We investigate the relationship between large-scale atmospheric flow and the evolution of the most extreme windstorms affecting Western Continental Europe. The 25 most destructive Western Continental European wind storms are selected from a 43-year climatology. 22 of these storms are grouped as having a similar trajectory and evolution. We show that these storms typically occur during particularly strong and persistent positive NAO anomalies which peak approximately 2 days before the storms' peak intensity; the NAO pattern then shifts eastward to a position over the European continent when the storms strike Europe. A temporal composite of potential temperature on the 2-PVU surface suggests that this NAO shift is the result of simultaneous cyclonic and anticyclonic wave breaking penetrating further to the east than during a typical high-NAO event. This creates an extremely intense, zonally-orientated jet over the North Atlantic whose baroclinicity favours explosive intensification of storms while steering them into Western Continental Europe.
 Extreme windstorms are a major natural hazard and among the leading sources of insured losses in Europe [Berz, 2005]. Societal awareness and scientific interest in these storms increased sharply after some unusually damaging events of the 1990s, the most destructive of which—‘Daria’ (January 1990) and ‘Lothar’ (December 1999)—each caused around 100 fatalities and over US$8 billion in inflation-adjusted insured losses, comparable to those caused by a typical US hurricane event.
 Detailed case studies have been dedicated to several of these storms individually, particularly ‘Lothar’ [Ulbrich et al., 2001; Wernli et al., 2002; Rivière et al., 2010]. Here we take a comparative approach, seeking to identify features common to destructive European storms. To keep the problem manageable, we confine our attention to storms affecting a relatively compact region encompassing France, Germany, Benelux and Denmark which we refer to as Western Continental Europe (WCE). This is a densely-populated, intensely industrialised area where most of the insured losses of the 1990s storms were incurred.
 Rather than focus on the detailed evolution of the storms themselves, we concentrate on the structure of the large-scale flow in which the storms are embedded. In doing so, we take an important cue from the previously cited case-studies of this storm which show that the evolution of ‘Lothar’ was strongly affected by the presence of a pre-existing large-scale low-pressure system located to the north of the storm itself. This deep, large-scale depression played a triple role: it helped steer ‘Lothar’ into WCE; it generated an intense baroclinic jet streak on its southern flank which helped explosively intensify ‘Lothar’ just before it made landfall on Europe; and it created a strong background surface pressure gradient across Europe on which ‘Lothar’ superimposed, leading to extremely strong surface winds.
 Our aim here is to understand whether such large-scale conditions are also associated with other destructive storms affecting WCE. We begin inSection 2 by defining an objective index of storm destructiveness over WCE. Direct inspection of the top 25 storms' synoptic evolution, discussed in Section 3, shows that the large-scale flow in 22 of them is indeed qualitatively similar to that of ‘Lothar’ described above.
 We then further examine the characteristics of the large-scale circulation associated with these 22 extreme windstorms. Large-scale variability in the Euro-Atlantic sector is dominated by the North Atlantic Oscillation (NAO), and extreme European storms are known to occur more frequently during moderately positive phases of the NAO [Pinto et al., 2009; Donat et al., 2010]; Section 4 examines the relation of the extreme storms with the NAO. Furthermore, several recent papers have shown that the NAO can be dynamically interpreted in terms of Rossby wave breaking [Benedict et al., 2004; Rivière and Orlanski, 2007; Woollings et al., 2008; Kunz et al., 2009]; we follow this approach in Section 5to provide a dynamical perspective on the evolution of the large-scale conditions associated with these destructive storms. Finally,Section 6 summarizes our conclusions.
2. Extreme North Atlantic and European Cyclones
 The present study is based on the European Centre for Medium Range Weather Forecasts 40-year reanalysis product (ERA40) [Uppala et al., 2005], and specifically on the following 6-hourly fields: mean sea level pressure (MSLP), potential temperature on the 2-PVU surface [1 potential vorticity unit (PVU) ≡ 10−6 m2 s−1 K Kg−1], and the 10 m and 250 hPa wind fields. Applying a recently-developed cyclone identification and tracking algorithm [Hanley and Caballero, 2011] to the 6-hourly MSLP field at 1.125°resolution for all winters (October-March) over the period 1958–2001, we compute a climatology of cyclone tracks over the Euro-Atlantic sector. It should be noted that while previous studies [e.g.,Ulbrich et al., 2009] have shown that the results of automated cyclone tracking may depend strongly on the methodology used, recent work within the IMILAST project [Neu et al., 2012] has shown that tracks of intense cyclones are typically captured in a similar way by different tracking methods.
