a. Observational data

Observed atmospheric fields used in this study are derived from the National Centers for Environmental Prediction–National Center for Atmospheric Research (NCEP–NCAR) reanalysis (Kalnay et al. 1996), which spans from January 1948 to December 2002. Most of the data are available on a 2.5° × 2.5° grid at standard pressure levels.

For monthly mean fields, the monthly climatology was first subtracted for each variable and deviations from the climatology; that is, anomalies are used in the analysis. Since the AO/NAO is prevailing in boreal winter, we mainly focus on the monthly anomalies from December to February. The anomalies on interannual time scale associated with the NAO and AO are obtained by linear regressions onto the monthly indices for the NAO and the AO, respectively. The NAO index is a conventional one, defined by a normalized SLP difference between Lisbon, Portugal, and Stykkisholmur, Iceland, (Hurrell 1995) while the AO index is adopted from the principal component associated with the leading EOF of the monthly 1000-hPa height (equivalent to SLP) anomaly, provided by the Climate Prediction Center. Regressed anomalies corresponding to one standard deviation of each index are hereafter referred to as either the monthly NAO or AO anomalies.

As for daily average fields, the climatology is defined by a 55-yr mean value for each calendar day, then daily anomalies are obtained as in the monthly anomalies. A four-pole, Butterworth low-pass filter, which retains periods longer than 10 days, is applied to daily anomalies in order to remove transient synoptic disturbances. Then the EOF analysis is performed on the filtered SLP anomalies over the Atlantic sector of 20°–90°N, 120°W–30°E during 1 November through 30 April in order to extract the NAO signature in the daily low-frequency fields. As will be shown in section 4, the leading EOF well represents the NAO pattern, and the associated principal component (referred to as the PC1) reveals quite high coherence with the conventional NAO index such that the correlation coefficient of the December–February average of this PC1 time series with the Hurrell index exceeds +0.87. All analyses of daily filtered data are made with PC1 as a reference time series. While variables that we analyze contain upper-level irrotational and nondivergent flow fields as calculated from the velocity potential and streamfunction on a sigma surface in addition to geopotential height and zonal and meridional winds at pressure levels, only some of them are presented in the present paper.

b. Barotropic model

The model follows a barotropic vorticity equation linearized about the observed 300-hPa climatological flow, as expressed by

View Expanded

where J denotes a Jacobian operator and ψ and ψ′ are respectively the basic state and perturbation streamfunctions; f is the Coriolis parameter, while an anomalous vorticity source induced by the divergent part of the circulation is symbolically represented by S′. The model includes a linear damping that represents the Rayleigh friction and a scale-selective biharmonic diffusion. The coefficient ν is determined such as to damp the smallest-scale eddy in 1 day while the coefficient α is set at (10 days)⁻¹, which ensures that the system is marginally stable.

Equation (1) can be solved with a given S′ on a spherical domain in which the perturbation streamfunction is represented by spherical harmonics of T42 horizontal truncation.

a. Downstream extension of the NAO

Monthly NAO anomalies in the 300-hPa geopotential height and 925-hPa temperature from December to February are presented in Fig. 2, together with the corresponding AO anomalies. Among these anomaly maps a common feature is found over the Atlantic and surrounding regions: north–south dipole in the height anomalies that characterizes the NAO, low-level warming over the eastern half of the United States and Scandinavia, and cooling over the Labrador Sea and northern Africa. The AO anomalies reveal another center over the North Pacific where positive height anomalies extend from East Asia, which is connected with the low-level warming over the region through the hydrostatic constraint (Figs. 2d–f). It is this pattern of the AO that has been recognized as the hemispheric variability. On the other hand, the NAO anomalies in December are confined to the Euro–Atlantic sector as they match the regional nature of the NAO (Fig. 2a). It should be noted, however, that the NAO anomalies in February show a wider horizontal structure that is quite similar to the AO anomalies (Fig. 2c). This “downstream extension” of the NAO is subtle in January (Fig. 2b) but clear in March as well (not shown), consistent with the high correlation between the AO and NAO indices during the months (Fig. 1). The similarity between Figs. 2c and 2f indicates that there exist NAO-covariant interannual fluctuations over East Asia and the North Pacific during a certain month, contrasting with previous arguments that point out the lack of such an interbasin connection associated with the AO/NAO.

