Introduction

Given the significant risks posed by near-surface extreme wind speeds associated with midlatitude cyclones to lives, livelihoods, and infrastructure1,2,3,4, it is imperative to understand their physical changes, including magnitudes and patterns, under human-induced global climate change. This understanding is crucial for the development of effective climate adaptation strategies and the mitigation of associated risks.

Midlatitude cyclones, often referred to as windstorms, have long been recognized for controlling extreme wind speeds and weather patterns in the midlatitudes5, and many studies have investigated the response of midlatitude cyclone tracks, intensity, and mesoscale features to a warmer climate in both the NH and the SH6,7,8,9,10,11,12,13. In the SH, a consistent and uniform poleward shift of midlatitude cyclone storm tracks has been projected by the end of the 21st century12,14. Similarly, in the NH, projections of midlatitude storm tracks in a warmer climate generally showed a poleward shift of storm tracks in the western North Pacific, along with an intensification over the British Isles and Northwestern Europe15,16, though with less certainty compared with the SH. Moreover, a few recent papers suggest that, under anthropogenic climate warming, the most severe cyclones are expected to further increase in intensity13,17, though there is no strong consensus10. A slightly more robust agreement exist among scientists on the future projections of the overall number of midlatitude cyclones, which are expected to decrease in both hemispheres under a warmer climate12.

A recent investigation13 using models from the Sixth Climate Model Intercomparison Project (CMIP6) initiative18 explored future global changes in extratropical storm tracks, cyclone intensity, wind speed, and structure. The study found a projected reduction of approximately 5% in the number of cyclones by the end of the 21st century, accompanied by a 4% increase in the number of extreme cyclones and higher peak relative vorticity at 850 hPa. These projected changes in cyclone frequency and intensity have implications for the associated wind speeds6,12. Specifically, the increase in the number of extreme cyclones, coupled with higher cyclone-associated peak relative vorticity, has been linked to an increase in 850 hPa wind speeds in certain regions of the NH, such as the British Isles and Northwestern Europe13.

Idealized global warming simulations based on changing the optical thickness of the atmosphere have also indicated potential increases in wind speeds and wind gusts associated with midlatitude cyclones19. This effect is likely attributable to the larger latent heat flux associated with warmer climate simulations, which amplifies the positive lower-tropospheric potential vorticity, and this amplification, in turn, enhances cyclone intensity and associated surface winds and gusts17. Indeed, results from a single climate model of CMIP6 have demonstrated an increased likelihood of extreme wind speeds in the UK region, attributed to the more frequent occurrence of midlatitude cyclones with sting jets, leading to higher winds and gusts at the surface20,21. Similarly, in the SH, simulations from CMIP5 models projected an increase in the wind intensity of extreme winter midlatitude cyclones, aligning with the findings regarding cyclone tracks discussed earlier12.

Despite the growing attention given in the last decade to future climate projections of midlatitude cyclones and their synoptic and mesoscale characteristics12,13, there exists a research gap between those studies12 simulating the response of midlatitude cyclones and associated winds to anthropogenic warming using an idealised modelling set up consisting of an aquaplanet, where either the optical thickness17 or Sea Surface Temperature (SST)22 is increased to simulate warmer climates, and those studies investigating climate projections of midlatitude cyclones using the complex fully-coupled CMIP59 and CMIP613 models. Furthermore, the few studies that have specifically examined how the intensity of the extreme near-surface wind speed events would change have primarily approached the topic from a wind energy perspective23,24, often disregarding the associated meteorological drivers, e.g., midlatitude cyclones (but also mesoscale convective systems, tropical cyclones...). Both of these research areas have been highlighted by a recent review on the future of midlatitude cyclones as key potential areas for future research directions12.

In this study, we aim to bridge this research gap by establishing an objective and systematic understanding of the global contribution of midlatitude cyclones to near-surface extreme wind speeds, and how their geographical distribution and intensity may change in the future, using idealised simulations of a warmer climate obtained with a realistic high-resolution general circulation model. Towards this aim, we used the latest GFDL high-resolution, ≈ 50 km grid-spacing, General Circulation Model (GCM), with the Atmospheric Model version 4 (AM4)25 which has demonstrated its ability to realistically simulate cyclone events, as well as other mesoscale phenomena like tropical storms and mesoscale convective systems26. We conducted simulations with the AM4 model spanning the entire 71-year period from 1949 to 2019, for a present-day and a future warmer climate scenario with a homogeneous SST increase of 2 K, which is an idealised representation of anthropogenic climate change27. This simulation setup offers the advantage of conducting climate sensitivity experiments with a warmer SST while retaining the benefits of a state-of-the-art climate model, run, and analyzed, at the high resolution of ≈ 50 km grid-spacing. The analysis of the AM4 climate simulations is based on all year data.

