Present‐Day Regional Antarctic Sea Ice Response to Extratropical Cyclones

Both atmospheric warming and poleward moisture transport increase the likelihood of sea ice surface melt. In the Southern Hemisphere, short‐lived extratropical cyclones (ETCs) are responsible for a bulk of total heat and moisture transport toward high latitudes. Although these storms form ubiquitously in the midlatitudes, moisture availability and temperature characteristics vary by source region. In this study, we assess atmospheric, oceanic, and sea ice concentration (SIC) anomalies associated with austral winter ETCs over different Antarctic regions using ERA5 reanalysis data. Between 1990 and 2019, we find a total of 514 ETCs, with greater storm frequency in the eastern hemisphere groups. Compared to the climatology, sea ice melts (grows) behind the warm (cold) front of each system and is negatively correlated with atmospheric poleward moisture transport, temperature, meridional winds, and sea surface temperature for all ETCs. We find that Bellingshausen storms move moisture and warm air furthest poleward over their lifespan. However, East Weddell and East Antarctic ETCs are responsible for greater absolute poleward moisture transport than Bellingshausen and Ross systems. More intense ETCs correspond to greater SIC through Day 1, suggesting that SIC impacts ETC strength, regardless of ETC region. From cyclogenesis to cyclolysis, sea ice extent declines underneath composite ETCs, trends are generally not significant. Overall, while sea ice response produced by ETC‐induced atmospheric and oceanic changes varies regionally, the long‐term impacts of ETCs on regional sea ice are negligible over the study period.

• Bellingshausen extratropical cyclones (ETCs) induce greater atmospheric moisture transport and warming at high latitudes than other cyclone groups • Sea ice concentration (SIC) change is best related to 2-m temperature anomalies for all cyclone locations • ETC intensity corresponds to greater SIC between cyclogenesis and day 1 Supporting Information: Supporting Information may be found in the online version of this article.

10.1029/2023JD038914
2 of 20 pattern changed between 2014 and 2018, during which time Antarctic sea ice loss exceeded three decade's worth of total Arctic sea ice reduction (e.g., Eayrs et al., 2021).
Short-and long-term changes in sea ice result from numerous atmospheric and oceanic processes.In the Antarctic, one of the main drivers of SIE change is the polar jet (Maksym et al., 2012).Westerly winds induce sea ice edge advancement (Eayrs et al., 2019) and produce polynyas that enhance atmosphere-ocean energy exchange and cause refreezing during austral autumn and winter.As the sea ice edge moves poleward from spring into summer, easterly winds enhance its retreat.Polar jet strength, position, and waviness are modulated by teleconnections like the Southern Annular Mode (SAM; Sen Gupta & England, 2006;Screen et al., 2018;D. Yang et al., 2020), El Niño-Southern Oscillation (ENSO; Turner, 2004;S. Yang et al., 2018), the Amundsen Sea Low (ASL; Raphael et al., 2016), and the effects of combined teleconnection phases (Clem et al., 2017;Fogt et al., 2011).As a result, frequency, distribution, and intensity characteristics of regional short-lived weather phenomena like atmospheric blocks (Oliveira & Ambrizzi, 2017) and extratropical cyclones (ETCs; Pezza et al., 2012;Wei & Qin, 2016) are affected by teleconnection phase.
ETCs form within the polar jet (Uotila et al., 2011) and move southeastward toward the Antarctic coast over the course of days.These storms are important to Antarctic climate because they are cumulatively responsible for the majority of all meridional moisture and energy transport between the middle and high latitudes (Tsukernik & Lynch, 2013;Gorodetskaya et al., 2014;Papritz et al., 2014;Kurita et al., 2016;Reboita et al., 2019;H. Wang et al., 2020;Naakka et al., 2021).Storm fronts circulate air masses around the low pressure center and affect regional cloud and precipitation processes, temperature, surface energy, and sea ice (Nicolas & Bromwich, 2011;Binder et al., 2017;Y. Wang et al., 2019;Alexander et al., 2021).Near the Antarctic coast, cooling behind the cold front is enhanced by Antarctic Ice Sheet katabatic winds (e.g., Mathiot et al., 2010;Thompson et al., 2020) as sea ice grows (melts) in the cold (warm) season.Conversely, poleward moisture transport occurs behind the warm front along the so-called warm conveyor belt (WCB), which forces air aloft behind the warm front and enhances local cloud cover (Dacre et al., 2019;Woods & Caballero, 2016).ETC warm sector air promotes net surface energy enhancement via turbulent and longwave heating.During austral winter when solar energy is limited at high latitudes, these fluxes are important for promoting local sea ice melt (e.g., Kusahara et al., 2019).Over the satellite era, ETC prevalence and intensity have increased (e.g., Grieger et al., 2018;Pezza & Ambrizzi, 2003;Screen et al., 2018).Although Coupled Model Intercomparison Project models simulate decreased ETC counts in all future warming scenarios (e.g., Chang, 2017;Priestley & Catto, 2022), the ETCs that do form are projected to be more intense.Stronger ETCs could enhance short-lived sea ice motion and thermodynamic sea ice change.
ETC source region also impacts polar moisture availability and warming for days to weeks at a time.Most Antarctic moisture is sourced from Southern Hemisphere middle and high latitudes (H.Wang et al., 2020).However, dominant moisture source regions and moisture availability that are essential to short-lived high latitude changes vary seasonally and spatially.During austral winter, moisture source maxima are in the South Pacific, South Atlantic, and South Indian Ocean midlatitudes (Sodemann & Stohl, 2009;Singh et al., 2017;H. Wang et al., 2020).The Southern Ocean is a large contributor to Antarctic moisture in austral summer when the sea ice edge retreats and local evaporation produces greater total-column atmospheric moisture (H.Wang et al., 2020).In terms of moisture availability, studies using fully coupled global climate models with moisture tagging capabilities (Singh et al., 2017;H. Wang et al., 2020) conclude that the South Pacific source region is the greatest contributor to Antarctic atmospheric moisture and precipitation.However, studies using Lagrangian back-tracing techniques (Reboita et al., 2019;Sodemann & Stohl, 2009) highlight that dominant ETC and planetary circulation moisture source regions vary around the Antarctic.Spatial heterogeneity and moisture availability of source regions imply that transient sea ice response to ETCs could be regionally dependent.
Transient regional sea ice changes in the context of ETCs are important to better paint a picture of Antarctic atmosphere-sea ice-ocean interactions.Many studies have examined Antarctic-wide cyclone impacts on the atmosphere and sea ice (e.g., Dacre et al., 2015Dacre et al., , 2019;;Naakka et al., 2021;Tsukernik & Lynch, 2013;Uotila et al., 2011), but none have compared these processes on regional scales.In this study, we aim to better describe regional ETC properties and their short-term impacts on sea ice during austral winter using meteorological, oceanic, energy flux, and sea ice reanalysis data.Given the study area breadth and our analysis methods (described in Section 2.3), we do this with the intention of focusing on transient sea ice response and do not explicitly examine the role of ETCs on long-term sea ice trends.First, we describe the data and analysis methods we use in Section 2. In Section 3, we present and explain our results in the context of regional variability.We then wrap up our findings and discuss the implications of this work in Section 4.

