Atmospheric response to changes in Arctic sea ice thickness



[1] Experiments with an atmospheric GCM are used to determine the effect of anomalous Arctic sea ice thickness on the atmospheric circulation. Ice thickness data are taken from a hindcast simulation with an ocean-sea ice model under NCAR/NCEP forcing. Ice conditions from 1964–1966 and 1994–1996 represent extreme cases of largest and smallest ice volume, respectively, during the last 50 years. The atmospheric response to the 1990s thinning of Arctic sea ice comprises a reduction in sea level pressure in the central Arctic and over the Nordic Seas. High pressure anomalies evolve over the subtropical North Atlantic and over the subpolar North Pacific. Similar signals occur at 500 hPa. Realistic sea ice thickness changes can induce atmospheric signals that are of similar magnitude as those due to changes in sea ice cover.

1. Introduction

[2] Many climate parameters in the northern hemisphere exhibit multidecadal trends [e.g., Polyakov et al., 2003]. Over the last five decades, the 500 hPa geopotential height decreased over much of the Arctic, the subpolar North Atlantic and the North Pacific. Arctic Ocean ice area and extent have a very pronounced downward trend for the period of satellite observations with recent summers setting new record lows for ice concentration [Serreze et al., 2003]. Winter sea ice extent increased in the Labrador Sea while it decreased in the Greenland and Barents seas. Ice thickness also appears to decrease over large parts of the Arctic Ocean [Rothrock et al., 1999; Wadhams and Davis, 2000]. Most of the changes in SST and sea ice extent in the North Atlantic appear to be related to the increasing strength of the NAO [Polyakov et al., 2003].

[3] Different hypothesis have been proposed to explain the trend in the NAO and the changes in atmospheric pressure over the Arctic [Hurrell et al., 2003]. Several recent studies have investigated the reaction of the NAO to changes in sea ice extent and concentration [Alexander et al., 2004; Kvamstø et al., 2004; Magnusdottir et al., 2004]. Sea ice extent conditions that are typically observed during the positive phase of the NAO induce an atmospheric response that contains a negative NAO signal. NAO-related SST anomalies, on the other hand, excite a weak positive phase NAO response.

[4] The effect of changes in sea ice thickness have so far not been considered because comprehensive observational data are lacking and the heat fluxes associated with changing ice thickness are deemed insignificant compared to the huge differences in heat fluxes between ice covered and ice free regions. However, differences in winter sea ice concentration are restricted to rather small areas and the overall anomalous heat flux from the ocean to the atmosphere is thus limited. On the other hand, substantial changes in ice thickness cover large areas. While the changes in local heat flux are small the total, area integrated anomalous heat flux might thus be larger than that associated with changes in ice concentration alone.

[5] Here, I shall investigate the response of an atmospheric model to changes in sea ice concentration and sea ice thickness in the Arctic and the North Atlantic. In lieu of observed ice thickness changes, I use results from an ocean-sea ice model that was forced with NCEP/NCAR forcing for the period 1948–1998 [Köberle and Gerdes, 2003]. The goal is to explore the relative importance of sea ice concentration and thickness trends for the atmospheric response and the possible attribution of the observed trend of high latitude northern hemispheric atmospheric flow to these changes.

2. Experimental Setup

[6] The set of experiments in this study consists of four simulations run over 40 years each, where different composite seasonal cycles of sea ice conditions (concentration and thickness) are prescribed as lower boundary condition together with a climatological seasonal cycle for SST. In the first experiment (THI95), monthly sea ice conditions averaged over the years 1994–1996 are used. The second experiment (THI65) employs conditions derived from 1964–1966. Two additional experiments with sea ice thickness averaged for the period 1948 through 1998 and different composite sea ice concentration fields from 1994–1996 (CON95) and 1964–1966 (CON65), respectively, address the question of how much of the differences can be attributed to the differences in ice thickness compared to those in ice concentration.

