Houssais.indd 99Houssais & Herbaut 2003: Polar Research 22(1), 99–106 Variability of the ice export through Fram Strait in 1993–98: the winter 1994/95 anomaly Marie-Noëlle Houssais & Christophe Herbaut The origin of the large positive anomaly of the Fram Strait sea ice export which occurred in winter 1994/95 is analysed on the basis of a model sim- ulation of the Arctic sea ice cover over the period 1993–98. The overall intra-annual and interannual variability in the model is in good agreement with observational estimates and the 1994/95 anomaly is well reproduced with an amplitude amounting to half of the mean winter value. Model results suggest that, concomitant to anomalous export velocities, larger than usual ice thickness in the strait contributes to the outstanding ampli- tude of the anomaly. Analysis on the ice thickness evolution in the strait indicates that the thick ice advected in Fram Strait at the end of the fall of 1994 originates in the anomalous cyclonic wind stress which prevailed during the preceding summer. This anomalous wind stress resulted in per- sistent convergence of the ice fl ow against the northern coasts of Canada and Greenland and in the formation of a large thickness anomaly north of Greenland. The anomaly then feeds the Fram Strait ice fl ow during those following winter months when the local wind forcing in the strait favours ice drift from the north-west. Our results suggest that short-term wind stress variations resulting in local thickness changes to the north of Fram Strait can lead to substantial variability of the Fram Strait ice export. M.-N. Houssais & C. Herbaut, Laboratoire d’Océanographie Dynamique et de Climatologie, UMR CNRS- ORSTOM-UPMC, Université Pierre et Marie Curie, 4 place Jussieu, 75 252 Paris Cedex 05, France, marie- noelle.houssais@lodyc.jussieu.fr. Fram Strait is the major exit for sea ice out of the Arctic Ocean. A volume export on the order of 0.1 Sv approximately counterbalances the net ice production within the Arctic Ocean. Anom- alies of this ice export are likely to impact on the Arctic Ocean sea ice distribution, especial- ly on the multi-year ice area (Vinje 2001), while downstream from Fram Strait such anomalies can result in anomalous volume of melted ice, possi- bly contributing to large surface salinity anoma- lies (e.g. the Great Salinity Anomaly; Dickson et al. 1988). When occurring in the Greenland–Ice- landic–Norwegian seas or, ultimately, in the Lab- rador Sea, which are important deep convection areas, such salinity anomalies may have large consequences for the thermohaline circulation of the world ocean. Observations show large variations in the Fram Strait ice export on time scales from days to years. Area fl ux variations are primarily attrib- uted to ice velocity changes induced by varying atmospheric forcing. Using ice velocities derived from satellite passive microwave imagery over an 18-year period, Kwok & Rothrock (1999) showed that in winter (October–May) 72 % of the var- iance of the ice area fl ux can be explained by the sea level pressure gradient across the strait. According to Vinje et al. (1998), the weakening of this gradient in summer over the period 1990–96 mostly explains the 50 % reduction of the monthly area fl ux in summer, while the almost doubling of the annual area fl ux between 1990/91 and 1994/ 100 Variability of the ice export through Fram Strait 95 should also be attributed to a change of the gradient between these two periods. Reliable estimates of the ice volume fl ux through Fram Strait are sparse due to diffi culties in collecting measurements of the ice thickness. The 1990–96 time series reconstructed by Vinje et al. (1998) suggests strong intra- and interan- nual variability of the volume fl ux with a stand- ard deviation of about 30 % of the mean over that period. The standard deviation of the annual mean ice thickness is only 10 % over the same period, implying that interannual variability of the ice fl ux is largely accounted for by fl uctua- tions of the ice velocity. Some models also reveal strong correlations between the local wind forc- ing and the ice volume fl ux through Fram Strait (Hakkinen 1993; Harder et al. 1998). Still, ice thickness anomalies