Cyclone Detection
Extratropical cyclone detection algorithms are numerous and vary widely in how they flag individual storm systems (e.g., Neu et al., 2013).In this work, we use the TempestExtremes (TE) feature detection framework (Ullrich & Zarzycki, 2017;Ullrich et al., 2021) to find individual austral winter 1990-2019 ETCs.TE is ideal because it detects low pressure centers (i.e., mean sea level pressure (MSLP) minima) at each time step and then tracks and connects these minima over time to find individual ETC trajectories (Ullrich & Zarzycki, 2017).Individual ETCs are eliminated by TE if they do not meet user-specified duration, pressure gradient, travel distance, or temperature requirements.TE software and installation instructions can be found at https://github.com/ClimateGlobalChange/tempestextremes.
In our TE setup, we modify some default TE thresholds to focus on synoptic-scale Antarctic ETCs because they have larger impacts on underlying surface energy than mesoscale cyclones, which tend to be influenced by surface processes (Uotila et al., 2011).We restrict the TE search domain to latitudes poleward of 40°S because most Southern Hemisphere ETCs form in the midlatitude jetstream and spiral poleward (Uotila et al., 2013).We also increase the required pressure change (from 2 to 4 hPa) and distance outward from the ETC center (from 4° to 6°) because we find this better detects synoptic scale systems than other pressure gradient combinations (not shown).Like Ullrich and Zarzycki (2017), we retain storms with a duration greater than or equal to 2 days with ≤2 missing time steps (i.e., 12 hr missing) and we allow TE to merge MSLP minima separated by less than 2° in grid space.For further information and details regarding TE setup and usage, see Ullrich and Zarzycki (2017) and Ullrich et al. (2021).
Although we set up TE to detect ETCs in the Southern Hemisphere middle and high latitudes, some detected storms have lesser impacts on Antarctic sea ice than others.Because we are focusing on atmosphere-sea ice impacts, we filter out ETCs that are less impactful from our final storm inventory.First, to better ensure that included ETCs pass over sea ice, all included storms must move poleward of 60°S at least once in their lifespan (e.g., Grieger et al., 2018).We also require ETCs to move southward by at least 0.7°S day −1 because meridionally moving ETCs produce more poleward moisture transport than zonally oriented systems (e.g., Dacre et al., 2019).
To assess regional differences in ETCs and their impacts on underlying sea ice, we cluster the remaining storms by location using the Gaffney clustering algorithm (Gaffney et al., 2007).Briefly, the algorithm separates spatial data like ETC tracks into a user-specified number of groups based on their shape (i.e., polynomial or spline) and location (i.e., latitude/longitude) using regression and clustering techniques (Gaffney et al., 2007).This code is publicly available in Matlab (http://www.datalab.uci.edu/software/CCT/).Here, we separate detected ETCs into four mutually exclusive clusters using only trajectory longitudes at each time step and characterizing trajectory shape using a quadratic curve-fitting scheme.We focus solely on longitude for cluster location because high latitude Southern Ocean moisture is minimal during austral winter (H.Wang et al., 2020).We test cluster counts from k = 3 to k = 8 and find that k = 4 provides the best ETC track cluster definition for our detected storms (not shown).We define our resulting cluster regions as follows (Figure 1, counterclockwise from 180°E): East Antarctic, East Weddell, Bellingshausen, and Ross.
Interestingly, the original Ross cluster also includes ETCs over the western extent of the Weddell region (not shown).To maintain a more consistent Ross cluster region, we exclude any trajectories with cyclolysis (i.e., end point) longitudes between 90°W and 150°E.The resulting cluster-mean locations, storm counts, and intensity characteristics are further described in Section 3.1.

