Figure  shows the peak cyclone relative vorticity distributions for DJF and JJA in the NH and SH. For all seasons, and all future experiments, the model mean distributions lie within the uncertainty estimation of the historical models, indicating a relatively small change in cyclone vorticity. Estimations of the median vorticity are very similar for all of the future SSPs relative to the historical models in the winter seasons. However, in the summer seasons a decrease is evident, with this being notably larger in the NH.

All distributions exhibit a similar shape for the SSPs as in the historical simulations. Despite this, all seasons except NH JJA show a reduction in the number of cyclones near the peak of the distribution. This confirms that the reduction in cyclones noted in Fig.  is most likely to be seen for moderate-strength cyclones and not extreme cyclones (as in Fig. ). Since a key question is how the intensity of the highest-impact storms will change, the changes in the 90th percentile of the vorticity distributions are shown in the inset of each panel in Fig. . The model mean 90th percentile increases slightly with increased warming for NH DJF and both SH seasons yet decreases for NH JJA. However, the uncertainty estimation on the 90th percentile for each SSP has significant overlap with the historical model estimation. Therefore, there are different responses in the median and 90th percentile for the winter seasons and SH summer, with extreme cyclones increasing in intensity with no change (or a slight decrease) in the median intensity.

In Fig.  distributions of cyclone minimum MSLP are shown and calculated the same way as for relative vorticity in Fig. . The MSLP distributions have quite different shapes, with the NH DJF distribution having a negative skew (Fig. a) and NH JJA being normally distributed (Fig. b). In NH DJF (Fig. a) the distributions of the four SSPs are similar to the historical models, but with a reduction in the number of cyclones with MSLP at the peak of the distribution. There is a shift toward lower pressures in NH DJF, with lower medians and also a lower 90th percentile (inset). The medians show more of a shift than for vorticity, although the 90th percentiles show less of a shift (Fig. a). In NH JJA there is a slight shift toward lower intensities (higher MSLP), although the distributions are again very similar with less of a shift in the median than for relative vorticity.

In the SH the distributions of MSLP in the SSPs is largely different compared to the historical experiment. In DJF (Fig. c) there is a large reduction in frequency of cyclones with average or above-average MSLP, leading to a reduction in the median. Furthermore, there is an increase in cyclones with the lowest pressures, resulting in a lower 90th percentile and the average distribution nearly lying outside the uncertainty from the historical simulations. These shifts in the overall distribution, the median, and the 90th percentile are also apparent in JJA (Fig. d), with a shift toward lower pressures (higher intensity) in the SSPs that is exacerbated in the highest-emission scenarios. This is unlike the distributions of relative vorticity (Figs. c, d), which show only an increase in the 90th percentile of the distribution and minimal changes (or a slight weakening) in the median.

For MSLP there are larger changes in the median of the distribution than for relative vorticity, with this being particularly notable in the SH. This is likely a result of the poleward shift of cyclones and the significant influence of the large-scale pressure distribution on cyclone MSLP . As cyclones shift toward higher latitudes (Fig. ), they will be moving into an environment characterized by deeper MSLP and stronger pressure gradients. This is also likely the reason why less change is seen in the 90th percentile of MSLP in NH DJF when compared to relative vorticity, as the change in latitude of maximum track density is less apparent than in the SH (Fig. a–e).

Recently, demonstrated that CMIP6 models under-estimate the strength of the system relative circulation at 850 hPa by up to 2 m s-1. This under-estimation was identified on the poleward flanks of the cyclone in the region where the cold conveyor belt (CCB) would be expected to be found. The models used in this analysis show similar biases, albeit slightly smaller in magnitude (Fig. S1 in the Supplement), due to the increased number of higher-resolution models used here (Table ).

