Introduction
Stewart (Reference Stewart1978) and Baker (Reference Baker1986) suggested that climate change in the northern hemisphere may be related to sea ice and ocean circulation in the Arctic; i.e., there is feedback between the atmosphere and sea ice/ocean. In this paper, decadal time-scale variations, which have been observed to exist and may significantly influence human life, are studied.
Ikeda (in press) reconsidered correlations among decadal oscillations of northern Hhmisphere atmospheric circulation, ice cover and ocean circulation. Decadal oscillations of the Ice cover are observed in the Barents Sea and Labrador Shelf, and are out of phase between these two regions. Heavy or light ice cover in these regions may be attributed to atmospheric conditions; when a low (high) anomaly exists over the Eurasian Basin, Europe is warm (cold) and Labrador is cold (warm) (van Reference Van Loon and RogersLoon and Rogers, 1978). Sea surface temperatures in the Barents Sea and Labrador Shelf also follow this seesaw pattern. Hence, ice cover anomalies in the Barents Sea and Labrador Shelf could be attributed to atmospheric pressure anomalies over the Eurasian Basin.
The question raised in this paper is whether there is feedback by which the sea ice and the ocean affect the atmosphere. In the ocean–ice–air feedback process, air temperature and atmospheric circulation may be modified by the degree of sea-ice coverage, which reflects solar radiation efficiently and reduces the heat flux from the ocean to the atmosphere at high latitudes by insulation. Ocean circulation and sea-ice distribution are in turn affected by wind stress and air temperature. A crucial area for the ocean–ice–air feedback may be the Barents Sea, which is the northernmost ice-free area.
In this paper, historical data are presented and an hypothesis of air–ice–ocean interactions is proposed. A two-level Arctic Ocean model is used to examine oceanic response to variable wind stress.
Historical Data
Historical data have been discussed thoroughly by Ikeda (in press) and only a summary is presented here. Monthly anomalies of areal sea-ice extent in the Barents Sea are taken from Mysak and Manak (Reference Mysak and Manak1989) and plotted in Figure 1a. The anomalies, which are smoothed by a 25-month running mean, exhibit decadal oscillations. Annual air temperatures for the Barents Sea, taken from Kelly and others (Reference Kelly, Jones, Sear, Cherry and Tavakol1982), are shown in Figure 1b. The Barents Sea experienced a temperature minimum around 1968, reflecting the dominant cooling trend in the entire Arctic. No decadal oscillation is observed before 1965.
The mean atmospheric circulation in the northern hemisphere can be represented by a westerly jet at mid-latitudes. To examine variations, atmospheric pressure at the sea surface has been analyzed by Trenberth and Paolino (Reference Trenberth and Paolino1981). The first EOF of the annual mean pressure field has a dominant low anomaly over the Eurasian Basin and two minor high anomalies over the Pacific and Europe. The smoothed time series in Figure 1c exhibits decadal oscillations, with positive (negative) values corresponding to 0.5 – 1.5 mbar low (high) anomalies in the Eurasian Basin. In the period of the positive values, the meridional component of the mid-latitude jet contributes to northwesterly winds over Labrador and southwesterly winds in the Greenland Sea, as well as cyclonic wind-stress curl in the Barents Sea.
Monthly mean 700 mbar heights have been analyzed by Barnston and Livezey (Reference Barnston and Livezey1987). The time series are shown in Figure 1d for the first monthly EOFs in the three winter months (December–February), which are similar to the first annual EOF of the sea surface pressure. The interannual variations are similar to those of the annual EOF after 1965, whereas no decadal oscillation is visible before 1965 in the winter EOF. Ikeda (in press) has found that the time series are insensitive to heights at which the data were collected.
There is an inverse correlation between the Barents Sea ice and air temperature after 1966, whereas ice variability before 1966 is not explained by air temperature. The air temperature is more highly correlated with the winter EOF of the atmospheric pressure (correlation coefficient r of 0.75) than with the annual EOF (r = 0.45). Thus, air temperatures over the Barents Sea are dominated by winter temperatures, which are determined by the winter atmospheric circulation. Barents Sea ice is more highly correlated with the annual mean atmospheric circulation (r = 0.79) than Barents Sea air temperature (r = −0.52) and the winter circulation (r = −0.66). These correlations suggest that the dynamic mechanisms that drive the ocean with a long time-scale are more important for Barents Sea ice than thermodynamic mechanisms affecting sea ice in any one season.