 To extract the most destructive storms from this climatology, we compute a destructiveness index for each track as follows. We define WCE as the land grid points enclosed by the quadrilateral with vertices (48.7°N, 7.0°W), (57.5°N, 15.0°E), (48.0°N, 19.125°E), (40.65°N, 0.0°E), shown in Figure 1b. A cyclone centre at a given time step is said to affect WCE if a circle of 1000 km radius around its centre encloses a land grid-point within the perimeter. For those cyclones which meet this criterion, we compute a destructiveness index followingKlawa and Ulbrich  and Pinto et al.  by summing the significant power dissipation over the affected land points at which v > v98, where v is the surface wind speed at a given point and time and v98is the climatological 98% percentile wind speed at that point. The 98% percentile is calculated based on all time-steps from October-March over the period 1958–2001. The use of a 98% percentile threshold serves to control for local adaptation to wind climate, focusing on the winds most likely to cause damage [Klawa and Ulbrich, 2003]. The storms are then ranked based on the peak value of the destructiveness index achieved over the cyclone's lifetime.
 A plot of the ranked destructiveness index (Figure 1a) shows a roughly exponential drop in destructiveness with rank, similar to the drop observed when ranking monetary loses over Germany for the period 1997–2007 [Donat et al. 2011, Figure 5a]. The top 25 storms, listed in Table S1 in Text S1 in the auxiliary material, dominate and include the well-known historical storms of the period, lending confidence that the peak destructiveness index does indeed select the storms with the greatest socioeconomic impact on WCE.
 Individual tracks of the top 25 storms are shown in Figure 1b. The tracks are overlayed on the climatological track density of extreme North Atlantic cyclones, defined as those cyclones whose tracks fall partly or entirely in the Euro-Atlantic sector (70°W-40°E, 30°N–75°N) and which belong to the top 10% intensity percentile as measured by lifetime-peak MSLP gradient (measured as the average MSLP difference between the cyclone's 1000 km circle radius and its central pressure). Cyclone track density is computed by counting the total number of extreme cyclone tracks falling each winter within a circle on the sphere centered at the given grid point with radius 840 km (equivalent to 7.5°latitude) and averaging over all winters in the data set. The pattern of extreme cyclone track density in this figure is in close agreement with similar density plots shown in previous work [e.g.,Pinto et al., 2009, Figure 4a]. Note that the 25 most destructive WCE storms all have tracks shifted well south of the peak storm density region, suggesting that these storms affecting WCE are unusual and distinct from typical intense North Atlantic storms.
3. Synoptic Evolution of the Top 25 Storms
 The evolution of both surface and upper-level fields over the 5 days preceding peak destructiveness is shown for all top 25 storms in theauxiliary material (Figures S1a–S1y). To highlight the features of interest to the present study, we examine here the most destructive storm according to our index, ‘Daria’, which struck WCE on the 25th January 1990. The top row in Figure 2 shows the surface pressure field. Around 5 days prior to peak destructiveness (Figure 2a), a deep low is present south of Greenland, which intensifies while migrating eastwards, reaching a position to the north of the British Isles at 2.25 days before peak destructiveness (Figure 2b). It is at this time that our tracking algorithm identifies the genesis of ‘Daria’ just off the coast of Newfoundland. The storm subsequently travels north-eastward, becoming embedded in the strong westerly flow on the southward flank of the pre-existing large-scale depression over the North Atlantic (Figure 2c). At peak destructiveness (Figure 2d), ‘Daria’ superimposes its SLP gradient on the already strong background gradient over WCE, leading to the extremely destructive surface winds there.
