Past Evolution and Recent Changes in Western Europe Large-scale Circulation

Detecting trends in regional large-scale circulation (LSC) is an important challenge as LSC is a key driver of local weather conditions. In this work, we investigate the past evolution of Western Europe LSC based on the 500 hPa geopotential height fields from 20CRv2c (1851-2010), ERA20C (1900-2010) and ERA5 (1950-2010) reanalyses. We focus on the evolution of large-scale circulation characteristics using three atmospheric descriptors that are based on analogy – characterizing the geopotential shape stationarity and how well a geopotential shape is reproduced in the climatology – together with 5 a non-analogy descriptor accounting for the intensity of the centers of action. These descriptors were shown relevant to study precipitation extremes and variability in the Northwestern Alps in previous studies. Even though LSC characteristics and trends are consistent among the three reanalyses after 1950, we find major differences between 20CRv2c and ERA20C from 1900 to 1950 in accordance with previous studies. Notably, ERA20C produces flatter geopotential shapes in the beginning of the 20th century and shows a reinforcement of the meridional pressure gradient that is not observed in 20CRv2c. We then focus 10 on the recent changes in LSC from 1950 to 2019 using ERA5. We combine the four atmospheric descriptors with an existing weather pattern classification to study the recent changes in the main atmospheric influences over France and Western Europe (Atlantic, Mediterranean, Northeast, Anticyclonic). We show that little changes are found in Northeast circulations. However, we show that Atlantic circulations (zonal flows) tend to become more similar to known Atlantic circulations in winter. Anticyclonic conditions tend to become more stationary in summer – a change that can potentially affect summer heatwaves. 15 Furthermore, Mediterranean circulations tend to become more stationary, more similar to known Mediterranean circulations and associated with stronger centers of action in autumn, which could have implications for autumn extreme precipitation in the Mediterranean-influenced regions of the Southwestern Alps.

a non-analogy descriptor accounting for the intensity of the centers of action. These descriptors were shown relevant to study precipitation extremes and variability in the Northwestern Alps in previous studies. Even though LSC characteristics and trends are consistent among the three reanalyses after 1950, we find major differences between 20CRv2c and ERA20C from 1900 to 1950 in accordance with previous studies. Notably, ERA20C produces flatter geopotential shapes in the beginning of the 20th century and shows a reinforcement of the meridional pressure gradient that is not observed in 20CRv2c. We then focus 10 on the recent changes in LSC from 1950 to 2019 using ERA5. We combine the four atmospheric descriptors with an existing weather pattern classification to study the recent changes in the main atmospheric influences over France and Western Europe (Atlantic, Mediterranean, Northeast, Anticyclonic). We show that little changes are found in Northeast circulations. However, we show that Atlantic circulations (zonal flows) tend to become more similar to known Atlantic circulations in winter. Anticyclonic conditions tend to become more stationary in summer -a change that can potentially affect summer heatwaves. Vautard and Yiou (2009) show that circulation changes poorly explain changes in surface climate in summer but they well control changes in winter, although this control seems to be weakening in the last 30 years. At a more local scale, decreasing autumn and winter precipitation from 1951 to 2000 in Southern France appears to be explained by a decrease in the occurrence of weather types driving precipitation over the region, while the increasing trend in Northeastern France is only partly explained by changes in weather type occurrence (Boé and Terray, 2008). In the British Isles, the decreasing trend in summer precipitation since 1850 appears to be related to more positive phases of the summer NAO pattern (Fig. 6b of Folland et al., 2009). Focusing on extremes, Horton et al. (2015) show that 44 % of the increase in summer hot extremes over Europe can be explained by an increase in the occurrence of blocking high pressure systems over Central Europe for the period 1979period -2013period . Iannuccilli et al. (2021 show that part of the increase in extreme precipitation over Central Italy in winter and spring can be explained by changes in occurrence of the circulation types. In the Southwestern Alps, the increasing extreme precipitation from 1958 70 to 2017 in autumn appear to be associated with a strengthening of the Mediterranean influence on extremes (Blanchet et al., 2021a, b).
The aforementioned studies dealing with trends in regional LSC mainly employ weather patterns classifications. However, part of the trend may also lie in changes in the characteristics within a given weather pattern -whether dynamical or thermodynamical -, as discussed in Boé and Terray (2008) and Iannuccilli et al. (2021). In this paper, we propose a contribution 75 to fill this gap by employing four atmospheric descriptors that characterize the 500 hPa geopotential height field and that allow the consideration of dynamical trends within the main atmospheric influences. These descriptors were introduced in previous works, and they were shown to explain precipitation variability and extremes in the Northern French Alps (Blanc et al., 2021b;Blanchet and Creutin, 2020). They characterize i) the stationarity of a flow direction (celerity), ii) how well a flow direction is reproduced in the climatology (singularity, relative singularity), and iii) the intensity of the low and the high 80 pressure systems (Maximum Pressure Difference). The four atmospheric descriptors are first employed to study the long-term evolution of Western Europe LSC from 1851 to 2010 using different reanalyses products. They are then combined with an existing weather pattern classification over the period 1950-2019 to address recent changes in the characteristics of the main atmospheric influences affecting Western Europe. Finally, the implications of these changes for local weather conditions are discussed.

