Abstract

Gridded sea surface temperature (SST), sea level pressure, 10-m wind field, and ice concentration data from the winters 1982–98 are used in a study of the large-scale variability of the SSTs in the Nordic Seas. Mean fields are extracted and areas of maximum variability identified. A complex principal component analysis is applied to identify coherent structure of variability in the SST field. The leading mode, accounting for 39% of the variance, reveals a band of high correlations at increasing phase lags along the west coast of Norway, across the Greenland Sea, and south through the Denmark Strait. The SST data contain two cycles of low-frequency variability during the 17 winters studied, having approximately 5-yr and 12-yr periods. For the eastern parts of the Nordic Seas, propagation speeds agree with other transport estimates. This indicates that SST anomalies are representing upper ocean heat anomalies advected by the mean flow from north of Scotland toward the Barents Sea. This is not true for the Greenland side of the Nordic Seas, where propagation speeds are unrealistically high. Composite maps of cold and warm years show that forcing from the wind field anomalies is likely to produce the observed SST anomalies here. No links are found between the SST anomalies in the Nordic Seas and anomalies south of the Iceland–Scotland Gap. It is therefore believed that the upper ocean heat anomalies are mainly created inside the Nordic Seas region.

1. Introduction

The mild climate in northern Europe is due to the northward flow of warm Atlantic water (AW). This water mass originates in the tropical Atlantic and enters the eastern parts of the North Atlantic as the North Atlantic Current (NAC). An estimated amount of 8 × 106 m3 s−1 of this slowly flowing water mass reaches the Nordic Seas (Worthington 1970), as defined by Hurdle (1986), including the Greenland, Iceland, and Norwegian Seas, and the western part of the Barents Sea (Fig. 1). The AW provides the heat and salt source for the Nordic Seas and the Arctic Ocean, and it plays a crucial role in the thermohaline circulation of the North Atlantic.

Fig. 1.

Map of the Nordic Seas based upon the ETOPO5 dataset. Isobaths are drawn at 250, 500, 1000, 1500, and 3000 m. Gray arrows illustrate the flow of warm Atlantic water, black arrows the flow of cold, polar water. The nomenclatures refer to names of currents defined in section 1.

Fig. 1.

Map of the Nordic Seas based upon the ETOPO5 dataset. Isobaths are drawn at 250, 500, 1000, 1500, and 3000 m. Gray arrows illustrate the flow of warm Atlantic water, black arrows the flow of cold, polar water. The nomenclatures refer to names of currents defined in section 1.

The inflow of AW to the Nordic Seas occurs through the Faroe–Shetland Channel, by a more indirect route across the Iceland–Faroe Ridge, and at the eastern side of the Denmark Strait. Inside the Nordic Seas, the major part of the AW is flowing along the shelf break off Norway, known as the Norwegian Atlantic Current (NWAC). Two main branches are flowing into the Barents Sea as the North Cape Current (NCC), and into the Fram Strait as the West Spitsbergen Current (WSC). Several branches of AW leave the main flow to enter the North Sea shelf or the central Norwegian and Greenland Seas.

The other main water mass in the Nordic Seas is the cold and fresh polar water (PW), leaving the Arctic Ocean at the western side of the Fram Strait. This water mass is flowing as the East Greenland Current (EGC) along the entire eastern coast of Greenland, with a major part entering the North Atlantic through the Denmark Strait. Two main branches are separating from the EGC to flow into the central Nordic Seas: the Jan Mayen Current, (JMC), making the southern rim of the Greenland gyre, and the East Icelandic Current, (EIC), flowing into the southern part of the Nordic Seas. Reviews of transport estimates for the currents in the Nordic Seas are given by Hopkins (1991) and Simonsen and Haugan (1996).

