The link between Rossby wave breaking (RWB) and the four wintertime weather regimes over the North Atlantic domain is studied in this paper. Using the 40-yr ECMWF Re-Analysis (ERA-40) data, frequencies of occurrence of anticyclonic and cyclonic wave-breaking (AWB and CWB, respectively) events are computed. Each weather regime has its own characteristic pattern of RWB frequencies. CWB events are found to be most frequent for the Greenland anticyclone weather regime whereas AWB events occur more for the Atlantic ridge and the zonal regimes. Time-lagged composites show that the RWB events characterizing each weather regime occur more often during the formation of the regime rather than during its decay. This suggests a reinforcement of the weather regime by RWB. An exception is the blocking weather regime, which is destroyed by an increase of CWB events south of Greenland.
Weather regime transitions are then studied using the low-frequency streamfunction tendency budget. Two types of precursors for the transitions have been identified. One is related to linear propagation of low-frequency transient eddies and the other to nonlinear interactions among the low- and high-frequency transient eddies. The latter has been related to the anomalous frequencies of occurrence of RWB. Two transitions are more precisely analyzed. The transition from blocking to Greenland anticyclone is triggered by a decrease of AWB events over Europe as well as a strong CWB event south of Greenland. The zonal to blocking transition presents evidence of two distinct precursors: one is a low-frequency wave train coming from the subtropical western Atlantic and the other, which occurs later, is characterized by a decrease of AWB and CWB events over western Europe that cannot continue to maintain the westerlies in that region.
Because of its major role in medium-range weather forecasts, the atmospheric low-frequency extratropical variability has been widely studied in the last decades. One concept of midlatitude low-frequency variability is the weather regime (WR), which corresponds to a recurrent and quasi-stationary state of the large-scale atmospheric circulation persisting over one or several weeks (Vautard 1990). Different methods can be applied to obtain the weather regimes over a given geographical domain and lead usually to similar patterns (Michelangeli et al. 1995, hereafter M95). Using two different methods, M95 found four weather regimes over the Atlantic sector and three over the Pacific sector. One classical weather regime is blocking, for which the formation, maintenance, and decay are still a matter of debate (e.g., Pelly and Hoskins 2003; Altenhoff et al. 2008). Initially, most studies focused on the maintenance of weather regimes and in particular of blocking (e.g., Shutts 1983; Vautard and Legras 1988). They found that synoptic waves tend to maintain or reinforce it. So far only a few studies have analyzed the onset to blocking. More recently, Nakamura et al. (1997), Michelangeli and Vautard (1998), and Altenhoff et al. (2008) found a wave train as a precursor for blocking, while Croci-Maspoli and Davies (2009) emphasized the role of cloud diabatic effects. But no studies have looked at all the weather regime transitions on the whole and tried to categorize them according to their different dynamical properties. The purpose of the present paper is to analyze all the weather regime transitions in the Atlantic sector and to identify the different types of precursors.
A potential precursor can be Rossby wave breaking (RWB). For the last few years, many studies have emphasized the link between RWB and the main teleconnections such as the North Atlantic Oscillation (NAO) or the Pacific–North American (PNA) patterns, which correspond to another notion of low-frequency atmospheric variability (Wallace and Gutzler 1981). RWB is a nonlinear phenomenon occurring when Rossby waves attain large amplitudes and is usually defined by a large-scale and irreversible overturning of the potential vorticity (PV) contours on isentropic surfaces (McIntyre and Palmer 1983). It results in a PV mixing in the wave-breaking (WB) region. Thorncroft et al. (1993) identified two RWB types appearing at the end of two distinct baroclinic wave life cycles (LC1 and LC2). They showed that the LC1 ends with an anticyclonic WB (AWB; featuring a wave tilting in the southwest–northeast direction) on the equatorial side of the jet stream whereas the LC2 ends with a cyclonic WB (CWB; featuring a wave tilting in the southeast–northwest direction) on the polar side of the jet. Using composites of extreme NAO phases, Benedict et al. (2004) have shown that these phases are triggered and reinforced by RWB. The positive phase is formed by two consecutives AWBs, one near the western coast of North America and the other over the subtropical North Atlantic. The negative phase of the NAO is the result of a single CWB in the North Atlantic. Since then, several studies have confirmed the dynamical link between RWB and the NAO in different ways (Rivière and Orlanski 2007; Martius et al. 2007; Woolings et al. 2008; Strong and Magnusdottir 2008; Kunz et al. 2009).
RWB therefore plays a crucial role in the modification of low-frequency anomalies, which are directly related to jet fluctuations. Indeed, during AWBs, eddy momentum fluxes are essentially poleward and the zonal flow is accelerated (decelerated) north (south) of the latitude of breaking whereas during CWBs, momentum fluxes are mainly equatorward and accelerate (decelerate) the zonal flow south (north) of the latitude of breaking (e.g., Rivière and Orlanski 2007; Strong and Magnusdottir 2008). Furthermore, AWBs act to split the North Atlantic eddy-driven jet and the subtropical jet whereas CWBs tend to merge them. In the present paper, the ability of RWB to trigger weather regime transitions is investigated.
