Arctic Seasonal Variability and Extremes, and the Role of Weather Systems in a Changing Climate

The effects of global warming are strongly amplified in the Arctic, causing rapidly rising temperatures and ongoing dramatic loss of sea ice, which itself is subject to large interannual variability. We investigate changes in seasonal‐mean temperature and precipitation, its variability and extremes, using large‐ensemble climate model data with a representative concentration pathway 8.5 forcing scenario for historical (S2000) and end‐of‐century projections (S2100). Our results reveal regionally and seasonally dependent changes in Arctic interannual temperature and precipitation variability that are strongly linked to sea‐ice loss. We show a doubling in precipitation variability over the Arctic Ocean and a significant reduction in temperature variability in the Barents Sea. Extremely warm seasons in S2000 rank among the coldest seasons or become unrealistic in S2100. We further show the key role of large‐scale weather systems for shaping seasonal temperature and precipitation extremes in the Arctic which persists under climate warming.

2 of 10 in Arctic temperature variability, especially in winter (Deng & Dai, 2022;Reusen et al., 2019;Screen, 2014). The further reduction of minimum sea-ice extent leads to an increase in sea-ice variability, which in turn is expected to affect surface temperature and precipitation variability (Bintanja & Selten, 2014;Boer, 2009;Mioduszewski et al., 2019). Due to the episodic nature of (increasing) poleward moisture transport, precipitation variability is expected to increase .
There is, however, a considerable lack of knowledge regarding Arctic variability on the seasonal timescale and its change in a warming climate. While several studies investigated changes in the mean state of surface air temperature and precipitation in various climate models (Koenigk et al., 2013;Overland et al., 2014), only few provide robust estimates of changes on the seasonal timescale  and of the occurrence of seasonal extremes relative to the underlying trend. Using over 1,000 years of large-ensemble climate model data, we provide a robust evaluation of seasonal-mean conditions and variability of surface air temperature and precipitation in distinct Arctic regions and different seasons. We further shed light on the question whether present-day extreme conditions such as strongly increased wintertime air temperatures could become the new normal in the near future, as suggested by recent studies (Landrum & Holland, 2020;Overland, 2021). In the second part, we investigate the role of cyclones and anticyclones for the occurrence of such seasonal extreme conditions and whether this relationship will change in future climate states. Better understanding the driving mechanisms behind Arctic seasonal extremes is crucial to assess the characteristics of (future) extremes and their associated impact on the Arctic environment.

Data and Methods
For this study, we use simulations with the fully-coupled Community Earth System Model version 1 (CESM1; (Hurrell et al., 2013)) which are initialized by using restart files from the CESM large ensemble project (CESM-LE; (Kay et al., 2015)). Based on the original CESM-LE data consisting of 35 ensemble members, the simulations used in this study are initialized as described in Röthlisberger et al. (2020), resulting in a 105-member ensemble for a historical period ( S2000; 1990-2000) and an end-of-century period (S2100; 2091-2100). Additionally, simulations of 38 members for two mid-century periods (S2040; 2031-2040 and S2070; 2061-2070) are used. In total this results in more than 1,000 years of data for S2000 and S2100 each, and 380 years of data for S2040 and S2070. Except for S2000, the simulations are set up to project changes under the representative concentration pathway 8.5 (RCP8.5) forcing scenario. We use monthly and seasonal-mean surface-level fields of 2 m temperature, precipitation (sum of rain and snow) and sea-ice concentration (SIC) at approximately 1° horizontal resolution. Further, the classical definitions of the seasons, that is, winter (December-February, DJF), spring (March-May, MAM), summer (June-August, JJA) and autumn (September-November, SON), are applied.