Data
We use daily 500 hPa geopotential height fields over a 32°× 16°region to represent Western Europe LSC (rectangle in Fig. 1a). The 500 hPa geopotential ranges from 4, 800 m to 6, 100 m, giving information about the location and the intensity of the low and the high pressure systems in the middle of the troposphere. We extracted the 500 hPa geopotential height from three different reanalyses covering different periods (Table 1). 90 We use the 20CRv2c reanalysis from NOAA-CIRES (Compo et al., 2011). 20CRv2c provides information about the state of the atmosphere since 1851 with an horizontal resolution of 2°. It only assimilates surface pressure observations using an ensemble Kalman Filter, and it is composed of 56 individual members that are equiprobable as well as a mean member. Sea arrows represent the wind anomalies at 500 hPa. In this study, the 500 hPa geopotential height is considered over the Western Europe region, represented by the black rectangle. (b) Schematic illustration of the atmospheric descriptors based on analogy. Each map represents the 500 hPa geopotential height field over Western Europe for a given day. Following days are represented on the same trajectory, but all trajectories are part of a single historical trajectory. The distance between each map represents here the difference in geopotential shapes/flow direction between individual days, using the Teweles-Wobus score. The celerity is the distance between a day D and the day before D-1 (dotted arrow).
The singularity is the mean distance between a day D and its closest analogs (solid arrows; three analog days for illustration, but 111 analog days in our study). The relative singularity is the singularity normalized by the distance to the farthest analog (dashed arrow; third analog day for illustration, but 111th analog day in our study). The day D considered here is 12 December 1978. surface temperature and sea-ice distributions are used as boundary conditions. In this article, we use two individual members -arbitrary member 1 and member 2 -as well as the mean member to derive whether significant differences are observed 95 between the individual and the mean members.
The twentieth century reanalysis ERA20C from ECMWF is also used (Poli et al., 2016). ERA20C provides a higher spatial resolution than 20CRv2c with a 1.125°grid, but it ranges over a shorter period . In addition to surface pressure, ERA20C also assimilates marine wind observations using a 4D-Var assimilation technique. ERA20C is single-member. It is forced by sea surface temperature, sea-ice cover, atmospheric composition and solar forcing.

100
Finally, we use the ERA5 reanalysis which is the most recent reanalysis product of ECMWF (Hersbach et al., 2020). ERA5 ranges over a more recent period (1950-today) but it provides atmospheric variables with a high spatial resolution of 0.25°.
ERA5 assimilates surface observations, upper-air observations and satellite observations (referred as "full-input") using a 4D-Var scheme. ERA5 relies on the radiative forcing of CMIP5 including total solar irradiance, ozone, greenhouse gases and some aerosols including stratospheric sulfate aerosols. It takes sea-surface temperature and sea-ice cover as boundary conditions. A 105 10-member ensemble with reduced resolution is available, but we use the high resolution realisation of ERA5 which is referred to as the main reanalysis.

