Abstract
Observations show that Arctic cyclones can significantly affect the marginal ice zone (MIZ). Understanding how these effects may evolve in the future is critical for accurately predicting future ice loss, yet cyclone-sea ice interactions in global climate models are understudied. This analysis compares output from the Community Earth System Model version 2 Large Ensemble (CESM2-LE) with observed intense summer cyclone impacts on the MIZ. We find that CESM2-LE reproduces observed net impacts but exhibits compensating biases, where fewer intense cyclones reach the MIZ but decrease ice area more than observed. In a future emission scenario, CESM2-LE predicts more frequent intense cyclones, but as the Arctic warms and late summer ice cover lessens, fewer storms will interact with the retreating ice edge. Subsequently, the largest effects shift earlier in the year, and ice loss from these storms declines after present day, suggesting intense summer cyclone impacts have reached a maximum.
Plain Language Summary
Intense Arctic summer cyclones have the ability to greatly reduce the area of sea ice along the edge of the ice, where the sea ice is thin and already broken up. This study compares previous results found using observations with climate model output to understand how well the climate model represents cyclone characteristics and sea ice changes, and how these components will change in the future. We find that the model predicts fewer cyclones reach the ice edge compared to observations, but that their impact is too intense. Interestingly, these competing impacts compensate, causing the net ice loss in the model to be similar to what has been observed in recent decades. In the future, the climate model predicts more intense cyclones in the Arctic, but as ice cover decreases, cyclones are less likely to move over ice, leading to less cyclone-induced ice loss overall.
Key Points
-
CESM2-LE reproduces observed recent cyclone impacts on MIZ ice but exhibits compensating biases: too few storms exerting too strong impacts
-
Total MIZ summer ice loss by storms is predicted to decline after present day, suggesting intense cyclone impacts have reached a maximum
-
As late-summer sea ice recedes in the future, fewer storms reach the MIZ and the largest decreases in ice area occur earlier in the summer
1 Introduction
The role of summer cyclones in altering the Arctic ice edge is complex, with a variety of competing factors leading to either the preservation or additional loss of sea ice within the marginal ice zone (MIZ). For example, enhanced winds can lead to ice breakup, advection, or ridging, which can change an ice floe's susceptibility to melt (Brümmer et al., 2008; Itkin et al., 2017). Additionally, cyclones can alter the surface energy balance and subsequently ice cover by increasing cloud cover and altering precipitation patterns (Kapsch et al., 2016; Screen & Simmonds, 2012). These impacts are increasingly important as summer sea ice extent shrinks and seasonal ice cover becomes more common (Haine & Martin, 2017; Landrum & Holland, 2020; Stroeve et al., 2012). In particular, the MIZ has widened in recent years due to the decline of thick multiyear ice (Strong & Rigor, 2013). With greater areas of open ocean, increased heat exchange between the ocean and atmosphere can change large-scale atmosphere and ocean circulation (Stroeve & Notz, 2018). Subsequently, understanding the factors that alter the ice edge is increasingly necessary.
Through a combination of observational and reanalysis data sets, many of these cyclone impacts have been observed. In early summer months, enhanced cloud cover has been suggested to slow seasonal ice melt (Finocchio et al., 2020; Schreiber & Serreze, 2020). Later in the season (July and August), thinner ice becomes more susceptible to enhanced wind forcing and accelerates ice area loss across the Arctic (Finocchio & Doyle, 2021; Finocchio et al., 2022). However, these cyclone impacts can vary in the MIZ, where the ice is potentially more responsive to atmospheric forcing. Meridional wind direction in the MIZ strongly relates to the associated changes in ice area, where enhanced southerlies decrease (and northerlies increase) MIZ ice area (Mundi & L’Ecuyer, 2025). Mundi and L’Ecuyer (2025) also found that June, July, and August storms locally decrease MIZ ice area while September storms generally increase ice area relative to climatological trends likely due to different ice states prior to the storms. Since the 1980s, an increase in June and July storms caused a corresponding shift toward more destructive storms.