To account for regional differences caused by varying sea-ice extent, weather system frequencies and impacts from lower latitudes, we perform our analyses for two distinct regions: (a) the Kara and Barents Seas, which are strongly affected by recent changes in SIC (Cavalieri & Parkinson, 2012), by mid-latitude cyclones, and by water inflow from the North Atlantic, and (b) the Arctic Ocean, which is less directly influenced by mid-latitude weather systems. A threshold for the seasonal-mean SIC over all simulated years in S2000 (SIC S2000 ) of 0.5 is used to distinguish between areas where primarily sea ice is present (SIC S2000 ≥ 0.5) and areas where primarily open ocean is present (SIC S2000 < 0.5). In our study we analyze the region of the Arctic Ocean where SIC S2000 ≥ 0.5 (AO) and the part of the Kara and Barents Seas where SIC S2000 < 0.5 (KBS) to contrast a mainly ice-covered and a mainly open ocean area. Results for a third region KBI (SIC S2000 ≥ 0.5 in the Kara and Barents Seas) are shown in the supplement. Separate regions are identified for each season based on SIC S2000 in the respective season ( Figure  S1 in Supporting Information S1). It is important to note that the same regions are used for each time period (S2000, S2040, S2070, S2100), irrespective of the changing sea-ice coverage.
In order to evaluate seasonal and regional changes in the Arctic under a warming climate, we analyze spatio-temporal averages of 2 m temperature (T) and precipitation (P) in a T-P phase space. To that end, we first take the seasonal mean before calculating a spatial average for the introduced Arctic sub-regions. The analysis of changes in this phase space has been rerun for several CMIP6 models ( Figure S3 and Tables S1-S17 in Supporting Information S1). The results reveal that CESM1 is in very good agreement with more recent climate models regarding changes in the seasonal-mean and variability of T and P, which enables us to focus on large ensemble CESM1 results in this study.

10.1029/2022GL102349
3 of 10 Based on 6-hourly sea level pressure data, weather system identification schemes (Sprenger et al., 2017) are applied for S2000 and S2100, enabling the investigation of synoptic features such as cyclones and anticyclones. Here, cyclones and anticyclones are defined as objects covering the area around a sea level pressure minimum and maximum, respectively, delimited by the outermost closed sea level pressure contour (Wernli & Schwierz, 2006). Each object is described by a two-dimensional binary field with a value of 1 at grid points inside the object, and 0 outside. By time-averaging these binary fields, we calculate the seasonal fields of cyclone frequency (f c ) and anticyclone frequency (f a ). For example, a cyclone frequency of 30% at a given grid point indicates that at 30% of all times, a cyclone is present at that grid point. Spatio-temporal averaging over a pre-defined area yields seasonal-mean cyclone frequencies for specific regions. Similarly, the seasonal-mean weather system frequency anomaly for a certain region and season is calculated as the deviation of the spatially averaged seasonal-mean weather system frequency of that season from the climatology.
For the composite analysis shown in Figures 2 and 3 and Figures S8-S31 in Supporting Information S1 we extract the 50 wettest and driest, and warmest and coldest seasons in each region, based on spatially averaged P and T, respectively. The composites are compiled based on the weather system frequency anomalies per category (warm: T + , cold: T − , wet: P + , dry: P − ). We further show standardized seasonal-mean weather system frequency anomalies which are calculated as: where we use the mean over the 50 most extreme seasons per category (f x,50 ) and over all seasons (f x,clim ), spatially averaged either over AO and KBS or over a rectangular box that is defined in the Nordic Seas for f c and in Western Siberia for f a , to analyze important remote anomalies. denotes the standard deviation of f c or f a over all seasons, spatially averaged over the respective region/box. For example, for wet seasons in KBS, a ′ of +2 means that the colocated cyclone frequency is increased by +2 standard deviations compared to the climatology of all seasons.