Another and more classical candidate for driving such transitions is a low-frequency wave train excited by tropical convection (Hoskins and Karoly 1981). Renwick and Revell (1999) clearly linked the occurrence of blocking in the southeastern Pacific to the propagation of Rossby waves forced by tropical convection using observational evidence. Franzke et al. (2011) have shown that RWB anomalies and low-frequency wave trains coming from the tropics are both important for the development of the PNA phases. Similarly, Cassou (2008) proposed that these two dynamical ingredients may explain the link between the Madden–Julian oscillation (MJO) and the NAO.
In the next section, the four weather regimes of the Atlantic sector are described and an index determining their intensity at each day is computed. Our automatic WB detection algorithm is also described in section 2. RWB patterns characteristic of each weather regime are shown in section 3. Time-lagged composites are made to determine the role of RWB in the onset and decay of each weather regime. In section 4, weather regime transitions are more specifically studied using RWB diagnostics and the low-frequency streamfunction tendency equation. Finally, a summary of the results is provided in section 5.
2. Data and methodology
In the present study, the 40-yr European Centre for Medium-Range Weather Forecasts (ECMWF) Re-Analysis (ERA-40) dataset (Uppala et al. 2005) is used. Various daily-mean fields with a resolution of 1.5° × 1.5° are examined, including the geopotential, the horizontal wind components, the absolute vorticity (AV), and the PV on isobaric and isentropic surfaces. Our analysis covers the period from 16 October to 15 April (called extended winter) for years from 1958 to 2001 corresponding to a total of 7837 days or 43 extended winters.
We also use the National Oceanic and Atmospheric Administration (NOAA) daily-mean interpolated outgoing longwave radiation (OLR) (Liebmann and Smith 1996), which covers the years 1974–2001 and has a 2.5° × 2.5° resolution. It will be used as a proxy for convection.
a. Weather regimes
The same dynamical cluster algorithm as that developed by M95 is applied to the low-frequency (periods greater than 10 days) geopotential at 500 hPa. The main steps of this algorithm are briefly recalled. An empirical orthogonal function (EOF) analysis is first performed on the 500-hPa low-frequency geopotential field in the Atlantic domain (28.5°–79.5°N, 79.5°W–28.5°E). Twenty EOFs, which explain about 95% of the variance, are retained. A partitioning algorithm is applied to the 500-hPa low-frequency geopotential field in this reduced EOF subspace. Given a number of clusters k, the principle of this algorithm is to partition the 500-hPa low-frequency geopotential field into k clusters in order to minimize the sum of the variance inside each cluster. The right number of clusters is determined using a classifiability index c(k), which is a measure estimating the resemblance of the different partitions obtained from different random initializations and which depends on k. If c(k) obtained for our real data field is significantly larger than c(k) obtained for random built fields, it means that the partition of the real data is relevant. By applying this test, k = 4 and k = 8 were found to be the two best suitable numbers of clusters k. Note that k = 4 is also the number found by M95 using other datasets and is therefore chosen in the present study. The present partition of the ERA-40 dataset is the same as that described in Rivière (2008). It leads to the four WRs initially described by Vautard (1990). They are called the blocking (B), Greenland anticyclone (GA), Atlantic ridge (AR), and zonal (Z) regimes and their main features are represented in Fig. 1. Each day of our extended winters belongs to a given cluster and corresponds therefore to a unique WR.
Regime B is characterized by a strong Scandinavian high and a very high-latitude jet in the eastern Atlantic (Fig. 1a). This WR is generally responsible for advection of cold air in western and central Europe. GA features a north–south-oriented dipolar anomaly in the low-frequency streamfunction with a high centered over Greenland (Fig. 1b). The Atlantic jet is zonally oriented, more to the south than usual, and connected with the subtropical African jet in the eastern Atlantic. AR features a strong anticyclone over the central North Atlantic with a jet shifted from the south to the north over the eastern coast of North America and then zonally extended to Scandinavia (Fig. 1c). Finally, Z is characterized by a north–south-oriented dipolar anomaly in the low-frequency streamfunction (Fig. 1d). The jet has a southwest–northeast orientation from North America to England and, when it is intense, is responsible for most of the European winter storms. GA and Z streamfunction anomalies strongly project onto the negative and positive phases of the NAO, respectively, and are often referred to in the literature (e.g., Cassou 2008). Each panel of Fig. 1 indicates the frequency of occurrence of each WR. Regime Z is the most frequent WR during the 43 extended winters.
To analyze RWB frequencies of occurrence for the different stages of each WR, a WR index is computed similarly to a principal component of an EOF. A low-frequency geopotential anomaly φL is defined by first removing the climatological geopotential mean from the total geopotential φ and then by applying a 31-point Lanczos filter (Duchon 1979) with a cutoff period of 10 days to the geopotential anomaly . The difference between φ′ and φL is called the high-frequency geopotential anomaly φH. Therefore, . The WR index is defined as follows:
where t denotes a day belonging to the extended-winter period and NT is the total number of such days (i.e., 7837 days). PWR(t) is the projection of φL onto the WR low-frequency geopotential anomaly :
where (λ, ϕ) are respectively the longitude and the latitude, NH is the Northern Hemisphere, is computed by compositing φL for all the days belonging to a given WR, and is the wintertime climatological mean of the projection. Note that each index of a particular WR is defined for all the 7837 days and not only for the days belonging to this WR.
