GEOPHYSICAL RESEARCH LETTERS, VOL. 27, NO. 16, PAGES 2569-2572, AUGUST 15, 2000

Meridional flow variability over the Nordic seas in the Oscillation framework Paul Skeie Geophysical Institute, University of Bergen, Bergen, Norway

Abstract. An anomalous recurring pronounced than during cold Siberian months. Since their pattern of high relevance for the of the Nordic Seas focus was on Siberian temperatures, they did not stress and Siberia is identified. It is found as the second Empiri- the meridional structure in the Nordic Seas of the Siberian cal Orthogonal Function (EOF) of monthly sea level “warm minus cold” SLP composite (their Figure 2c). In this pressure (SLP) anomalies poleward of 30◦N where the lead- note, further evidence of the association between anomalous ing EOF is the Arctic Oscillation (AO). The most prominent northerly winds over the Nordic Seas and Siberian warm centre of action of the circulation pattern is located over the months is presented. Barents Region. This “Barents Oscillation” (BO) is shown to have a high temporal correlation with the sensible heat Data and analysis methods loss of the Nordic Seas (r=0.76). The BO also correlates to Eurasian surface air temperature (SAT) anomalies with Monthly SLP, SAT and surface sensible heat flux (SHF) ◦ r=0.72 after the AO related SAT variations are removed by anomalies 1958 to 1999 polewards of 30 N are constructed means of a linear regression. Two sets of SLP composites are for the winter months December through March (DJFM) constructed where one is based on low and high Nordic Seas based on data from the NCEP/NCAR reanalysis [Kalnay ◦ ◦ heat loss months and the other is based on warm and cold et al., 1996]. The horizontal resolution is 2.5 × 2.5 for ◦ ◦ Eurasian months. Patterns reminiscent of the BO emerge in SLP and SAT data and 1.9 × 1.9 for SHF data. An EOF the two composites when AO related variability is removed. analysis is performed on the non-normalised SLP anoma- lies. Then SLP and SAT anomalies are regressed upon Introduction the principal components (PCs) so that maps show the anomaly associated with one positive standard deviation Oceanic convection and the variability of the Arctic of the PCs. Time series of the areally averaged Nordic Ocean ice export through the Fram Strait and the Barents Seas SHF anomaly and of Eurasian SAT anomalies are con- Sea is related to persistent anomalous northerly winds lead- structed. Then the monthly SLPs are stratified according ing to surface heat loss, sea-ice advection, and formation. to these time series and SLP composites are formed based On synoptic timescales, events with strong off-ice flows are on months when the time series are more than one standard commonly referred to as Cold Air Outbreaks (CAOs). In deviation off from the mean. this study, a new perspective on CAOs in the Nordic Seas is presented, a perspective which is inextricably linked to the Arctic Oscillation (AO) framework. The AO is a deep Results atmospheric mode of variability with signature extending The two leading EOFs are shown in Figure 1 where the throughout the troposphere and far into the stratosphere leading EOF is the AO, accounting for 36.8% of the vari- [Thompson and Wallace, 1998; hereafter TW98]. The rather ability. The AO is well separated from the other EOFs ac- zonal AO pattern is found as the leading empirical orthog- cording to the criterion of North et al., [1982]. It is plotted onal function (EOF) of monthly and seasonal extratropical in the phase with positive SLP anomalies in the Arctic (the SLP anomalies in the Northern Hemisphere. At low levels, low phase, using the sign convention introduced by TW98). the AO is linked to variations in the strength of the west- As shown in TW98 and in Figure 1c, variations in the AO erlies across the North-Atlantic, in CAOs from the Cana- are associated with a seesaw in temperatures between the dian Arctic towards the Labrador Sea and in advection of eastern seaboard of Canada and Eurasia. In general, the marine airmasses onto the Eurasian continent. The winter AO SLP pattern has weak gradients in the Nordic Seas, the mean SLP distribution in the Nordic Seas is characterised northern parts in particular. The important meridional flow by a trough stretching from the Icelandic Low and north- variability, with many implications for the Arctic and Nordic eastward, into the Norwegian and Barents Seas. In their Seas, may consequently be controlled by residual SLP vari- search of monthly anomalous flow configurations explaining ability not described by the AO. However, the time series the warm Eurasian of the 1980s, Rogers and Mosley- of 167 months is too short for additional modes to clearly Thompson, [1995] constructed SLP composites of warm and stand out, and the quality of Arctic SLP data further back cold months near the 90th meridian in Siberia. They found is questionable since SLP data from drifting stations only that during warm Siberian months, the north-eastward ex- began to be incorporated on a regular basis from the early tension of the Icelandic Low and the stormtrack was more 1950s [Hastings, 1971]. It turns out that EOF 2, accounting for 10.3% of the Copyright 2000 by the American Geophysical Union. variance, has a major centre of action over the Barents re- gion (Figure 1). The second EOF is not well separated from Paper number 2000GL011529. the third and fourth EOFs (accounting for 8.9% and 8.2% 0094-8276/00/2000GL011529$05.00 of the variance respectively). EOFs that are poorly sepa- 2569 2570 SKEIE: MERIDIONAL FLOW VARIABILITY

