Response of the Southern Hemisphere extratropical cyclone climatology to climate intervention with stratospheric aerosol injection

Little is known about how climate intervention through stratospheric aerosol injection (SAI) may affect the climatology of the Southern Hemisphere extratropical cyclones under warming scenarios. To address this knowledge gap, we tracked extratropical cyclones from 2015 to 2099 in a set of projections of three international projects: the Assessing Responses and Impacts of Solar Climate Intervention on the Earth System with Stratospheric Aerosol Injection (ARISE), the Stratospheric Aerosol Geoengineering Large Ensemble (GLENS), and the Geoengineering Model Intercomparison Project (GeoMIP/G6sulfur). Comparisons were performed between no-SAI and SAI scenarios as well as between different timeslices and their reference period (2015–2024). Among the findings, both no-SAI and SAI project a decrease in cyclone frequency towards the end of the century although weaker under SAI scenarios. On the other hand, cyclones tend to be stronger under no-SAI scenarios while keeping their intensity more similar to the reference period under SAI scenarios. This means that under SAI scenarios the climatology of cyclones is less affected by global warming than under no-SAI. Other features of these systems, such as travelling distance, lifetime, and mean velocity show small differences between no-SAI and SAI scenarios and between reference and future periods.


Introduction
Extratropical cyclones are a response of the atmosphere to attain thermal equilibrium.They develop mainly due to near-surface horizontal temperature gradients, a predominant feature in mid-latitudes, and transport heat and water vapor towards the poles and cold and dry air towards the tropics (Peixoto and Oort 1992).Despite their important role in the climatic system, many cyclones can also cause extreme weather events such as intense precipitation, strong winds, and abrupt temperature changes.Over the ocean, air-sea momentum exchange is responsible for maritime agitation, which can lead to the occurrence of storm surges and giant waves, causing disruptions to navigation, operations on oil platforms, and the destruction of coastal ecosystems and infrastructure (Rocha et al 2004, Gramcianinov et al 2020, Faria et al 2023).
In extensive databases, such as reanalysis and model outputs, cyclone climatologies are obtained using objective methods based on the mean sea level pressure (MSLP), relative vorticity, or geopotential height (Walker et al 2020).For the Southern Hemisphere, centenary reanalysis (ERA20C) from 1900 to 2010 indicates a negative trend in the frequency of extratropical cyclones (Marrafon et al 2021).This signal is also projected in future warming scenarios by global and regional climate models (Bengtsson et al 2009, Michaelis et al 2017, Sinclair et al 2020, de Jesus et al 2021, Reboita et al 2021a, Priestley and Catto 2022).In contrast, the frequency of stronger cyclones (systems that reach central pressure lower than 980 hPa in some period of their lifecycle) increases in reanalyses (Pezza andAmbrizzi 2003, Reboita et al 2015) and in climate projections (Reboita et al 2015(Reboita et al , 2021a)).For instance, along the eastern coast of South America the frequency of explosive cyclones (i.e.cyclones with pressure dropping by ∼24 hPa/24 h) is projected to increase mainly near Uruguay and south of Brazil (Reboita et al 2021b), which can cause even more damage to coastal areas.
Despite significant progress in understanding the role of climate change in the climatology of extratropical cyclones, there is a lack of studies focusing on the impact of climate intervention on these systems.Climate intervention (also known as climate geoengineering) appears as an aggressive approach to reduce global warming since the climate system is intentionally modified.It involves deliberate manipulation of the physical, chemical, or biological processes of the Earth system with the intention of tempering the harmful effects of anthropogenic greenhouse gas emissions (AMS.American Meteorological Society 2022).Climate intervention encompasses two categories: (1) removing CO 2 from the atmosphere, known as carbon dioxide removal (CDR), and (2) reflecting sunlight, known as solar radiation modification (SRM) or solar geoengineering.One of SRM approaches involves the injection of sulfate aerosols (or their precursor sulfur dioxide-SO 2 ) into the stratosphere to enhance solar energy reflection; this approach is known in the literature as stratospheric aerosol injection (SAI).By reducing the amount of solar energy entering the climate system, the Earth's surface would cool on average.This concept draws from observations of past volcanic eruptions.For instance, the eruption of Mount Pinatubo in 1991 injected 20 million tons of SO 2 into the stratosphere, resulting in increased sunlight reflection and a globally averaged surface air temperature cooling of ∼0.3 • C for a period of 3 years (National Research Council 2015).