Previous modeling studies investigating cyclone-sea ice interactions have tended to focus on individual case studies such as “The Great Arctic Cyclone of August 2012” (Lukovich et al., 2021; Simmonds & Rudeva, 2012; Zhang et al., 2013), or the record cyclone in January 2022 (Blanchard-Wrigglesworth et al., 2024), which were unprecedentedly large and intense cyclones. Other studies have examined the effects of many cyclones in models by investigating the winter cyclone impacts in phase 5 of the Coupled Model Intercomparison Project (CMIP5; Cai et al., 2020) or by using reanalysis to force ice-ocean models (Clancy et al., 2022; Kriegsmann & Brümmer, 2014). Alternatively, regional and global climate models have been used to project increases in the total number of Arctic cyclones, though these trends are regionally variable (Akperov et al., 2015; Crawford & Serreze, 2017; Day et al., 2018; Nishii et al., 2015). Weaker and equatorward biases in northern hemisphere storm tracks may cause these increases to be underestimated (Harvey et al., 2020; Zappa et al., 2013), possibly due to the growing land-sea temperature contrast in the Arctic impacting modeled storm tracks differently than in observations (Day & Hodges, 2018) or biases in net poleward energy transport (Cox et al., 2024).
-
Do biases in intense cyclone characteristics influence modeled impacts on MIZ sea ice?
-
How are these impacts on the MIZ predicted to change in a warmer climate?
Historical storm impacts are compared with results found using satellite observations and reanalysis (Mundi & L’Ecuyer, 2025) to reveal differences in cyclone characteristics and their effects on the MIZ, which are then used to inform predictions about future cyclone-induced changes to the MIZ.
2 Data and Methods
2.1 Large Ensemble and Observational Data Sets
The Community Earth System Model Version 2 (CESM2) is a free running, fully coupled climate model with approximately 1 horizontal resolution (Danabasoglu, Lamarque, et al., 2020). CESM2 utilizes the Los Alamos Sea Ice Model version 5.1.2 (CICE5; Hunke et al., 2017) to represent sea ice, the Parallel Ocean Program version 2 (POP2) for the ocean component of the model, and the Community Atmosphere Model version 6 (CAM6) to represent the atmosphere. Sea ice concentration and sea surface temperature (SST) values are output daily, and mean sea level pressure values are available 6-hourly. From 1850 to 2014, CESM2 is forced by CMIP6 historical forcing and, from 2015 onward, by the Shared Socioeconomic Pathways (SSP) forcing scenario SSP3-7.0 (Rodgers et al., 2021). The large ensemble consists of 100 members, where 50 members use standard CMIP6 historical forcing and 50 use smoothed biomass burning emissions in the historical period, which are known to impact Arctic sea ice trends (DeRepentigny et al., 2022). For this study, we used 40 of the 50 members with consistent smoothed forcing. We present results from the early satellite era, 1982 through 1991, and the modern-day Arctic, 2010 through 2019, to establish model performance relative to observations and then extend the analysis through 2100 to understand predicted future sea ice changes.
Hourly, mean sea level pressure data from the European Center for Medium-Range Weather Forecasts ERA5 reanalysis product was used to track cyclones from 1982 to 2019 (C3S, 2023; Hersbach et al., 2020). Satellite observations are then used to define the surface conditions during the storm. The National Snow and Ice Data Center Climate Data Record product (Meier et al., 2021; Peng et al., 2013) combines passive microwave data from three microwave sensors using the NASA Bootstrap algorithm to report daily average sea ice concentrations across the Arctic at 25 km resolution. This product has missing data early in the record, only affecting about 3–5 storms (based on climatological ice conditions) in July and August 1984. To establish ocean characteristics during past cyclones, we use daily SST observations from the NOAA Optimum Interpolation High Resolution Data set (Huang et al., 2021), which combines in situ and satellite observations on a 0.25-degree resolution global grid from September 1981 through present.