Table 1. Summary of temporal correlation statistics for importantly, it will be shown that EOF 2-like patterns may DJFM. also be identified using composites. EOF 2 has a remark- a c c able meridional structure across the Arctic and the Nordic Time series SHF TEur NAO SCAND Seas. Anomalous northerly flows in the Nordic Seas associ- ated with EOF 2 will be concurrent with anomalous south- PC 1 0.23 0.63 0.68 PC 2 0.76 0.56 0.13 0.65 westerly flows in central Siberia. The strongest northerly PC 2-1b 0.72 flows in the Nordic Seas as described by EOF 2 took place PC 1-2b 0.76 during the March months of 1961 and 1968. PC 2 is dom- PC 1+2b 0.84 inated by intraseasonal and interannual variability and has no significant trend.

a The SAT anomalies associated with EOF 2 display a Anomaly averaged over the Nordic Seas dipole pattern between the Nordic Seas and central Siberia. bPC 2-1(1-2) : Variations coherent with PC 1(2) are re- One standard deviation below the mean of PC 2 is associ- moved by means of a linear regression prior to calculating ◦ the correlation coefficient. 1+2 : PC 1 and PC 2 are added ated with temperatures 3.5 K below the mean around 80 N, prior to calculating the correlation coefficient. 30◦E east of Spitsbergen and 2.5 K above the mean around c NAO is the North Atlantic Oscillation index [Hurrell,1995]. 60◦N, 90◦E in Siberia. TW98 related the AO time series SCAND is similar to EU1 [Barnston and livezey,1987] and ◦ to Eurasian SAT anomalies (TEur) over the domain 40 - was obtained from the Climate Prediction Center. 70◦N, 0◦-140◦E with correlation coefficient r=0.65 for the period December through March. This is in fair agreement with r=0.63 found in the present analysis. The temporal rated should in general not be payed too much attention correlation coefficient between EOF 2 and TEur is 0.56. Af- to in their own right, as they may fail to be reproducible. ter removing the coherent variability of the AO time series However, EOF 2 is not sensitive to extending the winter sea- from the TEur time series by means of a linear regression, son with Octobers, Novembers, Aprils and Mays, and more the residual TEur time series relates to PC 2 with r=0.72.

a)

2 PC 1 DJFM 2 PC 2 DJFM 0 0 r=−0.76 −2 −2 PC 2 1960 1970 1980 1990 2000 1960 1970 1980 1990 2000 −2 180oW 180oW b) SLP PC 1 SLP PC 2 0 2 2 0 90 90 E E o o o o

W W −2

90 90 SHFA Nordic Seas 1960 1970 1980 1990 2000 Year 85 0o 0o o o o N 180 W 180 W 80 c) SAT PC 1 SAT PC 2 o N 75 o N