should contribute as well to part of the volume export variability at Fram Strait. As noticed by Vinje et al. (1998), month-to-month variations of the prevailing wind stress direction are frequently observed at Fram Strait, alternatively bringing into the strait thick- er ice from north of Greenland or thinner ice from the eastern Arctic. The shape of the annual cycle of the volume fl ux may then be altered compared to that of the cross-strait velocity. Thickness anomalies may also be formed in remote areas of the Arctic Ocean and advected across long dis- tances, therefore integrating a complex time his- tory of thermodynamic and dynamic interactions. Sev eral scenarios in which ice thickness anoma- lies are formed in the Beaufort Sea or the Siberi- an marginal seas have been proposed to explain interannual variability of the Fram Strait ice export (e.g. Tremblay & Mysak 1998; Venegas & Mysak 2000). In this study, we focus on the large positive anomaly of the ice volume export observed in winter 1994/95. Model simulations suggest that this anomaly has been one of the largest occur- ring in the strait during the last fi ve decades (Hil- mer et al. 1998; Arfeuille et al. 2000; Vinje 2001); according to observations, it has been the most extreme event over the period 1990–96, with an extra ice export of ca. 0.06 Sv, representing more than 60 % of the mean (Vinje et al. 1998). The event has been associated with a con comitant ice thickness anomaly since ice draft measurements reveal monthly mean thickness values up to 4 m at that time. Ice thickness changes may have con- tributed to the export variability on other large export events (Arfeuille et al. 2000) but the 1995 event is the only one for which draft measure- ments are available. In this study, we use results of a model simulation over the period 1993–98 to try to understand the origin of the winter 1995 anomaly, as well as the mechanisms and time scales involved. Model design and experiments The sea ice model used in this study is based on a variable ice thickness distribution follow- ing Hibler (1980). Four ice classes are consid- ered, including open water. The ice growth rate is determined from the vertical heat conduction equation which is discretized according to the zero- and one-layer approximation for the snow and ice, respectively. The ice dynamics are char- acterized by a cavitative rheology (Flato & Hibler 1992) in which shear stress is neglected. The ocean model is based on the primitive equation, z co ordinate, rigid lid ocean code developed at the Lab oratoire d’Océanographie Dynamique and Climat ologie in Paris (Delecluse et al. 1993). The therm o dynamic coupling between the ice and the ocean assumes freezing ocean surface tempera- ture in ice-covered areas and implies heat and salt exchanges at the ice–ocean interface. The dyn amic coupling is such that the surface forc- ing viewed by the ocean is the wind stress forcing mod i fi ed by the internal ice force, while the ocean exerts a tangential friction force at the bottom of the ice. The domain covers the Arctic and adjacent seas, with the southern limit lying at about 40° N. In the vertical, the grid includes 30 levels, with level spacing increasing with depth from 10 m in the top 100 m to 500 m in the deepest levels. In the horizontal, the resolution is slightly anisotrop- ic due to the grid geometry with the “pole” lying over China to overcome the North Pole singular- ity. The horizontal resolution is about 80 - 100 km in the Fram Strait area and in the central Arctic and increases eastward to reach 40 km in the zonal direction in the Kara Sea. All model bound- aries are treated as closed boundaries. On the southern boundary in the North Atlantic sector the temperature and salinity fi elds are restored to climatology. The model has been forced by daily atmos- pheric forcing fi elds from the period 1993–98 extracted from the 40-year NCEP reanalysis (Kal- nay et al. 1996). The forcing fi elds are surface 101Houssais & Herbaut 2003: Polar Research 22(1), 99–106 wind stress, air temperature, specifi c humidity, pressure and wind speed together with incom- ing longwave and shortwave radiation. Since the shortwave radiation was revealed to