The 2007 sea ice minimum: Impacts on the Northern Hemisphere atmosphere in late autumn and early winter

Authors


Abstract

[1] Over the past several decades, the minimum Northern Hemisphere summer sea ice extent has decreased substantially. We present an analysis of the influence of declining Arctic sea ice cover on the atmosphere, specifically during the autumn/early winter following an extreme summer minimum event. Using ensemble simulations from the Weather Research and Forecast model (v 3.0.1), we compare the atmospheric response for the case of the extreme sea ice minimum of 2007 to the corresponding response for the more typical ice conditions of 1984, the year with median ice extent for the 1979–2008 satellite era. Increased open water enhances heat and moisture flux from the Arctic Ocean to the atmosphere during autumn. We characterize the fluxes both horizontally and vertically and determine the spatial breadth of their influence. The atmospheric response is characterized by a strong increase in 2 m temperature and decrease in sea level pressure locally and by remote responses in the atmospheric circulation throughout the troposphere characterized by a quasi-barotropic ridge/trough signal in North America. The circulation anomalies drive remote anomalies of temperature and precipitation over eastern North America and the North Atlantic. Advectively driven temperature anomalies, in turn, cause surface flux anomalies over remote regions such as the Great Lakes and the Gulf Stream. The maximum response, as measured by difference in 2 m temperatures over the polar cap from 70°N, occurs between 10 September and approximately 15 November. The persistence of the signal over a 2 month period implies the potential for seasonal predictability of the stronger atmospheric response features. In addition, we determine the significance of prominent features, finding 95% significance in some remote features as far away as the North Atlantic.

1. Introduction

[2] The influence of the Great Lakes, with a surface area of over 243,000 km2, on the surrounding atmosphere is well known. In September of 2007, the Arctic sea ice cover reached a record minimum extent in the satellite era from 1979 to 2008. Due to the delayed freeze-up, the largest ice anomaly occurred in October, when more than 3,000,000 km2 of anomalous open water interacted with the atmosphere. This anomaly represents about twelve times the area of the Great Lakes.

[3] As winter set in, the sea ice began to grow back though negative anomalies persisted well into November, leaving relatively warm water exposed to the subfreezing atmosphere into the beginning of the cold season (Figure 1) – a stark contrast to the normal sea ice evolution, which increases rapidly at the onset of the cold season. This exposure of the atmosphere to the open water, like cold, dry air over the Great Lakes, allows heat and moisture transport to the atmosphere. It is well known that these fluxes alter local temperature and mass distributions thereby influencing local dynamics, but what are the remote effects of such large heat and moisture fluxes over such a large area? Many studies using global climate models (GCMs) project an eventual ice-free Arctic Ocean in the summer. How will the steady decline in Arctic sea ice impact atmospheric circulation on the planetary scale? These questions motivate this study.

Figure 1.

The 2005–2008 Northern Hemisphere sea ice area and anomaly in millions of km2. Observed sea ice area (black), 1979–2001 climatological mean (blue), and anomaly (red).

[4] Effects of sea ice loss on the atmosphere are direct and indirect. Direct effects, such as increased ocean-to-atmosphere latent heat fluxes, occur locally over the sea ice anomaly. The heating of the atmosphere redistributes its mass, altering the circulation. The altered circulation leads to indirect effects such as anomalous temperature advection and the distribution of precipitation.

[5] Many studies of the impacts of an anomalously small Arctic sea ice cover have been conducted. Some of these studies, including several with an ice-free Arctic Ocean, were performed decades ago with less technologically advanced models. Though such studies are important, the following summary emphasizes the most relevant, recent studies, most of which were performed with updated Global Climate Models (GCMs). Most previous studies focus either on summer (June/July/August (JJA)) or winter (December/January/February (DJF)) sea ice anomalies. For example, Herman and Johnson [1978] performed one of the first of such modeling studies to examine remote effects of the January and February sea ice extent. They found that anomalies in the sea ice margin in the Atlantic sector of the Arctic Ocean sea ice pack do alter climate in surrounding regions as well as other regions in middle and high latitudes.

[6] Through the use of a spectral atmosphere GCM developed at the Meteorological Research Institute of the Japan Meteorological Agency, Honda et al. [1999] found that anomalously low January/February sea ice extent in the Sea of Okhotsk leads to anomalous turbulent heat fluxes from the ocean surface, which thermally generates a stationary Rossby wave train eastward as far as the western United States. This signal may be overshadowed by the remote influence of other large-scale modes, such as the El Niño−Southern Oscillation signal, in the real atmosphere. Honda et al. [2009] used an atmosphere-ocean coupled GCM and found the remote response to reduced sea ice in early winter is characterized as a stationary thermally generated Rossby wave forced by surface fluxes due to sea ice loss.