Results

Present climate of near-surface extreme wind speeds and associated biases

The present-day climate of extreme near-surface wind speeds and associated biases relative to ERA5 data are depicted in Fig. 1. Figure 1a–b illustrate the median of the 99th-100th percentiles, corresponding to the 99.5th percentile, of the present-day daily maximum 10-m wind speeds (DMWSs) in the NH (Fig. 1a) and in the SH (Fig. 1b), to which we will refer as ν99.5 in the following text, in the interest of conciseness. It is evident that ν99.5 exceeds 25 m s−1 in both the NH and SH, particularly in regions corresponding to the storm tracks of the North Atlantic28, North Pacific, and Southern Ocean. However, in the Mediterranean region, where a storm track also exists, the median extreme wind speeds tend to be lower. The discrepancy found in the Mediterranean region may be attributed to the AM4 model resolution of approximately 50 km which may not adequately capture the intense wind speeds associated with the smaller scale of cyclones in this area29. Additionally, it may be necessary to adapt the tracking algorithm utilized in this study30 to effectively detect some of the most intense Mediterranean cyclones, known as medicanes, in future studies that specifically focus on this region13. Importantly, the results presented in Fig. 1a–b are skewed towards winter and spring values in both the hemispheres. In fact, when considering only the number of extreme wind speed events occurring during November to March in the NH and during April to September in the SH, they account, respectively, for 71% and 73% of the total extreme wind speed events, making the results in Fig. 1a–b “biased” towards winter and spring seasons.

Fig. 1: Map and associated biases of extreme wind speeds in AM4 present-day climate simulations.
figure 1

Spatial distribution of present-day climate simulations for extreme near-surface wind speeds characterised as the median of the 99th-100th percentiles daily maximum 10-m wind speeds (DMWS), corresponding to the 99.5th percentile ν99.5, in the a Northern Hemisphere and b Southern Hemisphere. Bias of present-day climate simulations ν99.5 relative to ERA5 in the c Northern Hemisphere and d Southern Hemisphere.

Full size image

Before analysing the response of extreme wind speeds to global warming, the major simulation biases of the present climate of extreme wind speeds simulated by the GFDL-AM4 model are presented. Figure 1c–d illustrate the present climate extreme wind speed bias relative to the European Centre for Medium-range Weather Forecasts (ECMWF) fifth generation hourly reanalysis (ERA5)31, computed as the difference between AM4 and ERA5 ν99.5. Over the ocean, the AM4 model overestimates the corresponding ERA5 extreme wind speeds up to ≈ 4 m s−1. This could be explained by the higher temporal frequency of the wind-speed timeseries used by AM4 to compute the extreme wind speeds compared to ERA5. In fact, AM4 extreme wind speeds are obtained using DMWS values computed over the daily wind values output every model timestep, 30 minutes, while ERA5 extreme wind speed values are obtained using DMWS computed over the daily hourly mean wind values. Moreover, although both ERA5 and AM4 use the Monin-Obukhov surface layer theory and stability profile functions to estimate the 10-m wind, ERA5 uses an exposure correction approach to make the postprocessed 10-m wind compatible with SYNOP observations. Over inhomogeneous terrain, where the “aerodynamic roughness length” in the model is adjusted to provide sufficient drag at the surface which is dominated by the rough elements, the ERA5 approach corrects the otherwise low area-averaged wind speed, making it comparable with SYNOP observations, which are often taken over an open terrain (as specified by the World Meteorological Organization).

Looking at the present climate biases over land, they are smaller than over the ocean, and contained within ± 2 m s−1, with the exception of regions characterised by high orography, such as Greenland or Antarctica, where AM4 underestimates extreme wind speeds by up to 8 m s−1. Although 10-m hourly winds are not assimilated over land32, it is plausible that the AM4 model coarser resolution of ≈ 50 km compared to the ERA5 resolution of ≈ 30 km, combined with the lack of an exposure correction method and an excessive orographic gravity wave drag in AM4, explains the large underestimation of AM4 extreme wind speeds over mountainous regions compared to ERA5 values. Therefore, care should be taken when assessing the AM4 future climate projections of extreme wind speeds over high orography regions such as Greenland and Antarctica.