Data Analysis
After detecting ETCs and clustering them by longitude, we investigate their meteorological characteristics and their resulting surface energy and SIC responses.For most of our analysis, we examine daily averaged anomaly fields to remove diurnal and seasonal cycles.Month, day, and year associated with each ETC detected in 6-hourly data do not change for daily anomaly fields.We compute anomalies for a variable X at each latitude/longitude grid point between May 22 and September 10 using the following formula: X clim (n) is calculated using a 31-day moving average centered on day n over all years .Anomalies are calculated for all dates, regardless of whether ETCs are detected.
Because our aim is to compare atmospheric, oceanic, and sea ice response to ETC presence over different Antarctic regions, we group anomalies by cyclone cluster (Figure 1) and compute resulting centered composites with TE's NodeFileCompose (Ullrich et al., 2021).Centered compositing is ideal for analyzing ETC dynamical features of systems spread over large geographical areas (e.g., Priestley & Catto, 2022;Sinclair & Dacre, 2019).
NodeFileCompose aligns low pressure centers within each user-specified cluster and averages the gridded data fields over time.Because ETC intensities evolve over storm lifetime, we examine location-based ETC composites for each cyclone cluster for 3 cyclone time steps: (a) cyclogenesis (i.e., Genesis), (b) Day 1, and (c) cyclolysis (i.e., Lysis).We choose these time steps because all detected ETCs have a start time, an end time, and must last at least 2 days.For a specific cyclone cluster and time step, we compute centered composites over a 40° × 40° window (from −20° to +20° with 0° at the storm center).We choose to analyze a larger domain than previous centered composite analyses (e.g., Dacre et al., 2019;Sinclair & Dacre, 2019) so we can examine impacts of the ETCs as well as the potential influence of neighboring average circulation patterns.Hereafter, we use "composite" and "centered composite" interchangeably.Within each cluster, we examine a variety of atmospheric, oceanic, and sea ice variables.These include MSLP, 10-m zonal (u 10 ) and meridional (v 10 ) winds, 2 m-temperature (T 2m ), vertically integrated meridional moisture transport (VIVT), radiative and turbulent surface energy fluxes, SIC, and SST.ERA5 calculates vertically integrated meridional and zonal moisture transport, but because we are interested in ETC-produced poleward moisture transport, we focus on VIVT.For both v 10 and VIVT, we define southward flow (poleward) as positive.We choose this definition to emphasize that southward flow brings warmer, moister air to higher latitudes.
One of the main atmosphere-surface interactions we examine is surface energy exchange.We define total surface energy (E tot ) as follows: where all component fluxes directed from the atmosphere to the surface (i.e., downward) are positive (Hersbach et al., 2020).Net shortwave and net longwave energy fluxes (SW net and LW net , respectively) result by summing all upwelling and downelling component fluxes.Sensible and latent heat fluxes (SH and LH, respectively) depend on temperature and specific humidity differences between the surface and atmosphere.Greater wind speeds induce larger turbulent fluxes between the atmosphere and surface.Total surface energy anomalies (ΔE tot ) are calculated by summing anomalies of each quantity on the right-hand side of Equation 2.
We directly compare composited anomaly fields between ETC clusters using one-way analysis of variance (ANOVA).We calculate ANOVA statistics at the 95% confidence level for each latitude/longitude grid point to understand the spatial variability of storm differences over our specified composite domain.For each cyclone cluster, we also calculate cross correlations at each latitude/longitude grid point between days n − 5 and n + 5 of a given time step (Genesis, Day 1, or Lysis) to establish relationships between ETC-related SIC changes and meteorological and energy anomalies.We examine lagged correlations because the impacts of atmospheric and oceanic feedbacks on sea ice occur over different time scales (e.g., W. Yang & Magnusdottir, 2017).