[7] Later, Alexander et al. [2004] examined the winters (DJF) with the most (1982–1983) and least (1995–1996) Arctic sea ice cover during the satellite record from 1979 to 1999 using the atmospheric component of the Community Climate System Model (CCSM) version CCM 3.6. Using 50-member ensembles, they found that the atmospheric response to prescribed sea ice anomalies from these two years was locally robust and shallow. The results showed upward heat fluxes in excess of 100 W m−2, near surface warming, enhanced precipitation, and below normal sea level pressure where the ice extent was reduced. Where the extent was increased they found an opposite response. Remotely, the magnitude of the response was more modest than the local response, with sea level pressure (SLP) anomalies of 2–2.5 hPa and 500 hPa height anomalies of 15–20 m. Both concentration and extent anomalies were considered, and the response to concentration anomalies was approximately twice as large as to the extent anomalies. The large-scale response for the 1982–1983 positive anomaly experiment resembles the negative phase of the Arctic Oscillation/North Atlantic Oscillation (AO/NAO).

[8] Deser et al. [2004] discussed the direct and indirect response of the atmosphere to progressively larger prescribed sea ice cover anomalies in the North Atlantic and Greenland Sea. Simulations were made using CCM3. In the North Atlantic, a warm SST anomaly had a destabilizing effect and the response was approximately linearly proportional to the magnitude of the induced SST anomaly. In the Greenland Sea, the atmosphere had a higher mean static stability and the anomalous ridge in the free atmosphere did not penetrate as deeply with a cold SST anomaly. Similar to the direct response, the magnitude of the indirect response scaled linearly with the magnitude of the forcing but nonlinearly with the sign of the forcing. The indirect response was on the hemispheric scale and resembled the AO/Northern Annular mode. A companion study [Magnusdottir et al., 2004] focused on the storm track characteristics as well as the main features of the response.

[9] Using an unforced coupled atmosphere/ocean GCM (ECHO-G at T30 resolution), Sokolova et al. [2007] presented evidence that changes between positive and negative Arctic sea ice extent anomalies may influence Northern Hemisphere storm tracks and quasi-stationary planetary waves on synoptic to seasonal time scales. Using present-day initialization, they ran the model into the future 500 years, choosing the seven largest and seven smallest ice covers for comparative evaluation. With more sea ice the zonal wind increased north of 40°N and decreases southward. The difference between the zonal wind patterns they found for high and low ice phases resembles the positive AO patterns and suggests a link between high ice years and positive AO patterns.

[10] In August of 1995, the Arctic sea ice extent decreased to what was then the lowest extent in the satellite record. Bhatt et al. [2008] employed CCM3.6 to simulate April to October 1995 using climatological sea surface temperatures and prescribed ice extent or concentration from 1995 and from climatology based on monthly observations. The experiment included 51 ensemble members. Locally, the anomalously low ice extent induced surface heat fluxes to the atmosphere, warmer surface air temperatures, and a slight decrease in sea level pressure. Remotely, the atmospheric response was characterized by an anomalously high SLP in the North Pacific. With height, the response was equivalent barotropic, increased in amplitude, and was statistically significant up to 200 hPa.

[11] A recent study at the European Centre for Medium-Range Weather Forecasts [Balmaseda et al., 2009] compared many of the above studies including Alexander et al. [2004], Deser et al. [2004], and Bhatt et al. [2008] with 2007 and 2008 July/August/September simulations using the ECMWF model, run both with prescribed (2007, 2008) sea ice and coupled to a sea ice model. This study indicates that the atmospheric response to sea ice is highly dependent on the atmospheric mean state, including natural variability phenomena such as the NAM and ENSO. The study also showed that the atmospheric sensitivity to sea ice varied with the treatment of sea surface temperature.

[12] In addition to the many modeled studies, Francis et al. [2009] presented observational evidence that summer sea ice loss affects winter weather. They showed that increased ocean-to-atmosphere heat fluxes lead to a less stable and deeper boundary layer in the Arctic, with statistically significant temperature, sea level pressure, and precipitation anomalies throughout the Northern Hemisphere.

