1. Introduction

Boreal winter [Dec–Jan–Feb, (DJF)] atmospheric circulation exhibits strong low-frequency variability on interannual, decadal, and interdecadal timescales. The dominant mode of winter climate variability in the North Atlantic region is recognized as the North Atlantic oscillation (NAO); a large-scale seesaw in atmospheric mass between the subtropical high and the polar low. The pattern is also identified by out of phase temperature fluctuations between Greenland and Scandanavia. The anomaly pattern associated with the NAO was first reported by Walker and Bliss (1932). The NAO is not only the dominant mode of variability regionally in the North Atlantic, but also the dominant mode of atmospheric variability found for the entire Northern Hemisphere (NH) in the mid- to high latitudes. The associated hemispheric-scale pattern of variability is referred to as the Arctic oscillation (AO; Thompson and Wallace 1998). The hemispheric-scale mode has been shown to have a barotropic structure, from the surface, through the troposphere, and into the lower stratosphere, and is associated with hemispheric-scale surface temperature trends with warming over the landmasses and cooling over the oceans (Hurrell 1995, 1996; Thompson et al. 2000). In recent years its relation with low-frequency climate variability in the Arctic region, including sea ice concentration (Mysak and Venegas 1998), the strength of the Icelandic low (Serreze et al. 1997) and Aleutian low (Overland et al. 1999), and variations in the upper-layer circulation of the Arctic Ocean (Polyakov et al. 1999) have been demonstrated. An important question that has not been convincingly answered remains: What are the forcing mechanisms for the interannual to decadal variability in the NAO (AO) pattern?

Due to the long timescales of the leading mode of winter climate variability, the oceans have been favored as the forcing mechanism. Observational studies have shown an NAO-like dipole pattern of variability in sea surface temperature (SST), surface air temperature, and atmospheric circulation on the decadal timescale (e.g., Deser and Blackmon 1993; Kushnir 1994). The potential role of SSTs in the atmospheric circulation variability has also been investigated using coupled general circulation models (GCMs). Some numerical experiments are successful in reproducing the recorded atmospheric variability from the observed SST time series (e.g., Latif et al. 2000; Mehta et al. 2000; Osborn et al. 1999; Rodwell et al. 1999; Watanabe et al. 1999; Bretherton and Battisti 2000; Robertson et al. 2000). However, despite efforts from both observational and computational studies, no clear lead–lag and cause–effect relation has been established, which shows that the ocean is forcing the atmosphere in the North Atlantic basin. Such a caveat has been proposed by Bretherton and Battisti (2000). They argue that the reproduced atmospheric variability in numerical models is not physical but merely a stochastic outcome, and that predictability is limited to only a few seasons.

On shorter intraseasonal timescales, the NAO (AO) pattern has been related to waves originating from the troposphere (Perlwitz et al. 2000) or the stratosphere (Baldwin and Dunkerton 1999). The polar night jet shows the largest extratropical circulation variability in the wintertime NH stratosphere. In addition to the theoretical suggestions of Charney and Drazin (1961), and Matsuno (1970), other studies have been conducted on the coupling between the troposphere and the stratosphere, and its association with the leading mode of variability during boreal winter (Kodera et al. 1996; Kodera and Koide 1997). The structure and strength of the stratospheric polar-night jet influence the reflection and transmission properties of vertically propagating planetary waves forced in the troposphere (Boville 1984). The feedback can be bidirectional between the troposphere and the stratosphere (Hartley et al. 1998). In the case of a weak polar vortex, the upward propagation of tropospheric waves is favored and precedes the poleward and downward shift of negative zonal-mean zonal wind anomalies from the midlatitude stratosphere, and the formation of an annular pattern in the troposphere (Kuroda and Kodera 1999). Kodera and Kuroda (2000) discuss two different types of coupling between the troposphere and the stratosphere: one favored in late autumn/early winter and has tropospheric origin, while the other preferentially occurs in mid-/late winter and originates in the stratosphere. Baldwin and Dunkerton (1999), using lag correlation between the strength of the AO mode at various levels both in the troposphere and the stratosphere, show that the strength of the mode propagates downward from the lower stratosphere to the midtroposphere on a timescale of a few weeks in winter. Understanding of the dynamical mechanism involved in maintaining the AO is still on going. One possibility, as discussed by DeWeaver and Nigam (2000), is through the coupling between the zonal-mean flow and waves.