The climate variability in the Nordic Seas is closely connected to the North Atlantic Oscillation (NAO), a large-scale shift in atmospheric mass between the Icelandic low and the Azores high. The NAO index, which is the normalized difference in sea level pressure (SLP) between these two poles (Hurrell 1995), expresses the strength of the North Atlantic westerlies and is a good indicator of the climate in the North Atlantic region. During the last 4 decades, the NAO has shifted from being extremely weak in the 1960s, to being at a record high in the late 1980s and early 1990s (Fig. 2a). Superposed on the linear trend, there have been pronounced oscillations of 7–8-yr periods. Year-to-year variability in the SLP is large. From 1995 to 1996 the winter mean SLP over the central Nordic Seas increased by more than 22 mb, causing the large drop in the NAO index shown in Fig. 2b.

Fig. 2.

NAO winter index for (a) 1864–1998 and (b) 1982–98, updated from Hurrell (1995). Solid lines show the low-pass filtered index, resulting from applying a third-order Butterworth filter with cutoff period of 5 yr. The vertical axes show the normalized SLP at Lisbon, Portugal (Ln), minus the normalized SLP at Stykkisholmur, Iceland (Sn).

Fig. 2.

NAO winter index for (a) 1864–1998 and (b) 1982–98, updated from Hurrell (1995). Solid lines show the low-pass filtered index, resulting from applying a third-order Butterworth filter with cutoff period of 5 yr. The vertical axes show the normalized SLP at Lisbon, Portugal (Ln), minus the normalized SLP at Stykkisholmur, Iceland (Sn).

Hydrographic measurements from different branches of the AW flow show interannual and interdecadal variability in temperatures, having amplitudes exceeding 2°C (Dickson and Blindheim 1984). Variations in the heat transport into the marginal ice zone (MIZ) may have large impacts on the sea ice distribution, deep water renewal, biomass production, and local atmospheric climate, and due to albedo changes, it will modify the global radiation budget.

The objective for this paper is to improve the understanding of the origin and advection of the upper ocean heat anomalies in the Nordic Seas. To meet this objective, gridded sea surface temperatures (SSTs), together with SLP, sea ice, and 10-m wind field data, will be utilized. Only data from winter months December through March will be used. This is motivated by the strong seasonality of the thermocline. During winter, both weak vertical density gradients and strong wind forcing make a deep, homogeneous, upper layer, and the SSTs are therefore representing the heat storage of a significant part of the water column. In contrast, the upper mixed layer during summer is very shallow, only of the order of a few meters.

The paper is organized as follows: Section 2 contains a description of the datasets used in this study and the analysis techniques which are applied. In section 3 the SST variability in the Nordic Seas are analyzed in terms of coherent structures of variability. Origin and propagation of SST anomalies are discussed in section 4, before the paper is concluded with a short summary in section 5.

2. Data and analysis

The datasets used in this study are as follows. 1) Monthly mean SST fields blended from ship, buoy, and bias-corrected satellite data are from the Integrated Global Ocean Services System (IGOSS) Products Bulletin dataset (Reynolds and Smith 1994), gridded onto 1° × 1° grid boxes. 2) Monthly mean fields of SLP, sea ice cover, and 10-m wind field are obtained from the National Centers for Environmental Prediction–National Center for Atmospheric Research (NCEP–NCAR) reanalysis project (Kalnay et al. 1996). The SLP dataset is on a 2.5° × 2.5° grid, while the other datasets are on grids approximately 1.9° × 1.9°. 3) The NAO index is obtained from the Climate and Global Dynamics division of NCAR (Hurrell 1995).

For this study data from 17 winters from 1981/82 to 1997/98 are used. This period is regrettably short, but limited by the length of the IGOSS record.

Patterns of coherent structures in the SST winter data are analyzed in terms of complex principal components (CPCs). While standard (real) principal component (PC) analyses return patterns of standing oscillations only, the CPC method detects traveling waves and irregularly occurring features as well (Horel 1984). The following procedure is followed. 1) From all observations f(x, t) a complex dataset F(x, t) is formed, where the real and imaginary parts of F are, respectively, the observations f and the quadrature function of f (the Fourier components of f rotated π/2). 2) The complex covariance matrix is formed, containing all combinations of covariances between the time series at different grid points. 3) From the covariance matrix, the CPCs and the complex empirical orthogonal functions (CEOFs) are calculated. The complex dataset F can then be represented as

 
formula

where a(n, t) are the CPCs (time series) and (n) the CEOFs. The asterisk denotes complex conjugation. The CPCs are orthonormal [Σta*(n, t)a(m, t) = δnm], and the CEOFs orthogonal [Σx*(x, n)(x, m) = λnδnm], where δnm = 1 for n = m, and δnm = 0 for nm. The eigenvalues λn are proportional to the part of the total variance of F represented by the individual CEOFs. The standard errors for the eigenvalues are given by λn(1/N)1/2, where N is the number of independent time steps. If the standard error for an eigenvalue is small compared to the separation from the next eigenvalues, it follows that the sampling error is small, and the result can therefore be considered robust (North et al. 1982).