be unrealisti- cally large, it was modifi ed by adding a correction based on the difference between this fi eld and the corresponding fi eld in the European Centre for Medium-range Weather Forecasts (ECMWF) reanalysis (ERA15). Because only the year 1993 was available in the ECMWF data set, the cor- rection was calculated for this particular year and applied identically to the fi ve other years of our NCEP forcing. The model is initialized from rest with ocean temperature and salinity distributions from the PHC global ocean climatology (Steele et al. 2001). The sea ice–ocean coupled system is fi rst spun up for 20 years with a repeated mean annual cycle of the forcing based on the 1993–98 climatology. The model is then run in the inter- annual mode using the 6-year time series of the forcing fi elds. Fram Strait ice export variability during 1993–98 Figure 1a shows the mean annual cycle of the ice volume transport through Fram Strait as estimat- ed from 9-day averages of the model transport. The section runs parallel to a meridian of the model grid, approximately from 16° W, 78.5° N to 12.5° E, 80° N. There is a strong seasonal cycle in the transport with a minimum monthly mean of 0.03 Sv occurring in July and a maximum of 0.23 Sv in March. The winter maximum has, in fact, a two-peak structure with a secondary max- imum occurring in December as a result of the weakening of the transport in February. Over the period 1993–96 overlapping with Vinje et al. (1998), the overall structure of the model annual cycle bears a strong resemblance with the data, although the model shows less weakening of the transport in January and a more rapid decrease of the transport in June–July. The amplitude of the annual cycle is also larger in the model due to higher winter values, especially when con- sidering Kwok & Rothrock’s (1999) estimates. The model 6-year mean transport of 0.12 Sv is therefore overestimated when compared with the 0.09 Sv (1 Sv = 0.317 × 105 km3 yr-1) of Vinje et al. (1998) or with the 0.075 Sv of Kwok & Rothrock (1999), both calculated over a slightly different period (1990–96). The model transport estimate falls in the upper limit of previous model esti- mates obtained in different periods (e.g. Harder et al. 1998; Hilmer et al. 1998). The transport shows considerable variability at time scales from weeks to years (Fig. 2a). Year-to- year transport variations, obtained after remov- ing the mean annual cycle, exhibit very similar time evolution and magnitude as compared with the data. Over the overlapping 1993–96 years, signifi cant correlations of 0.72 and 0.69 are found between the low-pass fi ltered (1 month running mean) model anomalies and the data estimates by Vinje et al. (1998) and Kwok & Rothrock (1999), respectively. All time series show a pos- itive anomaly of the ice volume fl ux starting roughly in November 1994, culminating in Jan- uary 1995 and persisting through the following Fig. 1. Mean annual cycle of the Fram Strait (a) ice volume transport and (b) mean ice thickness estimated from the model run (1993–98) (solid line) and from Vinje et al.’s (1998) observations (1993 to July 1996) (dashed line). Also shown in (a) are Kwok & Rothrock’s (1999) transport estimates (dotted line). (a) (b) 102 Variability of the ice export through Fram Strait spring. The amplitude of the anomaly averages to 0.06 Sv over the duration of the anomaly, both in the model and in Vinje et al. (1998), but it is much smaller in Kwok & Rothrock (1999). Another positive anomaly occurs in late 1996, but its mag- nitude is only half of that in 1995, suggesting that the latter is indeed remarkable. Over the 1993–96 period, the mean ice stream thickness in the strait is 2.71 m. For the same period, Vinje et al.’s (1998) draft measurements give a value of 2.84 m at 5° W, which corresponds to a strait averaged mean ice thickness of 2.54 m. Despite the mean thickness of the ice stream is a bit high in our simulation, its annual variations compare well with observations (Fig. 1b). The decrease in January is not as marked in the model as in the observations, nor are the ice thickness- es measured in summer as small as the simulat- ed ones. The model year-to-year variations show some