90 90 70

E E o o o o o

W W N

90 90 65 o N

30 5 o o E W 20 60 o o o 0 0 W 6 o E Figure 1. Time series and SLP and SAT variations associated 0 o o 40 20 E with EOF 1 (left panels) and EOF 2 (right panels) of SLP. Nor- malised principal component (PC) time series are shown in a) and SLP and SAT anomalies regressed upon the PCs are shown Figure 2. Upper: Normalised PC of EOF 2 (note that the ver- in b) and c) respectively. In a), every second winter is shaded tical axis is reverted) and normalised time series of the anomalous grey and December values are drawn on the boundary between areally averaged sensible heat flux of the Nordic Seas for DJFM. grey and white. The normalised PC of each month of the winter Lower: Grid point correlation of the SHF with the PC of EOF 2. (DJFM) is drawn as a thin line. The heavy line is a 12 The labels have been multiplied with -10 and only the -0.5 and months (3 winters) running mean. In b) and c) positive contours -0.6 contours are shown. Areas where the standard deviation of are solid, the zero contour is dash-dotted and negative contours the SHF exceeds 90 Wm−2 are shaded grey. The box indicates are dashed. The contour spacing is 1 hPa in b) and 0.5 K in c). the averaging area of the sensible heat flux.

SKEIE: MERIDIONAL FLOW VARIABILITY 2571

W W 180 W 180 o

o HIGH HEAT LOSS LOW HEAT LOSS (27 high and 30 low SHF months). The low minus high a) b) composite in Figure 3c has similarities with the regression pattern of PC 2 in Figure 1b. In Figure 3d the difference

90 90 composite in 3c is reconstructed by first removing SLP vari- E E o o o o W W ations associated with the AO. Although discrepancies may 90 90 be found, there is a striking resemblance between the pat- tern in Figure 3d and the SLP regression pattern of PC 2 in Figure 1b.

0o 0o ThesameproceduremayberepeatedwiththeTEur time

W W 180 W 180

o o LOW MINUS HIGH series (25 warm and 26 cold months). The Eurasian warm LOW MINUS HIGH WITH AO REMOVED c) d) and cold SLP composites in Figure 4 has an AO-like struc- ture, confirming the notion that Eurasian SATs are strongly modulated by the strength of the zonal circulation. How- 90 90

E E ever, the SLP patterns have a well defined centre of action o o o o W W

90 90 over the Barents region, and when the AO related variabil- ity is removed from the “cold minus warm”-composite, a pattern with striking resemblance with EOF 2 again arises.

o o 0 0 Concluding remarks Figure 3. SLP composites of months with anomalously high The most prominent centre of action of the recurring and low sensible heat loss in the Nordic Seas. a) Anomalously anomalous circulation pattern identified in this study lies high, b) anomalously low, c) low minus high and d) low minus over the Barents region, hence a suitable name could be the high with SLP anomalies associated with the AO subtracted. The contour spacing is 1 hPa in a) and b) and 2 hPa in c) and d). “Barents Oscillation” (BO). All three main centres of ac- Positive anomaly contours are solid, negative are dashed and the tion of EOF 2, the Barents-, the western Arctic/Atlantic- zero isobar is dash-dotted. In c) the composite SLP of high and and the Pacific centres of action are supported by the com- low sensible heat loss months differ at the 99 % confidence level posites. While the shape of EOF 2 is restricted to being inside the dotted contour according to a Student’s t-test. orthogonal to EOF 1, the composites are not, hence, the composites lends credence as to the authenticity of EOF 2. The EOF analysis has been repeated using the NCAR SLP dataset [Trenberth and Paolino, 1980] for the period 1899- Conversely, if the coherent PC 2 variability is removed from 1947. The EOF 2 found in this early 20th century data the TEur time series, the AO PC correlates to the remaining does not support the BO pattern. Whether this is a true TEur variability with r=0.76. Combined, PC 1 and 2 relates sign of differing leading modes of variability or a result of to the TEur time series with an r=0.84. A summary of the poor data coverage over the Arctic in this dataset, remains correlation statistics is given in Table 1. an open question. Of previously identified Strong northerly winds lead to high SHF, but might as patterns, the BO is probably most closely related to EU1 well advect ice southwards and/or lead to local ice freezing [Barnston and Livezey, 1987] (see Table 1). that may in turn suppress the SHF. The Arctic sea-ice con- Most of the sea-ice formed inside the leaves centrations also exhibit decadal variations [Slonosky et al., through the Fram Strait between Spitsbergen and Green- 1997] and the varying ice-cover is reflected in the large stan-

