In our analysis of cyclone impacts on the Arctic sea ice cover we rely on a coupled model simulation, which enables us to decompose the sea ice changes into dynamic and thermodynamic contributions. The simulation was performed applying version 2.2 of the coupled regional climate model HIRHAM–NAOSIM. This version represents a further development of the base version 2.0, which is described and evaluated by Dorn et al. (2019). Version 2.2 includes new parameterizations and changed parameter settings. The differences between the two versions are listed in the Supplementary Material.

HIRHAM–NAOSIM is applied over a circum-Arctic domain using rotated latitude-longitude grids with horizontal resolution of 1/4° (∼ 27 km) in the atmosphere component HIRHAM and 1/12° (∼ 9 km) in the ocean–sea ice component NAOSIM. More detailed information on the model components and their coupling are given by Dorn et al. (2019).

The simulation was initialized on 1st of January 2019 and run through 31st of December 2020, driven by ERA5 reanalysis data (Hersbach et al., 2020) at HIRHAM’s lateral boundaries as well as HIRHAM’s lower and NAOSIM’s upper boundaries, which lie outside the coupling domain (defined as the overlap area of the components’ model domains). For NAOSIM’s open lateral boundaries, ORAS5 reanalysis data (Zuo et al., 2019) were used. HIRHAM was initialized with the corresponding ERA5 fields, while NAOSIM was started from rest with temperature, salinity, ice thickness, and ice concentration fields from ORAS5. HIRHAM’s prognostic fields were nudged to the corresponding ERA5 fields with a uniform nudging time scale of 16.67 h (which corresponds to a nudging of 1% per time step).

We complement the HIRHAM–NAOSIM simulation with in-situ observations obtained during the Multidisciplinary drifting Observatory for the Study of Arctic Climate (MOSAiC) expedition (Shupe et al., 2020) and ERA5 reanalysis data to demonstrate that the model is able to 1) capture the synoptic situation (Supplementary Figures S1, S2) and 2) produce a realistic spatial pattern of SIC changes (Supplementary Figure S3) during the cyclone passages. It should be noted that the ERA5 SIC field is highly smoothed and has limitations to represent the observed strong gradient in SIC across the MIZ (Renfrew et al., 2021). However, for this cyclone case, the spatial patterns of SIC changes based on ERA5 are in strong agreement with high-resolution satellite derived SIC data based on AMSR (not shown). To further evaluate the simulated sea ice thickness (SIT) with remote sensing observations (Supplementary Figure S4), we utilize merged CryoSat-2 and SMOS satellite data (Ricker et al., 2017b).

During February 2020, RV Polarstern was located close to the North Pole in the central Arctic (Figure 1). The supplementary model evaluation is based on data from the 10-m meteorological flux tower installed at the meteorological observatory (Met-City) (Cox et al., 2021a) located in approximately 500 m distance to the RV Polarstern, three autonomous Atmospheric Surface Flux Stations (ASFS) (Cox et al., 2021b; Cox et al., 2021c) situated in the Distributed Network in a distance of approximately 25 km to RV Polarstern (Shupe et al., 2022), and a microwave radiometer HATPRO (Humidity and Temperature Profiler) (Walbröl et al., 2022). We use data averaged to 3-hourly resolution of 2-m air temperature, vertically integrated water vapour (IWV; 0–10 km height), mean sea level pressure (SLP) and 10-m wind speed. For the comparisons with the simulation, we use the nearest model grid-cell. IWV in the model simulation is the integrated specific humidity over all vertical levels from the surface up to the top (10 hPa, approx. 35 km height). To further evaluate the simulated spatial patterns of meteorological variables, we use 3-hourly ERA5 gridded data of 2-m air temperature, IWV, and SLP. We also use daily SIC data from ERA5, which are based on satellite data (HadISST2 and OSI SAF; see Hersbach et al., 2020).

The main objective of this study is to separately quantify dynamic and thermodynamic sea ice changes during and after the cyclone passages to gain insights into the underlying mechanisms. We approach this by temporally integrating HIRHAM–NAOSIM’s dynamic and thermodynamic SIT and SIC tendencies for specific time periods in order to decompose the overall sea ice changes in its respective components.

The overall sea ice changes are given by the model’s continuity equations for SIT (h) and SIC (A) as ∂ h ∂ t = − ∇ ⋅ h v ⃗ + S h ice + S h ow , (1) ∂ A ∂ t = − ∇ ⋅ A v ⃗ + S A ice + S A ow + D A , (2) where v ⃗ is the ice velocity and S h ice , S h ow , S A ice , and S A ow are the thermodynamic growth rates, which are separately calculated for the ice-covered (superscript ‘ice’) and the open water part (superscript ‘ow’) of the grid cell. A detailed description of the thermodynamic growth rates is given by Dorn et al. (2009). The term D _A represents the formation rate of open water due to shearing deformation (ridging) and is given as D A = 0.5 Δ − | ∇ ⋅ v ⃗ | exp − K 1 − A , (3) where Δ represents the total deformation, determined by the strain rate tensor ϵ ̇ i j (see Hibler, 1979), and K = 20 is an empirical constant. Consequently, the dynamic SIT and SIC tendencies are defined as ∂ h ∂ t d y n = − ∇ ⋅ h v ⃗ , (4) ∂ A ∂ t d y n = − ∇ ⋅ A v ⃗ + D A , (5) and the thermodynamic SIT and SIC tendencies are ∂ h ∂ t t h d y n = S h ice + S h ow , (6) ∂ A ∂ t t h d y n = S A ice + S A ow . (7)

Generally, we analyze these sea ice tendencies in concert with simulated changes in sea ice drift and surface energy budget (SEB). Since we focus on the impact of atmospheric variability on the sea ice–ocean system, we define the SEB as the sum of atmospheric net radiative, sensible, and latent heat fluxes at the surface, with positive values corresponding to a surface energy gain. To cover the importance of the ocean for (thermodynamic) sea ice processes, we additionally provide information on upward oceanic heat fluxes when discussing thermodynamic sea ice changes in Section 3.3.1.