Method
Studying changes in LSC is carried out using atmospheric descriptors characterizing daily 500 hPa geopotential height fields.
An existing weather pattern classification is also employed to consider changes in LSC characteristics that are specific to the 110 main atmospheric influences.

Main atmospheric influences
We use the weather pattern classification of Garavaglia et al. (2010) from 1950 to 2019 to derive the main atmospheric influences affecting Western Europe. This classification into eight weather patterns was established to link daily rainfall field shapes over Southern France with synoptic situations. We aggregate the eight weather patterns into four atmospheric influences ac-115 cording to the origin of the air flow reaching the French Alps, as previously done in Blanc et al. (2021b) and Blanchet et al. in Blanchet et al., 2021b).

Atmospheric descriptors
Changes in LSC characteristics are investigated using four atmospheric descriptors introduced in previous works Blanchet et al., 2018;Blanchet and Creutin, 2020). These descriptors are based on daily 500 hPa geopotential height fields over Western Europe (rectangle in Fig. 1a). Three descriptors are based on analogy, that is on the comparison of daily 125 geopotential height fields between each other. The analogy is based on the Teweles-Wobus score (Teweles and Wobus, 1954), which measures the similarity in shape between geopotential height fields using North-South and West-East gradients. As the geopotential shape defines the flow direction, we can interpret the analogy in Teweles-Wobus score as the analogy in flow direction. The Teweles-Wobus score between days t k and t k is given by: where Adj ranges the set of adjacent grid points in horizontal and vertical directions in the region of study. A T W S k,k of 130 0 means that day k and k feature strictly identical flow directions. A T W S k,k of 1 means that day k and k feature strictly opposite flow directions. In practice, the TWS obtained in this study range between 0.04 and 0.88. Figure 1b provides a schematic illustration of the three descriptors based on analogy in flow direction. The first descriptor is the celerity that is understood as the celerity of deformation of the geopotential. It measures the stationarity in flow direction between two consecutive days. It is defined for day t k as the TWS between day t k and day t k−1 (dotted arrow in Fig. 1b): The lower the celerity, the more stationary the flow direction between two consecutive days.
The two other descriptors based on analogy are the singularity and relative singularity. They measure the way a flow direction is reproduced in the climatology. The singularity and relative singularity rely on the comparison of a given flow direction with its Q closest flow directions in the climatology, referred as its analogs. The singularity of day t k is defined as the mean TWS 140 between day t k and its Q closest analog days (mean of solid arrows in Fig. 1b): where A k range the Q closest analogs of day t k . A flow direction featuring a low singularity means that close flow directions are found in the climatology. The singularity cannot be directly related to the frequency of occurrence of a given flow direction since a geopotential shape is never perfectly reproduced (T W S k,k > 0). Very low singularities even appear to be rare in the 145 climatology, which means that the atmosphere spends much time exploring quite unseen patterns than very closely coming back to an already seen pattern Blanchet and Creutin, 2020).
The relative singularity of day t k is defined as the singularity normalized by the Teweles-Wobus score with the Qth closest analog day (mean of solid arrows normalized by the dashed arrow in Fig. 1b): . (4)

