radial search was performed for the first open water point and that SST was used (with a further check that SSTs got colder further north). If open water was present in the climatological sea ice dataset but not in 2007, the SST was set to the salinity-adjusted freezing point of sea water (−1.8 °C). Figure 2 shows the change in SST for September from this methodology. There is a large area of increased SSTs in the area of anomalous open water. These changes were done to ensure physical consistency; to reflect the fact that the surface is no longer ice covered, as well as to have a smooth transition of SSTs. If the climatological SSTs were used for this area of open water, the model would not have felt much of a difference in lower boundary conditions between open water and ice-covered water.

Figure 2.

The sea surface temperature (SST) differences used in the 2007 experiment to ensure physical consistency in the lower boundary conditions for September, the month with the largest amount of anomalous open water. Units are °C. Negative values are indicated by crosshatching.

3. Results

For the following discussion, the figures show the atmospheric response as results from the 2007 sea ice experiment minus the climatological sea ice experiment. A Student's t-test was performed to determine statistical significance of the differences. Results are evaluated for the standard seasons for spring (March, April, May—MAM) and summer (June, July, August—JJA). We define autumn as September, October, November, and December (SOND) and winter as January and February (JF). This choice recognized that the December atmospheric response in our experiments is much more similar to the autumn months than to January and February. In addition, since the experimental setup uses annually repeating sea ice and SST lower boundary conditions, JF and SOND represent time periods forced by continuous sea ice/SST data without an artificial break in lower boundary forcing at the end of the calendar year as would be the case for seasons defined by SON and DJF. This break at the end of the calendar year does represent a step change in forcing from December to January (this change is evident in Figure 1 comparing panel (b) to panel (e) which shows the sea ice difference for JF and SOND, respectively). In previous modelling work, it was found that atmosphere only models retain a ‘memory’ of the model initial conditions for roughly 1 week (Cassano, unpublished results). Therefore the atmosphere responds quickly to the change in lower boundary conditions that occurs from December to January.

To further relate the atmospheric response to the anomalous open water, correlation coefficients were calculated for the differences in surface variables discussed in this section and in the sea ice between the two experiments. The calculation was performed using all grid points north of 70°N that met or exceeded the threshold of 50% open water.

3.1. Atmospheric circulation

Annual mean SLP shows negative pressure differences (lower pressure in 2007) centred over the Beaufort and Chukchi Seas corresponding to the largest area of anomalous open water in 2007 (Figure 3(a)). These differences extend over a large area including Alaska and northwest Canada and extend northwest from the western Arctic to the anomalous open water in the Barents and Kara Seas. Positive SLP differences (higher pressure in 2007) are centred over the Sea of Okhotsk and also over the North Atlantic into northern Europe and western Siberia. However, none of these differences are statistically significant. During JF (Figure 3(b)), lower pressures in 2007 are found over the western United States as well as over much of Siberia with the strongest negative differences centred over the Barents Sea. Positive differences are centred over the Gulf of Alaska and statistically significant differences over the British Isles. The positive differences over the British Isles occur primarily in January while in February they are shifted to the southwest. The strong negative differences over the Barents Sea area are primarily present in February (not shown).

Figure 3.

Sea level pressure differences for the 2007 sea ice experiment minus the control experiment for the (a) annual mean, (b) JF, (c) MAM, (d) JJA, (e), SOND and (f) September. Cool (warm) colours are lower (higher) pressure for the 2007 sea ice experiment. Negative values are further indicated by crosshatching. The long (short) dashed line represents statistically significant differences at the 90% (95%) confidence level as described in the text. Units are hPa.

During MAM (Figure 3(c)), higher pressure is simulated for the 2007 sea ice experiment over much of the Arctic, with the largest and statistically significant pressure differences along the Barents and Kara Seas coasts. Lower pressure for 2007 is centred over Alaska eastward into the Yukon province of Canada with positive differences to the south in British Columbia. Negative pressure differences are also simulated in Europe south of Scandinavia and statistically significant differences along the northeastern coast of North America.

The smallest differences between the two model experiments occur in JJA (Figure 3(d)). Negative pressure differences extend over Alaska north to over the centre of Greenland and also over Scandinavia. Positive pressure differences are found over northeastern North America and over the Barents and Kara Seas. Throughout the summer, the low pressure signature for 2007 in the western Arctic becomes more prominent as the anomalous open water area grows.

The differences in circulation between the two experiments are considerably larger in SOND than during JJA (Figure 3(e)). Negative and statistically significant pressure differences dominate much of the Arctic. Positive differences extend from northeastern North America across the North Atlantic into Europe and Siberia. In September, the month with the greatest amount of open water, the response is higher pressure over Alaska extending west across Siberia into Scandinavia (Figure 3(f)). Negative pressure differences are found over Greenland extending north into the central Arctic. The response in October and November is lower pressure over much of the Arctic with higher pressure in northeastern North America eastward across to Europe and Siberia (not shown). In December, strong negative pressure differences are located over the Bering Strait extending east across Canada with higher pressure over much of Europe and Siberia (not shown).