Month-to-month change in the interannual variability of the NAO has also been reported by Barnston and Livezey (1987). Hurrell (1996) shows that the NAO has the largest influence on the hemispheric surface temperature variance among various teleconnection patterns. It turns out from Fig. 2 that the influence was mainly due to the downstream extension occurring in late winter. Then, how can the NAO induce the near-surface warming over a wide area of Eurasia? A process which likely happens is horizontal temperature advection, more specifically the zonal advection of the climatological temperature gradient by the anomalous zonal-mean zonal wind. With respect to the AO, Thompson and Wallace (2000) depict that this advection term generates an anomalous temperature tendency quite similar to the low-level temperature anomaly pattern shown in Figs. 2d–f. For example, the zonal-mean wind associated with the positive phase of the AO (cf. Fig. 2d) shows an enhanced westerly around 55°N, which advects warm climatological temperature over the Atlantic Ocean deeper into the Eurasian continent. It is well known that the pronounced zonally uniform components characterize the AO, and dynamical modeling work suggests the AO-like structure originated by a coupling between zonal-mean anomalies and the climatological zonal asymmetry, which involves the aforementioned temperature advection (Ting et al. 1996; DeWeaver and Nigam 2000; Kimoto et al. 2001; Watanabe and Jin 2004).

The anomalous zonal-mean wind associated with the NAO has a meridional profile nearly identical to that associated with the AO except for slightly smaller magnitude (not shown). Nevertheless, low-level temperature anomalies in the NAO regressions are confined, compared to the AO anomalies, to early winter, suggesting that the advection due to anomalous zonal-mean zonal wind is not a unique process that brings the near-surface warming over Eurasia although it does work (Xie et al. 1999). We explore another possibility for the temperature anomalies: an upper-level circulation anomaly that is dynamically induced, which balances the temperature field beneath. Since the January fields exhibit a mixture of December and February anomalies in terms of the downstream extension (Fig. 2b), a comparison is made in the following analysis between December and February anomaly maps in order to distinguish the NAO without and with the downstream extension.

As noted above, the AO/NAO-covariant circulation anomaly has a substantial projection on the zonally uniform field. The anomalous circulation as indicated by the zonal wind or the streamfunction anomalies often masks the regional feature because of their pronounced zonal mean components. Therefore we analyze the circulation anomalies as represented by the meridional wind, which may be a good indicator for the zonally asymmetric teleconnection patterns (cf. B02).

Figures 3a and 3b show the monthly NAO anomalies in the 300-hPa meridional wind in December and February, respectively. It is evident that the meridional wind anomalies are mostly confined to the Atlantic region in December whereas they extend to the entire midlatitude band in February. The anomalous northerly and southerly wind near Greenland corresponding to the deepened Icelandic low is similarly found in both regression maps. In February, they are continuous toward the Middle East besides emanating to East Asia and the North Pacific along the axis of the subtropical jet (thin contours in Fig. 3b). Such a zonally oriented wave train is identified as one of the dominant teleconnection patterns at subseasonal time scales (Kushnir and Wallace 1989; B02). B02 further interprets the pattern shown in Fig. 3b as quasi-stationary Rossby waves trapped on the Asian jet, which acts as waveguide (Hoskins and Ambrizzi 1993). Although it should be again emphasized that the anomalous meridional wind in Fig. 3b is found on interannual time scale, analysis in the next section shows that it does happen on an intraseasonal time scale as well. The climatological zonal wind at 300 hPa along 30°N (Fig. 3c) illustrates a seasonal evolution of the Asian jet around 60°E which is weaker in early winter and strongest in February. Thus the February basic flow may provide more favorable conditions than that in December in terms of the establishment of the Asian jet waveguide.

b. Anomalous Rossby wave source

Figure 4 displays S′ at 300 hPa associated with the NAO in December and February, respectively, superimposed on the climatological mean zonal wind. The anomalous Rossby wave sources are similar to each other over the Atlantic basin, namely, positive and negative sources to the west and east of Greenland, positive anomaly around the Azores and negative anomaly on the equatorial band. However, conspicuous differences are found over continents from Africa to East Asia. While few significant anomalies are observed in December, there are a number of significant anomalies along the Asian jet in February. In particular, S′ over southern Europe to the Middle East appears to dominate the others, and it is the region where the Asian jet is established.