150
The relative singularity measures the similarity of a given flow direction to its very close analogs in comparison to the farther analog. It measures in a way the degree of clustering of the closest flow directions. The relative singularity is closely related to the local dimension of Faranda et al. (2017a) although they employ an Euclidean distance instead of TWS. A flow direction featuring a low singularity and relative singularity is said to be almost similarly reproduced in the climatology, as close flow directions are found in the climatology (low singularity) but the closest flow directions tend to be even more resembling than 155 usually (low relative singularity). Blanchet and Creutin (2020) showed that the singularity and relative singularity are not very sensible to the exact number of days selected as analog in Eq.
(3) and Eq. (4), and that the selection of the closest 0.5 % days was a reasonable choice to link LSC characteristics with 3-day precipitation in the Northern French Alps. The period 1950-2010 is considered for the search of analog, as it is the common period of 20CRv2c, ERA20C and ERA5. Therefore, we use in the rest of this study Q = 111 160 days.
The celerity, singularity and relative singularity are based on the TWS, which only focuses on geopotential shapes (that is on flow direction) whatever the range of geopotential heights, which governs the strength of the flow. Therefore we complement the three above analogy descriptors with the Maximum Pressure Difference (MPD) as fourth descriptor. The MPD of day t k is defined as the range of geopotential heights over Western Europe (in meters): The higher M P D k , the larger the pressure difference between the low and the high pressure systems at day t k , i.e. the more pronounced the centers of action over Western Europe for this day. Although it reflects a pressure difference, the MPD over Western Europe appears to be poorly related to NAO (Blanc et al., 2021b).
Overall, each result of the present paper are expressed per season. The four seasons are defined as December-January- The atmospheric descriptors are first employed to study the long-term evolution of Western Europe LSC over the period 1851-2010. As most of the descriptors rely on analogy in geopotential shapes, we start the analysis by checking whether 175 the different reanalyses provide similar geopotential shapes over this period. Figure 2 shows the evolution of the Teweles Wobus Score (TWS) between the daily geopotential height fields got from different reanalyses. ERA20C and ERA5 have been systematically interpolated on a coarser horizontal grid using a bilinear interpolation to allow the computation of crossed TWS between reanalyses. 20CRv2c grid is used to compare 20CRv2c, ERA20C and ERA5; ERA20C grid is used to compare ERA20C and ERA5. Recalling that a TWS score of 0 represents two identical geopotential shapes, we observe that differences 180 in geopotential shapes between reanalyses are weaker after 1950 than before (20CRv2c, ERA20C) and that differences remain quite steady from 1950 to 2010 (20CRv2c, ERA20C, ERA5). Before 1950, larger differences are observed between 20CRv2c and ERA20C, especially at the beginning of the 20th century. Those differences in geopotential shapes are larger in summer and weaker in winter while spring and autumn feature a transitional behavior. As a reference, we add in Fig. 2 the TWS between days D and D-1 considered in Fig. 1b (celerity of 12 December 1978) that is equal to 0.28 and corresponds to the 69 185 % percentile of celerity. Differences in shape between ERA20C and 20CRv2c (in red) before 1950 are notable; they are close to a TWS of 0.28, which reflects significant differences in geopotential shapes (see the difference in geopotential shapes between days D and D-1 in Fig. 1b). Substantial differences in geopotential shapes are also observed between 20CRv2c members (in gray) from 1851 to 1880 but they always remain less pronounced than differences between reanalyses from 1900 to 1950.
Furthermore, it is interesting to note the larger differences between reanalyses and members during both World Wars due to the 190 weaker number of assimilated observations. Overall, the significant differences in geopotential shapes before 1950 combined with the non-stationarity of the differences along the 20th century may have implications on the long-term evolution of LSC obtained from 20CRv2c and ERA20C.
In order to better understand the differences in shape between 20CRv2c and ERA20C, we map in Fig. 3  hand, 20CRv2c shows mainly increases in the 500 hPa geopotential height between the two periods, with a more pronounced increase over Great Britain. On the other hand, ERA20C features an increase in the 500 hPa geopotential height mainly in Southern Europe while a decrease in observed in Northern and Northeastern Europe for almost every season. This decrease in 500 hPa geopotential height is located further North from Western Europe in winter (black rectangle). The increase in the meridional pressure gradient in ERA20C between the beginning and the end of the 20th century is in line with Bloomfield et al. 200 (2018) who show an increase in the Arctic Oscillation from October to March in ERA20C. This increase is not observed in two other observation products; it appears to be explained in ERA20C by a larger sea-level pressure in the North Pole in 1900 that decreases along the 20th century, while no trend is observed over Northern Europe (Fig. 4 of Bloomfield et al., 2018). The latter is not necessarily in contradiction with our results since Bloomfield et al. (2018) study sea-level pressure while we study the 500 hPa geopotential height. At higher levels, this increase in meridional pressure gradient is consistent with Ménégoz et al.