The future response of the 850 hPa wind speeds in extreme cyclones is examined for each winter and summer season in both hemispheres individually and for both a system and Earth relative perspective. Figure  presents the climate change responses for NH DJF. In a system relative perspective a strengthening of the wind speeds is seen for all SSPs (Fig. b–e). This strengthening is broadly contained between 5 and 15∘ of the cyclone centre and is progressively larger for the strongest climate change scenarios, with a peak increase of up to 0.7 m s-1 in SSPs 3-70 and 5-85. The strengthening is robust across most of the models, especially for SSPs 3-70 and 5-85 (Fig. d, e) and is concentrated on the forward and rearward flanks of the cyclone. The robust strengthening to the west of the cyclone centre in SSP5-85 is co-located with the maximum wind speeds associated with the CCB and formation of the low-level jet . It is evident from SSPs 2-45 and 3-70 (Fig. c, d) that there is also a strengthening on the SE edge of the cyclone in the warm sector, indicating possible changes associated with the inflow of the warm conveyor belt (WCB).

In an Earth relative perspective (Fig. f–j) the progressively stronger increase noted from Fig. b–e is also evident, with SSP5-85 having the greatest strengthening of the wind speeds. From this perspective there is a notable increase in strength on the southern and southeastern flanks of the cyclones, which would be situated within the warm sector of the cyclones. These anomalies are 10–15∘ away from the cyclone centre and peak at up to 0.9 m s-1 for SSP5-85 (Fig. j). Despite differences between the Earth relative and system relative wind speed changes, we find minimal differences in the propagation speed of the cyclones by the end of the century (Fig. a). In NH DJF there is considerable uncertainty on the sign of the individual model response and also non-linear responses across the experiments. The minimal change in propagation speed, coupled with a strengthening of system relative wind speeds, indicates a potential increase in wind risk by the end of the century.

In NH JJA (Fig. ) in the very core of the cyclone a strengthening of system relative winds of up to 0.9 m s-1 is seen for SSP5-85 (Fig. e), with a weakening identifiable outside this between 5 and 10∘ of the cyclone centre of up to 0.7 m s-1 and may be associated with the CCB . The weakening is concentrated on the forward flank of the cyclone, and for the more aggressive climate change scenarios (Fig. d, e) in the warm sector. All anomalies are most notable for SSP2-45 and above, and they also become more robust across the models with larger climate change. The reduction in wind speeds surrounding the core of the cyclone are likely a result of the reduced cyclone pressure gradient (not shown).

In an Earth relative perspective (Fig. g–j), a similar pattern of the biases is seen as in the system relative perspective. However a further weakening is evident on the southeastern flank of the cyclone that is between 5 and 10∘ of the cyclone centre and is up to 0.9 m s-1 in SSP5-85 (Fig. j). This further weakening of the wind speeds suggests a decrease in cyclone speed. This decrease is identifiable and is larger with increased rates of warming (Fig. b), although changes in the median are only by ∼0.1 m s-1. It is interesting that the response of cyclones in the NH is opposite in DJF and JJA, which is consistent with the changes in eddy kinetic energy and baroclinicity .

The SH winter (JJA; Fig. ) is consistent with NH winter (Fig. ) in that a strengthening of the cyclone wind speeds is seen. In a system relative perspective (Fig. a–e) a minimal increase is seen for SSP1-26 (Fig. b). However, for the remaining SSPs an increase is seen with the largest increases on the poleward flank of the cyclone centre (Fig. c–e). This strengthening is located in a similar location to the underestimation noted in the historical simulations relative to ERA5 (Fig. S1 in the Supplement and ), which is likely associated with the CCB. In SSPs 3-70 and 5-85 there is also a strengthening on the equatorward side of the cyclone, associated with the stronger pressure gradients (not shown). Furthermore, the largest strengthening is generally confined to within 4 and 8∘ of the cyclone centre, which is further away from the centre than the largest biases in the historical simulations, and also further away from the centre than the strongest wind speeds in Fig. a), indicating a possible expansion of the wind field. The strength of the increase is up to 1.8 m s-1 and is larger than in NH DJF in SSP5-85 (Fig. e).