An element of a CEOF can be written

 
formula

and is therefore the complex dataset F regressed upon the nth CPC.

In this paper the gridded data have been weighted by the square root of cosine of latitude, so equal areas are afforded equal weights in the analysis. Afterward, the CEOFs have been divided by the same weighting function, then multiplied by the standard deviation of the CPCs. The regression coefficients shown in the maps are therefore the complex SST winter anomalies associated with an anomaly of one standard deviation in the CPC. As pointed out by Thompson and Wallace (2000), an effect of the weighting function is that the functions plotted in the regression maps are not strictly orthogonal.

Both the CPCs and the CEOFs can be written in terms of real amplitudes and phase angles. As the phase angles are known only to within an additive constant, they are here chosen so that the first element of the CPC is real and positive.

3. SST variability in the Nordic Seas

Map of the winter mean SLP (Fig. 3a) shows the well-known Greenland high (SLP > 1018 mb), and the Icelandic low (SLP < 996 mb) that stretches as a trough into the Nordic Seas. A secondary low is found in the central Greenland Sea. Toward the southeast, the SLP gradually increases, giving the prevailing westerlies over northern Europe. Strongest gradients are found between Iceland and Greenland, where winds are northeasterlies. Year-to-year variability is largest in the northeastern part of the Nordic Seas, with standard deviation in winter mean SLP exceeding 6 mb.

Fig. 3.

(a) Winter mean SLP (mb) calculated from the NCEP–NCAR dataset. The shading indicates areas with standard deviation of winter mean SLP exceeding 5 mb (light shading) and 6 mb (dark shading). (b) Winter mean SST (°C) calculated from the IGOSS dataset. The shading indicates areas with standard deviation of winter mean SST exceeding 0.5°C (light shading) and 1°C (dark shading). The thick solid line shows the winter mean ice edge (50% cover) as calculated from the NCEP–NCAR data, while the thick dashed lines show the minimum and maximum extent of the winter mean ice edge. Linear trends are removed prior to calculations of the standard deviations.

Fig. 3.

(a) Winter mean SLP (mb) calculated from the NCEP–NCAR dataset. The shading indicates areas with standard deviation of winter mean SLP exceeding 5 mb (light shading) and 6 mb (dark shading). (b) Winter mean SST (°C) calculated from the IGOSS dataset. The shading indicates areas with standard deviation of winter mean SST exceeding 0.5°C (light shading) and 1°C (dark shading). The thick solid line shows the winter mean ice edge (50% cover) as calculated from the NCEP–NCAR data, while the thick dashed lines show the minimum and maximum extent of the winter mean ice edge. Linear trends are removed prior to calculations of the standard deviations.

From the map of the winter mean SST (Fig. 3b), the pathways of the main currents and the positions of the main frontal systems can be deduced (cf. Fig. 1). The SSTs are ranging from above 9°C west of Scotland to the freezing point in the MIZ. Along the MIZ there is a wide band of large interannual variability and several places where the standard deviation exceeds 1°C. A tongue of high variability is also found along the East Icelandic Current. Unfortunately, both in situ and satellite observations are sparse near the ice edge. As a result, Reynolds and Smith (1994) used sea ice data and set the SST to the freezing point in grid cells with more than 50% ice cover. Changes in the ice edge position may therefore have too much impact on the SST values. Plots of the sea ice boundary (Fig. 3b) show that some areas with large SST variations coincide with areas having large fluctuations in the sea ice edge. However, the maximum SST variations are generally found several grid points away from the MIZ.