dard deviation of the SHF near the ice edge as shown in

W W 180 W 180

o Figure 2. Grid point correlations between PC 2 and the o EURASIAN WARM EURASIAN COLD SHF are consequently greatest over areas with permanently a) b) open waters. In the Norwegian Sea, correlations between 0.6 and 0.7 are found over large areas. However, the largest 90 90 E E heat fluxes in the Nordic Seas are found in proximity of o o o o W W the ice edge [H¨akkinen and Cavalieri, 1989]. In Figure 2 90 90 the normalised PC 2 is plotted together with the time se- ries of the normalised areally averaged SHF anomaly in the Nordic Seas. The SHF is averaged over the domain 20.6◦W- o o

50.6◦E, 65.7◦N-82.8◦N, indicated in the lower panel of Fig- 0 0

W W 180 W 180

o o COLD MINUS WARM COLD MINUS WARM ure 2. There appears to be a close correspondence between WITH AO REMOVED these two time series, confirmed by the high temporal cor- c) d) relation coefficient of r=0.76. This is higher than the corre-

lation in any grid point, emphasizing the importance of the 90 90 E E o o o o large heat fluxes near the ice edge on the total sensible heat W W loss. The corresponding correlation coefficient for the AO is 90 90 low (r=0.23), consistent with the weak gradients in the AO SLP pattern along the ice edge.

The robustness of the link between northerly winds and 0o 0o the heat loss can be tested by constructing SLP compos- ites based on months when the SHF anomaly (as shown in Figure 4. Same as Figure 3, but for Eurasian temperature Figure 2) is more than one standard deviation off from zero anomalies. 2572 SKEIE: MERIDIONAL FLOW VARIABILITY

