Atmospheric response to the extreme Arctic sea ice conditions in 2007


  • Jonas Blüthgen,

    1. Earth and Space Sciences, Jacobs University Bremen, Bremen, Germany
    2. Now at Centre for Ocean and Ice, Danish Meteorological Institute, Copenhagen, Denmark
    Search for more papers by this author
  • Rüdiger Gerdes,

    1. Earth and Space Sciences, Jacobs University Bremen, Bremen, Germany
    2. Climate Sciences/Sea Ice Physics, Alfred Wegener Institute for Polar and Marine Research, Bremerhaven, Germany
    Search for more papers by this author
  • Martin Werner

    1. Climate Sciences/Sea Ice Physics, Alfred Wegener Institute for Polar and Marine Research, Bremerhaven, Germany
    Search for more papers by this author


[1] An atmospheric general circulation model driven with the observed 2007 extreme Arctic sea surface temperatures and sea ice concentrations responds with higher surface air temperature over northern Siberia and the Eastern Arctic Ocean (+3 K), increased heat uptake of the ocean in summer (+40 W m−2) and increased oceanic heat losses in fall (−60 W m−2) compared to a climatological scenario. A pronounced low sea level pressure anomaly over the Eastern Arctic (−200 Pa) reinforces a sea level pressure dipole over the Arctic that has been observed to become an increasingly important feature of the Arctic atmospheric circulation in summer. The anomalous pressure distribution contributes to sea ice transport from the Eastern Arctic and is likely to reinforce the original sea ice extent anomaly. The results thus support assessments of observational data over recent years that sea ice loss may feed back onto the atmospheric circulation in the northern hemisphere. The resulting late summer / early fall (JAS) atmospheric anomalies are very robust, they appear in virtually all of the 40 realizations of the experiment. However, we find no significant continuation of the atmospheric signal into the winter as has been suggested based on atmospheric observational data.

1. Introduction

[2] Arctic summer sea ice extent has experienced a long-term decline, the linear trend reaching a rate of 11.5% loss per decade (NSIDC Arctic Sea Ice News, available from In September 2007, the sea ice extent fell to the record low of 4.3 million km2. This is 2.6 million km2lower than the September mean for the period 1979–2000. Much of the eastern Arctic, the Chukchi Sea, and the Beaufort Sea were ice-free in late summer. The large area of open ocean area in 2007 allowed more solar radiation to be absorbed by the ice-free ocean mixed layer. Peak ocean surface temperature anomalies above +3.5 K were measured in the ice free eastern Arctic (0°E–180°E) [Steele et al., 2008].

[3] Among the mechanisms that have been proposed to explain the recent sea ice retreat, the anomalous sea level pressure (SLP) pattern of low pressure over Siberia and high pressure over northern Canada prevailing during the summer of 2007 seems to be of special importance [Ogi and Wallace, 2007; Wang et al., 2009; Overland and Wang, 2010]. It was responsible for transferring anomalous amounts of warm air from the Pacific to the Arctic and at the same time driving sea ice from the eastern Arctic to the Western Arctic and thus dynamically generating large sea ice-free regions in the eastern Arctic. An increasing magnitude of the Arctic atmospheric Dipole Anomaly (DA) in recent summers has been found byOverland et al. [2008], and Wang et al. [2009] while Zhang et al. [2008b]find the DA pattern becoming the dominating winter SLP pattern over the Arctic since the beginning of the century. Enhanced advection of warm Pacific Water associated with the meridional pressure gradient could also have contributed to the ice-free conditions in the Beaufort Sea [Zhang et al., 2008a]. The continuing reduction of sea ice volume in the Arctic [Kwok and Rothrock, 2009; Farrell et al., 2009; Haas et al., 2010] has enhanced the seasonal cycle of sea ice extent [Kauker et al., 2009].

[4] The sea ice and sea surface temperature (SST) anomalies of 2007 represent huge changes in the surface boundary conditions for the atmosphere compared to long-term mean conditions. It is of interest how the atmosphere responds to such perturbations and whether any feedbacks between the atmospheric response and the ocean-sea ice conditions exist.