[13] The sea ice boundary and embedded areas of open water within sea ice are not well resolved by coarse-resolution global climate models. Thus, we have chosen to diverge from the methodologies of the aforementioned previous studies in the following ways: (1) We employ the higher-resolution regional Weather Research and Forecast System (WRF) Advanced Research WRF (ARW) rather than a GCM. (2) We focus on the autumn months of October and November, while other studies focus on DJF or JJA. (3) Our sea ice and sea surface temperatures are prescribed to observations, while other studies run a coupled ocean/atmosphere/ice model introducing more degrees of freedom.

[14] Such a controlled study allows for the isolation of the impact of the lower ice area on the atmosphere. Though we discuss the impact of the sea ice area on the atmosphere, we are unable to determine impacts of a given sea ice anomaly's location.

[15] We seek to quantitatively and qualitatively describe how increased ocean-to-atmosphere latent and sensible heat fluxes contribute to changes in large-scale circulation and subsequent changes in precipitation. Following a description of our methodology, we will discuss the locations and magnitudes of changes in ocean-to-atmosphere fluxes, changes in circulation, and finally changes in the distribution of precipitation. We also include Appendix A to depict the biases of our ensemble against two popularly cited reanalyses.

2. Methods

[16] In order to set up a controlled experiment comparing atmospheric conditions induced by sea ice of 2007 and a typical year, we chose to use an atmospheric model with prescribed boundary conditions. Because the selection of the model and physics packages may influence the results of the simulations, we discuss the chosen model and physics package options here.

[17] The fine horizontal and vertical resolution capabilities of the WRF model with the ARW core version 3.0.1 make it the optimal model for this study. WRF-ARW is a mesoscale model with fully compressible nonhydrostatic equations (with a hydrostatic option) developed from an Eulerian solver. The vertical coordinate is a terrain following hydrostatic pressure coordinate. The flux-form nonhydrostatic Euler equations, including moisture, are integrated using the σ coordinate along the Arakawa-C grid. More information about the model itself can be found in its documentation [Skamarock et al., 2008].

[18] Physics option selection, guided by the work of Seefeldt and Cassano [2008], targeted the ideal physics packages for the Arctic region to ensure that important surface fluxes and precipitation were optimally captured. Seefeldt and Cassano [2008] completed a series of simulations testing some of WRF-ARW v3.0s available physics option to aid in the development of polar WRF. Simulations of January and June 1998 were compared against observations of 2 m temperature, shortwave fluxes, longwave fluxes, and the temperature at the surface from the Surface Heat Budget of the Arctic Ocean Experiment (SHEBA) [Persson et al., 2002] to determine the optimal physics package to be used in Polar WRF. Table 1 depicts the chosen schemes following the recommendations of Seefeldt and Cassano [2008] with one exception. For shortwave radiation, the Dudhia [1989] scheme, previously adopted by the MM5 model, is used because it proved to be both more stable and more computationally efficient in longer integrations than the Goddard shortwave scheme.

Table 1. Model Parameterization and Physics Packages Used
Parameterization/PhysicsChosen SchemeSource
Cloud microphysicsMorrison 2-MomentMorrison et al. [2008]
CumulusKain-FritschKain and Fritsch [1990, 1993]
Shortwave radiationDudhiaDudhia [1989]
Longwave radiationRRTMMlawer et al. [1997]
Land surface modelNoah LSMNCAR/NCEP
Planetary boundary layerYSU PBLHong et al. [2006]

[19] The domain for all simulations is a 330 × 330 Arakawa C grid on a polar stereographic projection (Figure 2). This domain covers roughly from 30°N to 90°N at 40 km horizontal resolution with 28 vertical levels. This relatively broad domain minimizes the influence of boundary conditions on the results in the polar regions and permits an evaluation of the effects of changes in the Arctic surface state on remote locations in middle latitudes. The NCEP/NCAR Global Reanalysis Project provides lateral boundary forcing. Output from the reanalysis is available every 6 h at 2.5° resolution and is interpolated to the model grid using the WRF Pre-Processing System. Reanalysis output for the period 1 September to 31 December for 2007 were used.

Figure 2.

The model domain is 330 × 330, extending from roughly 30°N to 90°N latitude at 40 km resolution with terrain height (meters) in color.