Cyclone Characteristics
Before exploring sea ice response to ETCs in different Antarctic regions, we investigate ETC meteorological properties (i.e., temperature, pressure, winds, and moisture) and their surface energy flux and SST responses.First, we examine TE-detected ETC frequency and intensity statistics.Regardless of cyclone region, we find that storm center MSLPs range from 975 to 990 hPa at Genesis and decrease through Day 1 (Table 1).We define ETC intensification between Antarctic regional storm clusters using the normalized cyclone deepening metric (Equation 2 in Schossler et al., 2020) and compare average deepening rates between ETC locations using one-way ANOVA statistics with a 95% confidence interval.We find that E. Weddell ETCs deepen significantly more than Bellingshausen and Ross composite storms.E. Antarctic cyclones also deepen significantly more than Bellingshausen systems.All composite ETCs either strengthen further or remain consistent through Lysis.
In addition to differences in central MSLP deepening, we also find differences in composite MSLP spatial configurations that play a role in between-cluster meteorological differences.Average E. Weddell and Bellingshausen ETCs occupy most of the 40° × 40° composite domain during Genesis and Day 1.In contrast, both E. Antarctic and Ross cyclones are accompanied by high pressure systems along the southern boundary of their respective domains.These high pressure systems limit the poleward extent of warm air advection on the east side of the ETC, which is confirmed by associated ΔT 2m and Δv 10 patterns (i.e., Figure 2, bottom right composite quadrants).
In the E. Antarctic composite case, this same high pressure system also limits southerly cold air advection on the west side of the average ETC.As a result, we see significantly greater (Figure S1 in Supporting Information S1) warming (cooling) in the southeastern (southwestern) regions of Bellingshausen and E. Weddell domains that align with larger Δv 10 magnitudes (Figure S2 in Supporting Information S1).The Ross ETC composite alone maintains a central closed isobar that extends through the southeastern composite quadrant and lies northeast of a high pressure center during Genesis.On Day 1, this ETC separates into two cyclonic systems.The newly formed cyclone lies southeast of the parent composite and produces winds that induce negative 2 m-temperature anomalies in the parent (i.e., central) cyclone's southeastern quadrant (Figure 2).
Composite storm MSLP configurations are relatively similar between all ETC locations during Lysis.All MSLP contour fields depict closed storm centers adjacent to high pressure systems at the highest domain latitudes (i.e., the bottom of the composite panel) that are representative of the persistently high MSLP over Antarctica.Combined cyclonic and anticyclonic wind anomalies reinforce each other in the southern half of each composite domain and push warmer air westward.We observe low pressure systems to the west of composite E. Antarctic and Ross ETC centers that limit the westward extent of southerly airflow sourced from the Antarctic Ice Sheet (Figure 2).Despite different circulation regimes, Bellingshausen and Ross ETCs exhibit greater cooling due west of their centers than E. Antarctic and E. Weddell ETCs.This may be partially attributed to higher Lysis latitudes for composite Bellingshausen and Ross systems (Figure 1, Table 1).Interestingly, Ross ETC warm sector ΔT 2m magnitudes are lower than those for the remaining ETC clusters for Genesis, Day 1, and Lysis time steps.We attribute this to differences in genesis latitude and ambient temperature and moisture characteristics.
Along with advecting warm air poleward over the Southern Ocean, ETCs are responsible for transporting large quantities of moisture to higher latitudes.In Figure 3, we evaluate total meridional moisture transport (i.e., non-anomalies) and total surface energy flux changes (Equation 2) for all ETC locations and time steps.Regardless of storm location, we find that total meridional moisture transport is similar between Genesis and Day 1 before markedly decreasing at Lysis, at which time ETCs approach the Antarctic coast and available moisture decreases (e.g., Singh et al., 2017).We find that E. Antarctic and E. Weddell composite ETC warm sectors produce greater poleward moisture transport prior to Lysis (160 − 200 kg m −1 s −1 ) than Bellingshausen or Ross ETCs (140 − 160 kg m −1 s −1 ), despite greater regional sea ice cover (see Figure 1 in Hobbs et al. (2016)).Although the South Pacific region, where both Bellingshausen and Ross ETCs form, has been shown to contribute more moisture to the Antarctic than the South Indian and South Atlantic Oceans (H.Wang et al., 2020), we hypothesize that total moisture transport is larger for E. Antarctic and E. Weddell cyclones because they form at lower latitudes (Reboita et al., 2019).E. Antarctic and E. Weddell composite ETCs also exhibit greater meridional movement (Table 1), which has been shown to be an important factor for ETC-induced moisture transport (Sinclair & Dacre, 2019).It is worth noting that, despite differences in absolute VIVT between western and eastern hemisphere ETCs, VIVT anomalies are similar between all ETC locations (Figure S3 in Supporting Information S1).This is because anomaly calculations (Equation 1) are performed at each latitude/longitude grid point and account for lower moisture availability at higher latitudes.Therefore, smaller absolute VIVT in Bellingshausen and Ross ETC warm sectors still accounts for a larger overall change (relative to the climatology) in VIVT as the composite systems move poleward.On Day 1, Bellingshausen ETCs produce greater meridional moisture transport at high latitudes in the composite southeastern quadrant than the other ETC clusters (Figure 3 and Figure S2 in Supporting Information S1).E. Antarctic, E. Weddell, and Ross composite ETCs do not extend as far south in their domain and coexist with high pressure systems to the south (Figure 2).Anticyclonic flow converges with ETC winds and inhibits the poleward extent of moisture and heat transport.However, because this high pressure system is not present in the Day 1 Bellingshausen centered composite, northerly airflow is not deflected westward as it moves to higher latitudes.By Lysis, E. Antarctic ETCs maintain the greatest warm sector moisture transport ∼80 kg m −1 s −1 while maximum Ross ETC poleward moisture transport is only ∼40 kg m −1 s −1 .Ross ETCs are responsible for less poleward moisture transport over all time steps because of lesser source moisture availability (Reboita et al., 2019).
Strikingly, total surface energy flux anomalies are well-aligned with absolute meridional moisture transport (Figure 3).ΔE tot increases behind ETC warm fronts and decreases behind the cold front.These anomalies increase marginally between Genesis and Day 1 and markedly decrease by Lysis, like column-integrated meridional moisture transport.We can explain these total surface energy anomalies by examining component radiative and turbulent energy fluxes (Equation 2), winds, temperature, and moisture.Because we focus on austral winter ETCs, net shortwave changes are much smaller than the remaining flux types, regardless of time step.The remaining radiative (ΔLW net ) and turbulent (ΔSH and ΔLH) components of ΔE tot increase (decrease) behind the warm (cold) front because of storm-induced temperature, wind, and moisture transport anomalies (Figures 2-4, and Figure S3 in Supporting Information S1, respectively) and associated cloud changes (not shown).Warm sector ΔLW net , ΔSH, and ΔLH (Figures S4-S6 in Supporting Information S1, respectively) are significantly larger for E. Weddell ETCs during Genesis and Day 1, while corresponding Bellingshausen flux anomalies behind the cold front are more negative than in other ETC clusters (Figure 4).Regardless of ETC composite region, turbulent flux anomalies contribute more to total surface energy change than radiative fluxes at each ETC location (Figure 4).Overall, ΔE tot and its component anomaly fields correspond well with meridional moisture transport, temperature, and storm-based wind anomalies.
SSTs respond to ocean circulation, winds, and other atmosphere-ocean thermodynamic-related interactions.
Unlike ETC-produced atmospheric changes, however, we find that SST anomalies are highly variable and quite small (−0.2 to 0.2 K) at each composite time step (Figure 5).Positive ΔSST roughly align with E. Weddell and E. Antarctic composite warm sectors for Genesis and Day 1 but remain in the northern half of the composite domain for Lysis.On the other hand, ΔSSTs do not mirror ETC frontal movement in Bellingshausen and Ross ETC composites.Instead, positive (negative) ΔSSTs are well-mixed during Genesis and become more concentrated in the southern (northern) half of their respective domains during Day 1.By Lysis, however, ΔSST ≥ 0 cover most of the southern ETC composite quadrants.Ross ETCs produce mostly negative ΔSSTs in the northern quadrants, while Bellingshausen ΔSSTs are generally positive (negative) in the northeastern (northwestern) storm quadrant.Despite these pattern differences by ETC location, there are no statistically significant differences in ΔSST fields (no significance hatching in Figure 5).
JJA ETC meteorological properties and their corresponding SST and surface energy flux responses vary over time and space within the Antarctic region.We find that regional MSLP configurations of composite storms and other neighboring pressure systems impact the extent to which winds advect heat and moisture poleward.SST changes do not directly correspond to winds in all ETC clusters and could also be influenced by regional ocean currents below the sea surface.Without additional information, this is difficult to determine.These ETC cluster-based differences in atmospheric temperature, moisture, and winds directly impact total surface energy, and along with SST responses, could have varying impacts on regional sea ice, which we explore in the next section.