land, representing a major fresh water source for the North from the NOAA-CIRES Climate Diagnostics Center, Boulder, Atlantic [Aagaard and Carmack, 1989]. An analysis of ob- Colorado, from their web site at http://www.cdc.noaa.gov. This served ice thickness in the Fram strait [Vinje et al., 1998], work was funded by project number 112900/720 at The Research indicates that the variability of the ice export is closely re- Council of Norway. lated to the variability of the local atmospheric forcing. The SLP difference across the Fram strait for the AO and BO is References roughly 0 and 2 hPa, respectively (Figure 1b). Taking the Aagaard, K., and E. C. Carmack, The role of sea ice and other width of the Fram Strait to be 550 km, 2 hPa is equivalent to freshwater in the Arctic circulation, J. Geophys. Res.-Oceans, 2ms−1 geostrophic wind. Hence, a one standard deviation 94 , 14,485–14,498, 1989. Barnston, A., and R. Livezey, Classification, seasonality and change in the BO time series in Figure 1a is associated with persistence of low-frequency atmospheric circulation patterns, a change in the average monthly geostrophic wind along the Mon. Wea. Rev., 115 , 1083–1126, 1987. −1 Fram Strait of 2 ms . Deser, C., and J. E. Walsh, and M. S. Timlin, Arctic Sea Ice This study seems to suggest that the month-to-month Variability in the Context of Recent Atmospheric Circulation variability of climate parameters such as SATs and SHFs Trends, J. Climate, 13 , 617–633 , 2000. Dickson, R., J. Lazier, J. Meinke, P. Rhines, and J. Swift, Long- in the Nordic and Labrador Seas are related to EOFs un- term coordinated changes in the convective activity of the correlated in time. The correlation between the BO and North Atlantic, Prog. Oceanogr., 38 , 241–295, 1996. the NAO index (see Table 1) is also low. This is some- Furevik, T, Annual and interannual variability of Atlantic Water what surprising since several investigators have reported of temperatures in the Norwegian and Barents Seas: 1980-1996, an out-of-phase relation between the climate of these ar- Deep Sea Res., 47 , in press, 2000. eas to the east and west of (e.g. [Dickson et al., Hastings, A. D., Surface basin, Tech. Rep. ETL-TR-71-5 , U.S. Army Engineer Topographic Laborato- 1996; Deser et al., 2000]). I will offer two possible explana- ries, Fort Belvoir, Va, 1971. tions for this apparent discrepancy: Firstly; even though the H¨akkinen, S., and D. Cavalieri, A study of Oceanic Surface Heat AO has a weaker SAT signal in the Nordic Seas than does Fluxes in the Greenland, Norwegian, and Barents Seas, J. Geo- the BO, the AO has apparently more variability on inter- phys. Res.-Oceans, 94 , 6145–6157, 1989. decadal timescales. The time integrated AO-forcing of the Hurrell, J., Decadal trends in the North Atlantic Oscillation: Re- interannually auto-correlated sea-ice may consequently be gional temperatures and precipitation, Science, 269 , 676–679, 1995. substantial. Secondly; on interannual timescales the ocean Kalnay, E., et al., The NCEP/NCAR 40-year reanalysis project, may communicate information about the phase of the AO Bull.Amer.Meteor.Soc., 77 , 437–471, 1996. from the strong AO impact (steep SLP gradients) region to North,G.R.,T.L.Bell,R.F.Cahalan,andF.J.Moeng,Sam- the south of , towards the Arctic via the inflow of pling errors in the estimation of empirical orthogonal functions, Atlantic waters to the Nordic Seas [Furevik, in press 2000]. Mon. Wea. Rev., 110 , 699–706, 1982. Rogers, J., and E. Mosley-Thompson, Atlantic Arctic cyclones The composites in Figure 4 seem to filter out all SLP and the mild Siberian winters of the 1980s, Geophys. Res. Lett., variability but that associated with the AO and BO, in- 22 , 799–802, 1995. dicating that the most systematic SLP difference between Slonosky, V. C., L. A. Mysak, and J. Derome, Linking Arctic Eurasian warm and cold months lies in the phase of the Sea-Ice and Atmospheric Circulation Anomalies on Interan- AO and the BO. Adding positive and negative phases of the nual and Decadal Timescales, Atmos. Ocean, 35 , 333–366, SLP regression patterns in Figure 1b to the long term Jan- 1997. Thompson, D. W. J., and J. M. Wallace, The Arctic Oscillation uary mean (not shown) reveals that variations in both the signature in the wintertime geopotential height and tempera- AO and the BO time series control whether the isobars run ture fields, Geophys. Res. Lett., 25 , 1297–1300, 1998. parallel or normal to the western seaboard of the Eurasian Thompson, D. W. J., and J. M. Wallace, and G. C. Hegerl, An- continent. This suggests that the BO modulates Eurasian nular Modes in the Extratropical Circulation Part II: Trends, SATs in much the same way as the AO; by affecting the J. Climate, 13 , 1018–1036 , 2000. Trenberth, K., and D. Paolino, The Northern Hemisphere Sea- advection of warm and humid marine airmasses across the Level Pressure Data Set: Trends, Errors and Discontinuities, western seaboard of the Eurasian continent. However, while Mon. Wea. Rev., 108 , 855–872, 1980. the trend in the AO explains roughly 50% of the observed Vinje, T., N. Nordlund, and A. Kvambekk, Monitoring ice thick- Eurasian warming 1968-97 [Thompson et al., 2000], the BO ness in Fram Strait, J. Geophys. Res.-Oceans, 103 , 10,437– time series exhibits no clear long term trend and can conse- 10,449, 1998. quently not explain the residual Eurasian SAT trend. P. Skeie, Geophysical Institute, University of Bergen, 5007 Acknowledgments. The author is grateful to Dave Thom- BERGEN, Norway. (e-mail: paul@gfi.uib.no) pson for valuable comments and to colleagues at the Geophysical Institute for reading and commenting on early versions of this manuscript. Thanks also to two anonymous reviewers for valuable (Received February 14, 2000; revised June 21, 2000; suggestions. The NCEP/NCAR reanalysis data was downloaded accepted June 22, 2000.)