Water temperature anomalies in the Kola section (33.5°E) in the Barents Sea have been re-analyzed by Leong (Reference Loeng1988). The time series of the water temperature in Figure 1e is highly correlated with the air temperature (r = 0.77). However, correlation between the water temperature and sea ice is low (r = −0.40). This is further evidence that Barents Sea ice is controlled more by a dynamic mechanism than by a thermodynamic mechanism. The salinity at a 300 m depth from Station Mike (66°N, 2°E) plotted in Figure 1f represents upper layer properties and shows decadal oscillations. By comparison with the annual pressure field, the salinity decreased while a low pressure anomaly over the Eurasian Basin was strong. In the time series, the salinity variations seem to have a phase lag of 3 years with atmospheric circulation.
Fig. 1. (a) Monthly anomalies of areal sea-ice extent in the Barents Sea; (b) annual air temperature in the Barents Sea (70–85°N, 20–50°E), where the original values are shown by solid circles, and lines show the values smoothed by the Hanning (3 year) filter; (c) the coefficient time series (open circles) of the first EOF of atmospheric pressure at sea-level along with the 3 year filtered time series; (d) the normalized time series of the first EOF of winter mean 700 mbar heights (solid circles); (e) water temperature anomalies at 0–200 m in the Kola section; and (f) salinity anomalies at 300 m, Station Mike (66°N, 2°E). On each of these plots the x-axis is in years.
The above analyses lead to the following hypothetical link between the sea ice and atmospheric circulation. The low-pressure anomaly in the Eurasian Basin tends to produce cyclonic wind-stress curl over the Barents Sea and Fram Strait area. Cyclonic ocean circulation generated by the wind-stress curl acts to replace cold Arctic water (including sea ice) by warm Atlantic water, reducing Barents Sea ice. Thus, low-pressure anomalies in the Arctic tend to reduce Barents Sea ice. The quantitative discussion given below shows that the observed wind-stress variations can produce the observed variations in ice cover. The correlation coefficients obtained above are sensitive to the sea-ice data in the 1950s, when the data had lower quality (Reference Walsh and JohnsonWalsh and Johnson, 1979). It should be mentioned that the importance of atmospheric circulation suggested here could be misleading.
A feedback mechanism from ice/ocean to atmospheric circulation is discussed next by referring to model studies. The general circulation model developed by the Canadian Climate Centre was run with two different ice covers (average and heavy), producing sea-level pressure higher on the extra ice-covered area by ~4 mbar (personal communication from Boer). As estimated using a linear wave model of Egger (Reference Egger1977), an extra ice-free area of 2 × 105 km2 in the Barents Sea for 3 months produces a low pressure anomaly of 3 mbar in winter or 1 mbar in the annual mean.
It is well known that the number of sunspots has a quasi-decadal oscillation. The number is highly correlated with the winter atmospheric circulation in the northern hemisphere (van Loon and Labitzke, Reference Van Loon and Labitzke1988). However, it is doubtful that this weak variation in solar activity can cause the atmospheric circulation anomalies by itself (Kerr, Reference Kerr1982) A possible mechanism to enhance the weak variation is positive feedback, by which responses of the atmosphere–ocean system are amplified, between the atmosphere and ocean.
Dynamic Ocean Model
In this section, an oscillatory response of the ocean to a wind stress is examined using a dynamic ocean model. The two-level model used in this paper is similar to that in Ikeda (Reference Ikeda1987). The model has a rigid lid and a frictionless bottom. Hydrostatic and Boussinesq approximation are used, and non-linear advection terms are omitted in the momentum equations. The velocity field can be divided into barotropic and baroclinic components. Acceleration and horizontal viscous terms are retained in the barotropic momentum equations. The wind stress is assumed to determine Ekman transport, which drives the barotropic component, in addition to bottom topographic effects with feedback from the baroclinic component. The hydrostatic approximation relates the pressure to the vertically integrated density, which determines the baroclinic component under a geostrophic assumption.
As shown in Figure 2, a simple model domain is used in this paper, including the Canadian Basin, Eurasian Basin, Barents Sea, and Greenland Sea. The southwestern boundary of the Greenland Sea is a closed boundary in the model. The upper level is 300 m thick, and the total depth is 3000 m except for the Lomonosov Ridge (1500 m), Fram Strait (2000 m), the Eurasian Basin (linear slope from 3000 m at the Barents Sea side to 4000 m at Lomonosov Ridge), and Barents Sea (300 m).
