We use km-scale climate model data from the COSMO-6 regional climate model, which was run at a 2.2 km resolution over continental Europe (for model domain, see Fig. ). In the current climate (CC) simulation (2011–2021), the boundary conditions are provided by ERA5 . The future climate (FC) simulation follows a PGW approach, using a GWL of +3 °C . A model member representative of convective summer precipitation of the MPI-HRES-ENS was chosen to compute the +3 °C climate change Δ . This climate change Δ is added to the ERA5 boundary conditions, mimicking the same 11-year period and its variability in a warmed climate. Further details on the climate simulations and their verification are described in , and .

To identify supercells trajectories, supercell tracks from are used. To identify supercells, first all thunderstorms are tracked based on 5 min precipitation data . Along these tracks, the hourly 3-dimensional (3D) pressure-level wind field is used to apply vertical vorticity and updraft velocity thresholds in the mid-troposphere, identifying the mesocyclone (for full documentation, see ). The tracking differentiates cyclonic and anticyclonic updraft vorticity signatures, separating RM from LM supercells.

Table  provides an overview of analyzed variables. Precipitation variables are stored at a 5 min resolution, whereas all other variables are stored hourly. 3D data is available on 8 pressure levels. A limited set of convective parameters was computed online from 60 model levels.

We use the cookie-cutter method introduced in . In this method, 3D environmental fields are extracted from the model data within a 60 km radius around each identified supercell at hourly intervals. Each extracted disk is rotated so that the hourly smoothed propagation direction of the storm is due right (0°). The environmental data is extracted at the time of the storm, to reflect in-storm conditions, as well as 1 and 2 h prior, to reflect pre-storm conditions. For each variable, we compute composites across all supercells or subsets of supercells for the current and future climate simulations separately. Composite vertical profiles of the pre-storm environment are obtained from the extracted disks, with a filtering of active convection via a precipitation filter of 5 mm h⁻¹ with a 5-gridpoint circular buffer. Aggregated analyses are performed within the whole 60 km radius, while spatial figures showing the storm structure are zoomed in to a 20 km radius. All statistical tests are performed with a two-sided Mann–Whitney U test (p > 0.05, ), a non-parametric, unpaired test. Spatial analyses are additionally processed with a false-discovery-rate (FDR) correction .

Analyses of storm motion make use of the Bunkers storm motion estimate for RMs (Eq. ) and LMs (Eq. ): 1VRM=Vmean+CRMk^×S‖S‖2VLM=Vmean-CLMk^×S‖S‖, where Vmean = 1Δp∫925hPa500hPaV(p)dp, S = V500hPa-V925hPa, k^ = (0,0,1), and S⟂ = k^×S‖S‖, here adapted to pressure-level model data. Originally both CRM and CLM were estimated at 7.5 m s⁻¹ , while suggests an updated CLM of 5 m s⁻¹. In contrast, suggests a combined C in Europe for both RM and LM of 4 m s⁻¹, smaller than any C tuned for the United States.

Figure  compares the storm structure of right- and left-moving supercells based on composites of precipitation, vertical velocity, and vorticity. Overall, the composites show strongly smoothed storm structures, owed to the averaging across many cases. RM storms exhibit the familiar right-leaning curvature of the 15 mm h⁻¹ precipitation contour, with the mesocyclone located at the inner corner of this curve (center of the 20 km radius). LM storms show a weaker and less elongated leftward curvature of the precipitation field. To identify the updraft, we show the 90th percentile of 500 hPa vertical wind and 500 hPa vertical vorticity (10th percentile of vertical vorticity for the LMs, due to the opposite sign of the vorticity). Midlevel updraft and vorticity extrema are both in the center of the composite. The updraft and vorticity signatures are of similar size and intensity in both storm types. The downdraft features are depicted by extracting the 10th percentile of the 700 hPa vertical wind and 500 hPa vertical vorticity (90th for the LMs). Similar to the updraft, the downdraft intensity and area are comparable in the RM and the LM. Precipitation largely falls in the respective downdraft regions. The vorticity in the downdraft area is more pronounced on the LMs than the RMs. Despite the less pronounced shape in the mean precipitation contour, LMs still show a clearly identified updraft area in the center of the composite, and a downdraft area at the right-hand storm flank, that mirror the structure of the RMs and rival them in terms of intensity e.g.,.