Correlation analysis on an annual basis and for the four seasons shows the relationship for the circulation response to the sea ice differences between the two experiments (Table 1, column 1). The correlations are weak for JF, MAM, and JJA, but there is a strong correlation on an annual basis (0.62) and during SOND (0.60).

Table 1. Correlation coefficients for the differences in sea ice and the experimental atmospheric response as described in the text
SeasonSLPLatent heat fluxSensible heat flux2 m Temperature
Annual0.62−0.89−0.75−0.85
JF0.28−0.94−0.79−0.67
MAM−0.03−0.94−0.81−0.78
JJA0.13−0.40−0.54−0.51
SOND0.60−0.95−0.84−0.93

During autumn (SOND), circulation differences are similar between those at the surface and those aloft. One significant difference is that strong low SLP in the Beaufort and Chukchi Seas for 2007 (Figure 3(e)) is much less pronounced aloft (Figure 4 (a) and (b)), with the lowest heights aloft for 2007 located near the Aleutian Islands in Alaska and over Greenland. This is most apparent in October (not shown) for which lower tropospheric heights are simulated in 2007 in the western Arctic, but not the strong lower pressure that is simulated at the surface.

Figure 4.

Geopotential height differences at (a) 500 hPa, and (b) 300 hPa for SOND. Units are meters (m). Negative values are indicated by crosshatching.

3.2. Surface temperature and surface energy fluxes

Positive differences in annual mean 2 m temperatures of up to 2.5 °C are found in the western Arctic in the area with the greatest amount of anomalous open water and increased SSTs (Figure 5(a)). Higher temperatures in 2007 are also found in the Kara and Barents Seas where anomalous open water was also present in 2007. Lower temperatures in 2007 are simulated over the Sea of Okhotsk, over the Canadian Arctic Archipelago, and along the northeastern coast of Greenland. On a seasonal basis, during JF there are large areas of statistically significant lower temperatures in 2007 particularly over the Sea of Okhotsk (Figure 5(b)), and much of Canada, northeastern Siberia, and the northeast coast of Greenland.

Figure 5.

2 m temperature differences for the 2007 sea ice experiment minus the control experiment for the (a) annual mean, (b) JF, (c) MAM, (d) August, and (e) SOND. Cool (warm) colours are cooler (warmer) temperatures for the 2007 sea ice experiment. Negative values are further indicated by crosshatching. The long (short) dashed line represents statistically significant differences at the 90% (95%) confidence level as described in the text. Units are °C.

The temperature response, though statistically significant, is much weaker in MAM (Figure 5(c)). There are negative differences over the Canadian Archipelago as well as some areas off the northeast coast of Siberia. There is also an area of cooler conditions (around 3 °C) off of northeastern Greenland but positive differences of about the same magnitude in the Barents and Kara Seas. The negative differences in the western Arctic are most apparent in late spring (April and to a larger extent May). The contrasting pattern of cooler/warmer conditions off the northeast coast of Greenland and in the Barents and Kara Seas is present in all three spring months and corresponds well with the anomalous sea ice conditions in this region (i.e. less open water off the northeast coast of Greenland but more open water in the Barents and Kara Seas).

The warmer conditions in 2007 in the western Arctic begin to be apparent during JJA, although in general, averaged over the entire season, the temperature signal is weak. Higher statistically significant temperatures in the western Arctic in 2007 become more prevalent as the season progresses becoming strongest in August (Figure 5(d)). Higher temperatures of up to 3 °C are simulated for this month in the western Arctic with similar responses in this area for July. In June, there is little temperature response in the western Arctic, and there are lower temperatures in north-central Siberia of up to 2 °C.

During SOND the warm signal in 2007 in the western Arctic becomes much larger and stronger (up to 10 °C, Figure 4(e)). These statistically significant positive differences extend across the west central Arctic to the Kara and Barents Seas. This warm signal is quite prominent throughout all four autumn months; up to 6 °C in September and 10 °C in October, November, and December. There are some statistically significant negative temperature differences along the northern and northeastern coasts of Greenland (corresponding with an area of increased sea ice coverage in this area).