[7] Sea ice thickness and concentration data are taken from a hindcast simulation with an ocean-sea ice model for the Atlantic and the Arctic oceans [Köberle and Gerdes, 2003]. Areas outside the domain (Pacific, Southern Ocean) are treated as described by Anderson et al. [2004]. The hindcast was forced with 50 years (1948–1998) of NCEP/NCAR reanalysis data [Kalnay et al., 1996]. Consistent with the available sea ice thickness observations, the model shows a long term declining trend in Arctic sea ice volume from maximum sea ice volume in the mid-1960s to the mid-1990s minimum. The increase in surface air temperature is responsible for the long term decrease in ice volume in that model while decadal variability in the wind stress is mostly responsible for ice export events that can drain substantial parts of the Arctic ice volume into the Nordic Seas and Baffin Bay. According to the model, major ice export events occurred in the winters 1958/59, 1967/68, 1981/82, and 1994/95. The last event has been directly observed in Fram Strait [Vinje, 2001] while the event of 1967/68 is generally regarded as the source for the Great Salinity Anomaly of the 1970s in the North Atlantic [Aagaard and Carmack, 1989; Häkkinen, 1993].

[8] With the above choice, extreme sea ice thickness fields that reflect the decreasing sea ice thickness trend in the Arctic from the mid 1960s to today are used in the experiments. The differences in winter ice concentration (Figure 1) between the 1990s and the 1960s are largest near the ice edge in the Labrador, Greenland, and Barents seas. The ice edge progressed into the Labrador Sea while the sea ice was retreating in the Greenland and Barents Sea. These changes in sea ice edge position and sea ice concentration are related to the increasingly positive index states of the NAO. A positive NAO is associated with anomalously high temperatures over the eastern Nordic Seas and the Barents Sea while temperatures are low over the Labrador Sea. Largest sea ice thickness differences, on the other hand, are found in the East Siberian Sea and projecting from there into the interior Arctic Ocean (Figure 1).

Figure 1.

(a) Difference in winter (JFM) ice concentration between the 1994 to 1996 composite and the 1964 to 1966 composite. (b) Difference in winter ice thickness between the 1994 to 1996 composite and the 1964 to 1966 composite.

[9] The above sensitivity experiments have their limitations in the representation of processes that would in nature feed back on the here prescribed SST and sea ice fields. For instance, with a prescribed climatological seasonal cycle we neglect SST anomalies that can arise in response to changes in ice cover due to changes in absorption of short wave radiation. Reduced summer sea ice cover will result in heating of the oceanic mixed layer. This heat will be released in fall and early winter, delaying the formation of new ice and heating the atmosphere.

[10] The atmospheric model is the GFDL AM2 as described by Anderson et al. [2004]. This grid-point model has a horizontal resolution of 2.5° × 2° and 24 levels. The model is run with seasonal cycle forcing at the top of the atmosphere and at the bottom. Four 40 year integrations are thus performed that differ in sea ice thickness and concentration at the lower boundary. Each year of the individual runs is regarded as an realization of the atmospheric state for the corresponding boundary conditions. In the following, differences between the ensemble means for 1990s forcing and 1960s forcing will be discussed. Since summer differences are relatively weak, I shall focus on winter results.

3. Results

[11] The ocean-atmosphere heat flux decreases as sea ice concentration increases in the Labrador Sea (Figure 2). On the other hand, heat flux increases by up to 50 Wm−2 along the East Greenland coast and in parts of the Barents Sea where sea ice concentration decreased between the mid-1960s and the mid-1990s (the difference between the heat fluxes in the ensemble means of THI95 and THI65). In addition to these regions, there are anomalous net upward heat fluxes of more than 10 Wm−2 in large parts of the interior Arctic. Heat flux anomalies exceed 10 Wm−2 over the East Siberian Sea and adjacent regions. These regions of positive heat flux differences coincide with the areas of decreasing sea ice thickness (Figure 1).

Figure 2.

(a) Net upward surface heat flux anomaly [THI95–THI65] for the winter season (JFM). Contour interval is 5 Wm−2. (b) Difference in surface air temperature (SAT) between the THI95 and THI65 ensembles.