[5] Response experiments with comprehensive atmospheric general circulation models (AGCMs) have been undertaken for high northern-latitude sea ice and SST winter anomalies [e.g.,Deser et al., 2004; Alexander et al., 2004; Gerdes, 2006]. These experiments were mostly aimed at identifying the response to extreme sea ice conditions associated with high positive phases of the North Atlantic Oscillation (NAO). Conclusions regarding feedbacks between ocean-sea ice state and the atmospheric circulation were mixed.Gerdes [2006] found some indication of a positive feedback involving sea ice thickness anomalies. In that case, the atmospheric response was significant in its main features, however, individual years of the ensemble could differ vastly from the ensemble mean.

[6] Deser et al. [2010]studied the atmospheric response to late 21st-century (2080–2099) projected Arctic sea ice loss in an AGCM, covering the complete seasonal cycle. Their setup is closer to ours than one would expect from future-climate scenario studies, as they did not change additional factors, such as CO2levels. However, in contrast to our setup, they kept SST at 20th-century (1980–1999) levels and included sea ice thickness changes, while thickness is set to constant 2 m in our experiments. Responses in atmospheric circulation were confined to the cold season (October–April).

[7] Few results from corresponding response experiments considering summer sea ice anomalies have so far been published [Bhatt et al., 2008; Honda et al., 2009; Balmaseda et al., 2010]. The summer atmosphere was often considered too noisy and devoid of large-scale, persistent patterns as to warrant such studies [Singarayer et al., 2006; Deser et al., 2010]. The lack of significant summer pressure patterns compared to the prominent winter pattern of the Arctic Oscillation (AO) was taken as indicative of a dominance of internally generated random variability. In experiments with prescribed sea ice concentration or sea ice extent anomalies from the low sea ice summer 1995, Bhatt et al. [2008] found no significant response in SLP or dynamic height at 500 hPa and 200 hPa over the Arctic. Honda et al. [2009] described a response of the atmosphere to Arctic fall sea ice anomalies typical for the previous record year 2005 in a number of individual simulations but only a weak response when considering the ensemble average over all 50 model experiments. Balmaseda et al. [2010]forced an atmospheric GCM with 2007 and 2008 daily analyzed sea ice cover and observed daily SST fields. For both years, they found a positive anomaly in the July–August 500-hPa geopotential height over the central Arctic and a negative anomaly belt at mid-latitudes. The anomaly projects onto the positive summer AO pattern, which was associated with lower cloudiness and increased downwelling shortwave radiation [Kay et al., 2008] and possibly indicates a positive feedback through sea ice export [Ogi and Wallace, 2007]. The strengthening of the zonal winds in summer contrasts with the weakening of zonal winds in fall derived from observations by Overland and Wang [2010].

[8] Here we report on experiments with the atmospheric GCM ECHAM5 [Roeckner et al., 2003], forced with monthly sea ice concentration and SST data for the year 2007. We concentrate on the late summer and early fall (JAS) atmospheric response to the SST and sea ice anomalies, which is highly significant and robust. The results indicate a potential positive feedback mechanism that hints at processes in addition to the well-known temperature-albedo feedback that may contribute to the observed rapid decline of Arctic sea ice cover.

2. Experimental Design

[9] Three ECHAM5 experiments with annually repeating surface boundary conditions were performed: (1) A control run (Clim) with sea ice concentration (SIC) and SST from the climatological (1979–1996) annual cycle of global SST and SIC of the Atmospheric Model Intercomparison Project (AMIP II) [Taylor et al., 2000]; (2) a simulation (Arctic07) with SST and SIC for January through December 2007 taken from the HadISST1 dataset [Rayner et al., 2003] north of 60°N, AMIP II climatology SST and SIC south of 40°N, and a linear interpolation of both datasets between 40°N and 60°N; and (3) a simulation (Global07) with global SST and SIC from the HadISST1 dataset for 2007. In ECHAM5, a constant sea ice thickness of 2 m is assumed for the ice covered parts of any Northern Hemisphere grid cell. We focus the analyses on the results of the Clim and Artic07 experiments, using Global07 results only to clarify the possible influence of non-Arctic SST and SIC anomalies.

[10] A T63 grid resolution (96 latitudes × 192 longitudes) was chosen for all three ECHAM5 simulations, and the vertical resolution chosen was 31 levels with a common terrain-following sigma coordinate system.