Forcing of the interannual pattern of variability associated with the NAO (AO) mode has also been attributed to other factors such as natural and anthropogenic changes in atmospheric chemistry (Kodera and Yamazaki 1994; Shindell et al. 1999; etc.). To date however, no firm conclusions have been reached.

Recently it has been shown that September, October, and November (SON) Eurasian snow cover and the leading mode of NH climate variability during the following DJF are significantly correlated (Cohen and Entekhabi 1999, hereafter CE99; Watanabe and Nitta 1998, 1999; Bamzai 1999). This significant linear lag correlation between Eurasian snow cover and NH climate in the cold season suggests potential predictive value, although the sources of variations in the Eurasian snow cover are not yet clear.

Important properties of snow and ice cover included high reflectivity and emissivity, strong thermal insulation, and, when melting, the consumption of latent heat (Kuhn 1989; Cohen 1994).

The relation between anomalous snow cover and the climate system, both at the surface and in the troposphere, has been demonstrated in observational studies. Walsh et al. (1985) and Leathers and Robinson (1993) presented the local and remote effect of anomalous snow cover in North America. Gutzler and Rosen (1992) surveyed subcontinental-scale correlation between snow and the climate, while Serreze et al. (1998) described a simultaneous relationship found in the United States. Numerical experiments using GCMs have also been performed to investigate the impact of anomalous snow cover on the climate. Yamazaki (1989) and Yasunari et al. (1991) focused on spring snow cover anomalies while Walland and Simmonds (1997) demonstrated the impact of midwinter snow.

The relation of Eurasian winter/early spring snow cover to the monsoon activity in the following warm season has been recognized for most of the past century and reported in both observational studies and numerical simulation studies. Hahn and Shukla (1976) showed a negative correlation between the two using 9 yr of data. Bamzai and Shukla (1999) and Kripalani and Kulkarni (1999), using different gridpoint snow depth data, also found the inverse correlation between wintertime western Eurasia snow depth and the summer Indian monsoon rainfall. Similarly in GCM studies with different resolutions and physics incorporated, Barnett et al. (1989) and Douville and Royer (1996) successfully reproduced decreased rainfall following winters with increased snow depth (water equivalent). More recently, Gutzler (2000) showed the same inverse relationship between summer monsoon rainfall and spring snowpack for the southwestern United States.

As mentioned earlier, besides the long history linking winter/spring snow cover with monsoon circulation, more recently, observational studies have linked autumn snow cover with winter atmospheric circulation (CE99; Watanabe and Nitta 1999). CE99 showed that anomalous Eurasian snow cover in autumn is significantly correlated with the mid–low-tropospheric atmospheric circulation variations in the following winter. They hypothesize that autumn snow cover anomalies influence the winter atmospheric circulation anomaly pattern through the strenghtening and the expansion of the Siberian high into the Arctic and North Atlantic sectors. Further, Cohen et al. (2001, hereafter CSE01) demonstrated that the anomalous sea level pressure and temperature patterns associated with the winter AO originate in the fall in Siberia, in the region of highest snow cover variability.

Previous studies linking fall snow cover with winter atmospheric circulation anomalies (CE99, CSE01) have only investigated the two-dimensional (latitude–longitude) atmospheric response to snow cover forcings. Also, posed physical mechanisms, linking snow cover and atmospheric circulation anomlies, have been limited to thermodynamic arguments. In this study, the nature of the relations between autumn Eurasian snow cover anomalies and the leading mode of interannual variability of the winter climate is investigated, not only horizontally but also vertically from the surface to the lower stratosphere, based on circulation statistical diagnostics and analysis of the observational data. Also a possible dynamic mechanism linking the two is presented.

In order to improve our understanding of the role played by snow on the large-scale climate, it is essential to have as coherent and as accurate an observational dataset of snow as possible, with large coverage in both time and space. At the time of this study the best available dataset of Eurasian autumn snow cover is the remotely sensed dataset by the National Oceanic and Atmospheric Administration (NOAA) (Robinson et al. 1993). This dataset is used in the present study and will be explained in section 2. Also in section 2, other datasets used in this study and their quality control are discussed. The main modes of climate variability are defined in terms of circulation anomalies in section 3. The results of correlation analysis between these modes of interannual climate variability and Eurasian snow cover variations are shown in section 4 and the changes in correlation as a function of season are presented in section 5. A discussion on the hypothetical connection between those variables follows in section 6, while section 7 summarizes the overall results of this study.