Both categories of climate intervention present risks and deserve much study before their real application (Robock et al 2009, Dykema et al 2014, National Research Council 2015, AMS. American Meteorological Society 2022, Ricke et al 2023).One way to assess the impact of climate intervention approaches on the climate system (basic variables and atmospheric systems) is through climate simulations/projections.Three international projects have conducted simulations/projections by using SAI and have made the data available: the Assessing Responses and Impacts of Solar Climate Intervention on the Earth System with Stratospheric Aerosol Injection (ARISE; Richter et al 2022), the Stratospheric Aerosol Geoengineering Large Ensemble (GLENS; Tilmes et al 2018), and the Geoengineering Model Intercomparison Project (GeoMIP) with the experiment G6sulfur; Visioni et al 2021).These experiments have some differences such as the simulated period, greenhouse gas scenarios, and the region where the particles are introduced in the stratosphere (Irvine et al 2016); more details about them are provided in section 2. Currently, the data of these three projects are largely used for assessing the potential impact of SAI on the climate system (e.g.Bednarz et al 2022, Camilloni et al 2022, Patel et al 2023).
While a general result is that SAI could effectively limit global warming (Moore et al 2015, Irvine et al 2019, Krishnamohan et al 2019), the impacts on various aspects of the climate system, such as in water balance and in the lifecycle of the atmospheric systems, remain unclear due to the different stratospheric processes and formulations of SO 2 injection in climate models (Kravitz et al 2014, Jiang et al 2019).For instance, Ricke et al (2023) reported that 'The expected hydrological effects of reducing insolation are among the most uncertain and consequential impacts of solar geoengineering' .
In terms of atmospheric systems without considering SAI, Chand et al (2022) found a negative trend in the frequency of tropical cyclones across all ocean basins using centenary reanalysis.This trend is also consistent in climate projections under warming scenarios, along with an increase in the tropical cyclone's intensity (Vecchi andSoden 2007, Walsh et al 2015).Under SAI scenarios, studies such as that of Jones et al (2017) have shown that aerosol enhancements confined to a single hemisphere could effectively modulate the North Atlantic tropical cyclone activity.For instance, sulfate SAI in the Southern Hemisphere would enhance tropical cyclone frequency relative to global aerosol injection, and vice-versa for injection in the Northern Hemisphere.
Specifically for extratropical cyclones, the authors do not know of studies applying the identification and tracking of individual cyclones aiming to compare scenarios with and without SAI.Until this date, the only two published studies on extratropical cyclones analyzed environmental conditions: one based on mean available potential energy (Gertler et al 2020) and the other on Rossby wave packets (Karami et al 2020).Gertler et al (2020) investigated scenarios with continued preindustrial conditions, 4xCO 2 , and 4xCO 2 plus SAI from GeoMIP.Under 4xCO 2 , storm tracks in the Northern Hemisphere are projected to weaken somewhat and greatly strengthen in the Southern Hemisphere.On the other hand, under the climate intervention scenario the storm tracks weaken in both hemispheres, but in the Northern Hemisphere the weakening is comparable to that from 4xCO 2 .Karami et al (2020) investigated the storm track response to the RCP8.5 scenario and to this same scenario plus sulfate SAI over the Middle East and North Africa region between 2050 and 2070.They compared GLENS projections with the present climate, and the main findings are: (a) increasing greenhouse gas concentrations result in the northward (poleward) shift of the storm tracks in all seasons, (b) under SAI scenario, there is a partial offset of the poleward shift of the storm tracks seen in the RCP8.5, consequently contributing to reducing the precipitation in the study area.
As at mid-latitudes the weather is primarily controlled by the development of extratropical cyclones and their associated fronts (Catto andPfahl 2013, Eisenstein et al 2023), it is crucial to explore the consequences of SAI on the climatology of these systems under warming scenario using a different approach from the previously described studies.Hence, this study aims to address the existing research gap by assessing how the main features of extratropical cyclone's climatology (frequency, intensity, trajectory, etc) over the Southern Hemisphere could change in the future under SAI scenarios provided by ARISE, GLENS, and GeoMIP/G6sulfur projects.