similarities with the data from January 1993 until the middle of 1995, with the most noticeable feature being the occurrence of a large thickness anomaly in winter 1994/95 (Fig. 2b). The details of the time evolution of the anomalies over that period however differ. In particular, the anomaly starts developing earlier in the data. Since the width of the winter ice stream varies little from year to year in the model, it has little impact on the year-to-year variations of the ice volume fl ux. This has been checked by noting that the area fl ux (not shown) correlates extremely well with the variations of the ice export velocity. However, the width of the ice stream appears to be over estimated in the model and most probably explains the too high simulated winter transports. A possible reason might be that the cavitative fl uid rheology, by neglecting shear stress along the Greenland coast, allows for overestimated off-shore component of the ice drift under the effect of the pressure gradient force. Despite an Fig. 2. Time series of the Fram Strait (a) ice volume transport and (b) mean ice thickness estimated from the model run (1993–98) (solid line) and from Vinje et al.’s (1998) observations (1993 to July 1996) (dashed line). Thick lines represent the raw time series of the model while light lines are anomalies with respect to the mean annual cycle. Also shown in (a) are Kwok & Rothrock’s (1999) transport estimates (dotted line). (c) Raw time series of the mean ice velocity in Fram Strait for the model (thick line) and the IABP gridded velocities (1993–97) (thin line). Model anomalies have been low-pass fi ltered with a 1-month running mean. In (b) the scale of the raw time series is to be read on the right axis. (a) (b) (c) 103Houssais & Herbaut 2003: Polar Research 22(1), 99–106 expected discrepancy due to this rheology effect, the mean cross-strait velocity compares very well with IABP gridded buoy velocities (obtained from the Polar Science Center, University of Washing- ton, http://iabp.apl.washington.edu) throughout the year (Fig. 2c). Still, the slightly different ori- entation (to the south-east) of the model section as compared with the data section (to the south) may hide part of the discrepancy. The winter 1995 event is associated with a strong anomaly of the export velocity which, as for the ice thick- ness anomaly, occurs earlier in the data than in the model. Note that the comparison covers only the 1993–97 period and that buoy velocities with the variance of the interpolated error greater than 0.5 have been excluded from the comparison. In view of the above assessment of the model variability, we consider that the outstanding large ice export which occurred in Fram Strait in winter 1995 was indeed associated with the export of abnormally thick ice. In the next section, we ana- lyse the ice thickness variability in order to deter- mine the origin of the 1995 anomaly. Origin of the winter 1995 thickness anomaly at Fram Strait Averaged over the November 1994 to February 1995 period, the positive ice thickness anoma- ly in Fram Strait amounts to 71 cm, with a peak value of 130 cm in mid-February (Fig. 2b). The time evolution of the anomaly reveals two succes- sive events of very thick ice, the fi rst occurring in November and the second in February. In Fram Strait, the patterns of the ice thickness variations are essentially governed by the advec- tion fi eld, except for the ice edge region where the thermodynamics also play a major role (not shown). The question is whether thickness anom- alies in the strait should exclusively be attributed to changes in drift direction advecting the mean ice thickness fi eld or if they should also be relat- ed to thickness anomalies formed upwind from the strait. The evolution of the 1994/95 thickness anomaly can partly be explained by changes in the dominant direction of the mean ice drift immedi- ately north of the strait (Fig. 3) which bring ice of different origins into the strait. From the mean ice thickness distribution it can be deduced that, in November, the ice drift favours advection of thick ice coming from northern Greenland, while in December–January, advection from the north- east tends to bring thinner