Climate Change in the T-P Phase Space
Analyzing seasonal-mean anomalies with respect to S2000 in the T-P phase space ( Figure 1) reveals a remarkable increase in T in AO and KBS in both winter and summer (see Figure S2 in Supporting Information S1 for other regions and seasons). The largest shift occurs in winter in AO with an increase of +25 K between S2000 and S2100 ( Figure 1a). In this region, the warmest seasons in S2000 rank already among the coldest seasons in S2040 and S2000 T values are not simulated anymore after mid-century. This rapid change does not even allow for a "new normal" to settle in, as S2000 warm extremes quickly turn into cold extremes and soon become unrealistic under a continued forcing scenario. Changes in T variability show considerable regional and seasonal differences. In winter, a relative increase of +24% occurs in AO (Figure 1a), in strong contrast to a reduction of −73% in KBS (Figure 1b), which is likely related to the increasing distance of this area to the sea-ice edge, causing smaller temperature fluctuations over the open ocean (Boer, 2009). Similar changes occur in summer, although with a much smaller amplitude of +5% in AO ( Figure 1c) and −21% in KBS ( Figure 1d).
P increases in all seasons and regions with a larger increase in winter and in areas with ongoing sea-ice retreat such as AO (Figure 1a). A more rapid increase of P in the second half of the century is found in winter as illustrated by the shift of distributions of P values in S2040, S2070, and S2100. This can be explained by the increase in surface evaporation and enhanced moisture transport caused by the ongoing sea-ice reduction and rising T values, pointing toward an intensification of the water cycle (Bintanja & Selten, 2014). In almost all seasons and regions, new wet extremes can be observed for S2100 that have been unrealistic in S2000. In AO, dry winters in S2100 match wet winters in S2000. This is also found in other regions that experience strong sea-ice loss (see Figure S2b in Supporting Information S1), confirming the strong shift toward a wetter climate in these regions. Furthermore, a doubling in P variability occurs in winter in AO with the wettest winters in S2100 receiving the fourfold amount of P compared to the wettest winters in S2000. In KBS, P variability increases more in summer (+32%), leading to the only case where both wetter and drier seasons occur in S2100 compared to S2000 (Figure 1d).
Local changes in SIC and sea surface temperatures might have an important effect on regional and seasonal changes in interannual variability of T and P (Johannessen et al., 2016;Semenov & Bengtsson, 2003), next to reduced dry static energy transport and enhanced moisture transport (Deng & Dai, 2022;Min et al., 2008;Screen, 2014) from lower latitudes. Figure 1e shows severe changes in sea-ice extent and seasonality in AO in the coming decades, with an ice-free ocean from August to November in S2100, compared to only minor decreases for the period March-May. Most simulations in S2070 do already show an ice-free Arctic Ocean in September, which is in line with recent CMIP5 and CMIP6 model simulations (Huang et al., 2017;Notz, 2020;Overland et al., 2014). Thus, the area covered by multi-year ice will diminish drastically throughout mid-century and as a result, most of the Arctic Ocean will only be covered by seasonal ice in S2100, which is typically thin and prone to strong interannual variability. At the same time, a reduction in the seasonality of T is observed in AO with a minimum in March ( Figure S5a in Supporting Information S1), together with a weak increase in the seasonality of P ( Figure S6a in Supporting Information S1). Especially in autumn and winter, the Arctic Ocean is projected to experience remarkable changes in monthly-mean T and SIC, causing increases in evaporation and, thus, precipitation, which further affect regional and seasonal T and P variability (Figure 1a and Figure S2j in Supporting Information S1) (Bintanja & Selten, 2014;Bintanja et al., 2020;Boer, 2009;Reusen et al., 2019). The increasing variation in the position of the sea-ice edge will further Figure 1. Seasonal-mean anomalies with respect to S2000 of T (in K) along the x-axis and P (in mm day −1 ) along the y-axis in (a, c) AO (blue area in Figure S1 in Supporting Information S1) and (b, d) KBS (orange area in Figure S1 in Supporting Information S1) in (a, b) DJF and (c, d) JJA for S2000 (dark blue), S2040 (light blue), S2070 (light green), and S2100 (dark green). Colored dots inside the solid contour show mean anomalies per period and solid (dashed) contours include 80% (95%) of all seasons. A kernel density estimation is applied to estimate the probability density function of T and P. Seasons outside of the 95% contour are shown for the S2000 and S2100 periods. The table shows the change in the seasonal mean between S2000 and S2100 (ΔX) as well as the standard deviation of anomalies in S2000 (σ X,S2000 ) and S2100 (σ X,S2100 ) for both T and P. Monthly-mean sea-ice concentration (SIC) for all four simulation periods are shown for (e) AO and (f) KBS. Colored shading shows the standard deviation of all simulated months relative to the averaged monthly mean. Note that in (a-d) spatial averaging is done over regions defined by seasonal-mean SIC, whereas in (e, f) monthly-mean SIC is used.