2. Data

3. Construction of interannual climate variability modes

Empirical orthogonal function (EOF) analysis was performed, using the singular value decomposition algorithm, on DJF-mean anomaly fields of MSLP and 500–50-hPa geopotential heights from monthly reanalysis data. Resultant principal components were used as reference indices of interannual variability of the NH climate. The time period, 1972–99, is chosen to coincide with the period of snow cover data observation. The analysis presented was performed on the original raw data. However we did repeat the analysis with the data after interannual trends were removed (not shown), and the results were unaffected.

The AO, as defined by Thompson and Wallace (1998), is the leading mode of variability of MSLP reproduced by EOF analysis (in our analysis it explains 35.6% of the total variance). The respective leading modes of 500- and 50-hPa geopotential height anomaly fields and their respective DJF indices are similarly calculated.

EOF analysis is also conducted on combined fields in order to isolate the leading mode of covarying climate variabilities at different atmospheric levels. The MSLP, and 500- and 50-hPa geopotential height fields are combined and designated as MSLP, z500, and z50, respectively. Furthermore the 500- and 50-hPa height fields are combined and designated as z500 and z50. The first EOF mode of these two combined fields, which explain 30.8% and 30.5% of the total variance, respectively, are similar to the barotropic pattern associated with the AO mode from the surface to the middle stratosphere, as shown in Kodera et al. (1996) and Baldwin and Dunkerton (1999).

As stated in the introduction, a strong association in winter between the AO index and the downward propagation of circulation anomalies from the middle stratosphere to the troposphere on a timescale of a few weeks has been demonstrated. In order to capture this downward propagation between 500 and 50 hPa, another reference climate mode is constructed. First, series of 30-day averages of 500- and 50-hPa geopotential height anomaly fields, poleward of 20°N, are calculated for the cold season from 1972 to 1998, whereby each 30-day period is advanced by 15 days from the prior period. Let p-dimensional standardized vectors, z_500,j and z_50,j, denote these arrays, for 500 and 50 hPa respectively, at the jth time period (j = 1, 2, … , N), where p is the total number of grid points (p = p_x × p_y; p_x is the numbers of grids in the x direction and p_y is the numbers of grids in the y direction), and N the total number of time periods (N = q × r and q is the number of time periods in a year, and r is the number of years). A 2p-dimensional vector,

View Expanded

defines a (simulataneous) combined array of the 500- and 50-hPa geopotential height anomaly field.

Second, the norm of projection of this combined array, [z^T_500,j z^T_50,j]^T, onto a subspace defined by the normalized leading EOF mode of the combined field, [e^T₅₀₀ e^T₅₀]^T, is computed as

View Expanded

where e^T₅₀₀ denotes the transpose of e₅₀₀. The above norm is decomposed into two parts according to the mutually orthocomplementary space, e^T₅₀₀z_500,j and e^T₅₀z_50,j, for 500 and 50 hPa, respectively. These constitute two different time series of length N, presenting the temporal changes in the loading of the leading mode pattern of the combined geopotential height anomaly field at the respective vertical level. For the next step, the lag cross correlation of these two time series is calculated for different values of lag. Figure 1 shows the lag cross correlation as a function of lag between the two series. The dashed (dotted) line indicates 95% (99%) confidence level corresponding to the calculated equivalent sample size (48) according to Zwiers and von Storch (1995). Lag correlation of +15 day (e^T₅₀z_50,j leading by 15 days) seems an appropriate estimate of the maximum lead correlation, considering that the significant points with more than 99% confidence level are 0, +15, and +30 day, and that the cross correlation is highly symmetric with respect to the +15 day peak. The timescale of the downward propagation from 50 to 500 hPa appears to be approximately 2 weeks. Finally, EOF analysis is performed on the lagged-combined field of z50 and z500 where z50 leads by 15 days, denoted by [z500, z50(+15d)]. The leading mode explains 20.3% of the total variance and is well separated by North's criterion (North et al. 1982). The spatial patterns at both levels are shown in Fig. 2 with DJF time series of the primary coefficients. The most remarkable difference of this spatial pattern from the simultaneous, barotropic pattern is found at 500 hPa where the zonally symmetric feature of the geopotential height pattern is strongly visible. This mode is referred to as the upper-AO mode, or the uAO. Also included in Fig. 2 is the northern edge of the 1000-m-elevation contour, shown by thick lines for the NH landmasses (except for Greenland).