The anomalous Rossby wave source in Fig. 4 can be decomposed into several terms, each of which is calculated from monthly NAO anomalies as

View Expanded

where the primes denote anomalies regressed upon the NAO index while overbars are climatological quantities. The anomaly is not so huge that the nonlinear term can be neglected, then we can specify the leading term between S^′₁ and S^′₃ resulting in S′ shown in Fig. 4b. The three terms are illustrated in Fig. 5, with a denser contour interval than in Fig. 4. Note that their sum is nearly identical to Fig. 4b. It is obvious that the first two terms (Figs. 5a,b), the convergence associated with the relative vorticity, are less dominant than the third term (Fig. 5c), which is mostly the same as a commonly used divergent forcing, −fD′ (e.g., Branstator 1985), where D′ is the divergence anomaly. Sardeshmukh and Hoskins (1988) notice that one should correctly evaluate the Rossby wave source using Eq. (4) but not by a simple −fD′ term in order for accurate diagnosis of the rotational flow. In the present case, however, the −fD′ approximation seems valid, which tells us that most of the anomalous Rossby wave source in Fig. 4b comes directly from the divergence anomaly associated with the NAO. In the subsequent section, diagnoses using the linear barotropic model are made for the Rossby wave source and circulation anomalies associated with the NAO.

c. Steady solutions in a barotropic model

Regressed NAO anomalies in February showing the zonally oriented wave train along the Asian jet (Fig. 3b) concurrent with the anomalous Rossby wave source near the jet entrance region (Fig. 4b) suggest that the downstream extension of the NAO is accomplished by energy propagation due to stationary Rossby waves excited by the vorticity source, perhaps, associated with the NAO. This inference is encouraged by a theoretical consideration of the Rossby wave reflection (Hoskins and Ambrizzi 1993), which depicts that the region where the meridional gradient of the absolute vorticity, ∂ξ/∂y, is large, and hence stationary wavenumber K_s is also large, tends to trap stationary Rossby waves. Because of the meridional curvature, subtropical jet streams such as the Asian jet have local maxima in K_s and therefore act as waveguides for quasi-stationary Rossby waves. Following this theory, the vorticity source around the Mediterranean Sea and Middle East may be capable of exciting quasi-stationary waves with a zonal wavenumber larger than K_s, which is roughly 5–7 along the winter Asian jet.

In this section, by means of a barotropic model we test whether the above processes do occur. The ability of the barotropic model is first validated in terms of the reproducibility of observed NAO anomalies. By assuming that the tendency term in Eq. (1) can be neglected on the time scale discussed in this section, a steady streamfunction response is calculated with the February climatological flow as a basic state and the anomalous vorticity source depicted in Fig. 4b, then compared to the observed counterpart (Fig. 6). For both the observed and simulated streamfunction anomalies the zonal mean component has been removed in Fig. 6 so as to visualize the waveguide patterns. Since the vorticity source anomaly is less reliable in higher latitudes due to larger errors in the divergent wind, the forcing is restricted between 0° and 60°N in computing the steady response. The observed 300-hPa streamfunction anomaly in February, denoted as ψ^*₃₀₀ (asterisk stands for the deviation from zonal average), reveals a zonally elongated pattern over most parts of the Eurasian continent whereas a wavelike feature from East Asia to the North Pacific (Fig. 6a). The steady ψ^*₃₀₀ response with given observed S′ reproduces such patterns well (Fig. 6b) although several discrepancies exist, for example, too strong anticyclonic response over the North Pacific and response to the north of the Aleutian Islands with sign opposite to the observation. These shortcomings must be caused by the following reasons: lack of the forcing due to submonthly Rossby wave source anomalies, steady state assumption, errors in high-latitude vorticity source, and simple representation of nonlinear dissipation included by the linear damping, in which the first two may be the principal causes. Nevertheless, the overall feature in Fig. 6b is similar enough to the observed anomalies in Fig. 6a, indicating that the barotropic model is capable of simulating the NAO-covariant circulation anomalies over Eurasia. It should be noted that the NAO pattern itself is not well reproduced in the steady response due to lack of transient eddy vorticity forcing, which has been shown to be dominant in driving the NAO pattern (cf. Hurrell and van Loon 1997; Feldstein 2003).

