a. Coupled GCM: ECHAM4/OPYC3

To study the link between the NAO and Arctic ice volume flux through Fram Strait we analyze a century-scale (300 yr) integration of the European Centre for Medium-Range Weather Forecasts (ECMWF)–Hamburg (ECHAM4)–Ocean Isopycnal Model (OPYC3) coupled atmosphere–ocean–sea ice model under present-day climate conditions (i.e., concentrations of greenhouse gases are fixed to the observed 1990 values; Roeckner et al. 1996, 1999). The atmospheric component, ECHAM4, is the fourth generation of a hierarchy of models that was developed at the Max Planck Institute in Hamburg, Germany, from the former ECMWF model. ECHAM4 has a horizontal resolution of T42 (approximately 2.8° by 2.8°) and 19 hybrid levels in the vertical. OPYC3 is a three-component model including an isopycnal interior ocean, a mixed layer component, and a dynamic–thermodynamic sea ice model that represents internal stresses by a viscous–plastic rheology (Oberhuber 1993). The horizontal resolution poleward of 38° is identical to that of the atmospheric model. The meridional resolution increases toward the equator (0.5°) to properly resolve the equatorial waveguide. Vertically, 11 layers were used. Zonal and meridional ice flux components are formulated as prognostic variables. An annual mean flux correction scheme has been applied to air–sea heat and freshwater fluxes over the oceans. In the presence of sea ice the flux correction is only applied to surface heat fluxes. Note that no flux correction has been applied to wind stress fields. Further details are given elsewhere (Roeckner et al. 1996, 1999; Bacher et al. 1998).

b. Kiel sea ice model (KSIM) integration

Climatological sea ice properties of the coupled GCM are compared with results from a realistic hindcast simulation of the Kiel sea ice model, an optimized dynamic–thermodynamic sea ice model with viscous–plastic rheology and 1° × 1° horizontal resolution (Harder et al. 1998; Hilmer et al. 1998; Kreyscher et al. 2000, and references therein). KSIM was forced with daily fields of near-surface winds and air temperatures taken from the NCEP–NCAR reanalysis project (Kalnay et al. 1996) over the period 1958–97. Oceanic heat fluxes into an oceanic mixed layer with fixed depth were prescribed as climatological annual cycles; oceanic near-surface currents were prescribed as annual means. Both oceanic forcing fields were derived from a coupled ocean–sea ice model integration of Hibler and Zhang (1994). Further details about the “KSIM integration” used in this study are given elsewhere (Hilmer et al. 1998; Hilmer 2001).

c. Sea level pressure fields: 1908–97

Observed interannual variability of the NAO during the twentieth century is studied using gridded SLP data (5° × 5°) over the Northern Hemisphere north of 15°N and for the period 1908–97 [updated version from Trenberth and Paolino (1980)]. Errors and discontinuities are reported for this dataset by Trenberth and Paolino (1980). These occur primarily before 1922, particularly over parts of Asia. Problems also exist north of 70°N due to sparse sampling and in low latitudes where the signal-to-noise ratio is relatively low.

d. The NAO index

Following the methodology for the observations (e.g., Rogers 1984; Hurrell 1995) the NAO index is defined as the difference between normalized SLP of the Azores high and the Icelandic low. The normalization (i.e., removing the mean and dividing by the standard deviation) has been done separately for each month. Last, winter averages (DJFM) were formed using the monthly NAO indices.

For the coupled GCM, variability of the Azores high (Icelandic low) is described by SLP averaged over the region 40°–42.5°N and 10°–15°W (65°–67.5°N and 17.5°–20°W). In this coupled GCM these regions show the strongest teleconnectivity (Christoph et al. 2000) and represent the centers of action of the first EOF of North Atlantic SLP anomalies (not shown).

The observed NAO index (see section 5) is based on monthly station data taken from the Azores and Iceland.

e. Sea ice export through Fram Strait

The ice volume flux (IVF) through Fram Strait is calculated for a zonal section in the coupled GCM connecting the northeastern tip of Greenland with Spitsbergen, Norway, according to

View Expanded

where r now denotes the position along the section Γ, υ(r) is the ice drift component perpendicular to this section, and h(r) is the ice thickness along Γ. (We use r instead of x to highlight that generally other choices than a zonal section for Γ are possible.) Furthermore, the southward drift speed (SDS) of sea ice through Fram Strait and the ice thickness in Fram Strait (hereinafter h) are determined by averaging over all grid points along Γ. Because of the usage of different grids in the coupled GCM and KSIM, the sections Γ differ slightly between the two models (Γ as used by KSIM is located slightly farther northward in comparison with the coupled model).