Figure 2 shows an example of index variations for the four WRs. As expected, the highest index at a given day corresponds in most cases to the WR identified by the cluster algorithm for this particular day. The correspondence is less true for transition days between weather regimes or for periods presenting a succession of different short-lived weather regimes (see, e.g., the period from 31 October to 15 November 1958). In other words, there exist periods when none of the four weather regimes is well established. The aim of the index is precisely to provide extra information on how well the day projects onto the pattern of the associated weather regime. This WR index allows us in particular to select periods of 11 consecutive days (from lag −6 to lag +4 days) where the index increases during the first 7 days (lag −6 to lag 0 days) and decreases during the last 5 days (lag 0 to lag +4 days) in a monotonic way. By construction, lag 0 corresponds to the day when the index is maximum during the selected period. An additional condition is that lag −1, lag 0, and lag +1 day belong to the same WR according to the cluster partitioning algorithm and their corresponding index is greater than 1.33. The selected periods are therefore periods for which a given WR dominates and exhibits monotonic growth and decay phases; 85 periods have been found for B, 88 for GA, 80 for AR, and 82 for Z. Examples of such periods are represented by the black arrows in Fig. 2.
b. Automatic wave-breaking detection method
In this subsection, the main principles of the RWB detection method are recalled but the reader is referred to Rivière (2009) or Rivière et al. (2010) for a detailed description of the algorithm. To discard subsynoptic scales, the grid step has been trebled by subsampling to obtain a field resolution of 4.5° × 4.5°. The algorithm is based on geometrical considerations without any condition in time and space scales. At each day, the algorithm detects systematically local reversals of a vorticity gradient on a given surface. The more appropriate vorticity gradient is the PV gradient on an isentropic surface because of the quasi-conservation of PV on such surfaces and this is the common field used in other WB detection algorithms (e.g., Martius et al. 2007; Strong and Magnusdottir 2008). In Rivière (2009), the algorithm was applied to the AV gradient on isobaric surfaces because of its easier computation. In the present paper, the algorithm is applied to the PV field on isentropic surfaces and leads to similar findings. Twenty-one contours from 0 to 10 PV units (PVU) with a step of 0.5 PVU are detected. The longitude and latitude of each point forming a contour of constant PV are obtained. Each contour is oriented from west to east, the longitude of the first and last points being 180°. A wave-breaking region is detected along the contour as a local segment oriented from east to west. If the latitude of the second point along the segment is lower (higher) than the latitude of the first point, then the wave breaking is of the anticyclonic (cyclonic) type (see Fig. C1 of Rivière 2009).
3. Frequency of occurrence of wave-breaking events in the upper-level troposphere
a. Wintertime climatology
Figure 3 shows the climatology of RWB frequencies averaged over four isentropic levels: 300, 315, 330, and 350 K. It is important to vertically average over various isentropic levels because the tropopause (i.e., the 2-PVU isosurface) intersects several isentropic levels at different latitudes and because AWB and CWB events do not occur at the same level. When potential temperature increases, AWB and CWB events become more and less frequent, respectively, as already noticed in Martius et al. (2007) and Rivière (2009). A dynamical explanation is provided in the latter study in terms of PV gradient asymmetries. Because of this difference in the vertical distribution of AWB and CWB, it is adequate to average the frequencies of occurrence over these four isentropic levels as is done in Fig. 3 as well as in the rest of the paper. Figure 3a represents the climatology of RWB frequencies over all extended winters. Both the Pacific and Atlantic sectors exhibit local maxima in the AWB and CWB density fields corresponding to the end of the two NH storm tracks. The most frequent AWBs extend from the eastern Atlantic Ocean to Asia with a maximum near the Iberian Peninsula. There is another maximum of AWB frequency over the western coast of North America. CWBs are most frequent over the North Pacific and the CWB density exhibits a secondary maximum south of Greenland. Similar findings have been obtained in Strong and Magnusdottir (2008). Frequencies found from the 200-hPa AV field are close to the frequencies found from 350- and 330-K PV fields (not shown). Therefore, since the AV field is easy to get from GCM outputs, it can be used to diagnose RWB and to interpret different climate scenarios as was done in Rivière et al. (2010).
b. Weather regimes
In this subsection, typical RWB frequencies for each regime are first presented and then more precisely analyzed during their growth and decay stage.