[11] All prescribed SST and SIC boundary fields were monthly mean values and linearly interpolated to the exact model time. The interpolation of the HadISST1 oceanic data onto the ECHAM5 T63 grid resulted in a few grid points with missing values, which were not land according to the ECHAM5 land-sea mask. Such grid points were filled with the corresponding values from the AMIPII SST and SIC climatology.

[12] All other boundary conditions (e.g., vegetation cover distribution, atmospheric greenhouse gas concentrations) were set to identical present-day values in all three ECHAM5 experiments.

[13] The three experiments were conducted for 40 years after initial spin-up. As atmospheric processes have generally a short persistence time, we regard the total simulation period as an ensemble of 40 independent annual cycles.

3. Results

[14] The results of the experiments are presented as differences between the ensemble means of the experiments with 2007 boundary conditions and the control run (Arctic07-Clim). An ANOVA approach (Analysis of Variance) [Von Storch and Zwiers, 1999] is used to detect those differences that significantly differ from zero and thus stand out from the natural fluctuations of the atmosphere. Substantial differences between the Arctic07 and Clim experiments are found from July through October, corresponding to the time when the differences in forcing were strongest.

[15] Net surface heat flux is computed as the sum of sensible heat flux, latent heat flux, net surface solar radiation, and net surface thermal radiation. Net surface heat flux anomalies change sign from summer to fall. In July and August (Figure 1a, August only) there is an anomalous heat gain by the ocean in Arctic07 of up to +40 W m−2in the East Siberian Sea and the Beaufort Sea. During this phase, the ice-free ocean gains heat mostly from the anomalous net solar radiation. Heat is stored in the oceanic mixed layer and released later in the year when the incoming solar radiation is no longer sufficient to counterbalance losses through long wave radiation, latent and sensible heat flux. In September and October (Figure 1b, October only), the warm ocean loses heat to the atmosphere in basically the same (newly ice-free) areas where anomalous uptake took place earlier in the year. Starting in September and peaking in October, the net surface heat flux contains a distinct negative anomaly between the East Siberian Sea and the pole of around −60 Wm−2 (Figure 1b), while control values in Clim are rather neutral or slightly negative (not shown) due to ice cover in these regions.

Figure 1.

Ensemble-mean net surface heat flux anomalies Arctic07-Clim for (a) August and (b) October. Positive absolute values indicate downward fluxes, i.e. an energy gain for the ocean. The masks to the right show the corresponding ANOVA result: Grey-shaded points indicate F-values exceeding the critical value (4.00 for 95% significance) where the local anomaly can be considered significantly different from zero.

[16] Near-surface air temperature (SAT) is strongly coupled to the prescribed SST values. For the Arctic07 experiment, significant anomalies are found over all ocean regions. In July, August, and September (Figure 2), SAT exceeds the corresponding values in the control run by up to +3 K over the Chukchi Sea and positive anomalies above +1.5 K cover most of the Arctic between 120°E and 90°W. The simulated SAT anomalies over ice-free areas are slightly lower in summer 2007 than corresponding NCEP SAT-anomalies (+3–6 K).

Figure 2.

(left) Ensemble-mean 2 m air temperature anomalies Arctic07-Clim averaged over the summer months (JAS) and (right) the ANOVA result, according to which the anomalies over the whole Arctic Ocean are significantly different from zero.

[17] The response in SLP (Figure 3) to the Arctic 2007 SST and SIC conditions consists mainly of a negative anomaly exceeding −200 Pa that covers the whole eastern Arctic. Between July and October, the center of the low pressure anomaly shifts longitudinally between the East Siberian Sea and the Kara Sea. The low pressure anomaly is statistically significant while the positive pressure anomalies surrounding the high Arctic are not significant according to our metric. A response consistent with the SLP anomalies can be found at higher levels in the atmosphere. Compared to the control run Clim, geopotential surfaces z850 and z500 are depressed over the eastern Arctic in Arctic07 by −25 m and −40 m, respectively (not shown). However, these anomalies fade earlier in the year than the SLP anomalies.

Figure 3.

Ensemble-mean SLP anomalies Arctic07-Clim averaged over the summer months (JAS) and corresponding ANOVA mask.