The leading mode of variability derived from a CPC analysis of the winter SST data is shown in Fig. 4. While 39% of the total variance in the complex dataset is accounted for by this mode, it explains more than 70% of the local variance in some of the areas with large variability (not shown). This CEOF mode is very robust, as the separation distance to the second CEOF (accounting for 22% of the variance) is three standard errors.

Fig. 4.

The leading CEOF mode of the winter mean SSTs 1982–98, accounting for 39% of the total variance. (a) Magnitude (°C) and (b) phase angle (°) of the complex SST winter anomalies regressed upon the standardized first PC. SST winter anomalies along the dashed line are shown in Fig. 5. The shading indicates confidence levels above 95% in (a), while in (b) it indicates areas where the SST contribution from the leading mode is positive in 1982. (c) Magnitude and (d) phase angle (°) of the standardized first PC.

Fig. 4.

The leading CEOF mode of the winter mean SSTs 1982–98, accounting for 39% of the total variance. (a) Magnitude (°C) and (b) phase angle (°) of the complex SST winter anomalies regressed upon the standardized first PC. SST winter anomalies along the dashed line are shown in Fig. 5. The shading indicates confidence levels above 95% in (a), while in (b) it indicates areas where the SST contribution from the leading mode is positive in 1982. (c) Magnitude and (d) phase angle (°) of the standardized first PC.

In contrast to the second CEOF mode (not shown), the leading CEOF mode shows a well-organized pattern. The correlation between the CPC and the local SST time series is above the 95% confidence level in a band stretching along the coast of Norway, across the Greenland Sea, and south to the Denmark Strait (Fig. 4a). Also, the eastern part of the Barents Sea and some isolated areas in the southern part of the Nordic Seas show high correlation. South of the Iceland–Scotland Gap, correlations are low.

The amplitudes of the regression coefficients have maxima in the Barents Sea, Greenland Sea, and in the Denmark Strait, the areas where the standard deviation of the winter mean SSTs is largest (Fig. 3b).

The phase angles of the regression coefficients are increasing in a cyclonic direction along the band of high correlation in the Nordic Seas (Fig. 4b). With the CPC vectors rotating in a clockwise direction (Fig. 4d), the SST signals move northward along the Norwegian coast, across the Norwegian and Greenland Seas, and south along the east Greenland coast. A 180° phase reversal is found across Iceland. The CPC is seen to consist of almost two complete cycles during the 17 yr that are studied. The two periods are, respectively, 5 and 12 yr, corresponding to the frequencies 72° and 30° yr−1.

Plots of the SST anomalies along the high correlation band (thick dashed line in Fig. 4a) give a more detailed picture of the coherent structure of variability (Fig. 5a). During the first cycle, the SST maxima north of Scotland are leading the corresponding maxima at the western coast of Norway (600 km) by 6 months, at the northern coast (1500 km) by 11 months, in the central Greenland Sea (2000 km) by 16 months, and in the Denmark Strait (3800 km) by 18 months. Thus the signal moves with speeds 4, 7, 3, and 60 cm s−1 over each of the four segments (upper dashed line in Fig. 5). During the second cycle, the time lags are more than twice as large, and the corresponding signal speeds reduced by more than 50%. For comparison, the observed SST along the high correlation band is shown in Fig. 5b. More structures are seen here, but most of the variability is evidently covered by the leading CEOF mode.

Fig. 5.

Time–distance (Hovmoeller) diagram of (a) winter SST anomalies reconstructed from the leading CEOF, and (b) measured winter SST anomalies, along the track shown as the thick, dashed line in Fig. 4a. The numbers above the plot refer to the distance marks. The dashed lines emphasize phase propagation. Units on the shading bars are in °C.

Fig. 5.

Time–distance (Hovmoeller) diagram of (a) winter SST anomalies reconstructed from the leading CEOF, and (b) measured winter SST anomalies, along the track shown as the thick, dashed line in Fig. 4a. The numbers above the plot refer to the distance marks. The dashed lines emphasize phase propagation. Units on the shading bars are in °C.