3. Experimental Design

[20] Our WRF simulations consist of two ensemble collections, one based on an experiment run and the other based on a control. Both experiments employ the NCEP/NCAR Global Reanalysis data set. The “experiment” run (hereafter referred to as 2007ice) is initialized with 2007 lateral boundary conditions for the atmosphere while sea ice and sea surface temperatures are prescribed to 2007. For the sake of comparison, our “control” (hereafter referred to as CNTLice) is initialized with 2007 atmospheric conditions and sea surface temperatures but 1984 sea ice condition such that the only difference between the experiment and control sets are the time varying ice cover. In 1984 the annual median ice coverage in the satellite record from 1979 to 2008 occurred, and thus provides our representation of a “normal” ice year. The contrast between 1984 and 2007 sea ice concentration is depicted in Figure 3. Because both sets of simulations use 2007 sea surface temperatures, the CNTLice case has sea surface temperatures along the ice edge that are warmer than may be realistic. No instabilities arose from the discontinuity. Output is archived at 24 h intervals at 00Z from the period starting with 1 September 1 0000 UT and ending with 31 December.

Figure 3.

SSMI measured 2 October 1984 and 2007 sea ice concentration (%) in color.

[21] Natural variability is part of any 4 month simulation period and can obscure the atmospheric response to a surface boundary anomaly. In order to extract the signal produced by the surface forcing, we make use of ensembles of simulations. Both the 2007ice and CNTLice cases were run 10 times each. Ensemble members were each initialized with small random perturbations created by randomly varying the damping coefficient at the top of the atmosphere from 0.15 to 0.25 for the first 24 h of the simulation. Simulations were then restarted and run with the same damping coefficient (0.2) throughout the remainder of the integration.

[22] Monthly ensemble averages (September, October, November, and December) were created for key variables. We emphasize results in October and November here, as these are the months with the largest anomalies and relatively cold airflow over the open-water anomaly area, which has a surface temperature of at least −1.8°C (the freezing temperature of seawater) until freezing occurs.

4. Results

4.1. Changes in Heat Fluxes and Atmospheric Moisture

[23] Arctic ocean-to-atmosphere latent and upward heat fluxes are important locally and, potentially, remotely. Results are shown here as the difference between the October–November 2007ice case and the CNTLice case. In all cases, the differences are between the 10-member means of the two ensembles.

[24] When the CNTLice October–November mean latent heat flux is subtracted from that of the 2007ice case as depicted in Figure 4, several areas stand out. As one would expect, there is a large positive difference over the location of the sea ice anomaly, over 40 W m−2. Other prominent areas show up due to changes in circulation over the domain. One such prominent area, located over the Gulf Stream region, is likely the result of a tendency for lower pressure over eastern North America and more cold air advection in the 2007ice case. In the Gulf of Alaska region, there is a tendency for less latent heat transport from the ocean to the atmosphere due to a propensity for higher pressure in this region (60% of ensemble members had pressures greater than control mean in this area). Another prominent area is the Great Lakes system, which also experiences an increase in cold air advection causing anomalous lake-to-atmosphere heat fluxes. At the sea ice anomaly, upward sensible heat (not shown) flux is over 40 W m−2 greater in the 2007ice case than the CNTLice case. Remotely, there is increased sensible heat flux on the order of 12–20 Wm−2 over the Great Lakes and Gulf Stream region in the 2007ice case.

Figure 4.

The 2007ice–CNTLice difference fields of October–November upward latent heat flux (W m−2).

[25] To evaluate changes in midlevel and low-level atmospheric moisture, we employ dew point temperature (Td) difference plots 2007ice−CNTLice at 925 hPa, 700 hPa, and 500 hPa (Figures 5a, 5b, and 5c, respectively). At 925 hPa (Figure 5a), the dominant feature is located over the sea ice anomaly with 2007ice Td more than 2°C greater than the CNTLice case. Higher in the atmosphere, at 700 hPa (Figure 5b) and 500 hPa (Figure 5c) this feature is replaced by negative Td changes, though small in magnitude.

Figure 5.

The 2007ice–CNTLice difference fields of October–November dew point temperature (Td in °C) (a) at 925 hPa, (b) at 850 hPa, and (c) at 700 hPa.

4.2. Circulation Changes

[26] Changes in latent and upward heat fluxes induce changes in the local atmospheric column heat budget. Basic meteorology shows that such increases in heat can induce the formation of a thermal low-pressure system at lower levels. If this is strong enough, it can affect temperature and circulation both directly and indirectly, even aloft. Beginning with sea level pressure and 2 m temperature (T2), this section examines the changes in the atmospheric temperature and circulation between the 2007ice and CNTLice cases. After addressing lower levels, standard atmospheric levels are similarly analyzed.