Response of near-surface extreme wind speeds to global warming

The response of extreme wind speeds under the simulated idealised global-warming climate conditions exhibit worldwide spatial variability as depicted in Fig. 2. Figure 2a–b illustrates the projected changes in midlatitude extreme wind speeds under an idealised global warming scenario for both the NH and SH. The results are represented as percentage differences between the warmer and present-day climate simulations of ν99.5, normalised per sea surface temperature (SST) degree warming. In the NH (Fig. 2a), there is notable increase in extreme wind speeds ranging from 2% to 3% K−1 in regions such as the North Atlantic, Western Europe, the west coast of the United States (US), and Canada, spanning latitudes between 50 and 70∘N. Conversely, there is a reduction of up to 3% K−1 in latitudes between 30 and 50∘N. However, the North Pacific exhibits a uniform increase of approximately 2% in extreme wind speeds. The British Isles and Northwestern Europe exhibit the largest increase in extreme wind speeds, both in terms of percentage per SST degree warming (3% K−1) and magnitude (approximately 1.5 m s−1, not shown). This observed increase is similar in magnitude to that projected by the end of the 21st century in a previous study specifically focused on these regions33. In the SH (Fig. 2b), a more symmetric pattern of change emerges, with extreme wind speeds projected to increase by up to 3% K−1 (approximately 1.5 m s−1 in absolute value, not shown) along the entire length of Antarctica, spanning latitudes between 50 and 70∘S. Similarly to the NH, regions between 30 and 50∘S are expected to experience a decrease of up to 3% K−1 in extreme wind speeds, with the exception of the Southern Pacific Ocean region, which shows an increase of up to 2% K−1. To assess whether the difference in the medians ν99.5 between the present-day and warmer climate distributions resulted from the actual response to +2 K of warming, thus being statistically significant, rather than merely an outcome of random variations, we applied the non-parametric Mann-Whitney U test to the distributions of extreme DMWSs events in each grid cell for both present and warmer climate simulations. We consistently obtained Mann-Whitney p-values p < 0.05 everywhere, indicating the statistical significance of the response to an idealised +2 K SST climate scenario.

Fig. 2: Map of extreme wind speeds in AM4 idealised warmer climate simulations.
figure 2

Percentage change in the median of the 99th-100th percentiles daily maximum 10-m wind speeds (DMWS), corresponding to the 99.5th percentile ν99.5, per degree of SST warming for the a Northern Hemisphere and b Southern Hemisphere.

Full size image

These findings suggest a notable and statistically significant poleward migration of extreme wind speeds in both the NH and SH. The poleward migration is more pronounced and consistent in the SH compared to the NH. These results align with previous observations of the shift in midlatitude cyclone tracks under a warmer climate as reported in various studies6,13,15. It can be hypothesized that the changing global patterns of midlatitude cyclones play a significant role in driving the changes in extreme wind speeds.

Frequency distribution of cyclone-associated near-surface extreme wind speeds in both present-day and warmer climate simulations

To objectively and systematically evaluate the contribution of midlatitude cyclones to extreme wind speed events, characterized by DMWSs exceeding the 99th percentile threshold, an ad-hoc algorithm is employed in both the present-day and future warming climate simulations.

Figure 3 a presents box plots that summarize the global area-weighted percentage of extreme winds over each grid cell that occur within 750 km of a cyclone centre in both present-day and warmer climate simulations. The distributions are presented for each grid cell, partitioned by hemisphere and simulation type. In the present-day climate scenario, the median frequency of cyclone-associated wind speed events is comparable in magnitude between the two hemispheres, with 83% in the NH and 88% in the SH. However, the SH distribution exhibits a slightly longer tail towards lower frequencies of cyclone-associated extreme wind speeds compared to the NH. This distinction is likely attributed to the presence of Antarctica’s orography, which acts as a barrier shielding the interior landmass points from midlatitude cyclones, resulting in lower occurrence frequencies in the SH.