a. Modeled mean state

Mean sea ice thickness and sea ice drift fields in the Arctic during wintertime as simulated by the coupled GCM are depicted in Fig. 1, those from the KSIM integration are shown in Fig. 2. Although the coupled GCM appears to realistically simulate the gross features of mean Arctic sea ice properties, some differences compared to the KSIM integration are evident. The anticyclonic Beaufort gyre as simulated by the coupled GCM, for instance, is biased toward the North Pole in comparison with observations (Colony and Thorndike 1984) and KSIM. This leads to thinner ice in the Beaufort Sea region but to thicker ice in the Kara Sea in comparison with KSIM. Moreover, differences between the two model integrations are also noteworthy for sea ice thickness north of Greenland and upstream of as well as within Fram Strait (Table 1). Whereas, because of its model grid structure, the maximum sea ice thickness in the coupled GCM is located directly over the North Pole, the KSIM integration reveals largest mean values north of Greenland and in the north of the Canadian archipelago, which is more realistic in comparison with the observations (Bourke and McLaren 1992, their Fig. 4b).

Long-term mean ice thickness (h) and SDS in Fram Strait from both model integrations are summarized in Table 1 along with modeled IVF. IVF and SDS are in good agreement, whereas the coupled GCM simulates a much smaller h in Fram Strait than KSIM. This difference is partly attributable to the sharper longitudinal ice thickness gradient within Fram Strait of the GCM, which in turn is caused by differences in the mean ice drift patterns. It may surprise that the mean IVF agrees quite well although the mean sea ice thickness differs considerably. (The length of the sections Γ is very similar.) To explain this item it is necessary to break down the mean IVF into different components. Both sea ice quantities can be decomposed using x = [x] + x* and x = x + x′, where x denotes either ice thickness or SDS, [x] is the average along Γ, x* denotes deviations from [x], the overbar stands for the temporal mean, and the prime denotes temporal anomalies. Applying this breakdown to Eq. (1) and temporal averaging leads to

View Expanded

The left-hand side of Eq. (2) is given in Table 1 and amounts to 0.086 Sv (1 Sv = 10⁶ m³ s⁻¹) for the coupled model. Whereas contributions from the second and fourth term on the right-hand side are negligible for the coupled GCM, the contribution from the first (third) term amounts to 0.064 Sv (0.025 Sv). For KSIM the mean IVF is almost solely governed by the first term on the right-hand side of Eq. (2). These differences explain why the mean IVF in the coupled GCM and KSIM agrees relatively well although the mean sea ice thickness differs considerably.

It is worth noting that any direct comparison of mean IVF between the two models is difficult to carry out. This is because large north–south sea ice thickness gradients occur in the long-term mean near Fram Strait, thus, leading to a strong sensitivity of the mean IVF to small changes in the latitudinal position used to define the Fram Strait section Γ.

When comparing sea ice climatologies from the coupled GCM with observational datasets or other model integrations it has to be kept in mind that an annual-mean flux correction scheme has been applied to surface heat fluxes in the presence of sea ice in the coupled GCM integration. Thus, some agreement between simulated and observed annual-mean Arctic sea ice cover may simply be due to flux correction. Note, however, that no flux corrections were applied to wind stress fields.

b. Modeled variability

The standard deviation of winter-averaged sea ice quantities in Fram Strait is also shown in Table 1 for the two models. Obviously, variability of sea ice quantities in Fram Strait during wintertime as simulated by the coupled GCM is in good agreement with those obtained from the KSIM integration. This similarity is not very sensitive to small changes in the latitudinal position of Fram Strait [Γ in Eq. (1)] because in both models gradients of sea ice thickness and southward drift speed variability are relatively small near Fram Strait (not shown).

For the whole Arctic, the spatial patterns of the interannual ice thickness variability (Fig. 3) agree quite well between both models, despite some differences between the corresponding long-term mean fields. Both patterns exhibit largest amplitudes of the wintertime standard deviation in the East Siberian and Beaufort Seas but relatively small variability within the central Arctic. Although a comparison with observational data is not possible, this structure agrees well with results of other modeling studies (e.g. Flato 1995; Arfeuille et al. 2000).