Sea Ice Concentration Changes
Here we characterize SIC anomalies (ΔSIC) corresponding to composite ETCs at each time step and compare these changes between all cyclone locations (Figure 6).For all time steps and ETC cluster composites, maximum ΔSIC magnitudes are ∼5%.Although this is lower than findings of other ETC-sea ice studies (Jena et al., 2022;Schreiber & Serreze, 2020;Vichi et al., 2019), we believe this occurs because we include low-intensity ETCs in our composite analysis.We also observe that ΔSIC fields are noisy relative to the ERA5 atmospheric anomalies.To filter out noise related to low magnitude SIC changes, we concentrate on ΔSICs that are greater than 2 standard deviations above or below the domain-averaged mean ΔSIC at each time step (outlined regions, Figure 6).Overall, we find ΔSIC < 0 to the east of the composite cyclone centers while ΔSIC > 0 to the west for Genesis and Day 1.This suggests that SIC declines (increases) relative to the climatology behind the composite ETC warm (cold) front.However, we also find that thermodynamic processes are not directly responsible for sea ice loss in southeastern composite quadrants (i.e., Figure 6) because, despite large ΔT 2m (i.e., Figure S1 in Supporting Information S1), 2 m-temperature and SSTs collocated with sea ice are below freezing (Figure S7 in Supporting Information S1).However, as we see from 2 m-temperature, SSTs, and ΔE Tot (i.e., Figure 4), thermodynamics could be responsible for ΔSIC > 0 (i.e., most of ETC composite southwest quadrants, and during Lysis, parts of the southeast quadrants), making the relative contribution of thermodynamic and dynamic processes on ETC-induced SIC response spatially variable.We explore potential atmospheric and oceanic mechanisms responsible for ΔSIC changes in Section 3.3.Regardless of cause, these ΔSIC patterns become less robust by Lysis as composite cyclone centers approach the Antarctic coast.
Grid-by-grid, one-way ANOVA analyses indicate that ΔSIC are not statistically different at the 95% confidence level between the ETC cluster locations (Figure 6; i.e., there is no significance hatching).However, visual inspection of ΔSIC by ETC location suggests small but meaningful differences.Although all ETC cluster composites How is Antarctic ETC presence related to regional sea ice cover?To answer this question, we calculate Spearman's rank correlation coefficients between SIC and MSLP, as well as SIC and seasonal storm counts (Table 2).We also examine regional and ETC-based SIEs and their linear trends over 1990-2019 (Table S2, Figure S8 in Supporting Information S1).All statistics are calculated at the 95% confidence level.
Domain-wide correlations between SIC and ETC MSLP are negative and generally statistically significant during Genesis and Day 1 (Table 2), regardless of ETC location.Negative correlation coefficients indicate that more intense ETCs (i.e., lower MSLP) result in greater SIC.As we have already shown (see Section 3.2), sea ice is mostly located where composite 2 m-temperatures and SSTs are below freezing (Figure S7 in Supporting Information S1).Given that greater ETC intensity can lead to faster wind speeds and larger air temperature changes, we think that corresponding sea ice drift is more extensive and likely leads to greater sea ice production as converging ice packs become more concentrated and open water produced by sea ice divergence refreezes.However, these relationships could also result from an ETC response to SIC during storm formation since higher SIC could increase  local meridional temperature gradients and baroclinicity.Therefore, we urge caution when interpreting causality between ETC intensity and SIC.
By Lysis, E. Antarctic and E. Weddell ETC intensity is positively correlated with SIC while Bellingshausen and Ross ETC SIC-MSLP correlations are weakly negative and not statistically significant.We hypothesize that larger SIC response to weaker ETCs could result from enhanced equatorward airflow over the sea ice surface.More specifically, katabatic winds associated with a climatological high pressure system (located along the southern border of all ETC Lysis composites; Figure 2) over Antarctica flow downslope over the ice sheet margins and push sea ice equatorward.However, because intense ETC winds counteract this katabatic flow and limit sea ice motion, sea ice formation could stagnate.We also find that seasonal storm counts are positively correlated with SIC E. Antarctic and Bellingshausen ETCs at all time steps, as well as Ross ETCs during Day 1 and Lysis (Table 2), but none of these relationships are statistically significant.
We also investigate sea ice edge fluctuations resulting from ETCs by examining SIE, which we calculate at each longitude by finding the lowest latitude (i.e., northernmost) at which SIC ≥ 0.15.Given that ΔSICs vary longitudinally (Figure 3), we calculate average SIE on the west and east side of individual ETCs for Genesis, Day 1, and Lysis over the 40° × 40° window defined around that ETC's center longitude/latitude grid point.We do not define individual ETC domain windows using the composite average ETC location (i.e., Table 1) because ETC geographical spread is large in each cluster.When no ETCs are detected in a specific region for a given year, the ETC-based average SIE is set as a missing value.These yearly ETC-based SIE trends are compared to the Antarctic-and regional-average SIE climatology trends (Table S2, Figure S8 in Supporting Information S1), where each region corresponds to a specific ETC cluster and is defined by the non-outlier westernmost and easternmost ETC trajectory longitudes (Figure 1; longitude values are listed in Table in Supporting Information S1).Overall, we find that SIE associated with ETCs in each cluster are similar to regional average SIEs (Figure S8 in Supporting Information S1; lines vs. shading) and that SIEs retreat marginally (i.e., move poleward) between Genesis and Lysis, especially for E. Antarctic and Ross ETCs.Ross ETC SIEs exhibit the largest interannual variability between JJA seasons while E. Antarctic ETC SIEs are generally consistent throughout the 1990-2019 period.For all ETCs and their associated regions, SIE interannual variability magnitudes are much larger than 1990-2019 trends.This agrees with previous research suggesting that Antarctic SIEs are characterized by large interannual variability and exhibit statistically insignificant trends (e.g., Y. Wang et al., 2019).In the next section, we investigate the roles of potential atmospheric and oceanic drivers on these SIC anomalies.