Fig. 3: Cyclone-associated DMWS frequency as aggregated statistics and spatial distribution for present-day and warmer climate.
figure 3

a Box plot of area-weighted cyclone-associated DMWS frequency exceeding the 99th percentile threshold in GFDL-AM4 model for present-day (blue) and warmer climate (red) simulations. Whiskers represent the 0.3 and 99.7% percentiles, box indicates interquartile range, and median, 25th, and 75th percentiles are marked. b Northern Hempisphere cyclone-associated DMWS occurrence, c Southern Hemisphere cyclone-associated DMWS occurrence, d difference between NH warmer and present-day cyclone-associated DMWS frequency, e difference between SH warmer and present-day cyclone-associated DMWS frequency.

Full size image

Spatially, the highest frequency of cyclone-associated extreme wind speed events is concentrated along the major midlatitude storm tracks in both hemispheres, as depicted in Fig. 3b, c. The North Atlantic, North Pacific, and SH storm tracks consistently exhibit between 80% and 100% of cyclone-associated extreme wind speed events, providing objective evidence for the influential role of midlatitude cyclones in driving wind extremes in the midlatitudes, as also demonstrated by a recent global analysis of compound wind and precipitation extremes using ERA534.

Under the warmer climate scenario, when considering the total number of midlatitude cyclone tracks obtained using the Hodges’ (1999) algorithm35, there is a decrease of approximately 4% compared to the present-day climate scenario. However, the examination of the aggregated area-weighted frequency distributions of cyclone-associated events (refer to Fig. 3a), indicates that the medians of their warmer climate distributions increase by approximately 2% in both hemispheres compared to the corresponding present-day values. Nevertheless, a more detailed analysis of the spatial distributions in Fig. 3d, e reveals notable changes. While the frequency of cyclone-associated extreme winds remains relatively constant in the central regions of the midlatitude storm tracks, it decreases closer to the tropical regions and increases towards the poles along the edges of the storm tracks.

These findings indicate a shift towards higher frequencies of cyclone-associated extreme wind speeds in poleward regions (in both hemispheres). In the NH (Fig. 3b–d), the largest shift (increases up to 10%) occur over inhabited areas such as Britain and Scandinavian countries. In the SH (Fig. 3c–e), a substantial shift, comparable in magnitude, (increases up to 8%) is observed in the seas and coastlines surrounding Antarctica. These increases are not uniform across all regions. For example, the Southern Indian Ocean (Fig. 3d) experiences decreases of up to 10% in the frequency of cyclone-associated extreme wind speed events, while the decreases are limited to approximately 2% in the Southern Pacific Ocean.

Poleward increase of cyclone-associated near-surface extreme wind speeds

Although, as discussed earlier, the overall poleward increase in the frequency of cyclone-associated extreme wind speed events suggests a shift in the cyclones tracks and intensity in both hemispheres, a further analysis of the changes in wind speed magnitudes associated with cyclones under a warmer climate is necessary. Towards this aim, we assessed the changes in the intensity of cyclone-associated extreme wind speeds under global warming by computing the percentage differences between the warmer climate and present-day climate simulations of cyclone-associated ν99.5, normalized per SST degree warming, as shown in Fig. 4.

Fig. 4: Map of cyclone-associated extreme wind speeds in AM4 idealised warmer climate simulations.
figure 4

Panels same as in Fig. 2a, b but considering only the median of the 99th-100th percentiles cyclone-associated DMWSs (ν99.5). a Northern Hemisphere and b Southern Hemisphere.

Full size image

Figure 4 shows a clear intensification of cyclone-associated extreme wind speeds around the poles. In the NH (Fig. 4a), the most significant intensification occurs over the North Pacific, the British Isles, and Northwestern Europe, with increases in cyclone-associated extreme wind speeds which reaches up to 3% per degree of SST warming (equivalent to approximately 1.5 m s−1 in absolute value, not shown) and are concentrated between latitudes 50∘N and 70∘N. Similarly, in the SH (Fig. 4b), the projected changes in cyclone-associated extreme wind speeds exhibit a more symmetrical pattern compared to the NH, with a roughly homogeneous increase in cyclone-associated extreme wind speeds of approximately 2.5% per degree of SST warming (localized between latitudes 50∘S and 70∘S) around the Antarctica landmass. In contrast, both hemispheres experience a decrease in the lower midlatitudes of up to 3% per degree of SST warming, between 30∘N and 50∘N or 30∘S and 50∘S. The Pacific ocean is the only region in both hemispheres that displays increases in cyclone-associated extreme wind speeds extending to the tropics. These findings mirror the results presented in Fig. 1c, d, indicating that the projected changes in extreme wind speed events under warmer climate conditions can be attributed to the changes in cyclone-associated extreme wind speed events. Additionally, statistical analysis of the distributions of the cyclone-associated extreme wind speeds at each model grid cell shows that they satisfy the Mann-Whitney U test (p < 0.05), indicating the statistical significance of the results presented in Fig. 4a, b.