The barotropic response captures two anomalous ridges over East Asia and the North Pacific, which accompany a wave train in the meridional wind as observed (not shown, but similar to Fig. 3b). A question is raised here: which part of the Rossby wave sources in Fig. 4b is most responsible for the downstream response? To answer the question it is useful to employ the so-called Green's function approach as introduced by Branstator (1985). Namely, we define the Green's function as

View Expanded

where λ and φ are longitude and latitude, respectively. The linearized operator of the model, that is, the left-hand side of Eq. (1), with the tendency term neglected, is symbolically represented by L and the function G is obtained as a steady vorticity response to the forcing δ, which is a delta function having the value f₀ only at λ = λ′ and φ = φ′. The forcing position is actually given on discretized grid points corresponding to the T42 horizontal resolution, then G consists of 128 × 32 = 4096 steady response patterns (no forcing over the Southern Hemisphere). We set f₀ = 1 × 10⁻¹⁰ s⁻² multiplied by sin φ to mimic the divergent forcing that dominates in S′ (Fig. 5).

Green's function, which depends not only on space but also on the forcing position, is now used to evaluate an optimal vorticity source for a specific streamfunction response. Given a target region denoted as T, the optimal source S_o is defined as

View Expanded

where A is the area extent of the target region and the integral is taken in terms of λ and φ: S_o is obtained in a sense of the streamfunction and normalized by a factor σ such as to have unit spatial variance. We choose the anomalous ridge over East Asia as the target region, which is indicated by a rectangular box in Fig. 6b, since it is one of the major parts in the NAO downstream extension. The region is slightly more southward than the simulated anomalous ridge while it covers the observed counterpart. It is found that the result is not very sensitive to slight meridional change in the target region.

Shown in Fig. 7 are the distribution of S_o respectively evaluated with December and February basic flow, superimposed on the basic-state westerly as well as the location of T. The optimal source defined by Eq. (5) implies that the positive (negative) vorticity source at centers of positive (negative) S_o is effective in exciting the positive streamfunction response over the region T. For both December and February basic flow S_o has distinctive positive maxima over the Mediterranean Sea, Bay of Bengal, and Siberia while a minimum over the Arabian Sea. The positive values over the Mediterranean Sea are stronger in February, consistent with a stronger Asian jet to the east, which manifests that even for the same vorticity forcing over the region the East Asian ridge is more strongly excited in February, possibly due to the jet waveguide being solidly established. A comparison between Figs. 4 and 7 enables us to estimate which part of the observed Rossby wave sources associated with the NAO plays a leading role for the eastward extension. It is easily identified that the positive vorticity source over the Eastern Mediterranean Sea coincides with another in February while it is found only in S_o during December, suggesting that the anomalous wave source over the region is a key linking the NAO to anomalous circulation over East Asia. The optimal vorticity source reveals that the East Asian ridge can be excited by the subtropical forcing over the Indian Ocean and the western Pacific as well, but it is not found in S′ associated with the NAO.

We also tested the sensitivity of the optimal source pattern to the choice of the target region. If the region T is defined by the average over 25°–45°N, 145°–175°E, which corresponds to the anomalous trough in Fig. 6, the optimal negative source emerges at the jet entrance where the positive source is found in Fig. 7. On the other hand, for T between 120° and 150°E, the optimal vorticity source reveals a quite similar distribution to Fig. 7 but the positive source near the Asian jet entrance is shifted eastward by 20° (not shown). This result implies that the zonal phase of the wave train as observed in Fig. 3b is determined by the position of the vorticity source but not other processes such as the zonal inhomogeneity in the time-mean flow. In fact, B02 identified a family of these wave trains, each of which has a different zonal phase, consistent with the above interpretation. Namely, the anomalous Rossby wave source near the Asian jet entrance, when it is produced in association with the NAO, will excite one of the waveguide teleconnection patterns attributed to the spatially fixed vorticity source.