The time series of the downward propagating signal of the leading variability mode, or the uAO index, is highly correlated with the AO index and 50-hPa geopotential heights (correlation coefficient of 0.838 and 0.877, respectively). This is consistent with the view that a large percentage of interannual variability of wintertime sea level pressure is related to the variability originating at upper levels in the troposphere and in the stratosphere.

4. Correlation analysis

5. Relation of winter climate with Eurasian snow cover

Results presented in the previous section showed that a statistically significant relation exists between autumnal Eurasian snow cover anomalies and the leading mode of wintertime circulation variability at several atmospheric levels including sea level. In this section intraseasonal changes or evolution of the local correlation between ESC_SON and MSLP are investigated and compared with those between ESC_SON and the midtropospheric 500-hPa geopotential height.

Thirty-day-mean MSLP maps regressed onto 27-yr ESC_SON in the cold season are shown in Fig. 4 for three selected periods: November, mid-December to mid-January, and February (i.e., early, middle, and late in the winter season). The local regression coefficient (contoured) is derived as a result of pointwise linear regression, and is equivalent to the departure from the mean of the local MSLP corresponding to one unit standard deviation of snow cover. The area whose absolute value of local correlation coefficient is more than 0.381 (0.487), corresponding to the statistically critical values of more than 95% (99%) probability, is shown in light (dark) shading. These maps delineate the location and intraseasonal transitions of the MSLP interannual variability patterns that are linearly related to fluctuations in autumn Eurasian snow cover. Similar maps of MSLP and surface temperature, regressed/correlated onto the uAO index, were shown in CSE01.

During December a large area of significant negative correlation emerges over the polar cap, which extends from Siberia to northwestern Canada (Fig. 4c). This negative area originates in the Kara Sea in late November. Another center that originates from central Asia in late November fills the western Siberian sector in December. These negative areas cover almost the entire polar cap, while being confined by the northern edge of the 1000-m topography isolines.

In the meantime, positive centers with more than 95% significance appear over both coasts of the North Atlantic in late December, coinciding with the expansion of the Arctic negative area into the Norwegian Basin and the Beaufort Sea. Another area of slightly positive regression coefficients appears in the North Pacific. This spatial pattern of Fig. 4c is reminiscent of the AO pattern. This is not surprising considering the moderate correlation between the two time series (r = 0.521; Table 1).

In late winter the significant, negative area has contracted and is mostly confined to the polar cap, flanked by two positive areas over the North Atlantic and the North Pacific (Fig. 4e). As the season progresses, the correlation pattern diminishes both in area and in strength as the correlation pattern decays.

The above-mentioned transition of the centers of action in regression maps of MSLP from late autumn to winter (Figs. 4a, 4c, and 4e) appears to support the hypothesis raised by CE99 that fall snow cover results in a strengthened and expansive Siberian high during the following winter. Therefore, intrusions of positive anomalous MSLP anomalies can be traced from Sibiera into North America and Europe during years of extensive Eurasian snow cover. We investigated the daily change of sea level pressure from the twice-daily IABP dataset over the Arctic region to examine whether this mechanism is at work using Hovmöller diagrams (not shown). In the years 1976 and 1988 the SLP anomalies evolved consistent with the variability in snow cover, as proposed by CE99. In 1976, the year with the largest autumn snow cover over the Eurasian sector, sporadic migration of above average surface pressure was observed from Eurasia over the Arctic Ocean and into North America in early winter. And in 1988, the year with the least Eurasian snow coverage in autumn, below normal MSLP was observed in the Arctic. However, the same analysis was inconclusive for other years, possibly due to the noisy nature of weather events on smaller scales in time and space.

Despite the uncertainty concerning the dynamics, this lag correlation of the winter MSLP with the autumn Eurasian snow cover is valuable in terms of predictability of winter climate in Europe and North America. The negative area over the Arctic and the positive areas appearing over Europe and North America in winter (Figs. 4c and 4e) are a statistical suggestion that an autumn of extensive (deficient) snow cover over the Eurasian continent will be followed by positive (negative) sea level pressure anomalies over the Arctic and negative (positive) sea level pressure anomalies over Europe and North America, with a corresponding southward (northward) shift in the storm tracks (Hurrell 1995).