Study area and data
The study area encompasses the latitudes southern 20S to avoid the inclusion of tropical cyclones in the climatology (Reboita et al 2015).Extratropical cyclones are identified using MSLP at every 6 h (0000, 0600, 1200 and 1800 UTC) from three climate modeling projects: ARISE, GLENS, and GeoMIP/G6sulfur (table 1).
ARISE projections were carried out with CESM2 global climate model (GCM), GLENS with CESM1 and GeoMIP/G6sulfur with MPI-ESM1-2-LR (table 1).From each GCM we obtained three members without SO 2 SAI (hereafter called no-SAI) and with SO 2 SAI (hereafter called SAI).As these projections are well-documented in the literature, just a summary of their main information (such as number of members and design of the experiments) is provided here.Considering the no-SAI projections, they follow different emission pathways, i.e.ARISE is under SSP2-4.5,GLENS is under RCP8.5 and GeoMIP is under SSP5-8.5 (hereafter ARISE SSP2-4.5 , GLENS RCP8.5 and GeoMIP SSP5-8.5 ).Although SAI projections in ARISE are named by ARISE-SAI-1.5 and in GeoMIP we are just using the G6sulfur experiment, in the present study, for brevity, we call these dataset only by ARISE SAI , GLENS SAI and GeoMIP SAI .In the three projects, SAI projections consider the SO 2 injection into the lower stratosphere at four off-equatorial locations (30 • S, 15 • S, 15 • N, and 30 • N) in ARISE SAI and GLENS SAI , and within a range of 10 • N and 10 • S across the single longitude band of 0 • in GeoMIP-SAI.Injection amounts at each latitude in the three experiments are controlled by a feedback algorithm, which aims to maintain the global mean surface temperature and its equator-to-pole and inter-hemispheric gradients at the baseline levels (Kravitz et al 2017).In ARISE SAI the baseline period was defined as the 2020-2039 mean, corresponding to the likely period when the real world will reach 1.5 K above pre-industrial conditions (Tebaldi et al 2021, MacMartin et al 2022).In GLENS SAI , the baseline period used in the feedback algorithm was 2010-2030 mean.
As the projects differ in relation to the greenhouse gas emission scenarios (table 1), it leads to a different magnitude of SAI (Richter et al 2022), and also a distinct spatial distribution of the aerosols in the simulations (Bednarz et al 2022, Fasullo andRichter 2022).For instance, Bednarz et al (2022) compared the distribution of SO 2 injections in ARISE SAI and GLENS SAI and found that GLENS SAI has the largest concentrations of sulfate in the North Hemisphere tropics while ARISE SAI in the Southern Hemisphere tropics (the physical explanation for these differences are discussed in Fasullo and Richter 2022).Bednarz et al (2022) also highlight that larger injection rates are needed in GLENS SAI to reach the same amount of global cooling as in ARISE SAI or to offset the end of the century RCP8.5 scenario.
All GCMs provide data with horizontal resolution of 1.25 • longitude ×0.9 • latitude, except MPI-ESM-LR (1.875 o longitude × 1.85 o latitude).We highlight that one limitation of this study is the availability of projections with 6-hour frequency needed to track cyclones.The data used here were the ones available when this study started.Since the focus of this study is on extratropical cyclones, which are synoptic systems (horizontal dimension on the order of 10 3 km), the horizontal resolution of the datasets do not need to be high.Thus, all projections were interpolated to 1.5 • longitude ×1.5 • latitude using the bi-linear interpolation method (Wahab 2017, Cerlini et al 2020).The ensemble mean for each model will be used since each project has considerable differences.Hence, it can be expected that the different projects would lead to differences in the spatial pattern of the extratropical cyclones' characteristics.