2.2 Cyclone Detection and Quantifying MIZ Area Changes
While many cyclone detection algorithms exist (Neu et al., 2013), we applied a simple scheme for identifying low-pressure systems. Following the detection of intense cyclones in Mundi and L’Ecuyer (2025), we applied a constant pressure threshold of 984 hPa to identify storms in ERA5 (similar to the 985 hPa threshold used by Rinke et al. (2017) and Lukovich et al. (2021) to define “extreme” Arctic cyclones). We then tracked the evolution of these systems by grouping low-pressure points with longitudinal thresholds and defined the area affected by the storm with the extents of daily 1,000 hPa contours. To include only long-lived cyclones that interact with the MIZ, storms are required to have a minimum duration of 2 days and the storm area must contain 20%–80% “pack” (concentrations greater than 80%) ice on the first day of the storm (removing storms that likely have minor effects on the ice).
To track changes in MIZ ice area from 1 week before the start of the cyclone to 2 weeks after, a fixed region was determined based on all grid cells containing between 15% and 80% ice concentration (Strong & Rigor, 2013) within the storm area from 1 day before to 1 day after the storm. To remove the seasonal (not caused by cyclones) change in ice area, the 3-week time series of MIZ ice area within the storm region was computed for each year of the corresponding decade and averaged. This climatology was then subtracted from the time series of ice area during the storm. SST values are computed using daily regions of 15%–80% ice concentration such that the variability describes the impact of SSTs where marginal ice is present.
2.3 Adapting to CESM2-LE
To account for the coarser resolution in CESM2, the pressure threshold used for detecting cyclones is increased to 986 hPa. This value was determined by scaling ERA5 data to the horizontal and temporal resolution of CESM2 and adjusting the pressure threshold to produce the same number of storms as the original ERA5 census, with comparable timings and durations. The storm area definition was adjusted similarly (increasing the daily contours to 1,002 hPa to reproduce the areal extents of the original ERA5 storms). Also, when considering MIZ-interacting storms in CESM2, differences in sea ice concentration from satellite observations necessitated finding similar area fractions of MIZ and pack ice in CESM2 by determining the concentrations with the lowest root mean square difference from 15% to 80% to establish which portions of the storm domain were covered by MIZ or pack ice. An example of these methods is shown in Figure S1 in Supporting Information S1.
3 Results
3.1 Differences in Cyclone Detection
Adjusting the pressure threshold to account for resolution differences, we expect similar storm counts for both CESM2 and ERA5 in the historical period. Figures 1a and 1b show the mean number of low-pressure systems detected (before applying the additional criteria listed in Section 2.2) across the 40 CESM2 ensemble members and the total number of ERA5 storms for two decades from June through September. CESM2 has more storms overall, with over 100 intense storms each decade on average, whereas ERA5 has around 90. However, once these low-pressure systems are sorted to meet this study's conditions (colored bars), fewer CESM2 storms remain. The first filter (shown in blue) determines if the cyclone reaches the ice edge (within 1,250 km) during its evolution. For ERA5, roughly 30% of storms are removed, while more than two-thirds of CESM2 storms fail this criterion, indicating an equatorward bias of intense storms in the model, consistent with Konstali et al. (2024) and Priestley and Catto (2021). The next filters are similar between the two data sets. About 70% of the remaining storms are removed due to storm characteristics (green; defined as not having a duration of 2 days) and approximately half of the remaining are removed for not sufficiently interacting the MIZ (purple; storm area containing 20%–80% high-concentration ice). Subsequently, the equatorward bias in storm development locations causes more storms to be removed in the first step, with CESM2 simulating considerably fewer MIZ-interacting cyclones than ERA5.