1) Characteristic RWB patterns for each WR
Each WR has a characteristic pattern of WB frequency as shown in Figs. 3b–e. These patterns are performed by averaging days obtained from the cluster algorithm. Regime B shows a maximum AWB frequency of occurrence over northern Europe and a maximum for CWB between the south of Greenland and Iceland. These two peaks of RWB densities correspond to negative zonal wind anomalies (cf. Figs. 1a and 3b) as expected from the RWB definition, which is a zone of local reversal of the PV gradient. Furthermore, in the longitudinal band 30°–40°W, where there is a maximum of occurrence of CWB events, the Atlantic jet is near 45°N (i.e., more to the south than usual), whereas in the longitudinal band 0°–20°E, where AWB events appear to be more frequent, the Atlantic jet is much more to the north than usual. This suggests that at a given longitude, the jet latitude is closely related to the nature of RWB. A higher- (lower-) latitude jet is closely related to more AWB (CWB) events than usual. This close relationship appears in other WR properties. It is a general property that can be interpreted in terms of spherical geometry and variations of the Coriolis parameter with latitude (Rivière 2009; Barnes et al. 2010). GA features a strong CWB frequency between Greenland and Canada whereas a weak peak of AWB frequency of occurrence appears over the United Kingdom (Fig. 3c). These CWB events maintain the jet to the south near 35°N over the whole Atlantic from the eastern coast of the United States to the Mediterranean region. GA is the regime that has the highest CWB frequencies of occurrence, which is consistent with the findings on the negative phase of the NAO (e.g., Benedict et al. 2004). AR is characterized by CWBs centered over Newfoundland and more AWBs in the eastern Atlantic away from the Iberian peninsula (Fig. 3d), as also shown by Santos et al. (2009). This suggests that the jet is shifted and maintained to the north in the central Atlantic by these AWBs. Regime Z is characterized by more frequent AWB events than CWB events at all longitudes (Fig. 3e). A peak of CWB frequency appears between Greenland and Iceland. The jet is southwest–northeast oriented from North America to England. These more frequent AWB events are consistent with the results on the positive phase of the NAO (e.g., Benedict et al. 2004). Note finally that for B, the longitudinal dipole of RWB densities with CWBs to the west and AWBs to the east is representative of the “Ω shape” of B mentioned by Altenhoff et al. (2008) and some other studies. In other words, the S- and inverse-S-shaped parts of the Ω structure correspond respectively to CWB and AWB features. It is also the case but with less importance for GA and AR.
2) Time-lagged composites of RWB frequencies
The purpose of the present section is to document further the time lags between RWB anomalies and the WR evolution as described by the index introduced in section 2a. The results are shown in Figs. 4–7, where light (heavy) shadings correspond to RWB events that are significantly less (more) frequent than in the climatology. For B (Fig. 4), the AWB frequencies (left panels) reach their maximum at lag 0 and statistically significant regions are much larger at lags −3 and 0 days than at lag +3 days. It suggests that these AWB events participate in the formation and maintenance of B rather than in its decay. In contrast, for CWB (right panels), its maximum frequency is stronger at lags 0 and +3 days than at lag −3 days and the statistically significant regions cover a larger area at lags 0 and +3 days. CWB events between Greenland and Iceland seem therefore to act in large part in the decay of B. For GA (Fig. 5), both AWB and CWB frequencies decrease with time from lag −3 to lag +3 days as well as the areas spanned by the statistically significant regions. It suggests a reinforcement of GA by CWB events since the latter RWB is the most frequent one. For AR (Fig. 6), both AWB and CWB frequencies reach their maximum at lag 0 when the index is also maximum but regions of statistically significance are decreasing with time. Note also that AWB frequency decreases around 40°–20°W from lag −3 to lag +3 days. Since these AWBs tend to reinforce the ridge over the central Atlantic, their frequency of occurrence disappears during the destruction of the ridge. For Z (Fig. 7), RWB frequencies have the same behavior as GA with maximum frequencies occurring at lag −3 days and then decreasing. Like GA, all RWB events seem to trigger Z. Note that in the climatology (Fig. 3a), there is a peak of AWB frequency over the western coast of the United States that is slightly higher for Z (not shown). These results for GA and Z, which can be assimilated to the two opposite NAO phases, are similar to those of recent papers on the NAO (e.g., Benedict et al. 2004; Woolings et al. 2008). To summarize, RWB frequencies usually reach the strongest values before or during the peak of the WR event, suggesting a reinforcement of the WR by RWB events. One exception concerns the CWB frequency pattern for B, which is more important during the decay stage of B. This result will be analyzed and confirmed in the next section.
4. Transitions between weather regimes
In this section, transitions between the different WRs are studied. A transition between a given regime a and a regime b is defined by a period of three consecutive days in the regime a followed by three consecutive days in the regime b. The day of transition T is the first day of the future regime b. The number of composited transition periods is displayed on each panel of Fig. 8. The preferred transition is from Z to B (96 periods), as Vautard (1990) has already noted, followed by the AR to Z transition (90 periods), then Z to AR (80 periods), and only after B to GA (76 periods). Using a Student’s t test, comparing the observed probability of a transition to the equiprobability, we found that these four preferred transitions are statistically significant at the 90% confidence level. This classification has been checked to remain valid in the case where the number of consecutive days for each regime is four or five. In what follows, two distinct transitions are more particularly studied: first, the transition from Z to B because it is the most frequent transition of the extended-winter period, and second, the transition from B to GA because it is the second preferred transition found by Vautard (1990), and also because the link between B and the negative phase of the NAO (i.e., GA in the present case) is the focus of numerous studies (see, e.g., Croci-Maspoli et al. 2007).