 The corresponding evolution at upper levels is illustrated in the bottom row of Figure 2. The shading shows potential temperature on the 2-PVU potential vorticity surface (θPV2). This field is a good approximation to the tropopause elevation for air originating poleward of 25°N; it conveniently summarizes most of the important upper-level dynamical structure in a single picture [Thorncroft et al., 1993]. The large-scale surface low over the North Atlantic which pre-exists ‘Daria’ and strongly influences its evolution is associated at upper levels with a large-scale intrusion of high-latitude, lowθPV2air which is turned cyclonically by the ambient flow, leading to a local reversal of the north-southθPV2gradient—the classic signature of cyclonic Rossby-wave breaking [Thorncroft et al., 1993]. Simultaneously, an intrusion of low-latitude, highθPV2air collides with the cold-air intrusion, turning anticyclonically and breaking over the subtropical North Atlantic. We emphasize that these concurrent cyclonic and anti-cyclonic wave breaking eventsprecedethe birth of ‘Daria’ and act to create favourable conditions for ‘Daria's destructiveness: specifically, the confluence of warm and cold air over the midlatitude North Atlantic creates a region of high baroclinicity associated with a very strong zonally-oriented jet, which acts both to explosively intensify ‘Daria’ as it passes through to the left-exit region [Rivière and Joly, 2006], and to steer ‘Daria’ into continental Europe instead of taking the more common route towards Iceland.
 Detailed inspection of the evolution of the top 25 storms indicates that 22 can be subjectively grouped as having an evolution qualitatively similar to that of ‘Daria’, including ‘Lothar’ (Figure S1g in the auxiliary material). For the rest of the paper we will refer to these 22 storms as ‘embedded storms’. The 3 excluded storms (Figures S1n, S1u, and S1y in the auxiliary material) all lack at least one of these ingredients. The tracks of the 22 ‘embedded storms’, plotted in red in Figure 1b, resemble the ‘westerly flow’ cluster objectively identified by Donat et al. .
 Preliminary work also suggests that ‘embedded storms’ are disproportionately represented amongst the most intense storms. Examining the evolution of the top 10, 25 and 50 storms, we find ‘embedded storms’ fractions of 10/10, 22/25 and 33/50 respectively.
4. Surface Pressure Composites and Relation With the NAO
 We compute a daily NAO index by projecting monthly SLP EOFs onto daily SLP fields, following the method described in Pinto et al. . A composite of this daily NAO index over the ‘embedded storms’ centred on the moment of maximum destructiveness (Figure S2 in the auxiliary material), shows that the NAO peaks approximately 2 days before the time of maximum destructiveness; thus, the storms typically make landfall when the NAO index is waning and has moderate amplitude, in agreement with previous work [Pinto et al., 2009; Donat et al., 2010]. Figure S2also shows that the ‘embedded storms’ typically occur during a period of anomalously persistent high NAO, with a much longer build-up to maximum NAO index and a more gradual decay compared to a climatologically high NAO event.
Figure 3 offers a spatially resolved comparison of surface pressure evolution composited over the ‘embedded storms’ and over climatological high positive NAO events. The most notable feature of Figures 3a–3c is the striking resemblance to the evolution of ‘Daria’ shown in Figures 2a–2d, despite the fact that we are compositing over 22 storms. A high-positive NAO pattern with a very deep, large-scale low situated near Iceland and an elevated high situated over the Azores, is prevalent prior to the birth of these storms (Figure 3a). The large-scale low then deepens further and migrates eastward, while to the south the high also intensifies but remains approximately stationary (Figure 3b). This is the point of peak daily NAO, where the storms have just begun to deepen. The large-scale low then moves further east to Scandinavia while the high shifts closer to the continent (Figure 3c); while this is happening, the storms make landfall causing high winds over WCE.
 For comparison, Figures 3d–3fshow the evolution of a climatological high-positive NAO event (defined as an excursion of the NAO index above a threshold value of 0.7σ – see Figure S2 in the auxiliary material). These figures show a more rapid intensification and decay of the pressure gradient in comparison to Figures 3a–3c. Notably, there is no sign of the pronounced eastward shift seen in the ‘embedded storms’ composite (Figures 3a–3c), in which the large-scale low essentially maintains its depth while moving east; rather, inFigures 3d–3f, it decays rapidly in place. This eastward migration of the large-scale low in the days prior to the destructive storm event agrees with previous statistical findings [Leckebusch et al., 2008].