influence the formation, propagation and intensification of cyclones in the Arctic and, thus, poleward heat and moisture transport, due to changes in baroclinicity and open water areas (Parker et al., 2022;Simmonds & Keay, 2009;Valkonen et al., 2021). In AO, this will lead to a general increase in T and P variability (Figures 1a  and 1c). In contrast, in KBS where sea ice will completely disappear until S2070 (Figure 1f), the increasing distance to the sea-ice edge and thus reduced potential for cold air advection will contribute to the reduction in T variability (Figures 1b and 1d).  Figure S8 in Supporting Information S1.

Role of Weather Systems for Extreme Seasons
On daily to weekly timescales, synoptic weather systems such as cyclones and anticyclones are strongly affecting Arctic surface temperature and precipitation, facilitating air mass exchange with the mid-latitudes (Laliberté & Kushner, 2014;Messori et al., 2018;Papritz & Dunn-Sigouin, 2020;Woods et al., 2013) and air mass transformations within the Arctic (Ding et al., 2017;Papritz, 2020;Pithan et al., 2018). In their warm sector, extratropical cyclones favor warm and moist air intrusions from lower latitudes into the Arctic, followed by the advection of cold and dry air behind their cold front, thus increasing regional variability of surface temperature and precipitation (Finocchio & Doyle, 2021;Webster et al., 2019). Similarly, polar anticyclones which are often associated with upper-level blocking are strongly influencing Arctic temperature variability as they drive subsidence-induced surface warming Papritz, 2020;Wernli & Papritz, 2018) while suppressing precipitation (Kautz et al., 2022). Variations in the frequency of weather systems further contribute to the formation of extremes on a seasonal timescale (Hartmuth et al., 2022;Papritz, 2020), however their role for Arctic interannual temperature and precipitation variability in a changing climate has only received little attention so far. Several climate models project changes in Arctic weather system frequencies and characteristics such as their intensity and intensification in a warmer climate (Akperov et al., 2019;Rinke et al., 2017;You et al., 2022), with possibly large implications for future T and P variability. We show seasonal-mean changes in weather system frequencies in CESM1 in Figure S7 in Supporting Information S1. However, in the following, we investigate the role of cyclones and anticyclones for the formation of seasonal T and P extremes by analyzing anomalies of cyclone frequency (f c ) and anticyclone frequency (f a ; see Methods) for the 50 warmest, coldest, wettest and driest seasons in S2000 and S2100, separately for the regions AO and KBS. This analysis therefore considers only the about 5% most extreme seasons in the two periods and is not necessarily representative for a mean trend.