205
(2020) who show an increase in the westerly component of moisture flux over the Northern half of Europe using a regional climate model forced by ERA20C from 1902 to 2010 (Fig. 5 therein). It is also in line with Rohrer et al. (2019) showing an increasing storm track activity in ERA20C along the 20th century over the North Atlantic/European domain. Figure 3 therefore highlights that the spatial differences in geopotential shapes between ERA20 and 20CRv2c come out as a reinforcement of the meridional pressure gradient between 1900-1930 and 1970-2000 in ERA20C, that is not observed in 20CRv2c. Finally, we plot the evolution of the four atmospheric descriptors over the period 1851-2010, considering a 5-year running average (Fig. 4). Overall, we observe a large interdecadal variability except for the celerity, that is broadly similar between the different reanalyses over the whole period. Except in summer, ERA5 produces larger values of celerity, singularity and relative singularity in comparison to the long-term reanalyses, as well as larger MPD values in every season (colored dots in 2010, Fig. 4). This result is consistent with Rohrer et al. (2018) who show that high-resolution reanalyses tend to produce larger 215 cyclone intensities and higher cyclone center densities, while full-input reanalyses tend to produce more intense blockings.
The higher spatial resolution of ERA5 as well as the assimilation of surface, upper-air and satellite observations generate more detailed geopotential shapes at 500 hPa, giving larger pressure differences (MPD) and weaker resemblances (celerity, singularity and relative singularity).
Over the period 1900-1950, major differences in descriptors trends are found between 20CRv2c and ERA20C. Differences 220 are larger in summer and weaker in winter, as already observed for differences in geopotential shapes (Fig. 2). This result is in    (2019), who show larger differences between 20CRv2c and ERA20C in summer than in winter regarding trends in the 500 hPa geopotential height variability over the North Atlantic/European domain (Fig. 4 therein). The differences in descriptor trends are considerable as they are clearly out of the range of the descriptor natural variability. Differences are more pronounced for the MPD, the relative singularity, and the singularity than for the celerity. ERA20C shows a strong trend 225 at having more closely reproduced flow directions from 1900 to 1950, especially from spring to autumn (decreasing singularity and relative singularity). ERA20C also shows a strong trend at having more marked geopotential shapes from 1900 to 1950 (increasing MPD), in accordance with a reinforcement of the meridional pressure gradient (Fig. 3). properties of the North Atlantic circulation. This is reinforced by the fact that the two individual members mostly share the same evolution in LSC characteristics even with quite different geopotential shapes (Fig. 2). Furthermore, it is interesting to note that differences in MPD between individual and mean member are rather weak over the whole period, meaning that averaging individual members leads to smoother but not flatter geopotential shapes. This reflects that the location and the intensity of the centers of action in individual members of 20CRv2c are similar over the whole period of the reanalysis, while 250 the other regions of the pressure fields are less constrained, and are thus more variable in shape. The fact that geopotential shapes are more marked in winter (larger MPD, see Fig. 7 of  makes it easier to capture the main pattern of the circulation even with few assimilated observations, leading to weaker differences in geopotential shapes between individual and mean member in winter before 1950 (Fig. 2). Overall, the lower number of assimilated observations in the beginning of the 20CR reanalysis and the differences between individual and mean members make it difficult to explain the increasing celerity We can also note the increase in MPD in autumn, pointing to an increasing intensity of the centers of action over Western Europe from September to November.
To summarize, the interannual and interdecadal LSC variability is consistent between the three reanalyses, but substantial differences in LSC trends are observed before 1950 in 20CRv2c and ERA20C. ERA20C feature less marked and quite different geopotential shapes in comparison to 20CRv2c in the early 20th century, as well as a clear increase of the meridional pressure 265 gradient until 1950. This result is consistent with the literature which shows that the pronounced trends in ERA20C might be driven by the increasing trend in the assimilated marine wind -a variable that is not assimilated in 20CRv2c. Furthermore, significant differences are also found between the geopotential shapes of 20CRv2c members before 1950, which is probably related the low number of assimilated data in the beginning of the reanalysis. The large differences in Western Europe LSC between long-term reanalyses hence make the study of LSC evolution difficult before 1950. In order to look in more details on 270 the trends in LSC characteristics after 1950, we focus on the distributions of daily descriptors instead of their seasonal mean and we distinguish the main atmospheric influences affecting Western Europe.