Sea Ice Change: Driving Mechanisms
Now that we have characterized large (i.e., 2 standard deviations from the domain-average) SIC changes by time and Antarctic region, we explore relationships between these changes and potential atmospheric and oceanic driving mechanism anomalies.We define potential relationships between drivers and SIC anomalies using cross correlation statistics.For Genesis, Day 1, and Lysis (Day 0 in each lagged correlation calculation), we compute maximum correlation coefficients and their corresponding lags from data between Day−5 to Day+5 for each grid point.We consider more lag times than previous lagged correlation analyses focused on transient weather systems (e.g., W. Yang & Magnusdottir, 2017) to examine possible signals over longer time scales.Maximum correlation coefficients (r max ) and their corresponding time lags at each latitude/longitude grid point are useful for assessing the spatial variability of these statistics across each composite domain.Because we aim to understand sea ice response to ETCs, we focus on negative lags (i.e., the driver occurs before the sea ice response).Although strong correlation suggests relationships between variables, we emphasize here that our findings do not necessarily provide evidence of causal relationships.Hereafter, any causal mechanisms we discuss are hypothesized based on the analyses outlined.
Figures 7 and 8 depict maximum correlation coefficient (r max ; left-hand column) and corresponding time lag distributions (right-hand column) between driver and sea ice anomalies.Any listed r max in this section refer to the median maximum correlation coefficient over the entire composite domain.Excepting Δu 10 , we find that median maximum correlations between all variables in Figure 7 and ΔSIC are negative for all ETC time steps.This further confirms that northerly warm, moist airflow within the cyclone's warm sector leads to thermodynamic (relative to the climatology) and pushes the sea ice edge poleward (e.g., Pezza et al., 2012;Reboita et al., 2019) while southerly winds induce equatorward sea ice drift and enhance thermodynamic growth behind the cold front.We note that within-composite correlation variability is high (i.e., many Figure 7 boxes have negative and .r max (left column) and their corresponding lag times (right column; units of days) between ΔSIC and driving mechanism anomalies by extratropical cyclone (ETC) time step (rows).For each variable, bars are labeled by location: E. Antarctic (A), E. Weddell (W), Bellingshausen (B), and Ross (R).r max is calculated at each latitude/longitude grid point, so each boxplot represents the spread of all r max values in an ETC composite domain at a genesis, day 1, or lysis.Median r max and maximum lag value lines in individual box plots represent the median relationship between ΔSIC and ΔVAR over all longitude/latitude grid points where sea ice is present for that ETC domain/time step combination.For each variable, bars are labeled by location: E. Antarctic (A), E. Weddell (W), Bellingshausen (B), and Ross (R).
positive correlations within their interquartile ranges), so caution is recommended when inferring spatial relationships from r max .Of all discussed ΔSIC drivers, ΔT 2m exhibits the strongest relationship with ΔSIC for almost all cyclone regions and time steps.The only exception is E. Antarctic cyclones during Lysis: ΔT 2m is negatively correlated with ΔSIC (r max = −0.67),but to a lesser extent than SST anomalies (i.e., r max = −0.83).Examining individual ΔE tot component relationships with ΔSIC, we find that the largest r max magnitudes result for non-shortwave flux anomalies (Figure 8).Although ΔLW net is negatively correlated with ΔSIC for all ETC locations and time steps (r max = −0.38 to r max = −0.83),r max for both ΔSH and ΔLH becomes less negative between Genesis and Lysis.In fact, ΔLH and ΔSIC are positively correlated for all but Bellingshausen ETCs during Lysis.Although this appears to suggest that enhanced latent heating at the surface increases SIC, domainwide r max variability is large.As a result, ΔE tot becomes less negatively correlated with ΔSIC between Genesis and Lysis.We also find that r max associated ΔE tot and ΔSIC are lower than most other driving variables at each ETC location.
Of all ΔSIC drivers, Δu 10 alone exhibits positive correlations with SIC change during Genesis that transition to lower positive/negative during Lysis (Figure 7).This relates to ETC proximity to the sea ice pack and climatological winds based on storm latitude (Eayrs et al., 2019).During Genesis when ETCs are close to/within the midlatitude jet, storm winds impact sea ice in the southern half of composite domains such that Δu 10 pushes sea ice from east to west, decreasing (increasing) SIC in the southeast (southwest) composite quadrant.As composite storms approach Antarctica, climatological winds transition from westerly to easterly and storms cover more sea ice area.Thus, ETC zonal winds affect sea ice concentrations to the north and south of composite storm centers, leading to an offsetting effect and contributing to less agreement among gridded maximum correlation coefficients within the domain (i.e., larger interquartile range).Like ΔE tot , Δu 10 correlation coefficient magnitudes generally rank lower than those for meridional moisture transport, 2 m-temperature, meridional wind, and SST anomalies.However, regardless of time step, Δu 10 helps explain regions of ΔSIC > 0 (ΔSIC < 0) to the east (west) of central composite meridians (vertical axis in each panel; Figure 6).
We also find varying lag times between ΔSIC and individual forcings.Like r max , lag times listed here represent the median of all composite grid points. 2 m-temperature and 10 m-zonal wind anomalies precede maximum ΔSIC by 0 days at Genesis and 1 day during Lysis (Figure 7).SIC response to ETC-induced SST changes is also 0 days for all time steps.In contrast, total surface energy, meridional wind, and meridional moisture transport anomalies precede sea ice change by 1-2 days (Figure 7; right column).Overall, these lag times highlight the complexity of storm-produced feedbacks between wind, heat, and moisture conditions and the underlying sea ice surface and justify our use of cross correlation statistics.As with r max , when all composite grid points are considered for specific ETC time steps, we find large lead and lag time variability.
Comparing r max between ΔSIC and ETC-related driving mechanisms in different clusters, we find the largest correlation coefficient magnitudes and greatest agreement (i.e., lowest interquartile range; Figures 7 and 8) in the Bellingshausen and E. Antarctic ETC locations.This is especially true for 2 m-temperature, total surface energy, meridional moisture transport, and meridional wind anomalies during Genesis and Day 1.We cannot necessarily attribute these regional differences to location-based storm intensification (Sinclair & Dacre, 2019), count statistics (Table 1), or intra-domain spatial variability since these quantities are similar between ETC clusters.Instead, we posit that E. Weddell and Ross sea ice-driver relationships are less robust because their corresponding ΔSIC have greater spatial variability (i.e., Figure 6).
Bellingshausen ETCs are a special case because correlation values are large (0.61 ≤ .88 for all variables and times).Not only does this suggest that all drivers are important to SIC evolution, but it also helps highlight the complexity of relationships at other ETC locations.Correlations between ΔSIC and ΔSST, ΔVIVT, Δv 10 , and ΔE tot are consistent for all latitude/longitude grid points considered in this analysis (Figure 7).As we see in other ETC clusters, Bellingshausen Δu 10 and ΔSIC correlation magnitudes are consistently lower than those for the other variables.These relationships in the Bellingshausen cluster could be partly explained by the semi-permanent ASL circulation reinforcing transient ETC processes (e.g., Fogt et al., 2012).Provided that the ASL is not a significant factor, these results indicate that Bellingshausen ETCs could be used to further our understanding of Antarctic ETC mechanical evolution.
Variable-sea ice anomaly relationships exhibit more spread in the remaining ETC clusters, with the greatest correlation coefficient spread occurring during Lysis.As we see with Bellingshausen ETCs, storm-induced SST, meridional surface winds, and meridional moisture transport changes exhibit similar relationship strengths with sea ice change.Not surprisingly, Δv 10 and ΔVIVT r max values are similar because of how ERA5 meridional moisture transport is defined (i.e., the column-integrated product of meridional wind and atmospheric moisture content).Considering meridional atmospheric variables collectively, we find that their correspondence to sea ice change is stronger than all but 2 m-temperature anomalies for Bellingshausen and Ross ETCs from Genesis through Day 1. E. Antarctic and E. Weddell cluster ΔSIC relate more closely to ΔSST than atmospheric moisture transport and winds, despite maintaining greater absolute meridional moisture transport in their composite warm sectors (Figure 3).This suggests that Bellingshausen and Ross SIC response to ETCs may be more sensitive to moisture changes than sea ice in the remaining cyclone locations.By Lysis, the combined influence of meridional moisture transport and surface meridional winds becomes more important for E. Weddell and less important for Ross ETCs.We also find that E. Weddell ETCs uniquely produce stronger relationships between ΔSIC and 2 m-temperature, total surface energy (and all component fluxes), meridional moisture transport, and meridional winds at Lysis than at previous time steps.Since the magnitude of ΔSIC is generally larger for Lysis than other times for all cyclone locations (Figure 5), this suggests that additional driving mechanisms (e.g., sea ice drift) and geographic constraints (i.e., the Antarctic Peninsula inhibiting sea motion to the west) may play a role in sea ice-atmosphere-ocean interaction complexity for E. Weddell systems as they form and initially intensify.
Lagged correlation analysis suggests relationships and their timing between ΔSIC and possible forcings.Our results indicate that most atmospheric and oceanic variable anomalies are negatively correlated with SIC change for all ETC locations.We find that ΔT 2m is best related to ΔSIC between 0 and 1 days after storm passage for all except E. Antarctic Lysis ETCs.However, secondary relationships vary between ETC cluster.Sea ice change in the Bellingshausen and Ross regions corresponds strongly to Δv 10 and ΔVIVT 0-2 days after the composite ETC passes through for all time periods.In contrast, E. Antarctic and E. Weddell systems produce stronger relationships between ΔSIC and ΔSST than meridional atmospheric processes.ΔE tot r max magnitudes are small during Lysis for all but the Bellingshausen composite and exhibit variable lag times during Genesis, Day 1, and Lysis, which we attribute to weak r max for all ΔE tot components (Equation 2).Δu 10 alone is positively correlated with ΔSIC during Genesis because negative easterly winds produced by ETCs at higher latitudes produce ΔSIC < 0. As ETCs move poleward and cover more sea ice area, climatological winds transition from westerly to easterly and r max values become more negative.These results highlight the complexity of atmosphere-ocean-sea ice interactions and how these relationships change between different Antarctic regions.