To summarize, differences between the two models show up primarily for the long-term mean state. The agreement between the two models is much better in terms of climate variability. Therefore we feel confident that the coupled GCM integration is useful to shed further light onto the link between the NAO and the sea ice export out of the Arctic through Fram Strait.

a. Long-term characteristics

A scatterplot of the IVF and the NAO index as simulated by the coupled GCM for 300 subsequent winters is depicted in Fig. 4. The linear correlation coefficient r = 0.10; that is, in the coupled GCM the amount of variance of IVF that can be linearly explained by the NAO is negligibly small. Moreover, neither SDS of sea ice nor sea ice thickness in Fram Strait are significantly correlated with the NAO index (r = −0.03 and r = −0.06, respectively). Last, visual inspection of Fig. 4 does not indicate any noteworthy nonlinear relationship between the two time series.

To take into account the possibility that the link between the ice volume export through Fram Strait and the NAO index is timescale dependent and/or not directly in phase, the link between the two time series was studied in the spectral domain by applying cross-spectral analysis (Fig. 5). Except for a few narrow peaks the estimated squared coherency is relatively low and the estimated phase appears to vary more or less randomly over the frequency range under study (about 2–100 yr). On secular timescales (about 100 yr), for instance, the NAO leads IVF, whereas IVF leads the NAO in a narrow frequency band around 17 yr. While we cannot not definitely exclude that these locally significant squared coherencies are due to some physical processes, we note that significant peaks for squared coherency in specific frequency ranges can also occur just by chance for two statistically independent time series. Moreover, none of the two time series reveals pronounced power in the frequency bands of significant squared coherency (not shown). Similar arguments hold for the narrow peaks on interannual timescales (Fig. 5). Thus, it is clear that the “coupling” between the two time series for narrow frequency bands—even if not purely accidental—does not contribute noteworthily to the total variance of the two time series. To summarize, there is no evidence for a noteworthy wintertime link between the NAO and Arctic sea ice export through Fram Strait in the coupled GCM. This holds for different timescales and phases.

In order to clarify why the NAO exerts no significant control on the Arctic sea ice export through Fram Strait in the coupled GCM, the spatial pattern of the NAO was estimated by regressing SLP anomalies onto the normalized NAO index (Fig. 6). In the North Atlantic region and over the Arctic the spatial pattern of the NAO as simulated by the coupled GCM resembles the observed one during the period 1958–77 (HJ00, their Fig. 4a), that is, during a period when a link between the observed NAO index and the modeled ice volume export through Fram Strait was missing. As argued by HJ00, the vanishing correlation between the two time series can be explained by the fact that the NAO pattern shows no anomalous meridional geostrophic wind components near Fram Strait that might have affected the SDS of sea ice in Fram Strait and thus the sea ice volume export through Fram Strait.

That this argument also holds for the coupled GCM becomes evident from Fig. 7, which shows the Arctic sea ice response during wintertime to a forcing by the NAO (i.e., sea ice thickness and sea ice drift anomalies that were linearly associated with the NAO index). High-NAO winters, for instance, are accompanied by anomalously low ice thicknesses around Spitsbergen, Norway, and Franz-Josef-Land, Russia, and above-normal ice thicknesses around the northeastern coast of Greenland, north of Greenland, and the Canadian archipelago, as well as in the Labrador Sea/Baffin Bay region. These anomalies, whose magnitudes amount to about 0.1 m near the centers of action, can be explained by the combined effect of NAO-related near-surface temperature (thermodynamic forcing) and near-surface wind speed (dynamic forcing) anomalies. Note that such a NAO-related dipole is also well known for observed sea ice concentration anomalies (e.g., Fang and Wallace 1994; Deser et al. 2000). The percentage of sea ice thickness variance that can locally be explained by the NAO in the Arctic, however, does not exceed 10% in the coupled GCM. In Fram Strait NAO-related positive and negative ice thickness anomalies cancel each other. This explains the relatively weak correlation between the NAO index and total sea ice thickness in Fram Strait. Sea ice drift anomalies that are associated with the NAO index (Fig. 7b) agree with what can be expected from a direct local wind stress forcing by the NAO (Fig. 6). Others than for long-term means (Fig. 1) anomalous sea ice drift vectors are almost parallel to the anomalous NAO-related isobars. Although NAO-related meridional sea ice drift anomalies are present in Fram Strait, the net sea ice drift in Fram Strait vanishes because of opposite anomalies at the western and eastern side of Fram Strait. Looking at Fig. 7 it becomes evident that the NAO forces thicker sea ice out of, and thinner sea ice into, the Arctic. This holds for high- and low-NAO winters and therefore has no influence on the linear relationship between the NAO and Arctic sea ice export. However, since the percentage of sea ice thickness and sea ice drift variance in Fram Strait that can linearly be explained by the NAO is very small, this effect is negligible. [Note that this is consistent with the vanishing effect from the fourth term on the right-hand side of Eq. (2).] In summary, in this coupled GCM the influence of the NAO on local sea ice properties in Fram Strait is such that the NAO cannot exert a significant control on the ice volume export through Fram Strait.