In an Earth relative perspective, the cyclones in SH JJA (Fig. f–j) exhibit a large and robust strengthening on the equatorward flank of the cyclone. Unlike in the system relative perspective, the Earth relative anomalies get progressively stronger with each SSP, resulting in widespread anomalies of over 1.8 m s-1 in SSP5-85 (Fig. j). Associated with this, we identify a robust increase in the cyclone propagation speed (Fig. d), with a median change of ∼0.6 m s-1. There is strong model agreement on this change (unlike in NH DJF), and increases are largest for SSP5-85.

As with the vorticity and MSLP distributions (Figs.  and ), the lower tropospheric wind speed response in the SH summer (Fig. ) is consistent with the response from the winter for extreme cyclones (Fig. ). In both a system relative and an Earth relative perspective the circulation strengthens for the larger climate change scenarios, with minimal changes identifiable in the SSP1-26 simulations (Fig. b, g). In the system relative perspective the wind speeds increase on the poleward flank of the cyclone, with the largest change being between 5 and 10∘ from the cyclone centre, which is outside the maximum from the historical simulations (Fig. a) and as in JJA is suggestive of an expansion of the circulation. The largest increase in system relative winds is by up to 1.4 m s-1 for SSP5-85 conditions (Fig. e). As with all other seasons in both hemispheres, the change in wind speeds in the Earth relative perspective is concentrated on the equatorward flank of the cyclone for SH JJA (Fig. f–j). The wind speed changes are larger for the stronger climate change scenarios and peak at up to 1.4 m s-1 under SSP5-85 conditions (Fig. j). As with SH JJA, these increases demonstrate a change in the cyclone propagation speed, and we are able to identify an increase in propagation speed (Fig. c), albeit of smaller magnitude than SH JJA.

It is notable from the lower tropospheric composites and propagation speeds (Figs. –) that the responses of extreme cyclones to climate change in the NH and SH summer seasons are again different, as they were in the vorticity and MSLP distributions (Figs.  and ). This response is likely a result of the enhanced pressure gradients in the SH (not shown) and the persistent poleward shift of the storm tracks in the SH in both seasons (Fig. ). This is a contrast to the NH summer, whereby the storm tracks do not shift poleward, and instead a reduction in cyclonic activity is seen around the entire hemisphere (Fig. f–j). This is likely due to the larger polar amplification in the NH and reduction in lower tropospheric baroclinicity across the hemisphere, which has a strong impact on the NH storm tracks and general circulation .

For all seasons the largest 850 hPa Earth relative winds are commonly found within 5∘ of the cyclone centre (Figs. –). The largest future changes, however, are often found further from the cyclone centre and can be 10–15∘ from the cyclone centre. For all seasons it is likely that there are changes in how broad the circulation is. To quantify the change in the broadness, a transect of the circulation is taken on the equatorward side of the cyclone, from the cyclone centre to the edge of the composite area, thereby intersecting the region of maximum wind speeds (dashed white line Fig. f). The fraction of this transect above a fixed wind speed threshold is calculated and plotted against the maximum wind speed of the cyclone along the same transect. Wind speed thresholds have been chosen to best represent the footprint of the historical composites.

For NH DJF (Fig. a) there are only small changes in maximum wind speed; however there are increases in the broadness of the area of above-threshold wind speeds, with the peak of the distribution being ∼0.5∘ broader in the SSP5-85 experiment than in the historical period. Conversely, in NH JJA (Fig. d) there is a considerable contraction of the area of above-threshold wind speeds by nearly 2∘ that is also associated with a lower maximum wind speed. In the SH we find an expansion in the area of above-threshold wind speeds in both DJF and JJA (Fig. S2). However, unlike in the NH, this is also associated with a slight increase in the maximum wind speeds at 850 hPa.