The steady meridional wind response to the pointwise vorticity forcing placed at the center of positive S_o over the Mediterranean Sea is shown in Fig. 8. As expected, a zonally oriented wave train is excited to the east of the forcing, strictly along the Asian jet. The position of southerly and northerly wind anomalies are in remarkable agreement with those observed in February (Fig. 3b). It should be noted that the wind response in December is quite similar to that in February (Fig. 8a), indicating that the vorticity source over the Mediterranean Sea can be optimal in both months, as suggested by Fig. 7. The above diagnosis implies that the lack of the downstream extension of the NAO in December results from the lack of S′ in an optimal region (cf. Fig. 4a), which is crucial for the waveguide teleconnection as shown in Fig. 8. This automatically leads to a conclusion that the stronger Asian jet in February is of secondary importance.

d. Mediterranean convergence and the NAO

In the previous subsections the convergence anomaly over the Mediterranean Sea was shown to play a key role in the NAO downstream extension. While detailed analysis of how the NAO produces the Mediterranean convergence anomaly will be performed in the next section, preliminary results are shown here. We especially focus on whether the February time mean state favors it. In other words, the NAO extremes that accompany the downstream extension are looked for in December and January as well, using an analysis other than the linear regression.

First, an index that measures the Mediterranean convergence (denoted as MC) is defined by taking the monthly convergence anomaly at 300 hPa averaged over 25°–45°N, 10°–30°E, referred to as the Mediterranean convergence index (MCI). The time series of MCI in each winter month are shown in Fig. 9, superimposed on the Hurrell (1995) NAO index. Correlation coefficients between the two indices are negligible in December (r = 0.14) while significantly increasing in January and February (r = 0.39 and r = 0.41, respectively), which support our hypothesis that the downstream extension appears when the MC is coherent with the NAO. It is interesting to note that the MCI is stably correlated with the AO index, ranging from r = 0.46 in December to r = 0.50 in February, which again matches the regression analysis that shows that the AO anomalies always accompany the signal over East Asia and the North Pacific (Fig. 2).

The NAO extremes can be classified into two categories by taking the concurrence of the MC into account, as summarized in Table 1. Although the NAO index and MCI are almost uncorrelated in December, about half of the positive NAO extremes as defined by the index value exceeding one standard deviation accompanies the positive phase in MCI. This ratio is roughly the same in the other winter months for the positive NAO/MCI extremes whereas the concurrence of negative NAO and MCI extremes is hardly found in December and January. While the cause of this phase asymmetry is not clear at the moment, it might result in a low correlation between the NAO index and MCI in December (Fig. 9a). The categorized statistics also suggest that the downstream extension may be observed in December–January if we choose positive NAO extremes adequately.

In a similar manner to Table 1, monthly composite NAO anomalies that consist of 10, 8, and 9 extremely positive years in December, January, and February (cf. Table 1) are separated into two: one based on years having positive MCI and another with negative or normal MCI years. Shown in Fig. 10 are those two composite anomalies in the 300-hPa streamfunction for each month (results are similar in the 500-hPa geopotential height). Composite anomalies that accompany the positive MC anomaly (Figs. 10a–c) have some common features for each of three months, that is, anticyclonic anomalies over East Asia, the North Pacific, and the Arabian Sea in which the former two correspond to the anomalous ridges in the NAO downstream extension (cf. Fig. 6a). Anomalous cyclonic circulation also tends to occupy a wider region over the polar cap. In contrast, the positive NAO composites that do not accompany the positive MC anomaly (Figs. 10d–f) reveal a pattern varying from month to month over those regions, although large anomalies sometimes happen such as in January. Although the statistical significance of anomalies is relatively tenuous except for the Atlantic sector, which would be due to insufficient sample size, Fig. 10 clearly represents that the NAO anomalies look like the hemispheric AO anomalies when the MC anomaly is simultaneously taking place. Furthermore, the NAO downstream extension identified in one category of the composite (i.e., Figs. 10a–c) but not in another (Figs. 10d–f), which becomes clearer in the meridional wind composites (not shown), indicates that it is not only preferred by the February climatological flow but may occur in the other winter months.