ice from the east- ern Arctic. An inter mediate situation predomi- nates in February 1995, which is characterized by a strong northerly fl ow turning north-easter- ly to the east of the strait. Although such monthly reversals in the drift direction are not exception- al and have been reported in other studies, what makes the 1994/95 winter anomalous with regard to the ice drift are the high velocities associated with the reversal, as revealed by the multi-year velocity time series shown in Fig. 2c. To identify the possible contribution of up- stream ice thickness anomalies to the ice ex port variability, the distribution of these anomalies at the end of September 1994—that is just before the appearance of the anomaly in Fram Strait— is shown in Fig. 4a. The distribution is character- ized by a well-developed positive feature which extends from the Beaufort Sea along the Cana- dian Archipelago down to the northern coast of Greenland. The feature is attributed to the strong convergence of the ice fl ux created by a persist- ent cyclonic circulation in the preceding summer Fig. 3. Ice velocity in Fram Strait in (a) November 1994 and (b) January 1995. (a) (b) 104 Variability of the ice export through Fram Strait (Fig. 4b). To check that this anomaly was indeed advected into Fram Strait and contributed to the thickness anomaly detected in the strait in November, a model experiment was performed in which the daily wind stress in summer (July– September) 1994 is replaced over the entire model domain by the wind stress from the mean annual cycle. Comparing Fig. 5a and c, the ice build- up appears to be greatly reduced to the north of Greenland, while the anomaly disappears in the strait from November through December, indicating that the latter likely originates in the anomalous ice thickness fi eld formed in the west- ern Arctic during the previous summer. On the other hand, the reappearance of the anom aly in February (Fig. 5b), after the slow de crease in December–January, refl ects the change in the drift direction by the end of Janu- ary. The fact that the anomaly does not disap- pear entirely in the sensitivity experiment may indic ate that some thick ice is being created to the north of the strait in December and January. Indeed, the strong easterly component of the ice vel ocity during these months (Fig. 3b) prevents ice from being exported to the Greenland Sea. Thick ice may also be advected from quite differ- ent regions of the Arctic without being too much affected by changes in the 1994 summer wind stress. The smaller thickness anomaly in the sen- sitivity experiment at that time (Fig. 5d), howev- er, suggests that some effect of the wind stress change of the previous summer persists through the winter. Discussion and concluding remarks The above analysis suggests that large anomalies of the sea ice export through Fram Strait such as the 1994/95 event can be associated not only with large ice export velocities but also with the pres- ence of abnormally thick ice. Several events, in which the contribution due to advection of thick- er than usual ice dominates over that due to faster than usual export velocity, have also been identi- fi ed by Arfeuille et al. (2000) in a model simula- tion over 1948–1998. In their study, however, the 1994/95 event was not associated with a concom- itant thickness anomaly in the strait. This may be due to their model missing part of the ice thick- ness variability, perhaps for the same reasons it overestimates the mean annual ice export. In view of the good correlation between our model thick- ness time series and the observations by Vinje et al. (1998), we are somewhat confi dent that thick ice anomalies were indeed present in Fram Strait in winter 1995. The fact that the northerly wind stress correlates very well with the ice export at that time (see Fig. 3 in Arfeuille et al. 2000) is not contradictory with this idea but only indicates that the anomalous transport is also associated with a velocity anomaly when the ice fl ow gets Fig. 4. Arctic (a) ice thickness anomaly (m) at the end of Sep- tember 1994 and (b) ice velocity in August 1994. In (a) the anomaly is calculated with respect to the mean annual cycle and the contour