As the broadness of the circulation is increasing (in NH DJF), we can quantify the area of the cyclone above a fixed threshold for the entire composite area and how this evolves for the different SSPs over time. Figure  shows time series of the winter and summer seasons in the NH and SH for the area of the cyclone composite above a fixed threshold of 17 m s-1, encapsulating the area of strongest winds of the composite cyclones. The differing responses in the DJF and JJA in the NH are notable in Fig. a and b, with the SSPs showing an increase in cyclone area above 17 m s-1, and a decrease in JJA, by the end of the century. The differences between the SSPs are not overly large in the NH, with SSP5-85 showing a median increase of 7 % in above-threshold winds by 2100 in DJF, which is similar to SSP3-70 (Fig. a). However, increases may be as large as 15 % in SSP5-85 based on the large spread of model projections, with almost half of the SSP3-70 and SSP5-85 estimations exceeding the historical 75th percentile in the final decade of the 21st century. The SSP1-26 and SSP2-45 scenarios generally show lower increases in the area of above-threshold wind speeds with median increases of 2 %–3 % by the end of the century. All the SSPs show a very similar average evolution until 2080 of a 2.5 %–5 % size increase, with the two most aggressive SSPs only separating from the two weakest SSPs after ∼2085. Interestingly, the SSP2-45 scenario indicates a decline after 2075, which is in line with modelled reductions in CO2 emissions toward the end of the 21st century in that SSP . Despite the differences in scenarios, all of the SSPs in Fig. a have a median extreme wind speed area lower than the 75th percentile from the historical simulations by 2100. In JJA all SSPs are very similar and generally show up to a 10 % reduction in the area of above-threshold winds throughout the century. Generally, SSP5-85 and SSP3-70 show the largest reduction, which may be as large as 25 %–30 %. Despite this, the inter-annual variability is especially large for all SSPs, and each has considerable overlap with a large amount of the historical distribution of cyclone area.

In the SH a larger signal, with greater differences between the SSPs, is identifiable for the area of above-threshold cyclone winds (Fig. c, d). For both DJF and JJA the higher-emission SSPs demonstrate a marked increase in cyclone area of above-threshold winds relative to the historical simulations. In DJF (Fig. c) the SSP5-85 simulations have an extreme wind speed area nearly 15 % larger by the end of the century, which is also considerably above the 75th percentile from the historical simulations. The SSP3-70 simulation demonstrates around a 10 % increase by the end of the century, which is comparable to the 75th percentile of the historical simulations. Both the SSP5-85 and SSP3-70 simulations are similar until around 2075, when the rate of increase in cyclone area levels off in SSP3-70. Accounting for the spread in the model simulations by the end of the century indicates increases may be as large as 20 %–25 % relative to the historical average, with a majority of the SSP3-70 and SSP5-85 distributions being above the historical 75th percentile. The SSP2-45 experiment is very similar to SSP3-70 until 2070; however the average area increase is lower at only 5 % above the historical average by the end of the century. The SSP1-26 experiment demonstrates the smallest by the end of the century, which is within the range of the 75th percentile and less than a 5 % increase in area compared to the historical simulations. The SSP1-26 simulation actually increases until approximately 2070, when it closely matches the SSP2-45 and SSP3-70 scenarios, with a decline then following, before levelling off by approximately 2085. By the end of the century the spread in the SSP1-26 simulations is very similar to that of the historical cyclone area.

In SH JJA (Fig. d) a similar evolution to DJF (Fig. c) is notable, albeit with a larger signal in the SSP5-85 and SSP3-70 scenarios. Both SSPs demonstrate large increases in cyclone area above the threshold wind speed compared to SSP2-45 and SSP1-26 and show an increase in an average area increase of at least 20 % by the end of the century. From 2090–2100 a majority of the distribution in both SSPs exceeds the 75th percentile of the historical simulations, and the 75th percentile of the SSP5-85 scenario estimations is 35 %–40 % larger than the historical average. For both these SSPs the cyclone area demonstrates a steady increase from 2040, by which all experiments already show an increase in size of 5 %–10 % above the historical average. Neither SSP2-45 nor SSP1-26 shows a marked increase from 2040–2100. SSP2-45 maintains an average of approximately 10 % increase in cyclone area by 2100, whereas SSP1-26 shows a very slight decrease to an average of a 5 % increase in size by 2100. The clearer signal in the SH and greater separation in the SSPs is (as mentioned previously) likely due to the reduced land influence in the SH allowing for a more robust response to the evolving climate forcings. In fact, the evolution of the cyclone size above the 17 m s-1 wind speed threshold bears considerable similarity to the evolution in CO2 concentration noted in the ScenarioMIP experimental design (; Fig. 3b).