[12] Positive surface air temperature differences (Figure 2) result over the Greenland and Barents seas and over most of the Arctic Ocean. Negative SAT differences prevail in the Labrador Sea and over Alaska. Most of these differences are consistent with the changes in local heating from the ocean and through thin ice.

[13] The difference in sea level pressure is negative for the THI experiments over most of the Arctic Ocean and the Nordic Seas (Figure 3). Positive differences are visible over the northern North Pacific with a center over the Bering Sea. Outside the above regions, the changes are small and of large spatial scale, giving some confidence that internal variability has been successfully filtered in the 40 year ensemble. The pressure anomalies are approximately equivalent-barotropic with negative anomalies of 500 hPa geopotential height (Figure 3) and higher (not shown) centered over the Arctic Ocean and extending over North America. Positive anomalies are present over north-east Europe and the northern North Pacific, centered over the Bering Sea.

Figure 3.

(a) Difference in winter (JFM) sea level pressure (SLP) between 1995 and 1965 as calculated from NCEP/NCAR reanalysis data using the linear trend over the reanalysis period 1948–2000. (b) Winter sea level pressure difference between the experiments forced with 1994–1996 sea ice concentrations (CON95) and those forced with 1964–1966 sea ice concentrations (CON65). (c) The corresponding difference in SLP for THI95 and THI65 where both ice concentration and sea ice thickness vary. Contour interval for Figures 3a–3c is 1 hPa. (d) The difference [THI95 − THI65] in the winter 500 hPa geopotential height. The contour interval is 5 m. Positive values indicate larger height for the ensemble forced with 1994–1996 sea ice conditions (THI95).

[14] Formally, the above features are highly significant because of the large ensemble size (40 realizations for each experiment). SLP and geopotential height anomalies have additionally been evaluated for 10 and 20 year long sub-intervals of the 40 year long experiments. The anomalies over the Arctic and the North Pacific turned out to be very robust. They are present in even the shortest (10 year) ensembles. On the other hand, relatively large variability is present over the North Atlantic and not all individual years resemble the ensemble mean. The sea ice thickness anomalies thus have little predictive power in the sense that a loss of sea ice from the Arctic cannot be used as a reliable predictor for large scale same winter SLP anomalies over the mid to high northern latitudes. The results can only be interpreted as indicative for long term trends due to persistent changes in sea ice conditions.

[15] The atmosphere sees different ice concentration and ice thickness distributions in the two experiments discussed so far. The heat fluxes and temperature anomalies of Figure 2 indicate that the changes in ice thickness in the interior Arctic, where ice concentrations in winter are very close to 100% in both cases, could be important. This conjecture is confirmed by the additional experiments where only the ice concentrations differ while the ice thickness is held at climatological seasonal values (CON65, CON95). The results are consistent with those of Alexander et al. [2004] and Kvamstø et al. [2004] who used similar changes in sea ice extent and/or sea ice concentration. SLP for mid-1990s sea ice conditions is higher than for mid-1960s conditions over Greenland, Canada, and the western Arctic Ocean. Where CON95 has a lower pressure over eastern Siberia compared to CON65, Kvamstø et al.'s result show only a weak trough extending from the North Pacific (their Figure 2). Alexander et al. [2004, Figure 7] show negative SLP over Europe and positive SLP anomalies over Greenland and Canada. The overall similarity of these results, all gained with different atmospheric models, provides some confidence in the robustness of these results despite the small amplitude of the response. We can conclude that the robust results of the experiment with both ice concentration and thickness differences, namely the low pressure over the Arctic and the high pressure anomaly over the northern North Pacific, must be attributed to the differences in ice thickness.