June through September storm counts for 2010–2019 (a) and 1982–1991 (b) in CESM (left) and ERA5 (right). The mean total annual count per ensemble member (n) is indicated at the top of each bar. Colors describe the filters applied to the census of cyclones and are discussed in text. “Remaining Cyclones” (red) are storms considered in this study, with the mean annual count shown in parentheses. (c–f) Distribution of ensemble annual-mean storm counts (dots and boxplot) compared with ERA5 (cross) for total storm counts (blue) and remaining cyclones (red). (g–j) Frequency of June–September mean sea level pressure values falling below 986 hPa. ERA5 data is scaled to match the resolution of CESM2.
Figures 1g–1j reinforce this result by showing the frequency of sea level pressure below the pressure threshold for both data sets. ERA5 has two distinct maxima, with storms tending to reach minimum pressure values along the eastern coast of Greenland and over the sea ice in the Central Arctic Ocean. While the modeled distribution has the same Icelandic low maximum, the frequency of central Arctic low-pressure occurrences is reduced. The secondary low-pressure maximum appears much farther from the ice edge, in the Bering Sea, which does not appear as strongly in ERA5. These biases in storm frequency over the central Arctic are consistent with other global climate models, such as CMIP5 models (Zappa et al., 2013) and the previous version of CESM (Crawford & Serreze, 2017).
3.2 Impacts of Intense Cyclones on MIZ Ice Area
In addition to storm location differences, CESM2 and ERA5 indicate different MIZ responses to intense cyclones. Figures 2a–2d show the net change in MIZ ice area relative to 1 week before the storm for early and late summer months in each decade. In all periods, the CESM2-LE composite-mean change in MIZ ice area is larger than the mean observed outcomes. Like ERA5, early summer storms in CESM2 tend to have greater negative impacts on the MIZ than late summer storms (where modeled September storms generally increase the local ice area like in observations; Figure S2 in Supporting Information S1). All 40 ensemble members considered show net negative MIZ ice area changes in early summer months. The spread of ensemble members captures the observed mean MIZ impacts in the more recent decade, but model responses diverge substantially from the observed time series in the earlier decade. CESM2-LE has a much larger spread in outcomes 2 weeks after the storm in late summer, where a decline in MIZ ice is simulated initially, followed by significant spread after the storm (particularly in the earlier decade), with some ensemble members indicating continued decline and others increasing ice area. In all month-decade groupings, modeled storms tend to have longer-lasting effects on MIZ ice area. Possible explanations for these differences could be larger areas of lower sea ice concentration or melt pond coverage accelerating the ice-albedo feedback with the CESM2 MIZ, or differences in the non-cyclone-induced changes to the sea ice, since CESM2 is known to underestimate sea ice thickness (and subsequently extent) in summer months due to thinner spring clouds (DeRepentigny et al., 2020; DuVivier et al., 2020). The tendency toward lower-latitude storms could also explain the too-strong impacts in the model, as fewer storms travel over a thicker ice regime in the central Arctic, which are known to have differing impacts, as suggested by Lukovich et al. (2021).
(a–d) Net change in MIZ ice area, relative to 1 week before the storm. Mean storm impact for each ensemble member is shown in gray, with the mean and standard deviation across members shown in red. The mean observational response is shown in blue. (e) Fraction of cyclones in each ensemble member that causes a net decrease in MIZ ice area 2 weeks after the storm's passage or (f) have a corresponding increase in SST during the storm lifetime. (g) Total net sea ice change for each month/decade grouping. Red dots and box plots show the distribution of ensemble member values. Blue crosses show the corresponding mean value for cyclones detected in ERA5. Asterisk depicts no statistically significant difference at the 99.9% level (e–f all show significant differences).
Nearly all modeled cyclones decrease MIZ ice area in early summer months, a much larger fraction than observed storms (Figure 2e), possibly because over 90% of storms also correspond to local SST increases (Figure 2f). In late summer, most modeled cyclone cases also decreased MIZ ice area, as opposed to only about half of ERA5 storms. Late summer storms in CESM2 also have coincident SST increases more often than observed. The combination of enhanced and prolonged sea ice loss with more ubiquitous SST warming highlights the importance of upper-ocean characteristics in representing sea ice changes (Blanchard-Wrigglesworth et al., 2024; Finocchio et al., 2022).