As for the 500-hPa geopotential height field, the regression pattern in late autumn and winter appears to be a superposition of several teleconnection patterns (Fig. 4b): the Pacific–North American, NAO, and European modes (Wallace and Gutzler 1981). Deser (2000) and Ting et al. (2000) concluded that the annular appearance of the AO mode is primarily a combination of two different modes of regional variability in the North Atlantic and the North Pacific sectors. Therefore, the atmospheric circulation variability in the troposphere, associated with the autumn snow cover, may also be a composite of different regional variabilities.

The intraseasonal transitions of the 500-hPa geopotential height variability, associated with SON Eurasian snow cover, follow a similar pattern to that of MSLP. Two centers of significant negative anomalies, one over the Kara Sea and another over northwest Canada, appear in November and eventually extend and merge over the Arctic Ocean (Figs. 4b and 4d). In midlatitudes between 40° and 60°N, four positive anomalies of a zonal wavenumber 4 appear near each coast of Eurasia, eastern North America, and in the North Pacific (Fig. 4d). The zonally symmetric feature persists throughout the winter season, while the midlatitude zonally asymmetric component of zonal wavenumber 4 is replaced by that of zonal wavenumbers 2 and 3. In late winter two other areas of negative correlations are observed over the subtropical central Pacific and the Arabian Peninsula. As will be stated in section 6, the intraseasonal change of this regression pattern of 500-hPa geopotential height field has good correspondence with changes in the vertical energy propagation of stationary Rossby waves during NH winter.

In the regression maps (Fig. 4) discussed above, statistical significance is tested locally at each grid point. Due to the spatial correlation of underlying data between adjacent grid points and the multiplicity problem of conducting independent tests, a question remains whether they are collectively significant. In order to investigate the global significance, a field significance test was performed and the results are summarized in Table 2.

Table 2 shows the percentage area in the extratropical NH (northward of 40°N), in which the correlation between the 30-day-averaged field (i.e., MSLP or 500-hPa geopotential height) in the corresponding period and ESC_SON index is significant at the 95% confidence level. This value is compared with the null distribution of the percentage area, which is produced by the Monte Carlo method, following the method of Wilks (1995). The test time series is produced to have the same first-order autoregression structure with the ESC_SON index.

At sea level, starting from December the correlation between MSLP and the snow cover starts to show field significance at the 95% critical level. On the other hand, snow cover exhibits field significance at the 95% critical level with the extratropical NH 500-hPa geopotential height field one period earlier (i.e., mid-November to mid-December) than it does with the MSLP field. This further demonstrates the apparent contradiction (even though the correlation coefficient between snow cover and heights decreases with altitude, the highest correlation coefficient is between snow cover and the uAO; see section 4a), which leads us to speculate that the effect of anomalous Eurasian snow cover first appears in the midtroposphere or higher and then propagates downward.

The 50-hPa geopotential height field, regressed onto the ESC_SON, shows a much simpler spatial pattern. The pattern is primarily zonally symmetric and the zonally asymmetric component is apparent only in zonal wavenumber 1 in mid- to high latitudes. A negative area resides over the high latitudes in early to middle winter and diminishes in area rapidly during late winter (not shown).

6. Mechanisms for SON Eurasian snow influence on DJF NH climate variability

In the previous two sections we showed that the interannual variability of ESC_SON starts to show a globally statistically significant relation with the extratropical Northern Hemisphere sea level circulation around December, which coincides with the start of the “winter regime” at sea level (see section 4b). It was also shown that the start of global significance of snow cover variability with sea level anomaly patterns follows the global significance of snow cover with the variability in 500-hPa geopotential heights by about half a month. Also ESC_SON correlates as highly with the uAO or the downward propagating, zonally symmetric signal from the lower stratosphere to the midtroposphere, as with the AO. The strength of this downward propagation (uAO) increases steadily in autumn to reach a climax in midwinter (Jan).

In general the significant lagged relationship between autumn Eurasian snow cover variability and wintertime NH atmospheric circulation variability raises three possibilities. First, that the significance shown is spurious and unstable, due to the short length of data taken from only 30 yr, and the two have no connection of physical/dynamical significance (or in a less extreme sense the relationship lacks temporal stationarity). Second, that the anomalous autumn Eurasian snow cover is, in fact, one of the major forcing terms that contribute to the total variability of winter Northern Hemisphere atmospheric circulation. The third possibility is that there are other common contributors that take place during or before midautumn that affect both the autumn Eurasian snow cover and the wintertime NH atmospheric circulation. For example it is possible that SST anomalies are forcing both the SON snow cover and DJF atmospheric anomalies. Another potential forcing of both SON snow cover and DJF atmospheric anomalies is the tropospheric circulation during SON.