increment is 0.25 m. Solid isolines indicate positive values; dashed isolines indicate negative values. The bold line is the isoline zero. Also shown in (a) is the model section which has been used for transport estimates in Fram Strait. The dark grey shaded area is the model domain. (a) (b) 105Houssais & Herbaut 2003: Polar Research 22(1), 99–106 aligned with the north–south strait axis. Our analysis of the generation of the thickness anomaly observed in 1994/95 suggests that some thickness anomalies in Fram Strait may have a short history, being generated in the area north of Greenland by intra-annual variations of the surface wind stress and then advected towards Fram Strait in a few months. This scenario dif- fers somewhat from those proposed in other stud- ies (Tremblay & Mysak 1998; Mysak & Ven- egas 1998; Arfeuille et al. 2000) in which Fram Strait ice thickness or concentration anomalies were found to be generated in remote areas of the Arctic (the Beaufort and Chukchi seas or the East Siberian Sea). In these scenarios, the thick- ness anomalies get advected towards Fram Strait from their source region, clockwise around the Beaufort gyre and / or by the Transpolar Drift. The long time scales involved in these journeys imply that the anomalies must survive a few sea- sonal cycles before reaching Fram Strait, which may substantially alter their amplitude. In con- trast, our simulation implies shorter advection time scales which minimizes the impact of sea- sonal thermodynamic processes and preserves most of the anomaly integrity. The build-up of the thickness anomaly is initi- ated by the anomalous surface wind in summer 1994. The cyclonic wind stress curl anomaly, which fi rst appears north of the Chukchi Sea in July, then gets stronger in August and moves to the North of Canada by the beginning of Septem- ber. Such cyclonic anomalies are common fea- tures of the Arctic summer but the 1994 summer one appears to be particular in terms of strength and duration which altogether lead to anomalous ice motion fi eld. The enhanced impact of summer (as compared with winter) anomalies of the wind stress on the Arctic ice thickness distribution has already been mentioned by Zhang & Hunke (2001) to explain winter anomalies of the ice growth rate in the Canada Basin. Although the impact in the present study is rather of a dynami- cal nature, the fact that the ice is more responsive to the atmospheric vorticity in summer certainly plays a role in our proposed scenario. Fram Strait ice volume export anomalies such as the one which occured in winter 1994/95 are potentially very important for climate variabili- ty. Melting of the exported ice may impact on the Fig. 5. Ice thickness change (m) in Fram Strait, taking the ice thickness distribution at the end of June 1994 as reference, in the reference experiment in (a) November 1994 and (b) February 1995, and in the sensitivity experiment in (c) November 1994 and (d) February 1995. The contour increment is 0.5 m. (a) (b) (c) (d) 106 Variability of the ice export through Fram Strait ocean circulation via perturbations in the fresh- water fl ux. This impact depends on the patterns of ice melting and therefore of the ice drift as the ice moves southward towards the Nordic and Labrador seas. The high southward drift speed in winter 1995 apparently favoured the export of most of the ice passing through Fram Strait towards Denmark Strait (e.g. Hilmer et al. 1998), limiting the input of freshwater to the Greenland Sea and inhibiting the Odden formation. In con- trast, the Fram Strait ice export anomaly which occurred in 1968 and was assumed to be asso- ciated with the export of thicker ice from north- ern Greenland (Walsh & Chapman 1990) led to the formation of a wide freshwater signal known as the Great Salinity Anomaly. It is possible that other exports of thick ice, not necessarily associ- ated with high drift speed or above normal pres- sure gradient in the strait, have led to freshwater anomalies in the past. Due to lack of observations, we do not know the frequency of these Fram Strait ice export anom- alies which are associated with thickness anoma- lies, and our simulation is too short to give insight into this aspect. The