5. Upper-Level Composites
 For a more dynamically oriented perspective on the evolution of the large-scale atmospheric conditions that favour the development of extreme WCE wind storms, we consider composites ofθPV2. This field has been effectively used in previous work to characterize the synoptic evolution of the NAO [Benedict et al., 2004; Woollings et al., 2008].
Figure 4 shows composites of θPV2 over the 22 ‘embedded storms’ and over the climatological high NAO events. As is clear from Figures 4d–4f, the latter are primarily driven by anticyclonic wave breaking, identified by a reversal of the north-southθPV2 gradient over southern Europe and the subtropical North Atlantic, which can be seen most clearly in Figure 4e; this is in close agreement with previous work [Benedict et al. 2004, Figure 3]. Note that there is clear evidence of Rossby wave breaking in the high NAO composite only around the time of peak NAO index (Figure 4e), with no evidence of wave breaking 2 days before (Figure 4d) or 2 days afterward (Figure 4f).
 In contrast, the ‘embedded storms’ composite (Figures 4a–4c) shows both strong cyclonic wave breaking to the north and anticyclonic wave breaking to the south occurring simultaneously during all stages of development. The areas where the storm composite differs at the 99% statistical significance level (using a Monte-Carlo sampling procedure) from the climatological high NAO composite are outlined in white inFigures 4a–4c, and highlight precisely the regions of wave breaking. Over the northern North Atlantic, cyclonic wave breaking is associated with an extrusion of low θPV2air originating from polar regions west of Greenland and reaching the British Isles and Scandinavia; this is roughly mirrored to the south by an anticyclonically-turning extrusion of highθPV2air originating in the subtropical eastern North Atlantic and extending all the way to the western Mediterranean. Overall, the pattern looks like a somewhat smoothed version of the upper-level evolution associated with ‘Daria’, shown inFigure 2.
 The collision of the cold air intrusion associated with cyclonic wave breaking with the warm air intrusion associated with anticyclonic wave breaking produces a region of very high upper tropospheric baroclinicity situated to the west of WCE. This area of high baroclinicity almost certainly contributes to the explosive intensification of our selected storms just as they are about to make landfall—a strong and focused upper-level jet has been previously shown to play an important role in the intensification process of such storms [Rivière and Joly, 2006].
 We have studied the evolution of the large-scale flow field typically associated with a selected subset of the most destructive windstorms affecting WCE in the period 1958–2001. This subset contains most of the major named European storms of the 1990's as well as many storms from previous decades. These storms typically occur during persistent moderately-positive NAO events which peak about 2 days before maximum destructiveness. The evolution of MSLP during these events shows an eastward shift of the NAO dipole pattern, leaving a deep, large-scale low positioned over Scandinavia at the time of maximum storm impact. At upper levels, this shift is associated with exceptionally persistent, simultaneous cyclonic wave breaking to the north and anticyclonic wave breaking to the south penetrating much further eastward compared to climatological high NAO events. These simultaneous wave breaking events serve to intensify and elongate an extremely strong zonally-oriented upper-level jet which helps to explosively amplify pre-existing depressions [Rivière and Joly, 2006] and then steers them into WCE; this is in contrast to climatological high NAO events where the jet steers cyclones on a path close to Iceland.
 A remarkable feature of the extreme windstorms of 1990 and 1999 was their tendency to occur in clusters, with extremes occurring in rapid succession separated by 2–3 days. Recent statistical analysis suggest this tendency for seriality is a generic feature of European storms [Mailier et al., 2006; Vitolo et al., 2009]. Our results may help explain this tendency: the exceptionally strong wave breaking events identified here create large-scale conditions favourable for extreme windstorms which persist for longer-than-synoptic timescales. Exploring the dynamical mechanisms underlying the observed wave breaking behaviour also remains a goal for future research.
 This research was supported by the Bert Bolin Centre for Climate Research at Stockholm University. The authors would like to thank two anonymous reviewers for their comments which helped to improve the manuscript.
 The Editor thanks Joaquim Pinto and an anonymous reviewer for their assistance in evaluating this paper.