To gain further insights into the physical characteristics of cyclones associated with extreme wind speed events, we examined the mean changes in cyclone track densities and intensities between the warmer and the present-day climate simulations. The results, illustrated in Fig. 5, reveal distinct patterns in terms of changes in cyclone intensity and track density.

Fig. 5: Map of changes in the midlatitude cyclones track density and intensity associated with extreme wind speeds under idealised warmer climate.
figure 5

Difference in midlatitude cyclone track mean intensity (associated with extreme wind speed events) between the warmer and the present-day climate AM4 simulations for the a Northern Hemisphere and the b Southern Hemisphere, overlaid present-day climate cyclone track intensity (black contours). Difference in midlatitude cyclone track density (associated with extreme wind speed events) between the warmer and the present-day climate AM4 simulations for the c Northern Hemisphere and the d Southern Hemisphere, overlaid present-day climate cyclone track density (black contours). Units of track densities are number of cyclones per 5∘ spherical cap per month.

Full size image

Figure 5a–b visually demonstrates that in the warmer climate simulation the cyclone track intensity of midlatitude cyclones associated with extreme wind speed events strengthens near the poles in both hemispheres and weakens as the distance from the poles increases. In the NH (Fig. 5a), the regions of the Atlantic Ocean north of the British Isles and Northern Europe, as well as the northern part of the Pacific Ocean, experience increases in cyclone intensity characterized by relative vorticity at 850 hPa (ξ850) of up to 0.5 × 10−5 s−1. Conversely, the southern parts of the North Atlantic and North Pacific exhibit an equal and opposite decrease in cyclone intensity, reaching values of up to − 0.5 × 10−5 s−1. Shifting our focus to the Southern Hemisphere (Fig. 5b), the changes in cyclone intensity under the warmer climate simulations demonstrate a more symmetric pattern. Poleward increases in cyclone intensity are observed around the Antarctic landmass between 60∘ and 70∘S, reaching maximum values of 0.5 × 10−5 s−1, which are comparable to those found in the Northern Hemisphere. Additionally, reductions in relative vorticity at 850 hPa (ξ850) away from the poles exhibit a symmetric distribution spanning the lower midlatitudes between 30∘ and 40∘S.

The comparison of Fig. 5a, b with c, d, which illustrates the changes in track densities of midlatitude cyclones associated with extreme wind speed events in the context of warmer climate conditions, reveals a robust spatial agreement with intensity changes, particularly around the poles. In the SH, the changes in track density of cyclones associated with extreme wind speeds exhibit a relatively homogeneous pattern, characterized by a nearly symmetrical increase in track density. In the NH, the changes in track density of cyclones associated with extreme wind speeds appear more varied. Specifically, the North Atlantic Storm track extends further into the British Isles and Northern Europe. On the other hand, the North Pacific storm track experiences a reduction in cyclone track density between the western coast of Canada and Japan within a narrow range of approximately 45∘N, but expands over the North Pacific in other areas. Notably, the decrease in track density along this narrow region corresponds to an increase in both cyclone intensity and cyclone-associated extreme wind speeds. This suggests that the intensification of extreme wind speeds is primarily driven by the cyclone intensity itself, as defined here by ξ850, rather than by changes in track density. Lastly, the results presented here are likely to differ across different seasons due to the seasonal variation in the distribution of track density36.

Discussion

Understanding changes in the pattern and magnitude of extreme wind speeds associated with midlatitude cyclones under a warming climate has great significance for the safety of lives, businesses, and infrastructure. Although it has long been known that midlatitude cyclones lead to extreme wind speeds at the surface, either through individual studies or storm composites5,37,38, a systematic investigation of their contribution to extreme near-surface wind speed events (also known as cyclone footprint) in both the NH and SH under present-day and warmer climate has been scarce12. Here, we investigated the contribution of midlatitude cyclones to changes in extreme wind speed events under an idealised future global warming scenario, obtained adding homogeneously 2 K in the SST to a present-day simulation covering the 71-year period of 1949–2019 using the state-of-the art high-resolution ≈ 50 km GFDL climate model AM4.