3.3 Net Impact
The preceding analysis suggests CESM2-LE under-represents the number of MIZ-interacting storms but predicts stronger-than-observed impacts. Figure 2g quantifies the implications of these competing biases on the accumulated MIZ impact of intense cyclones in CESM2 (defined as the 3-week change in MIZ ice area, minus climatology, summed over all storms). The observed storm impact falls within the range of CESM2 ensemble members in the more recent decade, however, in the earlier decade, all CESM2 ensemble members produce a too-negative ice response compared with observations. CESM2 may be demonstrating similar MIZ ice area net changes compared to observations due to compensating effects of reduced storm frequency and increased storm impacts.
4 Evolution of Intense Cyclone Impacts in Future Decades
Although CESM2-LE has biases capturing observed intense cyclone frequency and sea ice response, examining how these impacts may evolve in the future is still instructive. As emphasized above, both storm count and timing modify net MIZ ice area responses. CESM2-LE predicts increases in Arctic storm frequency overall, especially in late summer and in lower latitude bands (Figure 3b). However, CESM2 composite mean trends suggest MIZ ice area will decrease by almost 30,000 per decade and shift poleward (Figure 3a). A progression toward an earlier ice-free Arctic is also evident in future decades causing fewer storms to interact with the ice edge. In near-future decades, MIZ-interacting storms shift poleward. Then, after the 2040s, as September months become ice-free, storm interaction with the MIZ declines (Figure 3c). This transition causes a predominance of early summer storms reaching the MIZ, which tend to decrease ice area more effectively than late summer storms. In later decades, as earlier months become ice-free, the number of early summer MIZ-interacting storms also decreases (relative to the lower number of storms compared with observations in recent years; Figure 3d).
Monthly mean (a) MIZ ice area, (b) storm count, (c) number of storms that interact with the MIZ, per latitude band. (d) Annual storm counts (green) and MIZ area (brown crosses) for both CESM2-LE (no marker) and observed storms (circles). Dashed lines indicate storms that interact with the MIZ and solid lines show total storm count. (e) Observed annual mean MIZ latitude (green) and MIZ area (brown) over the satellite record.
Figure 4 summarizes this transition, showing CESM2-LE composite-mean net cyclone impacts on the MIZ for each month. In the early satellite era (1980s) and present day (2020s), the largest MIZ ice area decreases occur in June and July, with weaker changes in August and slight increasing effects in September, consistent with observations (Mundi & L’Ecuyer, 2025). Generally, storm impacts increase from the 1980s to present as the MIZ becomes more vulnerable to external changes. Intense cyclone impacts in July (and also all months combined) reach their maximum in the 2000s and steadily decline after the 2020s. The decreased ice loss in the 2010s likely results from fewer July storms (Figure S3 in Supporting Information S1). August and September storms have their largest influence on MIZ ice in the 1990s and 1980s, respectively, before transitioning to near-zero impacts around the 2040s. In the near and far future (2060s, 2090s), late summer intense cyclone impacts are minimized as ice-free months become more common. In the far future, August and September become completely ice-free, and the greatest changes to the MIZ occur in May, shifting the largest intense cyclone impacts earlier as less MIZ ice is available later in the season. These predicted changes are plausible given the poleward shift and decrease in MIZ area in recent decades (Figure 3e). As fewer intense storms reach the ice edge with this transition, the model's storm count bias is likely minimized. Since the modeled storm impacts are overestimated, the actual shift in storm impacts may be muted in comparison with this prediction.
Changes in MIZ ice area due to cyclones for May–September, and total change for all 5 months (columns) and four selected decades (rows). Thick black lines depict the ensemble and decadal mean sea ice extent on the first day of the month. Thin contours indicate the mean number of days of each decade with storms occurring in that location. Monthly and total time series on the right show the change in MIZ area for all decades from the early satellite era through 2100. Markers correspond to the four selected decades shown in the maps.