The middle troposphere is defined as 500 hPa in this analysis. In the interest of brevity, only the NH DJF changes will be discussed in detail, with changes from other seasons available in the Supplement (Figs. S3–S5) and discussed briefly at the end of this section.

In NH DJF (Fig. ) the spatial patterns of change across the models presents a similar pattern across all SSPs in both a system relative (Fig. b–e) and an Earth relative (Fig. g–j) perspective. In the system relative perspective there is first of all a strengthening of the wind speeds on the poleward flank of the cyclone centre that is ∼5∘ away from the centre. This is a region where winds are travelling from east to west (Fig. a). This element of the circulation in the mid-troposphere is likely associated with the upper reaches of the CCB , and this may indicate a strengthening of this airstream, which is robust across a majority of the models.

The other notable difference in the system relative winds is situated within the warm sector of the cyclones. For all SSPs there is evidence of a strengthening of the winds, which is progressively stronger with the more aggressive climate change scenarios and by up to 1 m s -1 in SSP5-85 (Fig. e). This increase is located between 5 and 10∘ of the cyclone centre on the equatorward and eastern flanks. These regions are associated with the strongest ascent , which therefore indicates a possible strengthening of the WCB. A large amount of the strengthening in the warm sector is due to an increase in the meridional component of the wind by over 1.8 m s-1 to the east and southeast of the cyclone centre (Fig. S6). This stronger warm sector motion may, in part, be driven by increased atmospheric moisture content under future climate conditions, which will result in larger condensational heating in ascending branches of cyclones and potentially increased rates of ascent.

One final feature to note across all the SSPs in the system relative framework is the weakening of the circulation on the rearward (western) flank of the cyclone between 5 and 15∘ of the cyclone centre. This region is behind the cold front and generally is associated with the descent of drier air . This weakening suggests that the descent rate may be weaker under future climates, which will again be exacerbated with larger levels of climate change.

In the Earth relative perspective (Fig. g–j) a very similar picture is presented as in the system relative perspective (Fig. b–e). The main difference is the increased strength of the changes in the warm sector in the Earth relative perspective, which is also slightly broader than in the system relative perspective. In Fig. b it can be seen that, as at 850 hPa, the area of extreme wind speeds becomes considerably broader. However, unlike at 850 hPa the maximum wind speed also increases at this level. This may be due to a larger role of moist processes in the mid-troposphere; however projected increases in the strength of the upper-level jet may also have an influence at this level.

In the NH JJA (Fig. S3) and both SH seasons (Figs. S4–S5) the pattern of change is broadly consistent with NH DJF (Fig. ), but with magnitudes and sign that are consistent with the biases noted at 850 hPa (Figs. –). For NH JJA (Fig. S5) there is weakening of the wind speeds in the warm sector of the cyclones, which is progressively larger for the more aggressive SSPs. Furthermore, the broadness of the extreme wind speeds also decreases (Fig. e). In the SH the responses in both seasons are very similar (Figs. S4–S5), unlike in the NH. Both SH seasons feature a pronounced strengthening of the wind speeds in the warm sector, with the largest anomalies being between 5 and 15∘ of the cyclone centre on the equatorward and forward flanks of the cyclone.

Both SH seasons also show a strengthening of the wind speeds on the poleward flank of the cyclone in the system relative framework, at a distance of approximately 5–7∘, which is likely associated with the strengthening of the CCB noted earlier.

As in the previous section, only the changes for NH DJF (Fig. ) will be discussed in detail due to the similar responses of the other seasons and magnitude that is consistent with anomalies at the other tropospheric levels. Changes for the other seasons are present in Figs. S7–S9.