mechanism described above, which implies anomalous cyclonic atmospheric circulation in summer, may be linked to the par- ticular period of the study which, according to Prosh utinsky & Johnson (1997), corresponds to a cyclonic circulation regime. Venegas & Mysak (2000) recently suggested that time scales on the order of 16 - 20 years characterize Fram Strait ice thickness anomalies when they are associated with anomalous westerly winds north of Canada and Green land. Other mechanisms, such as ice growth rate fl uctuations in response to thermody- namic atmospheric forcing, may also lead to ice thickness anomalies with long characteristic time scales (L’Hévéder & Houssais 2001). This low- frequency variability is likely to impact on the variability of the Fram Strait ice export. Acknowledgements.—This work was supported by a grant from the French Programme National d’Etude de la Dynam- ique du Climat. Computational support was provided by the Institut du Développement et des Ressources en Informatique Scientifi que. References Arfeuille, G., Mysak, L. A. & Tremblay, L.-B. 2000: Simula- tion of the interannual variability of the wind-driven Arctic sea-ice cover during 1958–1998. Clim. Dyn. 16, 107–121. Delecluse, P., Madec, G., Imbard, M. & Lévy, C. 1993: OPA version 7 Ocean General Circulation Model: reference man ual. Rapport interne LODYC 93/05. LODYC, Université Pier re et Marie Curie, Paris Cedex 05, France. Dickson, R. R., Meincke, J., Malmberg, S.-A. & Lee, A. J. 1988. The “Great Salinity Anomaly” in the northern North- Atlantic 1968–1982. Prog. Oceanogr. 20, 103–151. Flato, G. M. & Hibler III., W. D. 1992: Modeling pack ice as a cavitating fl uid. J. Phys. Oceanogr. 22, 626–651. Hakkinen, S. 1993. An Arctic source for the Great Salinity Anomaly: a simulation of the Arctic ice–ocean system for 1955–1975. J. Geophys. Res. 98(C9), 16 397–16 410. Harder, M., Lemke, P. & Hilmer, M. 1998: Simulation of sea ice transport through Fram Strait: natural variabil- ity and sensitivity to forcing. J. Geophys. Res. 103(C3), 5595–5606. Hibler, W. D. III 1980: Modeling a variable thickness sea ice cover. Mon. Weather Rev. 108, 1943–1972. Hilmer, M., Harder, M. & Lemke, P. 1998: Sea ice transport: a highly variable link between Arctic and North Atlantic. Geophys. Res. Lett. 25, 3359–3362. Kalnay, E., Kanamitsu, M., Kistler, R., Collins, W., Deaven, D., Gandin, L., Iredell, M., Saha, S., White, G., Woollen, J., Zhu, Y., Chelliah, M., Ebisuzaki, W., Higgins, W., Jan- owiak, J., Mo, K. C., Ropelewski, C., Wang, J., Leetmaa, A., Reynolds, R., Jenne, R. & Joseph, D. 1996: The NCEP/ NCAR 40-year reanalysis project. Bull. Am. Meteorol. Soc. 76, 437–471. Kwok, R. & Rothrock, D. A. 1999: Variability of Fram Strait ice fl ux and North Atlantic Oscillation. J. Geophys. Res. 104(C3), 5177–5189. L’Hévéder, B. & Houssais, M.-N. 2001: Investigating the var- iability of the Arctic sea ice thickness in response to a sto- chastic thermodynamic atmospheric forcing. Clim. Dyn. 17, 107–125. Mysak, L. A. & Venegas, S. A. 1998: Decadal climate oscil- lations in the Arctic: a new feedback loop for the atmos- phere–ice–ocean interactions. Geophys. Res. Lett. 25, 3607–3610. Proshutinsky, A. Y. & Johnson, M. A. 1997: Two circulation regimes of the wind-driven Arctic Ocean. J. Geophys. Res. 102(C6), 12 493–12 514. Steele, M., Morley, R. & Ermold, W. 2001: PHC: a global ocean hydrography with a high-quality Arctic Ocean. J. Clim. 14, 2079–2087. Tremblay, L.-B. & Mysak, L. A. 1998: On the origin and evo- lution of sea-ice anomalies in the Beaufort–Chukchi Sea. Clim. Dyn. 14, 451–460. Venegas, S. A & Mysak, L. A. 2000: Is there a dominant time- scale of natural climate variability in the Arctic? J. Clim. 13, 3412–3434. Vinje, T. 2001: Fram Strait ice fl uxes and atmospheric circula- tion: 1950–2000. J. Clim. 14, 3508–3517. Vinje, T., Nordlund, N. & Kvambekk, A. 1998: Monitor- ing ice thickness in Fram Strait. J. Geophys. Res. 103(C5), 10 437-10 449. Walsh, J. E. & Chapman, W. L. 1990: Arctic contribution to upper-ocean variability in the North Atlantic. J. Clim. 3, 1462–1473. Zhang, Y. & Hunke, E. C. 2001: Recent Arctic change sim- ulated with a coupled ice–ocean model. J. Geophys. Res. 106(C3), 4369–4390.