4. Origin and advection of the SST anomalies

The observed temperatures in the upper ocean will be a function of 1) the advection of heat by the ocean currents, 2) atmosphere–ocean heat fluxes, and 3) the heat exchange due to vertical mixing (convection). This can be illustrated by assuming a steady state, one-dimensional flow of an incompressible water column of constant mixing depth (h) and speed (u) along the x axis. The general heat equation (Gill 1982, p. 71) can be written

 
formula

where T is the temperature, Qatm the heat flux from the atmosphere, Qdeep heat flux from the deep ocean (by convection), ρ the density, and cp the specific heat.

In general, the upper ocean temperature response to any changes in the atmospheric circulation can be influenced by all three mechanisms. During years with anomalous strong wind forcing, the heat loss to the atmosphere will be amplified, but the advection speeds, mixing depths, and the rate of convection may also become anomalous large.

The coherent cyclonic movements of the SST signals in the Nordic Seas are in the direction of the mean flow in the area. This indicates that the SST signals are advected by the ocean currents. However, there is no correlation with reported SST anomalies in the North Atlantic (Hansen and Bezdek 1996; Sutton and Allen 1997), suggesting that the upper ocean heat anomalies are either generated inside the Nordic Seas or advected into the Nordic Seas deeper in the water column.

The estimated speed of the signals can be compared to other findings in the Nordic Seas: The great salinity anomaly of the 1960s and 1970s was advected by a speed of 3 cm s−1 (Dickson et al. 1988), sea ice anomalies between the Greenland and Labrador Seas by a speed of 3.2 cm s−1 (Mysak and Manak 1989), and anomalous warm water propagating from the Fram Strait to the Denmark Strait (Dickson et al. 1999) can be estimated to a speed of 2 cm s−1. Time series from various hydrographical sections on the AW side of the Nordic Seas indicate the same advection speeds (Blindheim et al. 1998; Loeng 1998). Thus it seems that the SST anomalies in the dataset (Fig. 5) have reasonable advection speeds along the Norwegian coast and into the Greenland Sea. However, the movement south to the Denmark Strait is too fast to be explained by advection alone. Nor can advection by ocean currents explain why the speed of the anomalies are reduced by more than 50% during the data period, as no documentation of a similar reduction in the background flow field exist.

If the leading CPC of the winter SST anomalies (Fig. 4) is rotated 116°, the correlation with the NAO winter index is at its maximum, 0.71. Thus maximum positive and negative correlations are found where the CEOF phase angle is, respectively, 116° and 296° (Fig. 6a). The general trend is that for winters having a positive NAO phase, the SST field shows positive anomalies in the southeastern parts of the Nordic Seas and negative anomalies at the western side. This is to be expected as the cyclonic circulation is strengthened, resulting in a transport of more warm, moist air in over the southeastern part of the Nordic Seas and more cold, dry air in over the northwestern part. But not all is explained by this, for example, the isolated warm areas in the Greenland Sea.

Fig. 6.

Linear regression between NAO winter index and (a) SST winter anomalies regressed upon the leading CEOF and (b) measured SST winter anomalies. Negative regression slopes are drawn as dashed lines, positive as solid lines. Equidistance is 0.05°C. Shaded areas show confidence levels above 95%.

Fig. 6.

Linear regression between NAO winter index and (a) SST winter anomalies regressed upon the leading CEOF and (b) measured SST winter anomalies. Negative regression slopes are drawn as dashed lines, positive as solid lines. Equidistance is 0.05°C. Shaded areas show confidence levels above 95%.

The regression between NAO and the measured SST anomalies show an even stronger relationship than the correlation based on the leading CEOF (Fig. 6b). Thus other modes of SST variability must also be forced by the NAO.

From the datasets composite fields of four cold years and of four warm years are calculated (Fig. 7). From the SST field reconstructed from the leading CEOF, mean temperature anomalies (T) along the high correlation band (dashed line in Fig. 4a) are calculated. The cold years (1982, 1987, 1988, and 1989) are defined by T < −0.2°C and similarly the warm years (1984, 1991, 1993, and 1995) by T > 0.2°C. For comparison, the standard deviation is 0.25°C.

Fig. 7.