Conclusion
In this study, we use ERA5 reanalysis data to highlight differences in regional Antarctic ETCs and their impacts on underlying sea ice for JJA (i.e., Austral Winter) 1990-2019.We detect individual ETCs using the pseudo-Lagrangian TempestExtremes framework (Ullrich et al., 2021;Ullrich & Zarzycki, 2017) and assign detected storms into one of four possible location-based groups using a multi-dimensional clustering technique (Gaffney et al., 2007).Within-cluster averages at cyclogenesis, Day 1, and cyclolysis are calculated by using centered compositing techniques to avoid signal weakening that might result computing traditional composites over large domains (Sinclair & Dacre, 2019).We mostly focus on anomalies, or environmental changes produced by ETC presence, to understand short-lived cyclone impacts and remove trends and seasonal cycles from the data.
Storm counts are generally similar between each ETC cluster, with the Ross cluster containing the lowest storm count (n = 68) after removing geographic outliers.Regardless of ETC location, composite storms form in the Southern Hemisphere midlatitude jet stream, moving southeastward and intensifying over time.As composite storms evolve, poleward heat and moisture transport occur to the east of the storm center.Moisture transport, surface wind, and total surface energy anomaly magnitudes are generally largest for Day 1 and decline sharply as ETCs undergo cyclolysis at higher latitudes.Comparing cluster composites, we find that E. Weddell ETCs induce greater warming, moistening, and surface energy inputs than other location-based composites prior to cyclolysis.The Bellingshausen ETC composite warm sector moves heat and moisture further poleward through Day 1 and induces polar warming earlier than other ETC groups.We also determine that absolute meridional moisture transport is greater for E. Antarctic and E. Weddell storms because they form at lower latitudes where atmospheric moisture is greater (Reboita et al., 2019).Differences in ETC atmospheric conditions and surface energy by location are ultimately induced by changes in the MSLP fields over individual composite domains.Regardless of ETC cluster, ΔSICs depict a longitudinal division: We find decreasing SICs in the eastern composite quadrants as warm, moist air pushes poleward that leads to increasing surface energy input and sea ice edge retreat via atmospheric drag.The opposite occurs to the west of composite storm centers as cold, dry winds from higher latitudes promote sea ice growth thermodynamically and dynamically (i.e., pushing sea ice equatorward).We find that maximum SIC responses (∼5%) occur during Lysis, at which point ΔSIC patterns become noisier as composite average ETCs approach the Antarctic shoreline.Overall, this east-west ΔSIC pattern is most apparent for E. Antarctic and Bellingshausen ETCs.
We find that SIC is negatively correlated with MSLP, indicating that either more intense ETCs result in a greater SIC because of wind-induced sea ice dynamics, or that greater SIC enhances baroclinicity where ETCs form and mature.These relationships are significant through Day 1 and weaken by cyclolysis.However, we do not find any significant correlations between SIC and seasonal storm count.Most noticeably for E. Antarctic and Ross ETCs, sea ice retreat is insignificant beneath ETCs as the composite systems move poleward between cyclogenesis and cyclolysis.Regardless of geographic scope or ETC presence, SIE interannual variability is much greater than 1990-2019 SIE trends.This implies that, despite the large atmospheric and oceanic changes that respond to ETC passage, the long-term impacts of ETCs on Antarctic sea ice trends are minimal.
We analyze lagged correlation coefficients at each domain grid point to quantify relationship strength between SIC responses to atmospheric and oceanic storm-induced driving mechanisms.E. Antarctic and Bellingshausen ETCs exhibit the largest correlations between atmospheric and oceanic drivers and sea ice response.ΔT 2m is the most negatively correlated with ΔSIC for all storm location clusters and time steps except E. Antarctic Lysis composites and generally precede sea ice response by 0-1 days.The relationship between sea ice and zonal wind anomalies varies greatly by ETC location and time step and is smaller than most other examined drivers, excepting total surface energy anomalies during Lysis.We find that meridional wind and moisture transport anomalies are better related to sea ice change for Bellingshausen and Ross systems, despite lower absolute meridional moisture transport, while E. Antarctic and E. Weddell systems are better correlated with SST anomalies, despite exhibiting greater absolute northerly moisture transport.Along with absolute meridional moisture transport characteristics, between-cluster changes in the relationship strength (r max ) between meridional moisture transport and sea ice change indicate that Bellingshausen and Ross sea ice could be more sensitive to atmospheric moisture than storms formed in the other cluster locations.
Although general cyclone characteristics and SIC response to ETC passage show similar overarching characteristics at each location, the magnitude, pattern coherence, and relative strength of atmosphere-surface interactions appears to be spatially variable.Because we use reanalysis data and cannot specify climatological conditions during our study period, disentangling responsible processes for sea ice change is challenging.Furthermore, we are unable to investigate oceanic salinity, density, and temperature profile characteristics because ERA5 does not provide these variables.Although we choose ERA5 SIC to be consistent with the other variables we analyze, we acknowledge its potential for biases relative to passive microwave and in-situ SIC observations (e.g., Global SIC Climate Data Record; Lavergne et al. (2019)).Given that ERA5 is the newest ECMWF reanalysis product, comparisons between observed and ERA5 SIC have not been published (to our knowledge).An interesting follow-on to our analysis would be to examine Antarctic ETC influences on SIC using passive microwave SIC data and assess how different SIC data affects the relationships highlighted here.Individual ETC tracks constituting each cluster collectively cover wide areas and undoubtedly exhibit variability with other storms in the same ETC group.We can generalize average storm properties and feedbacks using centered compositing techniques, but this process obscures individual storm features that could shed more light on intra-cluster storm evolution variability.One solution to this over-generalization would be to define a larger number of clusters, but this reduces storm counts in each cluster and could impact the robustness of the statistical techniques we use here.
Our results indicate regional variability in SIC response to austral winter ETC presence around Antarctica.However, our findings focus only on present-day processes.Because ETC frequency and intensity are projected to change in the future (Priestley & Catto, 2022), a logical extension of this work would be to investigate ETC-sea ice relationships simulated in various warming scenarios.Climate modeling would also provide a fuller picture of these interactions with more extensive variable options (especially in the ocean) than are currently available from reanalysis.tempestextremes and its features are described in Ullrich and Zarzycki (2017); Ullrich et al. (2021).Gaffney clustering software program is described in Gaffney et al. (2007) and is available for download at http://www.datalab.uci.edu/software/CCT/.