The density equations have non-linear advection terms, and horizontal and vertical diffusion terms. To simulate buoyancy-driven circulation in the Arctic–Atlantic, buoyancy and density (negative buoyancy) are given in the Canadian Basin and Greenland Sea, repectively. With this forcing, the Arctic water circulates clockwise in the upper level of the Canadian Basin and flows out via the Eurasian Basin to the Atlantic through Fram Strait. The Atlantic water enters the Arctic in the lower level as well as through the Barents Sea. The horizontal viscosity and diffusion coefficients are 103 m2 s−1, and a vertical diffusion coefficient is 10−4 m2 s−1. The numerical grid size is 100 km by 100 km.
Fig. 2. The ocean-model domain used in this paper.
Fig. 3. The upper level density, lower level density, upper level stream function and lower level stream function at (a) t = 27 years, and (b) t = 33 years, for the 12 year period case. The contour intervals of the stream functions are 0.2 × 106 m3 s−1 everywhere, except for the Greenland Sea, where they are 106 m3 s−1.
The initial density difference between the two levels is 0.5 kg m−3 in the Eurasian Basin. The initial density structure in the upper level is uniformly low (by −0.25) in the Canadian Basin and high (by 0.25) in the Greenland Sea. Although an equilibrium state is independent of the initial structure in the upper level, the solution converges faster with this initial condition than it does with a uniform upper level.
In addition to a steady wind-stress curl of 3 × 10−8 Nm2 over the Greenland Sea, an oscillatory wind-stress curl, with an amplitude of 10−8 Ν m−2, is applied over the Eurasian Basin, Barents Sea and Greenland Sea. The wind stress oscillates to be cyclonic in the first half period and anticyclonic in the second half. For a 12-year period, calculations are carried out for 36 years. For the third period (24–36 years), density structures, and stream-functions are shown in Figure 3 for the cyclonic (27 years) and anticyclonic (33 years) maximum wind stresses. The evolution of the solution is shown in Figure 4.
The solution tends to converge to an equilibrium state. As shown by the stream-function at the northwestern corner of the Barents Sea, Arctic outflow to the Barents Sea is reduced during the cyclonic period (Fig. 4a). This weaker Arctic outflow is exhibited also by the weaker upper level flow in the Eurasian Basin during the cyclonic period (Fig. 3). The variability of this flow is well related to the zonal density gradient in the lower level, which supports the southward upper level flow at the Barents Sea–Eurasian Basin boundary.
During the cyclonic period, the Barents Sea becomes denser by receiving more (warmer) Atlantic Water in the Greenland Sea. Here the higher density in the Barents Sea is interpreted as representing less ice. Thus, a cyclonic wind stress tends to reduce Barents Sea ice. The area with ρ < ρ0 – 0.35 kg m−3 in the Barents Sea, less by 2 × 105 km2 in the cyclonic period (Fig. 3), is comparable with the observed ice-cover variability (Fig. 1). The density variations have maxima and minima about 2 years later than the maximum and minimum of wind-stress curl.
Fig. 4. The evolution of (a) stream function at the northwestern corner of the Barents Sea, showing the Arctic outflow to the Barents Sea; (b) the meridional density gradients in the upper and lower levels at the Eurasian Basin–Barents Sea boundary; and (c) density gain in the Barents Sea.
Discussion
In the numerical calculation, the density variability in the Barents Sea has a 2 year delay to the wind stress, whereas there is no delay in the observations. A response time-scale in the Barents Sea depends on detailed flow field of the Arctic outflow; speed, width, vertical profile, and so on. The Barents Sea water is well mixed and gradually replaced by outside water in the present model. If the ice edge position in the Barents Sea responded more quickly to the Arctic outflow, the time lag would be reduced. Hence, baroclinicity, more detailed bottom topography and presence of sea ice should be included in a more comprehensive model of the Barents Sea. In this paper, however, a general circulation pattern in the entire Arctic is examined to show responses of Arctic outflow to the Barents Sea.
Coupling of a sea-ice model with the ocean model, inclusion of thermodynamics, more realistic geometry and consideration of the atmospheric responses to oceanic conditions will be included in future work.
Acknowledgements
The work in this paper was supported by the Federal Panel on Energy Research and Development.