The anomalous SLP pattern that is associated with high ice volume export events through Fram Strait in the coupled GCM is depicted in Fig. 8 (“Fram pattern”). This pattern is very similar to that obtained if the IVF time series from KSIM is used instead (HJ00, their Fig. 4c). The center of action of the Fram pattern is located over the Kara Sea giving rise to anomalous meridional (geostrophic) wind components near Fram Strait. Note that this SLP pattern shows no pronounced SLP anomalies near Iceland and the Azores, thus indirectly supporting the notion that in this coupled GCM it is not the NAO that is forcing sea ice export anomalies through Fram Strait.

b. Time dependence

Once the long-term characteristics of the link between the NAO and Arctic ice export as simulated by the coupled GCM are described, its time dependence is then investigated. The cross-correlation function (19-yr running window) of the NAO index and the ice export time series is shown in Fig. 9 along with the low-pass-filtered (19-yr running mean) NAO index. Out of the 282 overlapping chunks (19-yr length) the highest cross correlation between the NAO index and the sea ice export does not exceed r = 0.58. Note, that a correlation of r = 0.70 was found during the period 1978–97 from the realistic hindcast simulation using KSIM (HJ00) and that the percentage of variance explained by a linear relationship is quadratic in the correlation coefficient. The long-term average of the cross-correlation function is about zero and its standard deviation is σ = 0.25. Thus, provided that the coupled GCM performs realistically in simulating natural variability of the NAO and Arctic sea ice, the relatively high correlation between the NAO and the ice volume flux through Fram Strait as observed during the last two decades appears to be rather unusual.

A Monte Carlo test was performed in order to investigate how the cross-correlation coefficient is distributed under the null hypothesis that the NAO index and the ice volume export through Fram Strait as simulated by the coupled GCM are realizations of independent first-order autoregressive [AR(1)] processes. AR(1) parameters were estimated from the time series. Then, 10 000 realizations of the null hypothesis, each having a length of 19 yr, were generated; 2.5th and 97.5th percentiles of the estimated cross-correlation coefficients are given by −0.45 and 0.47, respectively. Seven excursions of the cross-correlation function above the 97.5th percentile occur in Fig. 9. Note, however, that on average 0.025 × 282 ≈ 7 excursions are expected to occur just by chance—even for independent estimates (of course, the cross-correlation coefficients in Fig. 9 are not independent). The standard deviation of the ensemble of cross correlations amounts to σ = 0.24, which is in close agreement with the estimate obtained from the data (σ = 0.25, see above).

As noted by HJ00, the observed increase in the covariability between the NAO index and the Arctic sea ice export through Fram Strait during the last four decades came in parallel with the secular increase of the NAO. The correlation between low-frequency variability of the NAO and the cross-correlation function between the two time series (Fig. 9) is rather weak (r = 0.31) in the coupled GCM. An additional Monte Carlo test was performed to assess the unusualness of a correlation coefficient of r = 0.31 taking into account the strong serial correlation (lag-1 correlation > 0.9) of the time series in Fig. 9. Results from the Monte Carlo test show that 2.5th and 97.5th percentiles are given by r_2.5 = −0.77 and r_97.5 = 0.77, respectively. Hence, there is no statistical evidence that interdecadal changes of the NAO do exert a significant control on the coherence between the NAO and the ice export on interannual timescales—at least in the control integration of the ECHAM4–OPYC3 model.

In summary, the time dependence of the link between the NAO index and the sea ice export through Fram Strait as simulated by the coupled GCM (Fig. 9) is consistent with what is expected from realizations of two independent processes. This conclusion is supported by the results of Ulbrich and Christoph (1999) showing that the spatial pattern of interannual NAO variability remained unchanged during the course of the control integration of the ECHAM4–OPYC3 model (resembling those shown in Fig. 6).