The changes in the wind speeds at 250 hPa in NH DJF can be broadly split into three distinct features in the system relative perspective (Fig. a–e). Firstly, 10–20∘ to the southwest of the cyclone there is an increase in the wind speeds, which is larger for the strongest SSPs. This is likely associated with a strengthening/broadening of the upper-level jet (Fig. c). Secondly, there is an increase in the wind speeds by up to 2.1 m s-1 to the south, east, and northeast of the cyclone centre that is ∼5∘ from the cyclone centre and rotating cyclonically around it. This is the region of largest ascent and divergence (Fig. a) and the region where models struggle to capture the strength of the wind speeds in historical simulations . This increase is situated above and slightly downstream of a similar increase at 500 hPa (Fig. b–e) and therefore likely associated with the slantwise ascending motion in the warm sector. This strengthening is associated with large increases in meridional motion (Fig. S6) and together with the results at 500 hPa suggests that the WCB in cyclones is likely to be stronger and progressively so with increased warming rates by the end of the century.

The third change at 250 hPa is an increase in wind speeds on the forward flank of the cyclone between 10 and 20∘ of the cyclone centre and extending in a southerly and southwesterly direction. The strengthening is mostly driven by a more negative meridional component of the wind (Fig. S6e), which is associated with the anticyclonic turning of air and likely part of the WCB outflow. It is likely that the changes in this sector of the cyclone are associated with those closer to the cyclone centre.

All these anomalies are largest for SSP5-85 (Fig. e) with strong model agreement across the SSPs, indicating that the warmer climate, which is characterized by a higher moisture content in the future, results in a stronger wind speeds in the warm sector and anticyclonic turning near/at the tropopause. The forcing of these changes is likely to be somewhat driven by changing rates of vertical motion in the cyclone, which can be inferred from the rate of horizontal divergence at 250 hPa (Fig. ). There are increases in divergence in the warm sector, with the largest increases downstream and poleward of the maximum in the historical simulations (Fig. e), indicating stronger ascent in the warm sector of the cyclones and larger outflow near the tropopause. Furthermore, there is a smaller increase in convergence behind the cold front, suggesting that the descending motion within the cyclones, which often originates near this level, does not undergo changes as large as in the ascending branch. Therefore, the asymmetry in the vertical velocity within extratropical cyclones may increase.

In an Earth relative perspective (Fig. f–j) a majority of the features noted in the system relative perspective (Fig. a–e) are also notable and consistent in their magnitude and positioning. As at other levels an increase in the strength of change, particularly on the equatorward flank of the cyclones, is notable, which is a result of the increasing speed of cyclones in the future. As in the lower and middle troposphere, a considerable broadening of the circulation is also identifiable at 250 hPa (Fig. c) alongside an increase in the maximum speed to the south of the cyclones.

Composites for the changes in the 250 hPa wind speeds for the NH JJA (Fig. S7) and also for DJF and JJA in the SH (Figs. S8 and S9) are now discussed briefly. For these other seasons a majority of the same features identified for NH DJF (Fig. ) are seen, with magnitudes of changes that are consistent with the same season in the middle and lower troposphere. In NH JJA a weakening of the wind speeds is again identifiable throughout the warm sector of the cyclone. Due to the considerably weaker nature of the wind in NH JJA, the separate changes associated with the anticyclonic motion are harder to identify. However, the meridional component of wind shows a marked increase to less negative values (weakening, Fig. S6j) to the southeast of the cyclone, indicating a consistent (yet opposite) behaviour to the change in NH DJF. Furthermore, there is a considerable weakening of the divergence (Fig. j), suggesting that the rate of vertical motion in the cyclones will be weaker.

In the SH both seasons (Figs. S8 and S9) are very consistent with NH DJF in that strengthening wind speeds associated with both the cyclonic and anticyclonic motions within the composite area are identifiable, with both only becoming evident for SSP2-45 and above. For both seasons in the SH and for NH JJA, the changes are largest in magnitude for SSP5-85 and progressively lower for each SSP below. All of the major changes are robust across the model ensemble and represent a clear picture as to how cyclones will change as a result of future climatic change.

For all levels in the troposphere the cyclones in the SH only start to show a consistent signal and pattern of change for SSP2-45 and above, whereas the pattern of the anomalies in the NH is usually evident (although often small in magnitude) even for SSP1-26.