Arctic Ocean
Figures 2a and 2b shows extreme season composite differences of winter f c anomalies in AO in S2000 (additional composites for all seasons, regions and weather systems are shown in Figures S9-S32 in Supporting Information S1). Wet and dry winters are characterized by a colocated (i.e., located in the considered region) positive and negative f c anomaly, respectively (Figure 2a), and vice versa for f a ( Figure S10 in Supporting Information S1). This demonstrates the essential role of cyclones for transporting moisture into the Arctic (Woods et al., 2013) and for the generation of lifting and precipitation. In contrast, for winter T extremes, the pattern of weather system anomalies is displaced relative to AO (Figure 2b). Such seasons are characterized by a weak dipole pattern with a positive f c anomaly in the Nordic Seas and a negative f c anomaly mainly over the Kara Sea for warm and vice versa for cold winters. Further, over Western Siberia and the Laptev Sea a pronounced positive f a anomaly is evident during warm winters and a negative f a anomaly during cold winters ( Figure S10 in Supporting Information S1). Thus, next to advection of relatively warm air from lower latitudes, subsidence-induced adiabatic warming in anticyclones is most likely another important process driving winter warm extremes in AO (Papritz, 2020). Figure 2e shows standardized wintertime-mean anomalies of f c and f a for T and P extremes, spatially averaged in AO and in boxes in the Nordic Seas (f c ) and over Siberia (f a ), respectively (for additional regions and seasons see Figures S8 in Supporting Information S1). The standardization allows comparing f c and f a anomalies across regions, seasons and time periods with different climatologies. In addition to the symmetry of wet and dry as well as warm and cold winters in terms of contrasting colocated f c and f a anomalies, Figure 2e illustrates the importance of colocated weather system frequency anomalies for winter P extremes and of more remote anomalies for winter T extremes.
Compared to winter, summer P extremes show even larger standardized f c and f a anomalies (Figure 2f), which are also more confined to the AO region. A colocation of weather system frequency anomalies is also observed for T extremes with a negative f c and positive f a anomaly during warm summers and vice versa for cold summers ( Figures S21 and S22 in Supporting Information S1). Thus, the large-scale configurations are very similar for wet and cold summers as well as for dry and warm summers, which implies that warming by subsidence and increased incoming radiation dominates over temperature advection during warm summer seasons (Wernli & Papritz, 2018).
Investigating now similar composites for the S2100 period reveals only small changes in the large-scale patterns of f c (Figures 2c and 2d) and f a anomalies ( Figure S10 in Supporting Information S1) associated with extreme winter seasons. A slight change in the dipole pattern of f c anomalies is found for T extremes (Figure 2d), with a weaker positive f c anomaly in the Nordic Seas (see also Figure 2e). The positive f a anomaly over Western Siberia during warm winters is more pronounced and more confined to the continent ( Figure S10 in Supporting Information S1). The amplitude of colocated f c and f a anomalies increases for wet winters whereas it slightly decreases for dry winters in S2100 (Figure 2e). For extreme summers, anomalies in weather system frequencies change generally less compared to extreme winters (Figure 2f).

Kara and Barents Seas
Similar to AO, winter P extremes in KBS show strong colocated f c and f a anomalies (Figure 3a). T extremes are again characterized by a dipole pattern, although with a larger amplitude compared to AO (Figure 3b). For warm winters, a strong positive f c anomaly occurs in the Nordic Seas and a concomitant negative f c anomaly over Siberia and Western Russia, and vice versa for cold winters. This is in line with previous studies about the importance of meridional transport for warm and cold winters into this region (Papritz, 2020;Woods et al., 2013).
Again, as in AO, the colocated weather system frequency anomalies are even more pronounced for summer P extremes than for summer T extremes (Figure 3f). At the same time, the dipole signal in f c anomalies for T extremes is smaller than in winter and also features a reversed dipole in f a (see Figures S25 and S26 in Supporting Information S1). A positive f a anomaly in the Kara Sea during warm summers, which is possibly linked to anomalous blocking, underlines the relatively larger contribution of subsidence to the warm anomaly in summer (Papritz, 2020).
A future amplification of anomalies can be observed for colocated f c anomalies during winter P extremes and f c anomalies located upstream of KBS during winter T extremes (Figures 3c-3e). No remarkable changes are found for extreme summers in this region.