Map of the mean SST winter anomaly, 10-m wind anomaly (arrows), and ice edge (thick line) for (a) cold winters and (b) warm winters. Negative SST anomalies are drawn as dashed lines, positive as solid lines. Equidistance is 0.25°C. Shaded areas show anomalous strong wind speeds. For selection criteria, see text.

Fig. 7.

Map of the mean SST winter anomaly, 10-m wind anomaly (arrows), and ice edge (thick line) for (a) cold winters and (b) warm winters. Negative SST anomalies are drawn as dashed lines, positive as solid lines. Equidistance is 0.25°C. Shaded areas show anomalous strong wind speeds. For selection criteria, see text.

Associated with the cold anomaly (Fig. 7a), there are anomalous easterlies off the Norwegian coast, strong northerlies in the Barents Sea, and northeasterlies over the Greenland Sea and into the Denmark Strait. The following forcing mechanisms seem to be responsible for the cold surface water. 1) Easterlies are transporting cold, dry air off Norway, increasing sensible and latent heat loss, and also the longwave radiation. 2) Northerlies are transporting cold, dry Arctic air off the ice, having a similar but much stronger effect than the off-coast wind farther south, as also the wind speed is strengthened. The four cold years derived from this mode happen to be the years in the data period with largest extent of sea ice in the Barents Sea (Loeng 1998). 3) Along the cold anomaly band at the western side of the Nordic Seas, stronger than normal northeasterlies are cooling the surface mostly by an enhanced sensible heat loss.

During the warm years (Fig. 7b), strengthened westerlies (positive NAO phase) are warming the eastern part of the Nordic Seas and the Barents Sea. Along the East Greenland MIZ, SST anomalies are positive despite strong northerlies and partly off-ice winds. A possible explanation is that the front between cold and warm waters is forced toward Greenland. The area of maximum SST anomaly is the position of the maximum wind stress curl anomaly, which may indicate that the warm anomaly is generated by upwelling of deeper and warmer water.

The 180° phase reversal across Iceland (Fig. 4b) is evident in the composite fields. The east–west dipole seems to be due to wind-forced variability in the East Icelandic Current, as proposed by Blindheim et al. (2000), who found a good correlation between the amount of PW in the southern Nordic Seas and the NAO winter index.

5. Summary and final remarks

In this paper datasets of SST, SLP, 10-m wind field, and ice concentration for the Nordic Seas have been analyzed for the winters ending in 1982–98.

By applying complex principal component (CPC) analysis, it is demonstrated that there is a cyclonic movement of SST anomalies in the Nordic Seas. No connections are found to SST anomalies in the North Atlantic, clearly indicating that the strong atmospheric forcing in the Nordic Seas is capable of masking any SST signal advected with the AW flow into the area. Two periods of, respectively, 5 and 12 yr have been found in the dataset. The first period corresponds to propagating speeds of 3–7 cm s−1 from the north of Scotland, along the Norwegian coast, and into the Greenland Sea. This is in agreement with other findings. Along the coast of Greenland, the analysis yields a speed of 60 cm s−1 which cannot be due to oceanic advection alone. Composite fields of cold and warm years show that the wind field may produce the observed variations in SST and ice cover in the Barents Sea and along the east Greenland coast.

From the results of this paper, the SST signals are propagating in a cyclonic direction in the Nordic Seas. In the southern part of the area, the SST anomalies are generated by anomalous atmospheric fluxes linked to the NAO phase or by subsurface advection of heat. The SST anomalies are then transported eastward and northward along the Norwegian coast by the mean flow of the ocean, ending up in the Barents Sea and Greenland Sea, where the original SST signal is amplified or masked by the local atmospheric fluxes.

It is uncertain to what degree the propagating of the SST anomalies will influence the atmosphere, and if the SSTs are actively participating in the 7–8-yr NAO period observed during the last four decades. Until time gives more data, output from a coupled atmosphere–ocean model with focus on the North Atlantic and the Nordic Seas, should be analyzed in order to 1) quantify the relative effects of oceanic heat advection and atmospheric forcing on the SST anomalies, and 2) quantify atmospheric effects of heat anomalies advected by the ocean.