4. Discussion

[18] As substantial internal atmospheric variations over the Arctic region exist, it is essential to assess the robustness of the response to changes in the boundary conditions. To this end, we perform ANOVA significance tests. Furthermore, we split the total ensemble of 40 realizations into randomly selected sub-ensembles of 10 members each. The deviations of the sub-ensemble means from the Clim control run are very similar to the difference between the overall Arctic07 ensemble means and the Clim values, indicating a rather robust response of the near-surface atmospheric circulation to the 2007 anomalies in sea ice concentration and SST.

[19] Almost all individual realizations show a negative SLP anomaly over the Eastern Arctic in July, August, and September (Figure 4). When early summer SIC and SST anomalies are detected in the Eastern Arctic, a low SLP anomaly can be expected over the area for late summer and early fall. In our experiment, such anomalies occur with a probability of 73%, 68%, and 80% for JAS, respectively, and 87.5% when averaging over the three months.

Figure 4.

Difference between regional JAS mean SLP of the individual ensemble realizations of experiment Artic07 and regional JAS ensemble mean SLP of the control experiment Clim [Pa] for the Eastern Arctic (30°E to 210°E, 70°N to 90°N; shaded area in right plot).

[20] The persistence of the atmospheric response from summer into early fall is remarkable when considering the seasonal differences in atmospheric boundary layer stability. In July, high near-surface stability limits turbulent heat flux, which breaks up in September, permitting larger heat fluxes and atmospheric responses to oceanic changes. This pattern was also found for 2007 in 1-day observationally constrained simulations with the Community Atmospheric Model (CAM4) [Kay et al., 2011]. While the boundary layer was thicker (up to +89 m) and the near-surface inversion strength lower in both July and September 2007 compared to the control, the difference was stronger in September, causing an intensification of the seasonal transition [Kay et al., 2011]. Using the high-resolution Weather Research and Forecast model (WRF),Strey et al. [2010] found that a positive surface air temperature anomaly above the Arctic Ocean was most pronounced in October (up to −8 K). In contrast, the dynamic atmospheric response is not as persistent later in the year (SON) in our simulations. As the experimental design of the study by Strey et al. [2010] and our work differs substantially, it will require further analyses to explain the differences both studies. For example, Strey et al. [2010] altered only sea ice concentration in their simulations, while we changed both sea ice and SST.

[21] In a fully coupled system, the SLP anomaly forced by reduced sea ice cover and higher SSTs in the eastern Arctic feeds back on sea ice and ocean. Low SLP over Siberia and the eastern Arctic leads to a sea ice export from the eastern Arctic and an accumulation in the western Arctic. This is a process that is believed to be important for the generation of the 2007 sea ice anomaly [Zhang et al., 2008b; Wang et al., 2009]. The sea ice transport associated with the surface wind anomalies constitutes a positive feedback as it reinforces the primary sea ice concentration anomaly. While the main SLP response in our experiment is robust only until September, we would expect the feedback to prolong both sea ice reduction and negative SLP anomalies over the eastern Arctic later into fall.

[22] Our results agree with the analysis of observed SIC and atmospheric reanalysis of Francis et al. [2009] who find low sea ice extent in September associated with increased 500 hPa height and higher SAT (+3 K) during winter (NDJ) following a summer with low Arctic sea ice extent. Francis et al.'s analysis is restricted to the period of satellite passive microwave SIC observations from 1979 through 2006 and thus does not reflect the strong summer SIC anomalies of 2007 and 2008.

[23] Evaluating reanalysis data for the low-sea-ice years 2002–2008,Overland and Wang [2010] could not in each year establish a direct correspondence between low sea ice extent in summer and the SLP field in autumn (OND). It is thus likely that the specific pattern and especially the intensity of the 2007 anomaly are necessary for a significant SLP response. Overland and Wang also derive an Arctic dipole pattern in the 1000–500 hPa layer thickness that persists into late fall following low sea ice extent in September. As mentioned above, such a persistence into late fall is only weakly present in our simulations.