Recent changes in Western Europe LSC from 1950 to 2019 at daily scale
We focus on the changes in Western Europe LSC from 1950 to 2019 thanks to ERA5. We take advantage of the atmospheric descriptors to study changes in the whole descriptor distribution at the daily scale, rather than only considering trends in mean 275 descriptor values over a season as we did in Section 4.1. To do this, we separate the period 1950-2019 into two sub-periods of 35 years each and we look at the changes in descriptor distribution between the two sub-periods. Both Kolmogorov-Smirnoff and Anderson-Darling tests are carried out to detect significant differences in descriptor distribution at 5 % level. The significant differences in descriptor distribution and the sign of the difference in average descriptor value between the two sub-periods are summarized in Table 2. Considering the whole climatology, significant differences in descriptor distribution are found in 280 summer and autumn for almost every atmospheric descriptor and in spring for the relative singularity. In summer and autumn, the significant differences in descriptor distribution share the same sign, pointing to a decreasing average celerity, singularity, relative singularity (summer only) and an increasing average MPD. Considering the main influences affecting Western Europe shows that the differences in descriptor distribution are not equally distributed over the different influences (Table 2) Figure 5 shows the descriptor distribution of Anticyclonic conditions (boxplots) as well as the differences in descriptor den-290 sities between 1985-2019 (referred as the present period) and 1950-1984 (referred as the early period), per season. Over the present period, Anticyclonic conditions are associated with significantly lower celerities in summer and autumn. The increase in stationarity concerns Anticyclonic conditions below the 25 % percentile of celerity in summer and to a lesser extent Anticyclonic conditions below the 50 % percentile of celerity in autumn. In winter, Anticyclonic conditions are associated with lower singularities over the present period. This correspond to a strong decrease in the largest singularities (above 75 %), meaning that 295 new anticyclonic patterns are less explored over the present period. Finally, Anticyclonic conditions are associated with more pronounced geopotential shapes in winter (larger MPD), the most pronounced geopotential shapes (above 75 %) getting even more pronounced in the present period. We study how these changes in LSC characteristics affecting Anticyclonic conditions are distributed spatially by comparing the 500 hPa geopotential height composites of the period 1950-1984 and 1985-2019, per season (Fig. 6). Reminding that Anticyclonic conditions are associated to high pressure anomalies centered over Ireland 300 (Fig. 1a), the marked increase in 500 hPa geopotential height over Western Germany in spring suggests more eastward Anticyclonic blocking in the present period. In winter, the increase in 500 hPa geopotential height over the Atlantic up to Ireland suggests a reinforcement of the position of Anticyclonic blocking together with more intense blocking, in accordance with a decreasing singularity and an increasing MPD.