The results of our objective attribution of extreme wind speed events to midlatitude cyclones demonstrate that midlatitude cyclones account for the vast majority of extreme near-surface wind speed events (between 80% and 100% of the extreme wind speed events) in the regions of the major midlatitude storm tracks, such as the North Atlantic, North Pacific, and the Southern Hemisphere storm tracks. In these regions the median value of the extreme wind speeds consistently exceeds 25 m s−1, which corresponds to the storm-intensity winds and Beaufort scale 10, indicating a potential for significant hazards.

Although, under the idealized global warming scenario, the frequency of cyclone-associated extreme events is roughly unchanged and the number of midlatitude cyclones decreases by roughly 4%, we observed a migration of the cyclone-associated extreme wind speeds in frequency and intensity towards higher latitudes. The spatial distributions of extreme wind speeds simulated by the AM4 model, align well with the changes in the cyclone-associated extreme wind speeds, implying that changes in midlatitude cyclone track density and intensity are the primary drivers of changes in the midlatitude extreme wind patterns. Overall, extreme wind speeds increase by roughly 4% K−1 towards the poles, accompanied by an increase in midlatitude cyclone track density of up to 5 × 10−6 km−1 s−1 and cyclone intensity of up to 0.5 × 10−5s−1 in both the NH and SH. A positive relationship among poleward shift of midlatitude cyclone storm tracks, cyclone intensification, and increase in greenhouse gas emissions (indicative of warmer climate) was also highlighted by a recent assessment of CMIP6 model simulations conducted under various Shared Socio-economic Pathways (SSP1-26, SSP2-45, SSP3-70, and SSP5-85) with a resolution of ≈ 100 km, twice as coarse as the ≈ 50 km employed in our study, and in line with the results from previous CMIP5 assessments concerning the NH9, and the SH midlatitude storm tracks39. Although there is no consensus amongst scientists regarding the exact causes of the poleward shift of midlatitude storm tracks also found in our study, several key underlying physical mechanisms have been identified in assessments of CMIP5 and CMIP6 model ensembles40,41. These mechanisms include, but are not limited to, stronger upper-level winds, increased atmospheric water vapor40, and tropical warming41. However, further research is needed to establish the precise link between these mechanisms and the poleward shift of the cyclone-associated near-surface extreme wind speeds demonstrated in our study, while disentangling it from other relevant feedbacks such as the expected increase in static stability under a warmer climate. Overall, although the poleward shift of midlatitude cyclone tracks has been amply discussed in the literature, its extension to cyclone-associated near-surface extreme wind speeds, demonstrated in our study, represents an important result, for a twofold reason: (1) for its significance, being driven in our study solely by the global mean anthropogenic warming represented by the prescribed +2K SST, and (2) for the research methods used, which combines an intermediate complexity simulation framework with an Eulerian analysis of the cyclone-associated extreme wind speeds.

The regions most significantly affected, in our study, by the poleward migration of midlatitude cyclones and the resulting increase in cyclone-associated extreme wind speeds (up to 3%K−1) are the British Isles, Northwestern Europe, and the North American west coast. These results confirm, for the British Isles and Europe, the increase in cyclone-associated 850hPa winds found in CMIP5 model ensembles under the midrange and high-emission Representative Concentration Pathway RC4.5 and RC8.59. Notably, according to idealized warmer climate simulations conducted using the AM4 model in a different study25, the British Isles, Northwestern Europe, and the North American west coast are also projected to experience an increase by as much as 6% K−1 in the cyclone-associated extreme precipitation, defined as exceeding the climatological 99th percentile. These regions have already been identified at risk of compound wind and precipitation extremes driven by midlatitude-cyclones in the present climate34, and compound coastal flooding under warmer climate, the latter compound hazard frequency being projected to increase by ≈ 2.5 times than in the present-day climate for latitudes above 40∘42. Consequently, in light of the results of our study and of previous findings, the British Isles, Northwestern Europe, and the Western Coast of North America turn out as the inhabited areas most vulnerable to a heightened risk of compound wind, precipitation, and coastal flooding hazards under a warmer climate. This is in contrast to the SH midlatitudes, where the limited presence of populated areas reduces the potential impacts of increased frequency of climate extremes. Future research on climate projections of compound extreme wind speeds and precipitation hazards should prioritize the quantification of risks associated with midlatitude cyclones, taking into account the findings of our study, which specifically examines extreme wind speeds, recent results from future projections of compound precipitation and coastal flooding extremes based on CMIP simulations42, as well as present-day reanalysis-based analyses like ERA534.