4. Discussion and Conclusions

[16] Sensitivity experiments as described here are intended to selectively explore the atmospheric response to certain changes in boundary conditions. Since the real atmosphere is subject to a great number of effects, it is not expected that sensitivity experiments reproduce the actually observed change. However, within the limitations of fixed boundary conditions that prevent certain feedbacks, an attribution of changes to certain forcing factors can be attempted. Previous experiments with anomalies in Arctic sea ice extent or concentration have been interpreted in terms of feedbacks between the sea ice state and the NAO because the dipole in winter sea ice cover between the Labrador Sea and the Barents Sea is associated with opposite phases of the NAO. In this sense, the results of Alexander et al. [2004], Kvamstø et al. [2004], and Magnusdottir et al. [2004] indicate a weak negative feedback between sea ice concentration anomalies and the NAO. The present experiments CON95 and CON65 with similar sea ice concentration anomalies yield consistent results. Going beyond the above mentioned studies, the experiments THI95 and THI65 explore the effect of ice thickness anomalies on the atmospheric circulation.

[17] Thickness changes are not that easily associated with any major atmospheric circulation pattern and feedback considerations are not as straightforward as in the case of winter ice extent anomalies. Köberle and Gerdes [2003] found that the long term decline of the Arctic ice volume was due to a temperature increase while the ice export from the Arctic did not show an increasing trend over the last 50 years. Thus, the results of THI65/95 could on the one hand be interpreted as the immediate reaction of the atmosphere to large ice export events like that leading to the Great Salinity Anomaly or the observed event of the winter 1994/95 [Vinje, 2001]. In this case, a certain feedback with the NAO could be postulated because after around 1970, ice export through Fram Strait had a strong correlation with the NAO [Hilmer and Jung, 2000]. Following this interpretation, a positive phase of the NAO leading to a large sea ice export through Fram Strait would induce atmospheric pressure anomalies that reinforce the positive phase of the NAO. It is interesting to speculate whether a threshold might be involved in the sea ice thickness forcing of the NAO. Only when the sea ice thickness falls below a certain value over a large enough area, will the atmosphere feel a significant change in surface heat flux. This could have been the case with the large ice export event of 1968 that removed 20% of the pre-existing sea ice in the Arctic Ocean. However, the fact that some realizations deviate substantially from the ensemble mean might preclude such an interpretation.

[18] On the other hand, the results can be interpreted, perhaps with more justification, as the atmospheric response to the long term decline of the Arctic ice volume. Ice thickness in the mid-1960s represents the maximum in ice volume during the last 50 years while the mid-1990s conditions are close to the minimum Arctic sea ice volume [Rothrock et al., 1999; Köberle and Gerdes, 2003]. The differences between the THI95 and THI65 ensembles include a dipole of negative SLP over most of the Arctic Ocean and positive SLP over the northern North Pacific. No such differences exist between the CON95 and CON65 ensembles and thus they can be attributed to the differences in ice thickness. The amplitude of the simulated SLP signal is comparable to the observed trend in SLP between 1965 and 1995. Maximum absolute values reach 4 hPa (Figure 3). However, the pattern of the THI95–THI65 SLP differences differs from the NCEP reanalysis SLP trend. Decreasing pressure over the central Arctic is found in response to sea ice thickness changes and in the observed trend. However, the experiment indicates a mass exchange between the Arctic and the North Pacific region while such an exchange is observed between the Arctic and the North Atlantic.

[19] The SLP difference between THI95 and THI65 projects on the NAO pattern. The reaction of the atmosphere to decreasing Arctic ice thickness contains a strengthening of the NAO. This mechanism thus must be added to the number of processes that have been suggested to drive the NAO. Furthermore, reduced future Arctic Ocean ice thickness as predicted by most climate models could contribute to an increase in the strength of the NAO.


[20] NCEP/NCAR Reanalysis data was provided by the NOAA/CIRES Climate Diagnostics Center, Boulder, Colorado, This work was conducted during a visit at the Geophysical Fluid Dynamics Laboratory, Princeton. I thank GFDL for the hospitality and financial support. Special thanks are due to Rhong Zhang, Mike Winton, and Isaac Held for help with the set up of the experiments and for discussions and suggestions.