The above question would be partially clarified with the help of longer coverage of snow cover data. Even then, no cause–effect relationship is established by statistical means alone. A numerical climate model of adequate temporal and spatial resolution and refined physical/dynamical components can be used to simulate the relative role of autumn snow cover on the interannual variability of wintertime circulation, deepening our understanding of the dynamical mechanism.

Another possibility is to analyze through more comprehensive diagnostics derived from the basic motion and thermodynamic fields available through the reanalysis. In this section, we provide additional observational analysis of a possible physical mechanism linking early-season snow cover and NH climate variability later in the season. We propose that SON snow cover may influence DJF tropospheric circulation by first impacting stratospheric dynamics. To pose the hypothesis, Cohen and Entekhabi (2001) reproduced the tropospheric anomaly patterns associated with the NAO or the NH dominant mode of winter variability in a general circulation model forced only by varying snow cover. In trying to reconcile those results with other studies showing that the dominant mode of variability is forced by the stratosphere (refer to the introduction), Cohen and Entekhabi (2001) hypothesized that changes in the latitudinal temperature gradient forced by snow cover may influence stratospheric dynamics. It is plausible that latitudinal temperature gradient changes forced by snow cover lead to anomalous potential vorticity gradients in the stratosphere, which then alter the wave refraction index. Changes in where propagating wave energy is deposited lead to the different observed phases of the AO. In this section, we further pursue this idea as a possible physical mechanism linking fall snow cover and NH climate variability during the winter season.

Using general circulation diagnostics and reanalysis data, we aim to establish the following dynamical mechanism connecting anomalous ESC_SON with wintertime Northern Hemisphere climate variability. Anomalously extensive snow cover in Eurasia in midautumn induces perturbed thermal forcing of the lower boundary (CSE01), resulting in the amplification of the orographically forced regional stationary Rossby wave response (Ringler and Cook 1999). In autumn the tropospheric stationary Rossby waves are allowed to propagate into the stratosphere under favorable zonal-mean westerlies. Later in winter, with the onset of strong tropospheric–stratospheric coupling (Kuroda and Kodera 1999), a poleward and downward shift of mean zonal wind is expected from the stratosphere into the troposphere. Resultant stratospheric and tropospheric height anomalies are consistent with opposite phases of the dominant mode of variability in the NH.

We calculate the wave activity flux that diagnoses the vertical propagation of the stationary Rossby waves (Plumb 1985). The wave activity flux is a three-dimensional extension of the Eliassen–Palm (E–P) flux, which illustrates the propagation of wave activity and the net effect of the waves on the mean flow. Figure 5 shows the zonal-mean wave activity flux and the zonal-mean wind, u, both regressed onto the ESC_SON index for selected 30-day periods. In November, an upward and slightly poleward energy flux appears in the midlatitudes in the troposphere and in the stratosphere (Fig. 5a). Weakly negative zonal wind anomalies appear in the mid- to high-latitude stratosphere. After 45 days the upward and poleward flux is enhanced both in the troposphere and in the stratosphere. In this period negative u is strongly enhanced (Fig. 5b) and is shifted downward in the following month (not shown). With the end of the winter season the energy flux becomes weak in the stratosphere and downward in the troposphere (Fig. 5c) while the center of easterly wind anomalies propagates down to the lower stratosphere and weakens.

Our results, presented in Fig. 5, resemble Kuroda and Kodera (1999) calculations of the leading propagating mode of the common variability between the vertical component of the E–P flux and the zonal-mean zonal wind. Also a marked resemblance is found between the regression pattern over the Eurasian and the Atlantic sectors of 500-hPa geopotential height field in Fig. 4 (panels b, d, and f) and Fig. 2 (panels a, b, and c) in Kuroda and Kodera (1999). They demonstrate the role of tropospheric planetary waves in the coupled variability of the NH wintertime troposphere–stratosphere. What has not been shown is the anomalous source of vertical wave propagation over Siberia, which takes place earlier than the formation of the AO pattern. We suggest this may result from anomalous cooling forced by extensive snow cover.