When interpreting the estimations of the poleward-increase in cyclone-associated near-surface extreme wind speed presented in our study, two caveats have to be mentioned. Firstly, the use of uniform sea surface temperature (SST) warming may not fully capture the complex effects of regional SST warming patterns43. Regional SST patterns can directly influence the locations and intensities of cyclone tracks. As emphasized in recent studies44,45, incorporating SST warming patterns that increase SST gradients in key cyclogenesis regions, such as the Gulf of Mexico or the Pacific Ocean region influenced by the Kuroshio Current, has the potential to amplify the poleward shift and intensification of midlatitude cyclone tracks. Additionally, accounting for SST warming patterns may also result in regional changes in the occurrence of tropical cyclones transitioning to midlatitude cyclones and subtropical cyclones, which often extend well beyond the tropics6. Although our study filters out cyclones with a genesis region in the tropics (between ± 30∘), it is likely that the increase in extreme wind speeds observed in the Pacific Ocean, closer to the tropics in the warmer climate simulation, is influenced by a combination of an increased number of hybrid cyclones and a poleward shift of tropical storm genesis regions46. Secondly, the AM4 simulations employed in our study do not consider the anticipated decline in sea-ice cover throughout the 21st century47, resulting from a warmer climate. Consequently, keeping the sea-ice concentration fixed and uniformly adding 2 K to the SST to simulate warmer climate conditions might lead to an artificially enhanced zone of baroclinicity at the sea-ice edge, as the local SST gradients are strongly enhanced, possibly affecting the poleward shift of extreme wind speeds associated to cyclones. In fact, more realistically, in a future warmer climate, surface baroclinicity would likely decrease around the sea-ice edge, as polar regions warm more strongly than the tropics (an effect known as Arctic Amplification in the NH), likely leading to the weakening of the upper-level westerlies48,49, through the thermal wind relation. This may, in turn, potentially decrease the frequency and strength of midlatitude cyclones, which grow over baroclinic instabilities, particularly in the NH, where sea-ice loss is projected to be most pronounced. On the contrary, some existing literature suggests that incorporating reduced sea ice conditions in high-resolution climate simulations would likely amplify the poleward shift of cyclone-associated extreme wind speeds discussed here for the NH12,50. Moreover, a recent study analysing the effect of Arctic sea-ice loss on extratropical cyclones found changes to be small compared to the climate internal variability, though statistically significant, indicating that sea-ice loss represents just one of the many important aspects of the global warming response of midlatitude cyclones, which range from expected changes in atmospheric stability, to latent heating, and vertical stability51. Finally, due to the complexity of factors affecting midlatitude cyclones genesis and life-cycle, it remains challenging to project the impact of sea-ice changes on midlatitude cyclone storm tracks under warmer climate12,48, and further research would be needed to ascertain the role of sea-ice loss on the poleward intensification of cyclone-associated near surface extreme wind speeds.

Future research could focus on conducting more km-scale global climate simulations to obtain more precise regional estimates regarding the poleward migration of cyclone-associated extreme wind speeds52, particularly in densely populated areas of the British Isles, Northwestern Europe, and the West Coast of the US and Canada. Enhanced understanding in this regard would be invaluable for informing strategies to mitigate the potential impacts of extreme wind events in vulnerable regions53. Furthermore, these simulations could delve into the mesoscale features of midlatitude cyclones that contribute to surface-level extreme wind speed events, expanding upon the results of regional studies conducted for instance over the British Isles54. By examining both present-day and warmer-climate scenarios, these investigations could yield more detailed understanding of the underlying dynamics and atmospheric conditions that give rise to extreme wind speed events.

Methods

Model

The climate model used in the study is the latest high-resolution version of the GFDL atmospheric climate model (AM4). It employs a cubed-sphere topology for its atmospheric dynamical core with 192 × 192 grid boxes per cube face corresponding to roughly 50-km horizontal grid spacing. This GFDL atmospheric climate model versions, AM4, has been used for GFDL’s participation in the CMIP6 High Resolution Model Intercomparison Project18,25.

To investigate the projected changes in wind speeds under human-induced global warming the AM4 model was run for a present day and a future climate idealised global-warming simulation, with a homogeneous SST increase of 2K. Details of the present-day and future warmer climate simulations are now given:

  • Present-day simulation. The present-day simulation is a climate integration of the 71-year period of 1949–2019. The model is forced by the observed daily SSTs, sea ice concentrations, and radiative gases following the CMIP6 HighResMIP specifications18.