To investigate the synoptic meteorological situation, we look at the composite of geopotential height at 500 hPa 2 h before storm-encounter, to ensure unperturbed conditions – 1 h prior and hour zero were tested and showed considerable influence of the presence of the storm. Both storm types deviate to the right and left of the isohypses, respectively, consistent with their deviant motion. The isohypses also indicate a low-pressure area lying to the left-hand side of storm motion and higher pressure being on the right-hand side. The gradient of geopotential height is greater in the RMs, while also being at a lower geopotential height. Severe convection in Europe typically occurs at the leading edge of an approaching trough and in southwesterly flow , consistent with the pattern of geopotential height. The greater height and weaker gradient for LMs indicate that they may be preferentially located closer to the downstream ridge. This behavior will be discussed further in Sect. .

Figure  shows the same composites of storm structure and synoptic environment for the future climate. The hallmark features remain the same, with left- and right-leaning storm contours, and deviant motion to the left and right of the geopotential height contours. The most noticeable changes are an increase in storm precipitation area for both LMs and RMs, as well as a distinct increase of the 500 hPa geopotential height by ∼ 3 dam. Both up- and downdraft areas appear slightly broader in LMs and RMs, though reach similar peak values in the composite.

Prior research suggests the biggest differences between RM and LM supercell environments in their respective pre-storm hodographs e.g.,. The environmental conditions are extracted 1h pre-storm encounter (see Fig. ), which yielded the most representative (i.e., least convectively contaminated and highest convective potential) results. The hodographs are rotated so that storm motion is aligned among all of them, yielding storm-direction-relative (SR) uSR- and vSR-components of the profile, uSR being along the propagation direction and vSR perpendicular to it, i.e. modeled storm motion lies on the vSR = 0 line.

The hodographs of RMs in the current climate (Fig. a) exhibit a slight clockwise curvature with altitude, while overall being largely straight, with storm motion lying to the right of the hodograph. Shear is largely concentrated at the 850–700 hPa level, whereas the near-surface change in wind speed is very gradual. The LMs' hodograph has a lower degree of counterclockwise curvature with slight veering in the low-levels and virtually identical deep-layer shear e.g.,. Both mean hodographs are mostly straight and do not contain substantial directional shear (Fig. ). At a 3 °C warming level (Fig. b), both hodographs lengthen and the 925–500 hPa shear increases by 1–2 m s⁻¹.

Modeled storm motion was extracted from the 5 min precipitation tracks and smoothed with a moving-average 5-timestep window. We compare modeled storm motion (vSR = 0 m s⁻¹) to the estimated Bunkers storm motion (, see Eqs.  and ) using both the CRM,LM = 7.5 m s⁻¹ and CLM = 5 m s⁻¹ constant. While the obtained velocity is comparable, albeit slightly smaller than the modeled storms, the estimated deviation from the mean flow is considerably larger than in the modeled storms. By refitting C to scale estimated storm motion onto the vSR = 0 m s⁻¹ line, our simulations suggest an RM offset of CRM = 5 m s⁻¹ and an LM offset of CLM = 1 m s⁻¹. In combination, these are closer to C = 4 m s⁻¹ as estimated by .

In addition to the composite hodographs, we investigate the mean propagation direction and speed in Fig. . In line with previous studies showing that severe convection in Europe happens overwhelmingly during SW flow , supercells track largely towards the NE (see Fig. a). There is a distinct separation of the mean track direction of right- and left-movers by 12°, with the mean track of RMs lying to the right of the LMs, as expected. The relatively small angle difference between the storm types suggests that they form in different systematic relative locations within the parent synoptic systems, where the hodograph shape and orientation, and subsequent mean wind vector are different. Both storm populations track 3° more northwards in the future climate (Fig. a). From Fig. b we further see that propagation speed is comparable between RMs and LMs, with an average of approximately 15 m s⁻¹. In line with the increase in deep-layer shear and the lengthening of the hodograph, the propagation speed increases by ∼1.5 m s⁻¹ at the 3 °C GWL.