Acknowledgments

This paper is based on ideas emerging during the author’s Ph.D. study at the Geophysical Institute, University of Bergen, Norway, and at the Polar Science Center, University of Seattle, Washington. The author would like to thank Arne Foldvik and Knut Aagaard for fruitful discussions and sober guidance during these years of work. Parts of the work have been supported by the Norwegian Research Council through the RegClim Project. The quality of this paper was greatly improved by suggestions from anonymous reviewers.

REFERENCES

REFERENCES
Blindheim, J., J. Aure, and E. Svendsen, 1998: Økosystemene i Norskehavet og på kysten. Havets Miljø 1998. Fisken og Havet, J. Aure, Ed., Institute of Marine Research, 23–31.
——, V. Borovkov, B. Hansen, S. A. Malmberg, W. R. Turrell, and S. Østerhus, 2000: Upper layer cooling and freshening in the Norwegian Seas in relation to atmospheric forcing. Deep-Sea Res., in press.
Dickson, R. R., and J. Blindheim, 1984: On the abnormal hydrographic conditions in the European Arctic during the 1970s. Rapp. P.-V. Réun. Cons. Int. Explor. Mer, 185, 201–213.
——, J. Meincke, S.-A. Malmberg, and A. J. Lee, 1988: The “Great Salinity Anomaly” in the northern North Atlantic 1968–1982. Progress in Oceanography, Vol. 20, Pergamon Press, 103–151.
——, J. Meincke, I. Vassie, J. Jungclaus, and S. Østerhus, 1999: Is variability of the Denmark Strait overflow predictable? Nature, 397, 243–246.
Gill, A. E., 1982: Atmosphere-Ocean Dynamics. Academic Press, 662 pp.
Hansen, D. V., and H. S. Bezdek, 1996: On the nature of decadal anomalies in North Atlantic sea surface temperature. J. Geophys. Res., 101, 8749–8758.
Hopkins, T. S., 1991: The GIN Sea—A synthesis of its physical oceanography and literature review 1972–1985. Earth Sci. Rev., 30, 175–319.
Horel, J. D., 1984: Complex principal component analysis: Theory and examples. J. Climate Appl. Meteor., 23, 1661–1673.
Hurdle, B. G., Ed., 1986: The Nordic Seas. Springer Verlag, 777 pp.
Hurrell, J., 1995: Decadal trends in the North Atlantic Oscillation: Regional temperatures and precipitation. Science, 269, 676–679.
Kalnay, E., and Coauthors, 1996: The NCEP/NCAR 40-year Reanalysis Project. Bull. Amer. Meteor. Soc., 77, 437–471.
Loeng, H., 1998: Økosystemet i Barentshavet. Havets Miljø 1998. Fisken og Havet, J. Aure, Ed., Institute of Marine Research, 13–17.
Mysak, L., and D. Manak, 1989: Arctic sea-ice extent and anomalies, 1953–1984. Atmos.-Ocean, 27, 376–405.
North, G. R., T. L. Bell, R. F. Cahalan, and F. J. Moeng, 1982: Sampling errors in the estimation of empirical orthogonal functions. Mon. Wea. Rev., 110, 699–706.
Reynolds, R., and T. Smith, 1994: Improved global sea surface temperature analyses. J. Climate, 7, 929–948.
Simonsen, K., and P. M. Haugan, 1996: Heat budgets of the Arctic Mediterranean and sea surface heat flux parameterizations for the Nordic Seas. J. Geophys. Res., 101 (C3), 6553–6576.
Sutton, R. T., and M. R Allen, 1997: Decadal predictability of North Atlantic sea surface temperature and climate. Nature, 388, 563–567.
Thompson, D. W. J., and J. M. Wallace, 2000: Annular modes in the extratropical circulation. Part I: Month-to-month variability. J. Climate, 13, 1000–1016.
Worthington, L. V., 1970: The Norwegian Sea as a mediterranean basin. Deep-Sea Res., 17, 77–84.

Footnotes

Corresponding author address: Tore Furevik, Geophysical Institute, University of Bergen, Allégaten 70, N-5007 Bergen, Norway.