It is noted that the NAO itself shows a slightly different structure between Figs. 10a–c and 10d–f. The former has a southern part (anticyclonic anomaly) split into the upstream center over the U.S. east coast and the downstream center over Europe, whereas the latter has a center over the North Atlantic Ocean. The composite anomalies that do not accompany the MC anomaly match the classical view of the NAO in terms of a regional north–south dipole. The circulation anomalies in Figs. 10a–c form an arching structure from North America to Europe and farther downstream, suggesting energy propagation along the path that may favorably cause the anomalous convergence over the Mediterranean Sea. The difference in the NAO structure visible in Figs. 10a–c and 10d–f is somewhat consistent with Kodera and Kuroda (2003) who point out the presence of two “modes” over the Euro–Atlantic sector: a regional pressure seesaw between Iceland and the Azores and a hemispheric seesaw between the Mediterranean and polar regions. However, it is hard to state the anomaly patterns, such as Figs. 10c and 10f, as a different “mode”; rather we hypothesize that their Atlantic part (i.e., NAO) is generated by a common dynamical process but with a slight difference, for example, in the forcing and the background state (e.g., Rodwell et al. 1999; Peng et al. 2002) that cannot be specified yet. This serves as a condition whether the MC anomaly, hence downstream extension, is accomplished.

a. Detection of the NAO events

To define the NAO events we chose PC1 time series, associated with the leading EOF to daily SLP anomalies over the Atlantic sector, as an index of the NAO on an intraseasonal time scale. The leading sectorial EOF accounts for 23.8% of the total variance and clearly shows the NAO pattern (Fig. 11a). This EOF pattern, in addition to the high correlation with the canonical NAO index on an interannual time scale (cf. section 2a), allows us to consider the PC1 as a good indicator of the NAO events; therefore we refer it to as the daily NAO index hereafter. The basic property of the NAO can be represented by an autocorrelation function of the daily NAO index as shown in Fig. 11b (black line), which exhibits a decorrelation time of about six days. This is relatively short but still longer than the other EOF patterns as confirmed by the autocorrelation averaged for the leading 30 PCs (white line in Fig. 11b). If the autocorrelation function of the daily NAO index is calculated with a 31-day moving window, it is found that the decorrelation time of the NAO event is not uniform during the cold months (Fig. 11c) such that the NAO has the longest persistence in February.

The way to define the NAO events is similar to Feldstein (2002, 2003), and appears rather conventional. Namely, we define a positive (negative) episode when the positive (negative) spatial correlation between the leading EOF and daily SLP anomalies over the Atlantic sector (i.e., the EOF domain) exceeds the 99% significance level and prolong for 4 or more consecutive days. Note that the results presented in this section are not sensitive to the threshold of persistence. The peak, or mature, day of each event is determined by the highest correlation during the episode. Figures 12a and 12b indicate the positive and negative NAO events during the 1949–2002 period, respectively, following the earlier definition. The number of events are 168 and 146 for positive and negative phases of the NAO, which seem to be large enough for the composite analysis in the next section. It is interesting to note that the positive events are more frequent as well as persistent than the negative events in the recent two decades, reflecting the well-known upward trend in the NAO index.

Following the mature event dates identified in Figs. 12a and 12b we would like to perform a composite analysis to various daily, low-pass filtered fields. Given the typical time scale of the NAO event, its time evolution will be depicted by taking 10-day lags before and after the peak day. For this purpose, the events are further selected by a criterion that, if two or more peaks are found with the interval less than 10 days, only the one that reveals the largest value is retained. This procedure consequently reduces the total number of positive and negative events to 122 and 116, respectively, as shown by Figs. 12c and 12d in which the duration of each event is made uniform at seven days for convenience. The long-term change in the persistence found in Figs. 12a and 12b, which is beyond the scope of the present analysis, has been filtered out although the decadal change in the frequency of occurrence is still visible in Figs. 12c and 12d. Since it was already shown in the previous section that the NAO downstream extension is concurrent with the upper-level convergence anomaly over the Mediterranean Sea, individual NAO events are separated into two categories by referring to the daily MCI as in section 3d but obtained from divergent field at σ = 0.258. If the MCI exceeding one standard deviation is observed during day 0–4 of an event with the same sign (i.e., positive MCI with positive NAO event), such an event is categorized into the NAO that accompanies the MC anomaly, and vice versa. The choice of day 0–4 comes from the fact that the preliminary analysis indicates, as will also be confirmed later, that a waveguide pattern in daily anomalies slightly lags the development of the NAO over the North Atlantic.