Extratropical cyclone tracking
Cyclones were identified and tracked using 6 hourly MSLP data with an objective method (algorithm) developed by Murray and Simmonds (1991a, 199b) and updated by Simmonds and Murray (1999) and Simmonds et al (1999).This algorithm has demonstrated reliable results in studies of extratropical cyclones over the Southern Hemisphere (Pezza and Ambrizzi 2003, Neu et al 2013, Reboita et al 2015, Grieger et al 2018).
Initially, the algorithm interpolates the MSLP from a regular (latitude-longitude) grid to a polar stereographic grid centered on the South Pole using the bicubic spline method, which eliminates anisotropy (it ensures that the grid resolution is uniform in all directions, mainly in the pole, Simmonds et al 2003).Next, the Laplacian of pressure (∇ 2 p) for each grid point is calculated.Grid points candidate to be a cyclone are identified where there is a local maximum of ∇ 2 p (which is associated with the minimum pressure) compared to that of the surrounding eight grid points.This process is carried out for all timesteps, and only systems with ∇ 2 p exceeding 0.2 hPa (lat) −2 are considered for the following analyses (Simmonds and Murray 1999).Once the algorithm has identified the candidate grid points to be cyclones in all timesteps, it is necessary to connect these points over a sequence of timesteps to track the systems.This procedure comprises three stages (Simmonds et al 1999): (a) predicting the subsequent position of each low-pressure center, (b) calculating the probability of an identification between the predicted cyclone and each cyclone identified at the new timestep (identified with ∇ 2 p), and (c) defining the position of the minimum of pressure in the new timestep based on the highest probability of association obtained in stage (b).In summary, the tracking procedure is based on projecting cyclone positions from one analysis time to the next and comparing the projected positions with those of the cyclone analysis at the new time (Simmonds and Murray 1999).
The algorithm provides the central pressure and ∇ 2 p for each timestep of a cyclone's trajectory (latitude and longitude).The ∇ 2 p (calculated between the center of the system and the neighborhood) can be taken as measure of the strength of the cyclone, and values greater than 0.7 hPa ( • lat) −2 are classified as strong systems, while values between 0.7 and 0.2 hPa ( • lat) −2 are considered weak (Simmonds and Murray 1999).Knowing the cyclone trajectories, the algorithm is able to create a grid with some statistical quantities computed over different time scales (monthly, seasonal, yearly etc) as specified by the user.These statistics are trajectory density (SD), central pressure (CP), radius (R0), and depth (DP) of cyclones.The SD corresponds to the normalized number of systems passing through a given area, which is calculated by summing contributions from all sampled positions (recorded along the tracks) and normalizing by an area of 10 3 (degrees latitude) 2 .CP represents the minimum pressure at the center of the cyclones, R0 indicates the distance between the cyclone center and the location where ∇ 2 p = 0, and DP is also a measure of cyclone strength.Although there is an expression to compute DP (see Simmonds et al 2003), it can be understood as the MSLP difference between the center and the region of the system with ∇ 2 p = 0 (cyclone external border); the values given by this variable are positive.Further details about these quantities are provided in Lim and Simmonds (2002).In this study, we computed the statistics on an annual basis.

Analyses
Extratropical cyclones were identified in each member of ARISE, GLENS, and GeoMIP/G6sulfur no-SAI and SAI projections.Climatologies were calculated using only cyclones with a lifetime equal to or greater than 24 h, and presented in terms of ensemble mean.Cyclone frequency is defined as the number of systems per month, season (DJF, MAM, JJA and SON) and year.
Trends and their statistical significance (α = 0.05), using Sen's slope and Mann-Kendall test (Mann 1945, Kendall 1975), respectively, were calculated for the annual time series (2015-2099) of cyclone frequency, initial pressure, minimum pressure along the lifecycle, lifetime, travelling distance, and mean velocity.The t-test at the 0.05 significance level (Wilks 2020) was conducted to determine whether differences exist between the averages of the no-SAI and SAI scenarios at the same timeslice.
As shown in table 1, not all projects have data before 2015.Hence, we considered the period 2015-2024 from no-SAI projections as the reference period.This allows us to analyze the difference between the future timeslices (2040-2059, and 2080-2099) and the current period (2015-2024).In addition, the differences between the no-SAI and SAI scenarios are analyzed for annual mean features of the cyclones and displayed in maps.In these maps, significance statistical tests for mean difference are not included due to the weakness of the tests for cyclone's properties on the grid, as these systems have high variability in space and time (Pezza et al 2008, 2012, Catto et al 2011, Reboita et al 2015, Gentile et al 2023).