We investigate the thermodynamic environment both in a composite skew-T log-p diagram (see Fig. ) and by comparing the distribution of composite environmental parameters (see Fig. ) throughout the extracted 60 km radius. The skew-T diagram (Fig. ) shows statistically significant differences between RM and LM in the current climate. LMs occur in significantly warmer conditions, with higher lifting condensation levels (LCL) and higher CAPE and CIN values than RMs. Low-level specific humidity is also slightly higher for LMs, but relative humidity is lower due to the large temperature difference of almost 3 °C. This poses the question, why LMs occur in such thermodynamically different conditions. Comparing the distribution of CAPE, CIN, deep-layer shear, and low-level atmospheric conditions for RM and LM (Fig. ), we see that LM-supporting conditions have lower variability than those of RMs, with a narrower interquartile range. Given that LMs occur more rarely, it appears that LMs require a more specific range of environmental conditions (Fig. ). In the seasonal and diurnal cycle, LMs occur more frequently during the convective peak season, and less frequently in the shoulder seasons, showing a sharper peak in the distribution, with shorter tails (see Fig. a). During the convective peak season, instability and temperature are generally greater than in the less favorable shoulder seasons and nighttime hours. Additionally, LMs occur in significantly higher mean surface pressure situations than RMs (see Fig. h), and their surroundings show smaller surface pressure gradients (not shown). This indicates that, from a synoptic perspective, they occur closer to the center of the ridge, further supporting the hypothesis from Sect. . Overall, severe convection in Europe predominantly occurs between an upstream trough and a downstream ridge . In this transition area, warm air advection may occur in the warm sector ahead of the cold front. Warm air advection induces a vertical clockwise turning of the geostrophic flow, due to thermal wind balance. Closer to the center of the ridge, wind speeds are lower and hence warm air advection is smaller, hence producing less synoptic forcing towards a clockwise turning hodograph, which favors RMs .

In a 3 °C warmer atmosphere, the thermodynamic profiles for both RMs and LMs show an increase in temperature and specific humidity (see Figs.  and ). Focusing on the vertical profiles of the RMs, separate factors contributing to both increases can be identified. With a surface temperature increase and a low-level specific moisture increase, the LCL remains approximately stable (see Table ). Due to the steeper moist adiabatic lapse rate at warmer temperatures, the level of free convection (LFC) increases slightly. The environmental lapse rate is stabilizing slightly; however, the steeper moist adiabat over the full vertical range from the surface to the LFC for CIN and from the LFC to the level of neutral buoyancy (LNB) for CAPE compensates this and increases the area of the integral. While relevant environmental intensity metrics increase for both right- and left-movers (Fig. ), the increase tends to be even greater for LMs, with larger changes in CAPE, surface temperature, and specific humidity. Overall, the composite thermodynamic profiles show typical features of severe convective environments in Europe, with a moist boundary layer and increasingly dry mid-levels, consistent with the possible presence of an elevated mixed layer .

Overall, RMs exhibit higher average intensities of 5 min precipitation rate, and extend over larger areas for all hazards, consistent with their larger size compared to LMs (see Figs.  and ). Specifically, the greater relative humidity in RMs is consistent with the higher intensity and area of precipitation (see Figs. b, g and f, g). As for hail size, LMs produce similarly sized hail as RMs, but over a smaller area (Fig. c, h). Notably, they lack the lower end of the maximum hail size distribution, with smaller sizes being less common (Fig. c, LM distribution narrower at smallest sizes). Given the higher temperature, this is consistent with the melting of hail .