These positive surface temperature differences are associated with large surface latent and sensible heat fluxes to the atmosphere over the areas of anomalous open water present in 2007 (Figure 6(a) and (b)). A statistically significant increase of up to 10 W/m2 in the latent heat flux and 15 W/m2 in sensible heat flux is simulated in the area of anomalous open water and increased SSTs in the western Arctic in the Beaufort and Chukchi Seas. There are also large differences in surface latent and sensible heat fluxes linked to anomalous open water in the Barents and Kara Seas. The larger surface fluxes in the Beaufort and Chukchi Seas are most apparent during the late summer, autumn, and early winter—a small signal begins to be apparent in July and persists through December. Positive statistically significant differences of up to 30 W/m2 in the latent heat flux are simulated for October and November (Figure 7(a) and (b)). The largest signal in the sensible heat flux also occurs in October (statistically significant differences up to 30 W/m2) and November (statistically significant differences up to 70 W/m2) (Figure 8(a) and (b)).

Figure 6.

Annual mean difference in the (a) latent and (b) sensible heat fluxes at the surface for the 2007 sea ice experiment minus the control experiment. Cool (warm) colours are negative (positive) heat fluxes to the atmosphere for the 2007 sea ice experiment. Negative values are further indicated by crosshatching. The long (short) dashed line represents statistically significant differences at the 90% (95%) confidence level as described in the text. Units are W/m2.

Figure 7.

Differences in the surface latent heat flux for the 2007 sea ice experiment minus the control experiment for (a) October and (b) November. Cool (warm) colours are negative (positive) heat fluxes to the atmosphere for the 2007 sea ice experiment. Negative values are further indicated by crosshatching. The long (short) dashed line represents statistically significant differences at the 90% (95%) confidence level as described in the text. Units are W/m2.

Figure 8.

Differences in the surface sensible heat flux differences for the 2007 sea ice experiment minus the control experiment for (a) October and (b) November. Cool (warm) colours are negative (positive) heat fluxes to the atmosphere for the 2007 sea ice experiment. Negative values are further indicated by crosshatching. The long (short) dashed line represents statistically significant differences at the 90% (95%) confidence level as described in the text. Units are W/m2.

Correlation analysis as described above relating the spatial relationship between the sea ice and SST differences and the differences in surface temperature and energy fluxes show a strong relationship between the anomalous open water and the heating of the near-surface atmosphere (Table 1, columns 2–4). For 2 m temperatures, correlations range from −0.51 during JJA to −0.93 during SOND, demonstrating the strong heating is associated with the anomalous open water. This is also the case for the sensible heat flux (correlations ranging from −0.54 for JJA to −0.84 for SOND) and the latent heat flux (correlations ranging from −0.40 during JJA to −0.95 during SOND).

4. Comparisons with observations

Our experiments with CAM3 point to a strong atmospheric response to the observed 2007 sea ice anomalies. To gain further insight, it is useful to compare the model simulated circulation to what was observed in 2007. Such a comparison allows us to identify what portion of the observed circulation in 2007 was a forced response to the sea ice during that year. This will indicate if the observed circulation pattern for summer 2007, which favoured sea ice loss (Stroeve et al., 2008), was enhanced or diminished due to the forced sea ice response. Differences in the simulated circulation between the 2007 sea ice experiment minus the climatological sea ice experiment were compared with the 2007 anomalies with respect to the 1950–2001 mean (chosen to correspond with the time period of the climatological sea ice) from the National Center for Environmental Prediction/National Center for Atmospheric Research (NCEP/NCAR) Reanalysis (Kalnay et al., 1996). As discussed, the most prominent feature of the observed average annual circulation was an anomalous high pressure centre in the Beaufort Sea, which was present in JJA and SOND. For the model experiment, the pressure was lower in the western Arctic in the 2007 experiment, and this negative difference extended over much of the central Arctic and Greenland.

For JF, there are several interesting similarities between the simulated circulation differences and the observed anomalies. In the model experiment, negative differences are simulated over much of the Arctic with centres over eastern Siberia and the Barents and Kara Seas (Figure 9, top left panel). The reanalysis data show negative pressure anomalies over much of the Arctic as well as over much of Europe, Russia south of the Kara Sea, and off the east coast of Canada (Figure 9, top right panel). The most notable contrast between the simulated SLP differences and the observed anomalies is over the British Isles in JF; while there are positive differences in the simulations, no such feature is present in the observations.

Figure 9.

Differences in the sea level pressure (SLP) for the 2007 sea ice experiment minus the control experiment (left panels) and observed SLP climatological anomalies (right panels) for JF (top panels) and SOND (bottom panels). Units are hPa. Negative values are indicated by crosshatching.

The observed anomalies and simulated circulation differences share some features in SOND, but there are also substantial differences (Figure 9, bottom two panels). For the model experiments (left panel), negative pressure differences cover much of the Arctic, with positive differences over much of Europe and extending across Siberia. This was largely the case for the observed anomaly (right panel) except for an anomalous high pressure zone in the Beaufort Sea.