Figure 1 .
Figure 1.Extratropical cyclone (ETC) trajectories separated into four location-based clusters.Thick lines and stars depict average ETC trajectories for each cluster.

Figure 2 .
Figure2.Absolute mean sea level pressure (contour lines) with 2 m-temperature and 10 m-wind anomalies (filled contours and arrows, respectively) for each storm time (rows) and extratropical cyclone location (columns).

Figure 3 .
Figure3.Total meridional moisture transport (contour lines; kg m −1 s −1 ) with total surface energy anomalies (filled contours and arrows, respectively) for each storm time (rows) and extratropical cyclone location (columns).Northerly VIVT is positive.

Figure 4 .
Figure 4. Average total surface energy and component flux anomalies taken over the west (solid bars) and east (hatched bars) of the composite domain for each storm time (rows) and extratropical cyclone location (columns).Units are W m −2 .

Figure 5 .
Figure 5. Composite Average ΔSST for each storm time (rows) and extratropical cyclone location (columns).Units are K.

Figure 7
Figure7.r max (left column) and their corresponding lag times (right column; units of days) between ΔSIC and driving mechanism anomalies by extratropical cyclone (ETC) time step (rows).For each variable, bars are labeled by location: E. Antarctic (A), E. Weddell (W), Bellingshausen (B), and Ross (R).r max is calculated at each latitude/longitude grid point, so each boxplot represents the spread of all r max values in an ETC composite domain at a genesis, day 1, or lysis.Median r max and maximum lag value lines in individual box plots represent the median relationship between ΔSIC and ΔVAR over all longitude/latitude grid points where sea ice is present for that ETC domain/time step combination.For each variable, bars are labeled by location: E. Antarctic (A), E. Weddell (W), Bellingshausen (B), and Ross (R).

Figure 8 .
Figure 8. Same as Figure 7, but for ΔE tot component flux anomalies.Units are in W m −2 .

Table 2
Sea Ice Concentration, Mean Sea Level Pressure, and Storm Count Correlations