[27] Figure 6a depicts T2 in °C. Over the anomaly and much of the Arctic Ocean, temperatures were greater in the 2007ice case than the CNTLice case by over 8°C due to the heat transport from the open water to the atmosphere. The large negative difference in pressure coinciding with the large positive temperature difference indicates a possible thermal low structure. Such a structure may be a response to the surface type: relatively warm water rather than thicker ice, which in some respects is more similar to snow covered land than water. As discussed in the Introduction, changes in heat fluxes impact temperature and circulation. Directly over the anomaly (Figure 6a), SLP is more than 5 hPa lower in the 2007ice case than the CNTLice over October and November. Over most of North America's West Coast, there are slightly higher SLPs on the order of 2 hPa, in the 2007ice case. Over the North Atlantic, the SLP is slightly lower (>−2 hPa) in the 2007ice case. The larger picture of the difference field shows a “trough, ridge, trough, ridge”-like pattern beginning in the North Pacific and extending to the eastern North Atlantic.

Figure 6.

(a) The 2007ice–CNTLice at sea level difference fields of October–November 2 m air temperature (colors) and sea level pressure (contours, 1 hPa interval). (b) Student's t statistic for temperature at 2 m depicted as a percent chance that the results are not random.

[28] Remotely, temperatures are cooler over the midwestern/northeastern United States, as would be expected from an increase in cold air advection due to the position of the higher- and lower-pressure centers. This is also the case in northern Scandinavia, which experiences a slight increase in cold air advection. The Student's t statistic for T2 is shown in Figure 6b, which depicts these areas as statistically significant at greater than the 95% level.

[29] A simple cyclone tracking algorithm was applied to the model output to investigate storm frequency. This subroutine located and counted relative pressure minima. Figure 7 depicts October–November cyclone frequency 2007ice minus CNTLice, thus indicating an increase in cyclone activity in the region along the eastern North American seaboard, with a maximum increase over the Canadian Maritime Provinces. This shows that the statistically significant pressure decreases over eastern North America (Figure 6) are likely the result of increased cyclone activity. It is important to note that changes in SLP may arise from increased cyclone frequency, increased cyclone intensity, or a combination thereof.

Figure 7.

Ensemble mean change in October–November cyclone frequency, 2007ice−CNTLice (storm count/grid cell).

[30] To determine the time when the difference between the 2007ice and CNTLice T2 is greatest, we include the time series depicted in Figure 8, which illustrates the temporal evolution of the mean atmospheric response to the negative sea ice anomaly. Over the polar cap north of 70°N, the difference (warming) between the experiment and control average T2 is much more prominent from days 10–75 after which time, natural variability increases. Thus from 10 September to approximately 15 November, the difference in T2 between the 2007ice case and the CNTLice case is maximized. While Francis et al. [2009] show that several atmospheric variables in November–January are related to September sea ice, our results indicate that the greatest influence is likely in October when the differences between the two sets of simulations are largest. It is important to note, however, that Francis et al. [2009] use several years of observational data, while this study focuses on just one year. The largest ice anomalies occur in different sectors in different years. Moreover, the atmospheric anomalies observed in a single year (2007 or 1984) are more likely to be influenced by natural variability than is a composite field. This information can potentially be exploited for long-range forecasts during low ice years, as it implies that an extreme summer sea ice retreat (with the minimum coverage in September) should have its maximum impact on the atmosphere about a month later, i.e., in October or early November.

Figure 8.

Time series of domain area mean temperature (K) for north of 70°N over the entire simulation (not just October–November).

[31] Figure 9a depicts composite differences of 850 hPa heights (contoured) and temperatures (colored). Analogous to the surface (SLP), two remote features stand out, one over western North America and one over eastern North America. Over western North America, 850 hPa heights are higher in the 2007ice case and temperatures are warmer. More significantly, in eastern North America heights are lower and temperatures are colder. The Student's t statistic map for 850 hPa heights (Figure 9b) shows that both of these features are statistically significant. In northern Scandinavia, temperatures are slightly cooler and the differences in this region are also significant. Results at 500 hPa (not shown) are similar, though key features broaden with height.

Figure 9.

(a) The 2007ice–CNTLice at 850 hPa difference fields of October–November air temperature (colors) and 850 hPa heights (contours, 2 m interval). (b) Student's t statistic near 850 hPa depicted as a percent chance that the results are not random.