For MAM, the observed circulation anomaly pattern is broadly opposite of the simulated SLP difference pattern (Figure 10; top two panels). In the experiment (top left panel), 2007 features positive pressure differences with respect to the situation with climatological ice over much of the central Arctic into northern Europe and over Greenland. Negative differences cover much of Alaska and northeastern North America. The observed circulation in 2007 (top right panel) for MAM features anomalous low pressure over much of the central Arctic and in the Gulf of Alaska. High pressure anomalies encompassed the Bering Sea area and much of Canada.

Figure 10.

Differences in the sea level pressure (SLP) for the 2007 sea ice experiment minus the control experiment (left panels) and observed SLP climatological anomalies (right panels) for MAM (top panels) and JJA (bottom panels). Units are hPa. Negative values are indicated by crosshatching.

Turning to JJA (Figure 10; bottom two panels), the 2007 simulation (lower left panel) has lower pressure over Alaska, part of the central Arctic Ocean, Greenland, and Scandinavia with higher pressure over northeastern North America and in the Kara and Barents Seas region compared to the climatological simulation. The observed circulation (lower right panel), which as discussed earlier was a strong driver of the 2007 sea ice anomaly, features anomalously high pressure centred over the Beaufort Sea and Greenland with negative anomalies along northern Siberia.

For MAM, the simulated circulation differences are nearly opposite to what was observed during 2007. The simulated circulation differences and observed anomalies were also quite different in JJA, particularly in the western Arctic. This suggests the atmospheric response to the sea ice conditions preceding (i.e. spring of 2007) and during the summer of 2007 was not responsible for creating the anomalous atmospheric circulation pattern that was conducive to the large observed sea ice loss. In SOND and JF, the patterns of simulated circulation differences and observed anomalies are more similar. The primary exception is that the simulations do not show high pressure in the Beaufort Sea that was particularly prominent during SOND. The similarities between what were simulated and observed in SOND and JF suggest that the forced atmospheric response to reduced sea ice was in part responsible for the observed atmospheric circulation anomalies during the autumn and winter of 2007.

5. Summary and conclusions

Using CAM3 we investigated the atmospheric response to the anomalous sea ice conditions observed in 2007 with two 30-year experiments: one using a repeating 2007 annual cycle and other using a repeating climatological annual cycle of sea ice. Therefore the primary difference between the two experiments was the sea ice concentration and extent (with some changes to the climatological SSTs to ensure physical consistency in the lower boundary conditions). Results showed a significant response to the anomalous open water observed in 2007. The circulation response was most prominent during SOND and JF with negative SLP differences simulated for 2007 compared to the control simulations covering much of the central Arctic, attended by tropospheric pressure height differences extending up to at least 300 hPa. The circulation response during MAM was positive pressure differences over much of the Arctic while the response was weak during JJA.

Circulation anomalies extended throughout the troposphere. Low SLP and anomalously warm surface temperatures are collocated with the anomalous open water in the western Arctic, directly forced by the absence of sea ice. In this same location at 500 hPa, the lowest geopotential heights are not located over the same region as the lowest SLP. This signature of weaker geopotential heights aloft is consistent with the formation of a warm core surface low that is forced by the increased heating to the atmosphere from the surface.

Differences in surface temperature and heat fluxes were most pronounced during SOND over the areas of anomalous open water and increased SSTs in the 2007 experiment in the western Arctic. Positive temperature differences of up to 10 °C were simulated for late autumn and early winter (i.e. October through December) in the western Arctic. In October and November, there are positive differences in latent and sensible heat fluxes to the atmosphere of up to 30 and 70 W/m2, respectively.

Comparisons were made between the simulated circulation response and observed anomalies for 2007. For spring and summer, the period preceding the largest open water anomalies, the simulated circulation response was quite different to what was observed and in spring was nearly the opposite of what was observed. For autumn and winter, the simulated circulation response was more similar than during spring and summer to the observed anomaly structure with the exception that the simulations did not show the high pressure anomaly observed during autumn over the Beaufort Sea. These results suggest the sea ice conditions preceding and during the summer of 2007 (i.e. spring and summer of 2007) were not responsible for creating the anomalous atmospheric circulation pattern known to have favoured summer sea ice loss. The similarities between the observed anomalies and simulated circulation differences in SOND and JF suggest that the forced atmospheric response to reduced sea ice was in part responsible for the observed atmospheric circulation anomalies during the fall and winter of 2007.

In summary, our experiments showed that the atmospheric response to increased open water in the Arctic was significant. This atmospheric response could have implications beyond the Arctic such as a decrease in the pole to equator temperature gradient (given the increased temperatures associated with the increase in open water, e.g. Francis and Vavrus, 2012), leading to a weaker jet stream and less storminess in the mid-latitudes (Sinclair and Watterson, 1999; Francis et al., 2009; Bader et al., 2011).

Acknowledgements

This work was funded under NSF grants ARC-0629412, ARC-0805821, and ARC-0901962. The authors thank two anonymous reviewers for their helpful comments and suggestions for improving the manuscript.

Ancillary