Trends
The annual frequency of extratropical cyclones over the Southern Hemisphere from 2015 to 2099 in each ensemble is depicted in figure 1(a).Additionally, to provide a view of the spread among the members of each project, the minimum and maximum annual frequency identified for no-SAI and SAI scenario members are shown.Up to 2050, the ensembles do not indicate a large difference in the annual frequency of cyclones between the no-SAI and SAI scenarios.However, from the 2050-decade, SAI scenarios indicate a higher number of systems (figure 1(a)), contributing to a smoother negative trend, and even positive one in GLENS SAI , when compared with that from no-SAI scenarios (table 2).The negative trends under no-SAI scenarios are consistent with the findings in the literature (Bengtsson et al 2009, de Jesus et al 2021, Reboita et al 2021a, Priestley and Catto 2022, Xu et al 2023).Table 2 reveals that, except for ARISE SAI and GLENS SAI projections, all the others exhibit statistically significant trends.In general, under the SAI scenarios, the frequency of extratropical cyclones is higher than under no-SAI scenarios and also higher than in the reference period (2015)(2016)(2017)(2018)(2019)(2020)(2021)(2022)(2023)(2024), except in GeoMIP SAI (table 3).When the t-test is applied to identify whether the averages between the no-SAI and SAI scenarios at the same timeslice are statistically different, most timeslices and projects present statistically significant differences except for GeoMIP SSP5-8.5/SAI in 2060-2069 and 2080-2089 (table 3).
In future warming scenarios, the decrease in the frequency of extratropical cyclones is related to many interacting processes resulting in a complex picture.These include tropical upper-tropospheric warming (Kumar et al 2022), which leads to an expansion of the Hadley cell and, consequently, a poleward expansion of the regions with higher MSLP and subtropical anticyclones (Reboita et al 2019); polar near-surface warming, which leads to the weakening of the horizontal temperature gradients and, consequently, baroclinicity in mid-latitudes (Frederiksen et al 2016) resulting in the poleward migration of the storm tracks; increasing the amplitude of large waves and decreasing the amplitude of short waves (synoptic waves; Schemm and Röthlisberger 2024), further negatively affecting near-surface cyclogenesis.It is suggested that with the decrease in global warming caused by SAI, these processes will undergo fewer changes and consequently affect cyclones less.However, additional investigation of these mechanisms is necessary and is beyond the scope of this study.
The trend of the annual mean of the central MSLP during cyclogenesis and the minimum pressure (cyclone deepest phase across its lifecycle) are presented in figures 1(b) and (c), respectively.In both figures, the no-SAI scenarios project a negative and statistically significant trend (table 2), indicating that cyclones will be deeper in the future since lower central pressure is the main indicator of more intense cyclones.For   4 and 5 also indicate that in all timeslices cyclones are weaker under the SAI compared to no-SAI scenarios (in other words, MSLP is higher in SAI scenarios) and the differences are statistically significant.This is also noted when comparing the timeslices of SAI scenarios with the reference period; only GLENS SAI projects systems slightly deeper than in the reference period.It appears controversial that with a decrease in baroclinicity in warming scenarios, extratropical cyclones can exhibit greater intensity.However, as shown in the literature, this is a consequence of higher moisture availability contributing to diabatic processes in the cyclone's environment (Catto et al 2019, Kodama et al 2019, Sinclair et al 2020, Reboita et al 2021b).
While there is a clear signal regarding future trends for the aforementioned variables, there are more uncertainties concerning traveling distance.While ARISE SSP2-4.5 and GLENS RCP8.5 scenarios project a negative trend for the travelling distance, GeoMIP SSP5-8.5 projects a positive one.On the other hand, ARISE SAI and GeoMIP SAI have opposite trends compared with no-SAI scenarios, and GLENS SAI projects a more intense negative trend than under no-SAI.Only ARISE SSP2-4.5/SAIdoes not have a statistically significant trend (table 2).As shown in table 6, cyclone's travel distances range between 3000 and 3300 km, indicating a small mean difference of only 300 km (10%) in the trajectory of the cyclones.Nevertheless, the differences are statistically significant, except in 2060-2069 for ARISE SSP2-4.5/SAI and in 2070-2079 for GeoMIP SSP8-8.5/SAI(table 6).
Extratropical cyclones exhibit a negative trend in their lifetime, but that does not exceed ∼4 h (4% of their total duration) between the reference period and the end of the century in both no-SAI and SAI scenarios (figure 1(e)).Despite this small value, the trends are statistically significant, except for the ARISE SSP2-4.5/SAIscenarios (table 2).Under SAI scenarios although cyclones are less deep (table 5) their duration seems not to be affected (table 7).
As mean velocity is a relation between traveling distance and lifetime, the small changes projected for both variables throughout the future (tables 6 and 7) result in small changes in mean velocity (figure 1(f) and ).The maximum difference between SAI and no-SAI scenarios is 0.5 m s −1 in GLENS RCP8.5/SAI in 2090-2099 (table 8), which implies slower cyclones under SAI scenario.Although trends are not visually apparent in figure 1(f), the calculated trend slope reveals a positive and statistically significant trend for all datasets except for the ARISE SAI , which shows no trend, and the GLENS SAI , which exhibits a negative and significant trend (table 2).In table 8, the t-test for average differences between the scenarios only indicated no significance for ARISE SSP2-4.5/SAI .Overall, the slight changes projected for travelling distance, lifecycle, and mean velocity until the end of the century under no-SAI scenarios are consistent with findings from studies, such as Reboita et al (2021b) and Sinclair et al (2020).