[32] This broadening of the main features with height continues aloft to 300 hPa. Figure 10a shows that the western ridge-like feature now covers a large area of the North Pacific and West Coast of North America. Similarly, the trough-like feature now extends from the Dakotas eastward to the Gulf Stream and Sargasso Sea. The t statistic in Figure 10b, is similar to that at 500 hPa, though the local feature is now almost completely diminished while the remote low over eastern North America dominates. In general, the local signal in terms of heights/SLP widens with height while the magnitude of the temperature decreases. Local features are over 99% significant up to 850 hPa and decrease in significance by 300 hPa. Remote features are less significant near the surface but are more significant aloft. Both the remote and local signals are vertically stacked with height. The location of the cold anomaly in T2 is located under the low height anomalies at 850 hPa and 300 hPa.

Figure 10.

(a) The 2007ice–CNTLice at 300 hPa difference fields of October–November air temperature (colors) and 300 hPa heights (contours, 5 m interval). (b) Student's t statistic near 300 hPa depicted as a percent chance that the results are not random.

4.3. Precipitation

[33] Continuing the argument that changes in heat fluxes impact circulation and, subsequently, precipitation, this section discusses changes in precipitation between the 2007ice case with less ice in the Arctic Ocean and the CNTLice case with more ice in the Arctic Ocean. This section shows changes in precipitation by first examining liquid precipitation, then solid precipitation via the snowpack.

[34] Over the Arctic Ocean, ensemble differences of total precipitation show a slight increase (20–40 mm) in October–November ensemble mean precipitation (Figure 11a). This feature is highly statistically significant (Figure 11b). Remotely, over the Gulf Stream region, we see a slightly larger increase in precipitation, which is also significant. This region stretches from the Midwest over the Great Lakes region into the Gulf Stream region, indicating a possible enhanced storm track through the region previously shown to have lower pressures. This feature is geographically quite similar to the path of increased cyclone frequency over the same region shown in Figure 7.

Figure 11.

(a) The 2007ice–CNTLice October–November total liquid precipitation in millimeters (colors) and (b) Student's t statistic as percent chance that the results are not random. (Note that the color bar is reversed for precipitation, showing negative differences in red and positive differences in blue.)

[35] In western North America where there was a tendency for more ridging (Figures 9 and 10), we see precipitation deficits in the 2007ice minus CNTLice difference field; Sewall [2005] obtained similar results using the NCAR model. Because the Student's t statistic field is noisy in precipitation fields, we only mention areas where there is a relatively broad consistent signal and we can postulate physical foundation for the significance.

[36] In October and November, the average snowpack (here shown as snow depth) is north of 45°N. Where we saw lower pressures over eastern Canada in the 2007ice experiment, the snowpack is slightly deeper (Figure 12). Over the anomaly, the snowpack is greatly diminished for 2007ice for several reasons. First, there is no sea ice on which snow might build up. Second, more precipitation in this and surrounding regions (e.g., Alaska) falls as rain or mixed precipitation rather than snow in the warmer 2007ice scenario. Along much of the coastlines in the western Arctic, there is up to a −20 cm difference in snowpack in the 2007ice case. This suggests that there may be less snowpack when sea ice is diminished in the Arctic Ocean, having possibly serious consequences for Arctic hydrology and changes in permafrost.

Figure 12.

October–November difference in snow depth (centimeters) 2007ice–CNTLice.

5. Summary and Discussion

[37] Anomalies of Arctic Ocean sea ice were shown in this study to impact the atmosphere's temperature and pressure remotely as well as locally mainly through changes in heat fluxes. Large increases of latent and sensible heat fluxes over the anomalously low Arctic Ocean sea ice extent impact local dynamics through local changes in temperature and pressure. These changes then impact local precipitation. As the anomaly was so persistent, changes in local dynamics feed into large-scale dynamics with time, creating the tendency for a large-scale ridge/trough system to build over North America in the ensemble October–November averages when the magnitude of the sea ice anomaly was largest. The pressure and height responses lead to advectively driven remote anomalies of temperature, precipitation, and surface heat fluxes.

[38] Appendix A explores the biases of the WRF-ARW as we applied it and such biases should be noted when considering interpretation of prominent features. Some biases are the result of more advanced treatment of sea ice, especially local biases. Such biases may impact the model's reproduction of features like the Beaufort High. Other biases, like those in the North Pacific may be attributed to the lack of data available for assimilation into the reanalysis. Past studies and the results allow us to make the following conclusions.