305
Atlantic circulations feature a decreasing celerity in summer between the two sub-periods (Fig. 7). Atlantic circulations feature a decreasing relative singularity in spring and to a lesser extent in summer, pointing to more Atlantic circulations featuring more resembling closest flow directions than usually. Finally, Atlantic circulation feature slightly more pronounced centers of action in autumn (increasing MPD) and more closely reproduced flow directions in winter (decreasing singularity) over the present period. The increase in the reproducibility of Atlantic circulations in winter is associated with a marked increase 310 in the most closely reproduced Atlantic circulations (below the 25 % percentile of singularity). This result is consistent with Yiou et al. (2018) who show that winter circulations over the North Atlantic tend to become more similar to already known patterns, with the dominant atmospheric patterns -mainly NAO+/zonal patterns -being trapped for longer times within the winter season. Looking at the spatial patterns of the differences, we observe that changes in 500 hPa height are quite weak from spring to autumn, although we can note a slight increase in the meridional pressure gradient in spring which could lead to 315 a slight increase in the zonality of Atlantic circulations (Fig. 6). Winter definitely shows the largest differences with a marked increase in 500 hPa heights over Northern Italy and a decrease in the Northwest of Great Britain. According to the anomalies associated with Atlantic circulations (Fig. 1a), this pattern reflects i) a northward shift of the Atlantic storm track between the two sub-periods, and ii) an increasing southwest component of Atlantic circulations. The latter is consistent with a decreasing singularity, the least singular geopotential shapes for Western Europe featuring west-to-southwest flow directions (Fig. 6 in

Mediterranean circulations
Mediterranean circulations feature clearly the largest differences in descriptor distribution between the early and the present period, with opposite differences across the seasons (Fig. 8). In summer and autumn, Mediterranean circulations become more stationary (lower celerity), more marked (larger MPD) and less singular in shape. These changes in LSC characteristics are 325 more correlated in autumn than in summer, in so far as they affect more often the same days, as illustrated by the shift in the 2D descriptors densities of Fig. 9 showing combined lower celerity, lower singularity, lower relative singularity and larger MPD in autumn. Nevertheless, we can note that differences in 2D descriptors densities are only significant at 10 % level for the combination of the relative singularity and MPD. This shift in LSC characteristics in autumn corresponds to more than a doubling (from 0.7 % to 1.7 %) in the proportion of Mediterranean circulations featuring among the most stationary, the most 330 closely reproduced and the most pronounced geopotential shapes (see the 10 % percentile in Fig. 10, right), and still a 30 % increase (from 7.9 % to 10.4 %) in the proportion of Mediterranean circulations featuring quite stationary, closely reproduced and pronounced geopotential shapes for Mediterranean circulations (see the 30 % percentile in Fig. 10, right). In summer, the shift affecting Mediterranean circulations concerns less extreme LSC characteristics, as shown by the 30 % increase for the 50 % percentile in Fig. 10 (left) and by the shift in density centers in Fig. 9.  In winter and spring, Mediterranean circulations tend to become more singular and less marked as well as less stationary (Fig. 8). The opposite patterns between autumn and winter changes could reflect a seasonal shift -the more marked, stationary and less singular Mediterranean circulations occurring in winter in the early period being shifted to autumn in the present period. The seasonal contrast of the differences in Mediterranean circulations is clearly visible in the maps of Fig. 6. Reminding that Mediterranean circulations are associated to low pressure anomalies over the near Atlantic (Fig. 1a), the large increase in 340 500 hPa geopotential height over the whole Northwestern Europe region in winter and to a lesser extent in spring confirms the weakening of Mediterranean circulations over the present period during these seasons. In summer and autumn, an opposite pattern is observed with a decreasing 500 hPa geopotential height over Northwestern Europe reaching further South in autumn, pointing to a reinforcing of Mediterranean circulations. The observed spatial patterns in summer and autumn -that is, an increasing pressure over Southern Europe and a decreasing pressure over Northwestern Europe -suggest an increasing zonality 345 of Mediterranean flows.  have shown that the singularity of Western Europe LSC is related to the zonality of the flow -the more zonal circulations being the more closely reproduced in the climatology. Here, the decreasing singularity of summer and autumn Mediterranean circulations together with the spatial patterns of the changes may suggest more frequent Southwestern flows and less frequent purely Southern flows at 500 hPa. This is fully consistent with the trends in summer 500   circulations associated with flow directions that, compared to the climatology, are among the most stationary (celerity < 10% percentile), among the most closely reproduced in the climatology (singularity and relative singularity < 10% percentile) together with among the most pronounced centers of action (MPD > 90% percentile). The percentage of change in the relative occurrence of such days between the two periods is shown by the darkgreen line associated with the right y-axis. Summer (autumn) Mediterranean circulation represent 573 days (841 days) in the early period and 605 days (957 days) in the present period. over the near Atlantic close to Ireland and a decreasing occurrence of low pressure anomalies centered over Northern Portugal (Extended Data Fig. 2 of Horton et al., 2015).