Conclusions
Our evaluations of the large-ensemble climate simulations lead to three important conclusions. First, our analyses reveal substantial seasonal and regional differences in the change of Arctic seasonal-mean temperature (T) and precipitation (P) between historical simulations (S2000) and end-of-century simulations using a RCP8.5 forcing scenario (S2100). In addition to the overall warming and wettening of the Arctic, a particularly strong T increase is observed in regions and seasons with the largest sea-ice loss. The most dramatic warming of +25 K occurs in winter over the Arctic Ocean. Seasons ranking among the warmest in S2000 are soon projected to rank among the coldest or become unrealistic. This is in particular the case for High Arctic winter, illustrating the exceptional and unprecedented changes that are projected to occur in the Arctic climate system.
Second, we observe remarkable regional and seasonal differences in the changes of interannual variability of T and P, including a pronounced increase in P variability over the Arctic Ocean in winter and a substantial reduction in T variability in the Kara and Barents Seas in most seasons. These changes in variability can plausibly be related to the severe increase in the seasonal cycle and interannual variability of SIC, as well as to increasing moisture transport toward the high latitudes .
Third, we conclude that in all Arctic regions, extremely wet and dry seasons are strongly driven by colocated anomalies in the occurrence of cyclones and anticyclones. For instance, a higher frequency of cyclones, either due to more passing systems or to more stationary cyclones, lead to more precipitation. In contrast, seasonal T extremes are rather driven by large-scale dipole patterns of weather system frequencies with the largest anomalies outside the region affected by the extreme season. These dipole patterns favor, for example, advection of warm air from distant areas during warm seasons. An exception occurs in summer, when colocated anticyclones are also important for the formation of seasonal warm extremes. Despite the predicted changes in the frequency and intensity of weather systems in the Arctic (Akperov et al., 2019;Rinke et al., 2017;You et al., 2022), it is remarkable that the large-scale patterns determining seasonal T and P extremes are largely unaffected by global warming.
As the Arctic sea-ice cover is projected to experience a strongly increasing seasonality throughout the coming decades, local interactions between sea ice and weather systems will further determine changes in T and P variability. Cyclones have been shown to strongly affect regional sea-ice variability due to both, thermodynamic and dynamic processes (Lukovich et al., 2021;Rinke et al., 2017;Simmonds & Keay, 2009). Inversely, changes in sea ice are further assumed to indirectly affect the formation and intensification of cyclones due to changes in baroclinicity and local heat and moisture fluxes (Overland & Wang, 2010;Valkonen et al., 2021). A better understanding of how these processes are interlinked as well as how they evolve under the projected Arctic warming will be key to improve our knowledge about Arctic variability and extreme seasons, and how they change toward the end of this century.

Data Availability Statement
The original CESM-LE data (Kay et al., 2015) are available from NCAR's Climate Data Gateway at https://www. cesm.ucar.edu/projects/community-projects/LENS/data-sets.html. Processed CESM model data including the weather system data is openly available via the ETH Research Collection (https://www.research-collection.ethz.ch/; doi: https://doi.org/10.3929/ethz-b-000586054). The CMIP6 model outputs used in this study are freely available from the Earth System Grid Federation (ESGF) (esgf-node.llnl.gov/search/cmip6). See Supporting Information S1 for a detailed listing of all CMIP6 data sets used in this study. All figures were produced with Python version 3.9.2 (Van Rossum & Drake, 2009), which is freely available at https://www.python.org/downloads/ release/python-392/. providing the CESM-LE restart files and are thankful to Urs Beyerle (ETH Zurich) for performing the CESM1 simulations. They further thank Michael Sprenger for calculating the weather system data and Matthias Röthlisberger and Mauro Hermann (all ETH Zurich) for constructive feedback and helpful discussions. Two anonymous reviewers are acknowledged for their thoughtful comments which helped to improve the manuscript. This research has been supported by the H2020 European Research Council (Grant 787652).