Annual cycle
The annual cycle of extratropical cyclones in both scenarios considering two periods (2040-2059 and 2080-2099) is shown in figure 2. There is a high frequency of cyclones between May and August, which is the period with higher baroclinicity in the Southern Hemisphere (Holton 2004, Frederiksen et al 2016) and an important cyclogenesis driver.In December, the frequency of cyclones reaches a minimum, which is followed by another one in February.When the ratio of cyclones per day for each month is computed the minimum in February is smoothed (figure not shown).Therefore, the decrease in the cyclone frequency from January to February is due to February having fewer days (28 or 29), which affects the count of systems.These results are consistent with climatologies obtained for present and future climates (Hoskins and Hodges 2005, Reboita et al 2015, Marrafon et al 2021, 2022).Some of the results from figure 1 are also evident in figure 2: a decrease in the frequency of cyclones towards the end of the century, but with a weaker decrease under SAI scenarios (figure 2(b)).

Spatial pattern
In this section, the spatial pattern of extratropical cyclones characteristics (figures 3-6) is presented in terms of comparisons between the future and the reference period, as well as between no-SAI and SAI scenarios.
The future timeslices under no-SAI scenarios indicate a decrease in the SD frequency, mainly in mid-latitudes, and an increase around Antarctica compared to the reference period (figures 3(a), (c), (g), (i) and (m)), which is known in the literature as the poleward shift of the storm tracks under global warming scenarios (Mbengue andSchneider 2013, Chemke 2022).In GeoMIP SSP5-8.5 (figures 3(a) and (c)) the decrease is more pronounced than in the other datasets near the continents (southeastern South America, and southern Africa and Australia).These patterns become stronger towards the end of the century (figure 3(c)).SAI scenarios also project a decrease in SD compared to the reference period, but this decrease is lower than that under no-SAI scenarios (figures 3(b), (d), (h), (j) and (n)), which compensates for the effect of global warming.This is clearer in the difference between SAI and no-SAI scenarios, where positive differences (indicative of higher SD under SAI scenarios) predominate mainly around Antarctica and in the latitudes of eastern Australia (figures 3(e), (f), (k), (l) and (o)).GLENS SAI shows an increase in the SD over the South Pacific compared to the reference period (figures 3(h) and (j)).This signal is weak in GeoMIP SAI and only appears in eastern Australia.Hence, the positive SD in GLENS is in general opposite between SAI and no-SAI scenarios, therefore affecting patterns of the other analyzed variables.This different signal in  GLENS SAI might be related with the spreading of aerosols in this projection, which has largest concentrations of sulfate in the tropical band of the Northern Hemisphere (Richter et al 2022), and their response to the atmospheric circulation.In this SAI scenario, Bednarz et al (2022) found a strengthening of the stratospheric zonal winds that extends downwards to the troposphere, resulting in a poleward shift of the eddy-driven jet and sea-level pressure anomalies, corresponding to the positive phase of the Southern Annular Mode (SAM).According to Reboita et al (2015), SAM positive phase is related with a tri-pole in the spatial distribution of cyclones: higher frequency near Antarctica and northward 45S and lower frequency between these two bands.On the other hand, Bednarz et al (2022) also indicated an opposite response in ARISE SAI (which has largest concentration of aerosol in the tropics of the Southern Hemisphere), that is, an equatorward shift of the eddy-driven jet and a sea-level pressure response resembling a negative phase of SAM that can make difficult the occurrence of cyclones northward 45S (Reboita et al 2015).
Figure 1(c) showed a general view of the MSLP trends in the cyclone's center in the Southern Hemisphere, but in a spatial analysis not the whole hemisphere may present the same CP trend signal.Indeed, both no-SAI (figures 4(a), (c), (g), (i), (m)) and SAI (figures 4(b), (d), (h), (j), (n)) scenarios project stronger systems southern 50S, i.e. towards Antarctica (negative difference in the figures), and weaker in mid-latitudes and near the continental coasts (positive difference).However, there is a difference in SAI compared to no-SAI scenarios: while the increase in intensity of the systems (negative difference) is lower under SAI than under no-SAI scenarios, the decrease in the intensity (positive difference), in general, is higher.This means that, on average, SAI scenarios project weaker systems.Of course there are differences among the projects.For instance, GLENS SAI projects lower MSLP over the Pacific Ocean than GLENS RCP8.5I , and also in comparison with GeoMIP SSP5-8.5/SAI .This is a consequence of higher frequency of cyclones projected by GLENS RCP8.5SAI over this ocean, which impacts the MSLP (figures 4(e), (f), (k), (l) and (o)).