a. Low-frequency streamfunction budget
1) Tendency equation
The vorticity equation can be written as
where ζ is the relative vorticity, t is time, v is the horizontal wind, and f = 2Ω sinϕ is the Coriolis parameter; also, R is a residual term that contains dissipation, external forcing, vertical advection, and the twisting term. Then, ζ and v are decomposed into their time-mean (noted by an overbar), low-frequency, and high-frequency (defined with the superscripts L and H as in section 2a) components such that
Including these different components of v and ζ in Eq. (3) and applying the inverse of the Laplacian operator ∇−2 to the latter equation leads to the low-frequency streamfunction tendency equation
and where is the residual term and η is the absolute vorticity (η = ζ + f). In contrast with the previously mentioned studies, there is no decomposition of the wind into a divergent and a rotational part. The term ξ1 represents linear processes (i.e., the low-frequency eddy vorticity advection by the climatological wind, the climatological absolute vorticity advection by the low-frequency wind anomaly, and the associated two divergent terms); ξ2 represents the nonlinear interactions among the low-frequency transient eddies; ξ3 represents the nonlinear interactions among the high-frequency transient eddies; ξ4 represents interactions between the low- and high-frequency eddy components; and ξ5 represents interactions between the high-frequency transient eddies and the climatological flow. These two latter terms are expected to be negligible and should be null if the low-pass filter were a step function. As in Feldstein (2003), the time derivative is computed with the use of a centered time differencing with a one-day time step.
To know what terms contribute to the formation of the future regimes during the transitions, projections of each term ξn onto ψL field of the future regime are computed. The latter is obtained by compositing ψL for the last day of the transition period (i.e., at T + 2 days) and is denoted hereafter as . It has been checked that the pattern closely resembles the composite of ψL for all the days of the future regime (i.e., ). The projection can be expressed as
where n identifies the projection of each term of Eq. (6). The projections are made over the North Atlantic domain (20°–80°N, 80°W–20°E) and at the 300-hPa level where the terms ξn are maximum.
In Fig. 8 are displayed the projections for all 12 WR transitions from 8 days before the day of transition T to 12 days after. When a projection is positive, it means that the projected term tends to favor the future WR and when it is positive before T it tends to trigger the transition. When the projection is negative, the projected term tends to destroy the future WR. Projections of the tendency term [left-hand side of Eq. (6)] and of the sum of all the terms [first term of the right-hand side of Eq. (6)] are quite similar, meaning that computation errors and the residual term are small. Figure 8 shows that the WR transitions can have two distinct behaviors. The first one is characterized by a positive and high projection of the nonlinear terms ξ2 + ξ3 before T associated with a negative projection of the linear term ξ1 during the whole period (e.g., Figs. 8d,f,h,j). The second one is characterized by a positive projection of ξ1 from T − 8 to T − 4 days followed by a higher positive projection of ξ2 + ξ3 about T (e.g., Figs. 8b,c,e,k). Other transitions have this latter behavior but with a slightly positive projection of ξ1. In general, two transitions out of three toward the same future regime have the same behavior. As expected, for all the WR transitions, the projections of the two last terms ξ4 + ξ5 oscillate around zero and show no significant influence during the transitions. Once the future regime is formed (i.e., after T + 2 days), the linear term is the main term that tends to destroy this WR. These results are in good agreement with those of Feldstein (2003), who showed that the linear term is responsible for the NAO decay and more specifically its divergent part. The latter arises as a response to vorticity advection in order to maintain thermal wind balance and is mainly anticorrelated with the nonlinear transient eddy fluxes. By separating the effects of ξ2 and ξ3, it is shown later that ξ2 also participates in the decay of the future regime while ξ3 acts to maintain it.
1) Blocking to Greenland anticyclone transition
As can be seen in Fig. 8d, the projection of ξ1 is small or negative, meaning that this term does not participate in the transition. The projection of ξ2 + ξ3 increases from 0 at T − 8 days to its maximum at T − 2 days and then decreases. At T − 6 days, the nonlinear terms tend to maintain B (not shown) and at T − 2 days, the nonlinear terms destroy B for the benefit of the future regime GA (Fig. 9a). The streamfunction anomalies present a dipolar structure that projects well onto the GA pattern (Fig. 1b). It clearly shows that the nonlinear terms trigger the transition.