[39] Dynamically driven increases in upward latent heat fluxes were found to be prevalent locally as well over the Gulf Stream and Great Lakes. Several other studies [Alexander et al., 2004; Deser et al., 2004; Bhatt et al., 2008] show increases in latent heat fluxes in the Gulf Stream region. Alexander et al. [2004] cite increases over 100 W m−2 in winter while this study found over 28 W m−2 more with 2007ice than CNTLice over October–November. The large disparity between our study and Alexander et al. [2004] is likely due to the differences in air temperature between autumn and winter, and the associated increased vertical temperature gradients and instability in winter. Coincident with Bhatt et al. [2008], these local increases in latent heat fluxes led to local increases in T2 and decreases in SLP.

[40] As shown in the Td difference fields, the confinement to low levels of the Arctic Ocean moisture anomaly is not surprising, as there is generally less moisture aloft to contribute to a large difference at 700 hPa on up; most downwelling longwave radiation comes from the lowest kilometer, so the enhanced moisture is likely contributing to the warmer temperatures over the ice anomaly in 2007. There is not much change over the northwest Atlantic, but there is over the eastern North America/Great Lakes region. Since there is not much of a local (surface) moisture source for these regions in October–November, the anomaly there must be advective, hence dynamical in origin. By contrast, the large moisture differences in the Arctic Ocean are found right over the anomalous surface source of moisture, pointing to a local (thermodynamic) forcing. Therefore the local response to sea ice loss is thermodynamic while the remote response is likely an advective/dynamic response rather than to be attributed to thermodynamic forcing such as increased water vapor.

[41] The lowering of pressure over the ice anomaly in the 2007 simulation is consistent with the response obtained by Higgins and Cassano [2009] (hereafter denoted as HC2009) in a global model run with reduced sea ice prescribed from projections for the late 21st century (2080–2099). In HC2009, a large decrease in sea level pressure (1000 hPa geopotential height) was increased with an increased frequency of strong low-pressure systems in the central Arctic. In this respect, autumn synoptic patterns extended into the winter. Similarly, it can be inferred that our results indicate an extension of summer patterns, as per HC2009's Figure 2 (panels (0,0) and (1,0)). By far, the low-pressure feature north of Alaska (shown in several panels of HC2009's Figure 6) is one of the most consistent features found in their reduced sea ice simulations. Here our results coincide well with HC2009's Figure 6 (panels (3,0), (3,1), (4,0), (5,4), and (6,4)), which explain greater than 40% of the contribution of the NDJF mean 1000 hPa geopotential height related self organizing map patterns. An evaluation by HC2009 of the drivers of the temperature changes (warming over area of reduced sea ice) showed that approximately half the warming could be ascribed to diabatic heating, while the other half was attributable to temperature advection and hence to synoptic weather patterns. By contrast, the thermodynamic component (sensible heating and evaporation) accounted for more than 98% of the precipitation, while dynamical processes contributed little.

[42] Results presented by Francis et al. [2009], an observational study, show reduced SLP over eastern North America and increased SLP over western North America when detrended. Studies examining other seasons [Alexander et al., 2004; Deser et al., 2004, 2010], identified remote signals similar to the Northern Annular Mode.

[43] The upper air difference fields showed higher heights over western North America and slightly lower heights over eastern North America, which maintained statistical significance over 80% up to 300 hPa. Significance of the local feature decreased with height while the significance of the remote feature was maintained vertically. This, too, is consistent with Bhatt et al. [2008], who found an equivalent barotropic signal that maintained significance up to 200 hPa. Though Singarayer et al. [2006] found that the only statistically significant sea ice driven anomalies occurred during winter, their geographic location of the sea ice anomaly was drastically different, with sea ice loss found mainly in the eastern Arctic rather than the western Arctic as occurred in 2007.

[44] Because changes in circulation dynamics are observed, we are more likely to see precipitation changes. Though the significance field for liquid precipitation was noisy, the results obtained here show several regions of significant features, such as (1) the Arctic Ocean feature found both in this study and by Alexander et al. [2004], and (2) the storm-track-like feature shown over the Great Lakes and Gulf Stream region. This feature is closely comparable to the area of increased cyclone frequency. Similar changes in midlatitude storm tracks were obtained by Sokolova et al. [2007].