Potential impacts on local weather
The potential impacts of the observed changes in LSC on local weather can be discussed, both for weather variability and extremes. Focusing on weather variability, the increasing MPD in autumn for both Atlantic and Mediterranean circulations 355 reflects more pronounced centers of action and suggests a stronger mean flow towards Western Europe. A previous study showed that autumn seasons associated with pronounced centers of action over Western Europe are associated with large precipitation amounts in the Northern French Alps (correlation of 0.68, see Fig. 4 of Blanc et al., 2021b). In this way, the strengthening of the centers of action in autumn could induce an increase in autumn LSC-driven precipitation in the Northern French Alps.
Focusing on weather extremes, the increase in stationarity of Anticyclonic conditions in summer could have potential impacts on summer heatwaves, as the persistence of summer high pressure systems is a key parameter driving temperature extremes (Jézéquel et al., 2018;Riboldi et al., 2020). Regarding precipitation extremes, previous studies showed that Mediterranean circulations largely drive extreme daily precipitation in the Southwestern Alps in autumn (Blanchet et al., 2021b), and that the Mediterranean influence on extreme daily precipitation in autumn has been reinforced over the last 60 years (Fig. 4 therein).

365
A considerable increase in extreme precipitation is also observed in the Mediterranean-influenced regions of the Southwestern Alps in autumn over the last 60 years -autumn being the season featuring the most extreme precipitations . The combined increase in strength and stationarity of Mediterranean circulations in autumn -increasing the air flow toward a given region -together with an increasing humidity in a warmer air may increase the moisture flux, which is relevant to explain extreme precipitation magnitude and occurrence (Tramblay et al., 2012). Finally, previous studies showed that in winter 370 and spring, the Mediterranean influence on extreme daily precipitation in the Southwestern Alps have significantly weakened over the last 60 years (Blanchet et al., 2021b). This is consistent with less pronounced and less stationary Mediterranean circulations in these seasons over the last 30 years.

Conclusions
We have studied the past evolution of Western Europe large-scale circulation based on the 500 hPa geopotential height fields 375 using different reanalyses products. We employed several atmospheric descriptors that are mostly based on analogy and that allow a quantitative characterization of daily LSC.
We first focused on large-scale circulation evolution from 1851 to 2010 at seasonal scale. We showed major trend differences before 1950 between 20CRv2c and ERA20C, in accordance with the literature. The two reanalyses feature quite different geopotential shapes in the first half of the 20th century, especially from spring to autumn. ERA20C produces flatter geopotential 380 shapes in the beginning of the 20th century and an increase in the meridional pressure gradient that is not observed in 20CRv2c.
In 20CRv2c, the lower number of observations that are assimilated in the second half of the 19th century could lead to the generation of smoother geopotential shapes and may be responsible for the differences in geopotential shapes between the individual members, especially between 1850 and 1880. Overall, the differences in geopotential shapes in long-term reanalyses make it difficult to study long-term trends in Western Europe large-scale circulation. 385 We then focused on the changes in large-scale circulation after 1950 when the different reanalyses agree, using ERA5.
The atmospheric descriptors have been combined to an existing weather pattern classification to study large-scale circulation changes in the main atmospheric influences affecting Western Europe. On the one hand, we have shown that little changes are observed for Northeast circulations. On the other hand, we have shown that winter Atlantic circulations tend to be more resembling to known Atlantic circulations over the last 30 years. Anticyclonic conditions associated with the most stationary 390 geopotential shapes in summer are more frequent in the last 30 years than in the middle of the 20th century, which could have