[24] The mutual interaction between September sea ice extent and Arctic basin cyclones (and thus the AD pattern) was stressed by Simmonds and Keay [2009], based on reanalysis and satellite sea ice concentration data for 1979–2008. Septembers with low sea ice extent and relatively warm ocean mixed layers later into fall are associated with increased energy supply to existing cyclonic systems through surface latent heat flux. The cyclones themselves disperse sea ice and generate open water areas [Schröder, 2005], maintaining the heat supply.

[25] Balmaseda et al. [2010]performed experiments similar to ours, forcing the ECMWF operational seasonal forecasting system with SIC and SST for the extreme Arctic summer sea ice years 2007 and 2008. Resulting July–August atmospheric anomalies strongly project onto the summer Arctic Oscillation pattern with stronger mid-latitude zonal winds. This differs from our results and from the observed relationships discussed above for fall, but also from our results for July and August. An explanation for the different atmospheric responses is offered by Balmaseda et al., who state that the atmospheric response to SIC anomalies depends on the mean state of the atmosphere, which in turn is governed by the large-scale SST field. While we used climatological SST and SIC in the control experiment, they changed only the sea ice condition, using observed 2007/2008 SST also in the control experiment. Another possible source of differences is the short integration period of the experiments by Balmaseda et al. (May–August).

[26] We find rather similar responses in the Arctic, whether we use SST anomalies that are confined to high northern latitudes or global SST anomalies for 2007. With global surface boundary conditions from 2007 (experiment Global07) the main atmospheric anomalies over the Arctic during summer are retained. In fall, a positive SLP anomaly is found over Bering Strait and the northeastern North Pacific that is not present in Arctic07 and thus not related to Arctic ocean-sea ice anomalies. A possible source of the signal is the equatorial Pacific, which shows a negative SST anomaly of up to −2 K throughout the year 2007. This cold phase of ENSO (La Niña mode) is associated with a negative phase of the Pacific North American pattern (PNA), which results in higher than average pressure above the Aleutian Islands [Straus and Shukla, 2002].

5. Conclusions

[27] Forcing the global atmospheric GCM ECHAM5 with observed SST and SIC for the extreme Arctic sea ice year 2007 yielded a very robust response in surface heat flux, SAT, SLP, and in the 500 hPa dynamic height field. The response in SLP and dynamic height is especially strong in July, August, and September (up to −200 Pa and −40 m, respectively). Significant signals in heat flux and SAT extend into December, powered by a still warm ocean mixed layer. The most pronounced feature of the SLP response is a low-pressure anomaly over the eastern Arctic (0°E–180°E). The anomaly resembles part of the anomalous Arctic SLP dipole pattern, which contributed to the extreme summer sea ice anomalies of 2007 and 2008 [e.g.,Wang et al., 2009]. In an adjoint sensitivity analysis for the September sea ice extent in 2007, Kauker et al. [2009] found that SLP in May and June contributed most strongly to the sea ice extent anomaly in September. Here, we find that the SIC and SST anomaly associated with the sea ice retreat themselves force a similar atmospheric response later in the year (JAS) that thus is suitable to prolong the SIC anomaly by promoting sea ice redistribution from the eastern to the western Arctic.

[28] With these numerical experiments we confirm the prolongation of existing anomalous atmospheric conditions into fall that has been proposed by Francis et al. [2009]. However, a significant dynamic response is only detected until September. This might be due to model biases, lack of feedback with the ocean and sea ice, or simply an increase of internal atmospheric variability starting in mid- to late fall. The latter can be detected in our simulations as an approx. twofold increase of the calculated standard deviation of the East Arctic index between September and November (not shown).

[29] The atmospheric model results and the indication of positive feedbacks between sea ice anomaly and atmospheric circulation is intriguing as it might provide an explanation for the recent eastward migration of the low pressure center from Iceland into the Siberian Arctic [Zhang et al., 2008b] and the prospect of predictability based anomalies in sea ice conditions. Clearly, there is a need to investigate these relationships in high-resolution coupled models that include additional feedbacks.


[30] The authors wish to thank Frank Kauker for his help with the ANOVA implementation, Christof Lüpkes and two reviewers for their invaluable comments, and the Geophysical Research Institute (GFI) in Bergen for support during the writing of this paper.

[31] The Editor thanks the anonymous reviewers for their assistance in evaluating this paper.