[45] Though GCMs and AOGCMs are commonly used for climate studies by researchers, there are some disadvantages. To combat the disadvantages of GCMs we employ a higher-resolution mesoscale model, WRF-ARW. Though researchers are increasingly using WRF for regional climate studies, it has not been as widely tested and used as many global climate models. However, the use of WRF-ARWs more advanced physics and higher horizontal, vertical, and temporal resolution make it a potentially superior tool for addressing the interplay between sea ice, air temperature and precipitation, especially in areas with complex land-sea boundaries and/or topography. In addition, we are more reasonably able to resolve features like the Great Lakes, which are vital to understanding our remote response.

[46] Several studies, including the work from the Intergovernmental Panel on Climate Change Fourth Assessment Report project continued decline in Arctic Ocean sea ice extent. Some studies even predict a seasonally ice free Arctic Ocean within the next few decades. If these projections verify and sea ice continues its decline, our study indicates regions that are likely to see changes in October–November weather patterns such as an increase in lake-effect snow downwind of the Great Lakes region. This increase in lake-enhanced snowfall is the result of a strong tendency for lower pressure over eastern North America, which increases cold air advection into that region. Such a tendency for lower pressures was observed in 70% of the 2007ice ensemble members. This result could ultimately be incorporated into a probabilistic framework for monthly or seasonal forecasting of remote anomalies during years with extreme sea ice minima. In addition, the increased fluxes of latent heat from the Gulf Stream to the atmosphere may have an impact on the nature of the downstream ocean currents and vertical mixing in the subpolar seas.

Appendix A:: What Happened in Nature?

[47] Using reanalysis output from the NCEP/NCAR Global Reanalysis, we examine October–November 2007 minus October–November 1984 using SLP, 850 hPa heights, and 300 hPa heights. Similar to modeled 2007ice−CNTLice SLP, the 2007–1984 SLP difference (Figure A1a) in eastern North America was slightly negative. This feature is quasi-barotropic, as seen in Figures A1b and A1c. There is also a region of higher pressure higher than that of 1984 along the west coast of North America into the eastern Pacific Ocean. Over Alaska, slightly west of the Arctic Ocean, SLP and heights were lower in 2007 than in 1984. While the latter features do not coincide with the modeled results presented within this paper, it must be noted that the observed fields for 1984 and 2007 represent single realizations of a system subject to large natural variability (addressed in our model experiments by the use of ensembles of simulations). It is important, also, to note that the modeled control set (CNTLice) uses 2007 atmospheric boundary conditions and only sea ice boundary conditions from 1984.

Figure A1.

NCEP/NCAR Global Reanalysis October–November average 2007 minus 1984 (a) sea level pressure (hPa), (b) 850 hPa geopotential heights (meters), and (c) 300 hPa geopotential heights (meters).

[48] Ten ensemble WRF-ARW October–November SLP, T2 (Figure A2), and 500 hPa height (Figure A3) biases from the NCEP/NCAR Global Reanalysis and the ERA Interim Reanalysis were calculated. SLP and T2 biases from NCEP (Figure A2a) are larger near the Canadian Arctic Archipelago and Greenland side of the Arctic Ocean. This larger bias can be explained by WRF's use of fractional sea ice. In general, biases are largest where data observations that would likely be assimilated into the reanalysis are sparse. Biases from the ERA Interim (Figure A2b) are smaller in magnitude than biases from the NCEP Reanalysis in general. Biases from both reanalyses are present in regions of topography, especially the Himalayas and Rocky Mountains due to the higher resolution of the WRF-ARW model. At 500 hPa, for both the NCEP and the ERA-Interim reanalyses (Figures A3a and A3b), there is a low height bias over the sea ice anomaly and a high height bias over the remote feature in eastern North America.

Figure A2.

WRF-ARW October–November SLP (hPa) and T2 (°C) biases from (a) NCEP/NCAR Global Reanalysis and (b) ERA Interim Reanalysis and a side-by-side comparison of October–November SLP (hPa) and T2 (°C) from (c) the NCEP/NCAR Global Reanalysis, (d) WRF-ARW, and (e) the ERA-Interim Reanalysis.

Figure A3.

WRF-ARW October–November 500 hPa height (meters) biases from (a) NCEP/NCAR Global Reanalysis and (b) ERA Interim Reanalysis and a side-by-side comparison of October–November 500 hPa heights (meters) from (c) the NCEP/NCAR Global Reanalysis, (d) WRF-ARW, and (e) the ERA-Interim Reanalysis.

Acknowledgments

[49] We would like to thank the three anonymous reviewers for their time and effort in giving us such invaluable comments. The National Science Foundation grant NSF ARC-0732650 supported this research. The NCEP/NCAR Global Reanalysis Project provided boundary forcing for this project.

Ancillary