Figure 6 shows a longitude–pressure cross section at 60°N (averaged over all latitudes between 50° and 70°N) of the mean wave activity flux difference between the five heaviest snow cover years (1972, 1973, 1976, 1993, and 1998) and the five lightest snow cover years (1979, 1980, 1988, 1990, and 1992) for the same three 30-day periods as shown in Fig. 5. In November, anomalous upward propagation is initiated around 60°E in the lower troposphere in years of extensive snow cover (Fig. 6a). Through early and midwinter the region of strong upward flux spreads over from 60°E to the date line (Fig. 6b). In late winter some remnant upward flux is still seen over eastern Eurasia, but the major fluxes over Eurasia are downward.

The horizontal distribution of the composite difference of the vertical wave activity flux component at the 100-hPa level is shown in Figs. 7a, 7c, and 7e. In November upward fluxes are observed over high-latitude eastern Siberia and the Gulf of Alaska, and a maximum is found over the Bering Sea (Fig. 7a). In midwinter, upward fluxes are strengthened and spread over a wide area of mid- and highlatitude Eurasia and the Gulf of Alaska (Fig. 7c). The maximum upward flux anomaly is found over eastern Siberia. In contrast, at the end of winter these high-latitude areas, once covered by upward flux anomalies, are largely covered by downward anomalies (Fig. 7e). While the region of upward flux anomaly is shifted to the south and another large area of upward flux anomaly is found over the northern North Atlantic. Presented in Figs. 7b, 7d, and 7f are the 5-yr composite differences of the horizontal distribution of Eurasian snow cover, for 1 month earlier than periods presented in the corresponding maps in Figs. 7a, 7c, and 7e. This demonstrates the geographical collocation of the region of greatest snow cover difference and the area of the greatest anomalous upward propagation, in the Eurasian sector. The 1-month lag is probably because it is the resultant heating anomaly that is forcing the changes in vertical propagation rather than the presence of snow cover itself; however, the reason for the lag of 1 month, found between the two fields, needs to be investigated further.

These results support the hypothesis proposed that connects autumn Eurasian snow cover anomalies and the leading mode of the winter Northern Hemisphere variability. However, the evidence is somewhat circumstantial. There are also gaps in the theoretical understanding on the dynamical mechanism of the tropospheric wave–stratospheric mean circulation interaction (Perlwitz and Graf 2001). The proposed hypothesis needs to be examined through further observational analyses, theoretical investigations, and simulation studies with numerical models of the global atmosphere with adequate resolution in the stratosphere.

7. Conclusions

Early season Eurasian snow cover (ESC_SON), from 1972 through 1998, shows statistically significant correlation with the leading modes of the Northern Hemisphere wintertime (DJF) extratropical atmospheric circulation variability with apparent barotropic characteristics from sea level to the lower stratosphere. The correlation coefficient between snow cover and the leading mode of variability of a given geopotential height level decreases with altitude, implying that snow cover may act as a lower boundary forcing to the atmospheric circulation in the troposphere. However field significance tests of the correlation between ESC_SON and atmospheric circulation fields show that the global significance is attained earlier at the 500-hPa level by half a month than at sea level, suggesting a downward propagation. This apparent contradiction may be resolved through a mechanism that incorporates the impact of snow cover initally on the stratosphere, which then progates to the midtroposphere and eventually down to the surface.

Further analysis shows that during the cold season the strength of the leading mode of variability propagates downward from the lower stratosphere to the middle troposphere on a timescale of a few weeks. The constructed variable of the downward propagating circulation anomaly, the uAO, is shown to have as high a correlation with ESC_SON as the AO.

Given the above results, a mechanism connecting ESC_SON anomalies and wintertime MSLP variability through upper-level dynamics is proposed. An anomalously extensive snow cover in Eurasia in midautumn acts as an additional thermal forcing at the lower boundary resulting in an amplification of the orographically forced stationary Rossby wave response over Eurasia. As autumn progresses, zonal-mean westerlies in the stratosphere are established and propagation of the tropospheric stationary waves are allowed into the stratosphere. Evidence of anomalous upward (downward) propagation of stationary Rossby waves is found over the Siberian sector in late autumn during years of more (less) extensive snow cover. As a result of tropospheric–stratospheric coupling, a poleward and downward shift of zonal-mean zonal wind is expected in winter. So that when extensive (reduced) ESC_SON occurs, during the following DJF, a weaker (stronger) polar vortex results in the stratosphere while in the troposphere over the Arctic positive (negative) anomalous pressures/heights are observed while at midlatitudes negative (positive) anomalous pressures/heights are observed.