To clarify the link between the nonlinear terms and RWB, the ξ2 and ξ3 patterns are compared with the CWB and AWB frequencies of occurrence at T − 2 days in Figs. 9b–d. Note first that the ξ2 + ξ3 pattern in zonal wind (Fig. 9a) presents a south–north-oriented dipolar anomaly in the western and central North Atlantic characterized by an acceleration south of 45°N and a deceleration north of that latitude. This anomaly can be related to the RWB densities (Fig. 9d) at the same time lag and more precisely to the more frequent CWB events than usual for B near 55°N south of Greenland and Iceland (cf. Fig. 3b). Note that the longitude of these enhanced CWBs is about 35°W more to the east relative to the typical CWBs of GA, which are located near 60°W (cf. Figs. 9d and 5d). Concerning AWB events, they are less frequent over northern Europe than when B is well established. These RWB anomalies are in agreement with those observed in Figs. 4c,f for the decay of the B index. By separating the two components of the nonlinear interactions, it can be noticed that ξ2 has stronger anomalies than ξ3 (cf. Figs. 9b,c) but both participate in the dipolar zonal wind tendency. This destruction of the westerlies south of Greenland and Iceland leads to the separation between the western part of the Atlantic jet and its high-latitude part north of Scandinavia. Furthermore, the latter part becomes less intense because of the reduction of the AWB events over Scandinavia that cannot maintain the jet to the north. This explanation is confirmed by looking at ξ2 (Fig. 9b), whose anomalies are negative north of Scandinavia and form a tripole anomaly at longitude 40°E. This tripole is missing for ξ3 (Fig. 9c) but appears in the sum of the two terms (Fig. 9a). Therefore, the high- and low-frequency nonlinear terms reflect the RWB frequency anomalies and can be locally related to them.
One question arises from the previous result. Since high-frequency eddies are well known to maintain the low-frequency anomalies where they are (Lau 1988; Vautard and Legras 1988), do they more maintain the preceding regime or participate more in the formation of the future regime? To clarify this aspect, projections of ξ2 and ξ3 onto the streamfunction anomalies of the preceding WR are compared with the projections onto the future WR anomalies in Fig. 10. The streamfunction anomalies of the initial and future WRs are respectively defined as the composite of ψL at T − 3 and T + 2 days and closely correspond to and . The ξ3 projection on the initial WR is positive over almost the entire studied period whereas the ξ2 projection is positive before T and becomes negative after T. The ξ2 projection favors the initial regime B until T − 3 days and then favors the future regime GA (Fig. 10a). This nonlinear term tends to trigger the transition toward the future regime. The ξ3 projection maintains B until T − 1 days and then favors GA (Fig. 10a). Also, ξ2 more than ξ3 tends to trigger the transition toward the future regime GA while ξ3 tends to maintain more the present-day WR in place. These results are consistent with Feldstein (2003)’s findings on the NAO. However, projections using the low-frequency zonal wind rather than the low-frequency streamfunction are made in Fig. 10b and give a slightly different result. The ξ2 term reinforces more the initial regime B until T − 6 days and then favors the apparition of the future regime GA. The same behavior is observed for ξ3. In terms of zonal wind anomalies, the high-frequency eddy feedback mainly acts as a precursor of the transition. This discrepancy between the zonal wind and streamfunction results can be related to geographical differences. The strongest streamfunction anomalies of ξ3 are located east of 20°W in a region where the high positive values project well onto the B pattern (cf. Figs. 9c and 1a). In contrast, the largest zonal wind anomalies of ξ3 appear west of 20°W where they bear a strong resemblance to the GA anomalies (cf. Figs. 9c and 1b).
The nonlinear terms are both precursors of the B to GA transition whereas the linear term is not important. The low-frequency and high-frequency nonlinear terms can be related to a strong CWB event to the south of Greenland and Iceland that kicks off the regime transition.
2) Zonal to blocking transition
The zonal to blocking transition, which is the most frequent during the studied period, differs from the previous one since ξ1 participates in the transition. Indeed, the ξ1 projection (Fig. 8c) reaches its positive maximum at T − 7 days (i.e., the ξ1 pattern projects well onto the characteristic low-frequency streamfunction pattern of B at that time). But while the B regime appears, the projection decreases and becomes even negative when the B regime is well established at T + 2 days. At T − 4 days, the projection of ξ2 + ξ3 begins to prevail and is maximum at T + 1 days. At T − 2 days, the ξ2 + ξ3 pattern (Fig. 11a) bears a resemblance to (Fig. 1a). Zonal wind anomalies in black contours on Fig. 11a form a dipole in the western North Atlantic with acceleration near Newfoundland and deceleration north of it. This dipole can be linked to the CWB events that occur southwest of Greenland (Fig. 11d). There is also a tripole centered near 50°N in the western North Atlantic and Europe with a strong deceleration surrounded north and south by accelerations of the westerlies. These negative zonal wind tendencies clearly destroy Z in that particular region over western Europe and favor the apparition of B. Note that the RWB anomalies at T − 2 days are significantly weaker than those found for a usual day of the Z regime (compare the values in Figs. 11d and 3e). This lack of CWB and AWB events south and north of England may explain the deceleration zone in the nonlinear tendency terms and therefore the disappearance of Z. This interpretation is consistent with our findings on the decay of Z (Figs. 7c,f). As shown for the previous transition, ξ2, the term due to nonlinear interactions among low-frequency transient eddies has higher anomalies than ξ3, the term due to nonlinear interactions among high-frequency transient eddies (cf. Figs. 11b,c). It is also true for the low-frequency zonal wind tendency anomalies represented in black contours in Figs. 11b and 11c. Indeed, the tripole centered near England is much more present in Fig. 11b than in Fig. 11c whereas the dipole at 40°W is present in equal amounts in the two nonlinear terms.