At the 3 °C GWL, all hazard intensities increase significantly for both RMs and LMs, apart from LM wind gust intensity (Fig. a–d). The hazard intensification contrasts with stable maximum updraft speeds in both RM and LM (see Fig. e), however, updraft area increases significantly (Fig. j). Larger updrafts are more resistant to midlevel entrainment and additionally, the mid-levels are significantly moister in future (Fig. h). The hazard intensity changes are larger for LMs, especially at the higher percentiles (not shown). Hence, despite LMs being a smaller subset of all supercell thunderstorms, their hazards are increasing more than those of the total storm population. Hazard areas also increase consistently for both storm populations, apart from the surface gust area of LMs, which has no significant difference. Given the overall smaller size of LMs, the absolute change in hazard area is smaller than for RMs, but the relative change is of a similar magnitude.

The lack of significant change in surface wind gusts is surprising, as thermodynamic considerations suggest that with the given change in environmental conditions, a more pronounced intensification should occur (expected ∼ 15 %, ). However, the variable investigated depicts the hourly maximum 10 m wind speed, and it is not at a 5 min temporal resolution, as opposed to precipitation and hail metrics. Moreover, 10 m wind gusts are parametrized and not diagnosed directly in the model. This may be responsible for part of this discrepancy. Another factor is the presence of considerable terrain over large parts of Europe, especially in the supercell occurrence hotspots of the Alpine region. Wind gust behavior in complex topography is considerably more complicated and less straightforward to infer, as local channeling and damming effects nonlinearly affect the gust amplitude.

To investigate the impact of local climates and their future changes on supercells, we use two subregions in the European domain: the Iberian Peninsula (IP) and the Northern Alps (NAL, see Figs.  and , regions defined in Fig. ). The climate of these two regions differs strongly between a dry, Mediterranean climate in IP and a moderate, strongly topographically influenced climate in NAL. The IP region is ∼ 6.5 times larger compared to NAL (ca. 1 000 000 km² vs. ca. 150 000 km²), with a moderate supercell occurrence (1420 mesocyclones and 20.8 % LMs.) The NAL is a comparatively smaller region, however, it has a higher supercell occurrence for its area with 447 detected mesocyclones in the current climate and a LM-ratio of 11.6 %.

Comparing the pre-storm environmental conditions of the current climate in both regions (see Fig. ), IP emerges as the region where supercells form in hotter and drier conditions (Fig. d, e, f); 850 hPa specific and relative humidity are considerably lower than in NAL (Fig. e, f). This combination of hotter, but drier conditions results in comparable instability (Fig. a), with the range of the CAPE distribution being greater in NAL, reaching higher values in the upper percentiles. CIN is greater in IP, driven by the dryness (Fig. b). In the regional decomposition, the pattern of LM occurring in hotter and relatively drier conditions also persists. In NAL we also find higher-CAPE conditions for LMs, but the dry conditions in IP compensate for the hotter temperatures and lead to the same instability range for RMs and LMs. This further supports the hypothesis of LMs occurring (a) closer to the ridge center in synoptically driven situations and (b) preferentially during the convective peak in the day and year, when conditions are most optimal.

In line with the hotter and drier conditions, the distribution of maximum hail sizes in IP lacks the small hail sizes (Fig. c, IP's distribution is narrower for small sizes than NAL's), consistent with melting signatures. However, upwards of the median, the percentiles are distributed very similarly to NAL (Fig. c, h). IP also has lower precipitation intensities that decrease for LMs in the future climate but increase for RMs (Fig. b, g). This suggests that for these storms, while the warmer air could carry more moisture, there is not enough specific humidity available to sustain higher precipitation rates or greater precipitation areas. The increase in RM precipitation rates may result from the larger shear, which increases storm area and organization, RH, and updraft area and may overcompensate for the slight decrease in updraft strength e.g.,. In both regions, LMs reach higher lightning activity, but in a smaller area than RMs (Fig. a, f). Lightning activity in NAL is slightly greater than in IP, possibly related to the greater moisture content. IP has a greater potential for wind gusts in area and speed (Fig. d, i). Strong gusts are more likely in drier conditions that can initiate and accelerate downbursts through evaporative cooling .

Overall, within these sub-regions, the hazard intensities between LMs and RMs can have similar variability as the climate scenarios. Moreover, LMs can have equal or higher intensity hazards, but tend to have them in a smaller area, as their size is overall smaller.