As cyclones are perturbations superimposed on a background of the global pressure field, which is characterized by a MSLP decrease from lower to higher latitudes (Sinclair 1994), it is expected cyclones with higher CP (weaker systems) in lower latitudes than in higher latitudes.Due to this fact, the real intensity of the cyclone can be masked when CP is analyzed.A more realistic measure of cyclone intensity is obtained through DP.It should be noted that weaker (stronger) cyclones have lower (higher) DP values; therefore, the DP differences in figure 5 will have opposite signals to CP in figure 4. Despite the previous consideration, when comparing both variables under no-SAI and SAI scenarios, we find good agreement between the spatial distribution of regions with more intense systems and weaker ones.But, a difference occurs between DP and CP over the South Pacific: the CP field shows a larger area with weaker cyclones than DP (for instance in figures 5(a), (c), (g), (i) and (m)).This may be related to the fact that the global pattern of pressure is projected to change in the future, as indicated by various studies suggesting a polar amplification of the Hadley cell, leading to higher pressures towards mid-latitudes (Reboita et al 2019 and their references).So, this background is being added to the cyclones environment leading to systems with higher MSLP values.However, in terms of real intensity, there are no great changes in DP over the South Pacific.In a nutshell, the DP confirms that in both future timeslices, extratropical cyclones can be weaker under SAI than no-SAI scenarios (figures 5(e), (f), (k), (l) and (o)).
No-SAI and SAI scenarios practically do not indicate great changes in the R0 in the areas with an increase in DP near Antarctica, but project a decrease in mid-to-low latitudes for the period 2040-2059 (figure 6).A signal of increasing R0 towards Antarctica and near the continents is projected for the period 2080-2099, mainly in southeastern South America, meaning slightly bigger cyclones in future.In general, SAI scenarios project cyclones with slightly lower R0 than the no-SAI scenarios (figures 6(e), (f), (k), (l) and (o)).

Conclusions
This study compared the impact of global warming in scenarios with and without stratospheric aerosol injection (SAI and no-SAI, respectively) on extratropical cyclone characteristics over the Southern Hemisphere using projections from three projects: ARISE, GLENS, and GeoMIP/G6sulfur.Despite differences in the projection configuration of the three projects, both no-SAI and SAI scenarios indicate trends in cyclone climatology in the same direction, giving confidence to the results.The main features in the extratropical cyclone climatology are summarized as follows: Frequency and annual cycle: both no-SAI and SAI scenarios for present and future climate exhibit an annual cycle of extratropical cyclone frequency in phase with that described in the literature, with winter being the most cyclogenetic season.The frequency of cyclones decreases towards the end of the century, but with a weaker decrease under SAI scenarios.Therefore, SAI scenarios compensate for the lower frequency of cyclones in the global warming scenario (no-SAI).
Intensity: the intensity of extratropical cyclones under no-SAI and SAI scenarios obtained from different approaches (initial pressure, minimum pressure during the lifecycle, and depth) indicates that cyclones will be stronger at the end of the century.However, under SAI scenarios cyclones are less intense compared with no-SAI, highlighting that under SAI cyclones have more similar intensity to the reference period.
Traveling distance, lifecycle and mean velocity: there are small differences between the reference period and the end of the century under both no-SAI and SAI scenarios.In addition, the differences between these scenarios are also minimal in the whole studied period.
Spatial patterns: although the three projects (ARISE-SAI, GLENS, and GeoMIP/G6sulfur) under both no-SAI and SAI scenarios have some differences in the spatial patterns of cyclone features (trajectory density, central pressure, depth, and radius), they agree that the extratropical cyclones are decreasing (increasing) in density and intensity in mid-latitudes (towards Antarctica), which is a poleward shift of the storm tracks.In addition, they show that the radius of the cyclones can be smaller mainly in mid-to-low latitudes and bigger around Antarctica.
As this is the first study to address SAI in cyclones' climatology-from the tracking perspective-there are no other studies for comparison.However, all described features concerning the no-SAI are consistent with the literature (as shown by the references throughout the text) and they bring important information to the decision makers since coastal areas, such as southeastern South America and New Zealand, are vulnerable to more intense systems in the next decades (e.g.floods, heavy rains, etc.).In addition, there is a consistent indication that the cyclone's climatology is less affected by global warming when SAI is considered.
In a subsequent study, we will evaluate cyclone synoptic patterns through composite analysis to elucidate the physical differences of these systems between no-SAI and SAI scenarios, which provide valuable insights in understanding the impacts of SAI on extratropical cyclones: vital systems to the thermal equilibrium of the planet.