  • Future warmer climate simulation. The future warmer climate simulation is an idealized global-warming experiment that is the same 71-year climate integration as the present-day simulation but with observed daily SSTs uniformly increased by 2 K. The warming simulation is used to explore the possible change in midlatitude-cyclones-associated extreme wind speeds in a future warmer climate.

Midlatitude cyclones tracking

The cyclone tracking was carried out using Hodges’ algorithm (1995)35. To identify cyclones, the algorithm employs the 6-hourly relative vorticity at 850 hPa, ξ850, as an input. The use of relative vorticity at the 850 hPa level is advantageous for detecting cyclones in the lower troposphere compared to mean sea level pressure. It enables early identification of cyclone systems and is less influenced by large-scale background effects, as ξ850 focuses on the smaller-scale synoptic range compared to mean sea level pressure55.

Initially, the algorithm spectrally truncates the relative vorticity field to T42 resolution. This step is necessary since a high-resolution relative vorticity field ( ≈ 50 km grid-spacing in AM4 configurations used in our study) tends to exhibit significant noise, making it less suitable for tracking purposes. To remove the influence of planetary scale waves, the algorithm masked wavenumbers ≤5.

Subsequently, the algorithm performs the tracking process by first initialising a set of tracks from the feature points and then refining the tracks by minimizing a cost function, which aims to achieve a smoother ensemble of tracks. The positive and negative anomalies in ξ850, corresponding to cyclones in the Northern Hemisphere (NH) and the Southern Hemisphere (SH) respectively, were tracked separately. Importantly, the tracking operates on a spherical domain, eliminating the need for projections that may introduce systematic biases. This approach allowed for the analysis of global data or large spherical regions.

Furthermore, the algorithm retained only those tracks from the ensemble of tracks that persisted for at least 48 hours and traveled a minimum distance of 1000 km from the cyclone points of origin. This criterion ensured that only mobile, well-developed, and long-lived cyclones were included in the analysis. Additionally, each cyclone was required to exhibit a maximum vorticity of at least 1 × 10−5 s−1 (scaled by − 1 in the SH).

To derive spatial statistics, spherical kernel methods56 were employed on the track ensembles directly on the unit sphere. This approach was chosen to avoid the biases of grid box methods and map projections. Furthermore, the initial identification of cyclone centers before tracking is performed on a polar projection to mitigate any latitudinal bias introduced by the meridian convergence near the poles30,56.

It is important to note that the choice of a tracking method based on different meteorological fields, such as Mean Sea Level Pressure (MSLP), is likely to yield different cyclone track density patterns. This divergence primarily arises because various tracking methods do not consistently identify the same weather systems. For example, a comprehensive intercomparison study of different midlatitude cyclone detection and tracking algorithms that used both relative vorticity and MSLP57 revealed that cyclones tracked using a relative vorticity-based approach, similar to the method employed in our study, tend to be displaced poleward compared to those tracked using pressure-based methods. However, it is worth noting that when focusing exclusively on deep cyclones, which are the ones most likely to generate extreme near-surface wind speeds (the primary focus of this paper), the disparities in tracks produced by relative-vorticity and pressure-based methods exhibit better agreement, as demonstrated in the same intercomparison study57.

Determining cyclone-associated extreme-wind speeds

Extreme wind speeds were defined as those exceeding the 99th percentile of the daily maximum 10-m surface wind speed, separately for the present and warmer climate experiments, and thus characterize the upper (or extreme) tail of the wind speed distribution at each model grid cell over the globe in each climate simulation performed. Considering only Daily Maximum Wind Speeds (DMWS) allowed to select only independent extreme wind speed events. An objective approach was used to associate extreme wind speeds and midlatitude cyclones by defining each DMWS event that exceeded the 99th percentile threshold as cyclone-associated if it was within a 750-km radius of the nearest midlatitude cyclone track point. The midlatitude cyclone track point had to match the time of the DMWS event to the nearest 6-hour interval. We performed some sensitivity analysis to determine the robustness of our findings to the radius size and found that changes of ± 50 km led to changes of only roughly ± 1% in the frequency of cyclone-associated events. This method has been used by a few recent studies to objectively and systematically associate modelled58 and observed59 extreme wind speed events to midlatitude cyclones.