With the definition described above, it turns out that about one-third of the total events accompanies the MC anomaly, as summarized in Table 2. Frequency of concurrence of the MC anomaly with the NAO event is generally higher in the positive phase, especially during February when more than half the positive NAO accompanies the MC anomaly. This feature is consistent with the similar statistics obtained with monthly mean data (Table 1), so that pictures similar to Fig. 10 can be drawn with daily fields as well (not shown).

b. Composite analysis

Results of analysis to the monthly mean fields let us expect that the NAO events with (without) the MC anomaly reveal (do not reveal) the downstream extension as represented by the waveguide pattern over Eurasia. This inference is tested by means of the lagged composite of the low-pass filtered data in reference to the NAO events shown in Figs. 12c and 12d. The composites for the NAO with and without the MC anomaly are calculated both for positive and negative phases, but the results are quite similar to each other except for the different number of samples (Table 2) and the sign reversed, so only the anomaly patterns for positive phase are presented.

Figure 13 illustrates the time evolution of the positive NAO events in terms of the 300-hPa meridional wind anomaly. Since “day 0” coincides with the peak of events, the evolution displayed corresponds to the decay period of the NAO event. In the NAO composite anomalies that accompany the MC anomaly (Figs. 13a–d), the waveguide pattern along the Asian jet is evident after day 2. In the counterparts that do not accompany the MC anomaly (Figs. 13e–h), however, a quadrapole pattern is found over Europe throughout the period, but no extension toward East Asia is observed. This result supports our argument based on the monthly data and, furthermore, identified that the establishment of the waveguide pattern lags the peak of the NAO by two days. Eastward propagation of wave packets associated with the decay period of the NAO is seen more clearly in the longitude–time section at 30°N (Fig. 14). A set of northerly and southerly wind anomalies over western Europe exists irrespective of concurrence of the MC anomaly. On the other hand, the NAO event with the MC anomaly (Fig. 14a) accompanies a fast propagation of wave packets, persistent even until day 10. These wave packets, as well as the slow phase propagation, are again consistent with the energy propagation of quasi-stationary, meridionally trapped Rossby waves (Hoskins and Ambrizzi 1993; B02). The slow phase propagation visible in Fig. 14a may be the major reason that the downstream extension can be identified in monthly mean data.

Two sets of the positive NAO composites are categorized by the MCI, so we anticipate that there are conspicuous differences in the upper-level divergent field as well (Fig. 15). In the NAO composites with the MC anomaly (Figs. 15a–d), a large positive anomaly in the velocity potential and associated convergent flows are found over the Mediterranean, which is exactly the region of optimal vorticity source in a steady limit (Fig. 7). Compared to those anomalies, the convergence signals in the counterparts (Figs. 15e–h) are weak, less persistent, and shifted to the west by about 25°. If the quasi-stationary Rossby waves in Figs. 13a–d can only be excited near the Asian jet entrance, as suggested in section 3c, the westward shift in the convergence anomaly appears crucial for the presence/absence of the NAO downstream extension. The anomalous convergence during the decay period of the positive NAO events, as shown in Fig. 15, has also been identified by Feldstein (2003) who points out, based on vorticity budgets, that the divergent term works to damp the NAO. We may therefore conclude that the NAO downstream extension is an energy dispersion accompanied by the decay of some of the NAO events. Feldstein also argues that the upper-level divergence anomaly results from the Ekman pumping near the surface. Our analysis indeed supports his argument, as revealed in Fig. 16, showing similar composites in SLP and boundary layer (σ = 0.995) divergent flows. The location of divergence (convergence) anomalies agrees well with the high (low) pressure anomalies in Fig. 16, consistent with the frictional divergent flow. Again, the NAO composite with the MC anomaly (Fig. 16a) reveals, at the peak stage, a center of positive SLP and divergence anomalies over the Mediterranean, which are not clearly detected in the counterpart (Fig. 16b). It is interesting to note that the high pressure and divergence anomalies appear over the same location, even at 5 days before the peak of the NAO (Figs. 16c and 16d) collocated with the upper-level MC anomaly (not shown), indicating that the signal of the NAO downstream extension, established during the decay stage (see Figs. 13a–d), is detectable even during the growth stage of the NAO. While thorough analysis of the whole NAO cycle is not in the scope of this study, the above results suggest as a medium-range weather forecast over East Asia that the NAO downstream extension, which affects surface temperature over the region, is to some extent predictable by carefully monitoring the developing stage of individual NAO events.