Figure 1 .
Figure 1.Extratropical cyclone trends over the Southern Hemisphere under no-SAI and SAI scenarios from 2015 to 2099: (a) annual frequency (number per year), (b) initial pressure (registered in the cyclogenesis, hPa), (c) minimum pressure during the whole lifecycle (hPa), (d) travelling distance (km), (e) lifetime (hours), and (f) mean velocity (m s −1 ).Bold lines indicate the ensemble mean and light lines indicate the minimum and maximum values obtained by the members of each ensemble.

Figure 2 .
Figure 2. Annual cycle of the frequency of extratropical cyclones under no-SAI and SAI scenarios considering (a) 2040-2059 and (b) 2080-2099 periods.

Figure 3 .
Figure 3. Cyclone tracking density (SD) projections for each project (GeoMIP/G6sulfur, GLENS and ARISE).The first to fourth columns indicate the difference between each timeslices and the reference period for (a), (c), (g), (i), (m) no-SAI and (b), (d), (h), (j), (n) SAI scenarios, and the fifth and sixth columns indicate the difference between SAI and no-SAI scenarios.

Figure 4 .
Figure 4. Similar to figure 3 but for cyclone central pressure (CP in hPa).

Figure 5 .
Figure 5. Similar to figure 3 but for cyclone depth (DP in hPa).

Figure 6 .
Figure 6.Similar to figure 3 but for cyclone radius (R0 in km).

Table 1 .
Mean characteristics of the three projects.The used period in the study is indicated by * .In the rows 'Ensemble size' the numbers in brackets indicate the members available online.For instance, GLENS no-SAI has 20 members from 2010 to 2030 but only the realizations 001, 002 and 003 are available.

Table 2 .
Slope of the trends calculated for the annual time series (slope year −1 ) of each ensemble shown in figure1.Trends statistically significant at the level of α = 0.05 are highlighted in bold.
stronger for initial pressure (table 4) and minimum pressure (table 5), respectively, than in the present climate.Under SAI scenarios, the ARISE SAI and GLENS SAI project a positive trend in MSLP (figures 1(b) and (c)), corresponding to weaker systems in the future (table 2).Only GeoMIP SAI projects a negative trend, but it has a smoother slope compared to the no-SAI scenario (table 2).Tables

Table 3 .
Mean annual frequency of extratropical cyclones over the Southern Hemisphere obtained by the ensemble of no-SAI and SAI scenarios from ARISE-SAI, GLENS and GeoMIP for different timeslices.The asterisk ( * ) indicates that the difference between no-SAI and SAI averages in the same timeslice is statistically significant at the level of 0.05.

Table 4 .
Similar to table 3 but for initial pressure (hPa).

Table 5 .
Similar to table 3 but for minimum pressure along the cyclone's lifetime (hPa).

Table 6 .
Similar to table 3 but for travelling distance (km).

Table 7 .
Similar to table 3 but for cyclone's lifetime (hours).

Table 8 .
Similar to table 3 but for mean velocity (m s −1 ).