a. Link between NAO and Arctic sea ice export

While the NAO doubtless exerts a significant control on a variety of different climate parameters in the North Atlantic and Arctic region (e.g., Hurrell and van Loon 1997; Dickson et al. 2000) including sea ice concentrations in the Labrador Sea and the Greenland/Icelandic/Norwegian Seas (e.g., Fang and Wallace 1994; Deser et al. 2000), our study suggests that a significant forcing of Arctic sea ice export through Fram Strait by the NAO during wintertime, as is evident for the last two decades (Kwok and Rothrock 1999; Dickson et al. 2000; HJ00), is rather unusual in a long-term context. One might criticize that our conclusions about the link between the two time series suffer from the fact that no long observational time series of the Arctic sea ice export was used. (Continuous observational estimates for the sea ice drift and sea ice thickness in Fram Strait are only available since 1978 and 1990, respectively.) While this is clearly somewhat unsatisfactory, we feel confident that our major conclusions, which are based on modeling results, are realistic. This is because the models appear to simulate key aspects realistically and—most important—the modeled strength of the link between the two time series agrees with what we do expect from our physical reasoning, that is, sea ice export anomalies through Fram Strait are largely driven by anomalous atmospheric flow fields.

While finalizing this manuscript an extended version of the NECP–NCAR reanalysis became available (1948–99). We used these new forcing data to extend the KSIM integration over the period 1949–99. The year 1948 was excluded from the analysis since it may be inflated by the model's spinup. [Details about this integration are given in Hilmer (2001).] The running cross-correlation function between winter averages of the NAO index and the modeled sea ice export through Fram Strait using a window length of 19 yr is depicted in Fig. 11. The change in the link between the two time series that took place around the late 1970s (HJ00) is clearly evident. Moreover, there is no indication for the existence of a significant link between the two time series during the 1950s—thus, further supporting the conclusions of this study.

While the focus of this study (and previous studies) was on the winter season (DJFM), from an oceanographic point of view sea ice exports through Fram Strait during the other seasons may be of importance too. We have also looked onto the link between the NAO and the sea ice export through Fram Strait during the nonwinter seasons. In the coupled GCM, there is no significant link (r ≈ 0.0) between the two time series during nonwinter seasons (MAMJ, JJAS, SOND). The KSIM integrations reveals negative but small correlation coefficients between the two times series during the other seasons. Thus, a rather weak link between the NAO and Arctic sea ice export through Fram Strait appears to be characteristic for nonwinter seasons.

b. Changes of interannual NAO variability

As found out by HJ00, recent changes in the link between the NAO and Arctic sea ice export through Fram Strait can be explained by changes in the spatial structure of interannual NAO variability. The eastward shift of the NAO's centers of interannual NAO variability, which appears to be unprecedended during the twentieth century (Fig. 10), led also to pronounced changes in NAO-related interannual variability of the number of deep cyclones, near-surface air temperatures, and net surface heat fluxes in the North Atlantic region (Jung 2000). Thus, it is important to unravel possible causes for the recent shift of interannual NAO variability.

A possible explanation is provided by the study of Ulbrich and Christoph (1999). (A detailed physical mechanism is still missing, however.) They analyzed a control and a scenario integration of the same coupled GCM as is used in the present study (ECHAM4–OPYC3) and found a systematic northeastward shift of the NAO's northern variability center under enhanced greenhouse gas concentrations (after 2020 or so). In the control integration the centers of action were rather fixed. Therefore, one might speculate that the observed recent shift of interannual NAO variability is caused by enhanced greenhouse gas concentrations. To further test this hypothesis, it is important, however, to investigate whether similar changes of interannual NAO variability are evident (i) from control and scenario integrations of other coupled GCMs and (ii) from different ensemble integrations of the same coupled GCM. To our knowledge, relatively few studies have dealt so far with possible changes of interannual NAO variability under enhanced greenhouse gas concentrations (e.g., Ulbrich and Christoph 1999; Paeth et al. 1999; Monahan et al. 2000).

Recently, Palmer (1993, 1999) proposed a nonlinear dynamical perspective on climate variability and change. From this perspective an external forcing may primarily appear as a change in the frequency of occurrence of fixed natural modes of atmospheric variability (e.g., NAO), rather than as a change in the location of natural modes. Evidence for the validity of this nonlinear paradigm is presented from observations by Corti et al. (1999) and from a coupled GCM integration by Monahan et al. (2000). On the other hand, the results from Ulbrich and Christoph (1999), HJ00, and this study suggest that the location of the centers of NAO variability can undergo secular changes (although presumably not very frequently under present-day climate conditions). We cannot exclude, however, that changes in the occupation statistics of more than one fixed natural mode of variability led to changes in the location of the NAO. From the above discussion it becomes clear that still some work remains to improve our understanding about (possible low-frequency changes of) the atmospheric attractor, both under present-day climate conditions as well as under enhanced anthropogenic emission scenarios (e.g., enhanced greenhouse gas and sulfate concentrations).