Projections of ξ2 and ξ3 onto the anomalies of the initial regime Z and future regime B (not shown) reveal that ξ2 triggers the present transition while the role of ξ3 in favoring this transition is less clear. The projection of ξ3 onto the B pattern is more important once the B regime is established.
The linear term ξ1 (Fig. 12a) exhibits a strong positive anomaly between Iceland and Scandinavia at T − 6 days, which means that ξ1 favors the appearance of B at that time, as mentioned previously. Following its definition, ξ1 reflects the linear propagation of the low-frequency anomalies. Its pattern in Fig. 12a suggests a large-scale wave train propagating from the Caribbean Sea and Central America toward Iceland because of the southeast–northwest orientation of its isolines. This wave train follows one of the preferred curved path discussed in Hoskins and Ambrizzi (1993; see their Figs. 6a and 13). Many studies have shown evidence of connections between anomalously deep convection in the tropics and anomalous circulation in the extratropics through a Rossby wave response using idealized numerical simulations (Hoskins and Karoly 1981; Sardeshmukh and Hoskins 1988; Hoskins and Ambrizzi 1993; Ambrizzi and Hoskins 1997) or observed data analysis (Rasmusson and Mo 1993; Tyrrell et al. 1996; Renwick and Revell 1999). The so-called Rossby wave source (RWS) is a fruitful quantity to locate the forcing of such waves and is expressed as follows:
where the wind is divided into a divergent vχ and a rotational vψ part, that is, v = vχ + vψ. Note that ∇2ξ1 is the sum of the low-frequency linearized component of the RWS and the low-frequency linearized rotational component, which can be written as
Figure 12 presents ξ1, ∇−2RWS1, and ∇−2ROT1 at 200 hPa. Note that ∇−2RWS1 exhibits a negative minimum at 30°N, 40°W corresponding to a zone of large upper-level convergence. But in this particular region, ∇−2RWS1 and ∇−2ROT1 cancel each other. The dipolar anomalies of ξ1 near 60°W between 20° and 50°N are created by the ∇−2ROT1 anomalies. There is therefore a shift between RWS1 and the effective anomalies in ξ1 (cf. Figs. 12a,b).
Following the study of Qin and Robinson (1993), we decompose RWS1 into a tropical component, which is related to the advection of the vorticity by the divergent wind, , and an extratropical component linked to the stretching of the vorticity by the divergent wind, . Figure 13a shows the tropical RWS1 in the region of interest. It presents a minimum that extends between 60° and 20°W in longitude and between 10° and 20°N. It is located to the north of a divergence region of the upper-level low-frequency divergent wind (see Fig. 13c), which corresponds to a minimum of the low-frequency OLR (Fig. 13e). The tropical RWS1 part is therefore directly linked to a zone of enhanced convection. The extratropical RWS1 part (Fig. 13b) presents a maximum in a region of convergence near 30°N, 40°W (see the low-frequency divergent wind in Fig. 13c and the positive OLR anomaly in Fig. 13e). The RWS1 (Fig. 13c) presents a south–north-oriented dipolar anomaly at 40°W, which corresponds to the sum of the tropical and extratropical components of RWS1 and to the intensification of the Hadley cell in that region. Note that RWS1 and RWSL show very similar shading patterns (cf. Figs. 13c,d), suggesting that nonlinear interactions among the low- and high-frequency transient eddies are small relative to the linear interactions in the RWS term.
To summarize, there is a region of enhanced convection in the tropics (15°N) that is associated with upper-level divergence and a region of upper-level convergence more to the north (30°N). It corresponds to a strengthening of the local Hadley cell, leading to a large RWS in its downward branch. These results corroborate those of Tyrrell et al. (1996) concerning the Southern Hemisphere (see in particular the schematic RWS in their Fig. 14b).
To confirm the Rossby wave propagation, the horizontal wave-activity flux WH derived by Takaya and Nakamura (2001) for a zonally varying basic flow on the pressure coordinates [Eq. (C5) of their paper] is computed for ψL at lag T − 6 days and plotted in Fig. 14. The phase velocity is taken as null, as in Takaya and Nakamura (2001) in their section 3b, since it concerns the low-frequency or quasi-stationary waves. Similarly to the previous study, the basic state for the quasi-stationary eddies is the wintertime climatological mean defined in section 2a. On the Fig. 14, the low-frequency streamfunction at 200 hPa (colored contours) exhibits a dipolar anomaly typical of the zonal regime (Fig. 1d), which is logical since we are still in this regime at that time. There is also a large WH divergence region (gray shading) covering the midlatitude Atlantic whose western part extends to the subtropics around 60°W in the same region as the peaks of ξ1 (Fig. 12a) and to the northeast of those of the RWS (Fig. 12b). North of that region, the WH vector, which is nearly orthogonal to the ψL isolines, is oriented northward from the high to the low anomaly of the ψL dipole. Then, WH presents a curved path in the northeastern Atlantic and converges mainly over Africa and weakly over Asia. The WH vector and its divergence therefore confirm the Rossby wave train originating from the western subtropical Atlantic toward Scandinavia.