5 Conclusions
As Arctic sea ice becomes less expansive and more vulnerable to external changes in recent years, intense cyclones are playing a more significant role in modulating sea ice cover. Storms can largely reduce MIZ ice area and potentially prompt the first ice-free day (Heuzé & Jahn, 2024). This ice cover loss significantly modifies the energy balance and contributes to the Arctic's amplified warming rate (Stroeve & Notz, 2018), highlighting the importance of predicting future changes in intense cyclone effects.
Our second research question asks how intense cyclone impacts are predicted to change in a warmer climate, and we find a temporary increase in MIZ ice area loss from the 1980s to present, as more susceptible sea ice becomes exposed to passing summer cyclones. Then, despite a predicted increase in future intense Arctic storms, shrinking ice cover and longer ice-free periods from continued climate forcings lead to fewer cyclones reaching the MIZ, reducing the impact of intense storms particularly in late summer (even when scaled by the MIZ ice area, not shown). As a result, the largest cyclone-induced ice area decreases occur earlier in the summer, transitioning from July to June around 2050, then to May toward the simulation's end, total ice-decreasing effects begin to decline after the 2020s, suggesting intense cyclone impacts on summertime sea ice may have reached a maximum.
Regarding our first research question on modeled biases of intense cyclone characteristics, CESM2-LE had fewer cyclones pass over the MIZ, but on average decreased ice area more than observed in past decades. Compared with observations, individual modeled storms were more likely to decrease MIZ ice area, with early summer storms corresponding to SST increases more frequently, highlighting the ocean environment's importance in cyclone-sea ice interactions. The modeled sea ice state and cyclone distribution may also impact the net effect of storms. However, these biases in storm count and impact magnitude reproduced similar net impacts in the model compared with observed storms.
While these two discrepancies partially compensate, these biases limit the ability to assess future changes in CESM2 fully, particularly under changing ice conditions. Better representing Arctic cyclones and their impacts can reduce the uncertainties described in this analysis. Further investigating the key relationship between changing sea ice conditions and intense cyclone development near the ice could increase confidence in the number of storms reaching the ice edge (Valkonen et al., 2025). Additionally, we highlight the importance of upper ocean conditions (including upwelling or wave dynamics, as discussed in Blanchard-Wrigglesworth et al. (2024)) in improving modeled cyclone impacts on the MIZ. By addressing these concerns, predictions of future cyclone impacts can be more accurate. Regardless of model biases, this study emphasizes the significance of the evolving summer sea ice cover when considering intense cyclone impacts.
Acknowledgments
CLM and TSL were supported by NASA CloudSat/CALIPSO Science Team Grant 80NSSC20K0135. AKD acknowledges support for this work from the National Science Foundation (NSF) Award 2043588. This material is based upon work supported by the National Center for Atmospheric Research (NCAR), which is a major facility sponsored by NSF under Cooperative Agreement 1852977. The CESM project is supported primarily by NSF. Computing and data storage resources were provided by the Computational and Information Systems Laboratory (CISL) at NCAR. We thank all the scientists, software engineers, and administrators who contributed to the development of CESM2.
Conflict of Interest
The authors declare no conflicts of interest relevant to this study.
Open Research
Data Availability Statement
All data sets used in this study are publicly available through open repositories. Output from the CESM2 Large Ensemble is available from the Climate Data Gateway (Danabasoglu, Deser, et al., 2020). Hourly ERA5 data on single levels are available to download from the Copernicus Climate Data Store (Hersbach et al., 2023). Sea ice concentration data are available from the National Snow and Ice Data Center (Meier et al., 2021). NOAA OI SST V2 High Resolution Data set data provided by the NOAA PSL, Boulder, Colorado, USA, from their website at https://psl.noaa.gov/data/gridded/data.noaa.oisst.v2.highres.html. Python scripts used for data analysis and plotting are available via Zenodo (Mundi, 2025). Storm counts used for comparison in the Supporting Information are available at Cassano et al. (2023).