The transition is first triggered by a wave train coming from the subtropical western Atlantic, possibly driven by a strengthening of the Hadley cell in the Atlantic and propagating eastward following a preferred curved path over North Atlantic and Asia. The second precursor is linked to RWB; it occurs during the transition itself and can be mainly related to a lack of AWB and CWB events in the eastern Atlantic.
The existence of a wave train triggering the blocking is well supported by making regressions of the low-frequency streamfunction, WH, and OLR fields onto the blocking index (Fig. 15). The time-lagged regressions (Fig. 15) clearly show a wave train coming from 30°N, 60°W propagating toward Scandinavia making a curved path and then toward Asia. This wave train is in agreement with that found by Nakamura et al. (1997). The anomalous high at 30°N, 60°W is over a divergence zone of WH, which emphasizes that this region is a wave train source. Furthermore, this anomalous high is close to a region of upper-level convergence (see the positive OLR anomaly), which itself is closely linked to an enhancement of the convection in the tropical Atlantic (see the negative OLR anomaly to the east of South America).
A new view on weather regimes and their transitions has been presented in the present study in terms of Rossby wave breakings. Each Atlantic weather regime has a characteristic pattern of RWB frequency. CWB events are found to be the most frequent in the North Atlantic during GA while AWB events occur more often during AR and Z. These results support the NAO findings discussed in the introduction but provide a more detailed picture of the link between RWB and the low-frequency atmospheric variability in the North Atlantic. A general rule on the relation between the jet latitude and the nature of the breaking has been observed. At a given longitude, the higher (lower) the latitude of the jet, the more frequent is AWB (CWB) relative to CWB (AWB). For example, for B, the jet is quite far to the south in the western Atlantic and is associated with more CWB events than usual, whereas in the eastern Atlantic the jet is centered at a very high latitude and more AWB events occur at these longitudes. This rule is consistent with the positive eddy feedback acting on the latitudinal fluctuations of the jet explained in Rivière (2009): a more northward (southward) jet renders AWB more (less) probable and CWB less (more) probable, which in turn helps to maintain the jet more to the north (south).
Time-lagged composites show that the different RWB events characterizing each regime occur more often during the formation of the regime rather than during its decay. It suggests a reinforcement of the weather regime by RWB. An exception is blocking, which seems to be destroyed by an increase of CWB events south of Greenland.
Then, the dynamical processes during transitions between weather regimes are studied using the low-frequency streamfunction budget. Two distinct precursors for the transitions have been identified: one is related to linear propagation of low-frequency transient eddies and the other to the two nonlinear interactions among the high- and low-frequency transient eddies. The former is not systematic but may appear a few days before the transition whereas the latter, which is shown to be directly related to RWB, mainly acts during the transition itself. The B to GA transition is triggered by RWB events alone. A strong CWB event south of Greenland favors the destruction of the Scandinavian high as well as a lack of AWB events more downstream. The Z to B transition presents evidences of two precursors. First, the linear tendency term, linked to the propagation of the low-frequency anomalies, projects well onto the future regime several days before its occurrence. A low-frequency wave train, initiated in the subtropical western Atlantic, has been identified as the key phenomenon to produce this tendency. Second, a lack of AWB and CWB events over western Europe does not allow the continued maintenance of the westerlies in these regions and favors the appearance of B.
By separating the effects of the high- and low-frequency transient eddy fluxes, it is shown that both correspond to a signature of RWB. However, they do not act at the same time and not necessarily at the same location either. During weather regime transitions, the nonlinear interactions among the low-frequency transient eddies favor the appearance of the future regime a few days before the transition while the nonlinear interactions among the high-frequency transient eddies occur mainly during the transition. Our analysis revealed that high-frequency eddy fluxes can maintain the initial regime in some regions while participating in the appearance of the future regime in other regions. Storm-track feedback onto the low-frequency flow is classically diagnosed from the high-frequency eddy fluxes (e.g., Lau 1988) but the present study suggests that synoptic RWB is related to both the high- and low-frequency eddy fluxes. Therefore, it is not obvious that the synoptic eddy feedback can be reduced to the high-frequency part and that it tends simply to maintain the low-frequency anomalies as they are. Future studies should investigate the ability of the storm tracks in the modification of low-frequency anomalies. For example, since the destruction of the blocking is kicked off by CWB events, it could be of interest to make the link with surface cyclogenesis south of Greenland.
The paper has benefited from discussions with Philippe Arbogast, Olivia Martius, Guillaume Lapeyre, and Alain Joly and from constructive comments of three anonymous reviewers. The authors thank Marie Boisserie for her English corrections in an early version of the manuscript. This work has been partially funded by a CNRS/INSU/LEFE/IDAO grant and an ANR grant (ANR-06-JCJC-133-01).