From February 9 to 25, 2020, a sequence of three cyclones travelled through the BKS (Figure 1), shaped the local weather conditions, and led to shifts in the position of the sea ice edge (15% SIC contour). The minimum SLP in the core of the cyclones was below 970 hPa (when crossing the sea ice edge) for all events, which is an extremely low value compared to climatological SLP conditions in the BKS region. Consequently, all three events can be classified as intense, stronger than normal cyclones (following Rinke et al., 2017). This classification is supported by a comparison of the intensity of cyclones during the MOSAiC expedition (and particularly in February 2020) with climatological cyclone conditions along the MOSAiC drift track (Rinke et al., 2021).

The strongest cyclone event (in the following referred to as cyclone 2) occurred during February 16–20 (Figure 1) with a minimum pressure of less than 960 hPa. The cyclone travelled through the southern Barents Sea, crossed the ice edge near Novaya Zemlya, and entered the central Arctic through the western Kara Sea. The associated advection of a comparatively warm and moist air mass on the eastern flank of the cyclone into the Arctic impacted the eastern Barents Sea, the Kara Sea, and eventually the central Arctic north of Franz Josef Land and Svalbard (Figure 2). On February 19, the 2-m air temperature in large parts of the central Arctic reached values slightly below the freezing point, which corresponds to an increase of approximately 20 K in only 2 days compared to February 17 (Figures 2A–C). Close to the North Pole, a rise in 2-m air temperature from −30°C to −10°C as well as high IWV of up to 6 kg/m² was observed when the cyclone hit RV Polarstern (Supplementary Figure S2). Both conditions were extremely anomalous and near-record breaking (Rinke et al., 2021).

Both before and after this particular cyclone event, the BKS region was affected by another cyclone with comparatively similar intensity and track (Figure 1, Supplementary Figures S5, S6). The first of the three cyclones occurred during February 10–13 (in the following referred to as cyclone 1), crossed the central Barents Sea between February 11–12, and entered the central Arctic close to Franz Josef Land. The last of the three consecutive cyclones (in the following referred to as cyclone 3) occurred during February 21–25, entered the Barents Sea on February 22, and followed almost the same path as the second cyclone for most of its lifetime. However, in contrast to cyclone 2, it decayed quicker, i.e., one day earlier than cyclone 2, after reaching the central Arctic between Svalbard and Franz Josef Land.

The comparison of the nudged coupled model simulation with MOSAiC data demonstrates that the observed atmospheric variability in the central Arctic during this series of cyclones is captured well by the model (Supplementary Figure S2). With respect to larger spatial scales, the simulated patterns of 2-m air temperature and IWV over the BKS region agree with those of ERA5 (Supplementary Figure S1). This confirms the validity of the atmospheric forcing in the simulation.

Analyzing the SEB for cyclone 2 (Figures 3A–C) reveals that starting with February 18, the advection of cold and dry air at the back side of the cyclone leads to a strong energy transfer from the ocean to the atmosphere in the Barents Sea. Over the open ocean, but also in parts of the marginal ice zone (MIZ), i.e., in the central Barents Sea and west of Svalbard, the SEB reaches values below −350 Wm⁻². On February 19, this advection of cold dry air further extends to the eastern Barents Sea. Comparing individual components of the SEB indicates that this intensification of the usually slightly negative wintertime SEB is almost exclusively driven by turbulent heat fluxes (Figures 3D–F). Changes in net longwave radiation associated with the cyclone passage (determined by an increase in longwave downward radiation) lead to a slightly less negative SEB in the Kara Sea and central Arctic (Figures 3G–I), but this signal does not reach the same order of magnitude as the negative SEB change due to turbulent heat fluxes.

The SEB change during all three cyclone cases (Figure 3, Supplementary Figures S5, S6) is in contrast to reports on strong positive SEB changes (energy gain of the surface) in ice-covered grid-cells during an extreme cyclone in December 2015/January 2016 (Boisvert et al., 2016) and during the record Arctic cyclone in January 2022 (Blanchard-Wrigglesworth et al., 2022). For our presented mid-February 2020 case, only small patches of slightly positive SEB are found during a very few days, i.e., during cyclone 2 on February 18 southwest of Novaya Zemlya (Figure 3B) and during cyclone 1 on February 12 over the Barents Sea (Supplementary Figure S5). Apart from that, the SEB remains negative in ice-free grid-cells and is close to zero in ice-covered grid-cells. Since both the December 2015/January 2016 cyclone and the January 2022 cyclone entered the BKS close to Svalbard on a more northerly route than the Mid-February 2020 cyclones, it can be supposed that there is a strong variability in the surface impacts of individual cyclones depending on their track and presumably also on further cyclone properties.

As a next step, we analyze changes in SIC from directly before the start of the first cyclone passage to the end of the third one (from February 9 to February 25). Figure 4A shows that the sequence of cyclones causes a strong decrease (increase) in SIC in the eastern (western) part of the study domain, particularly in the MIZ. Strongest changes exceeding values of 50% are found in the vicinity of Novaya Zemlya and Svalbard. The simulated spatial pattern of SIC changes shows strong similarities to satellite observations of SIC contained in the ERA5 reanalysis (Supplementary Figure S3), which confirms the suitability of the coupled model simulation for our study.

In this section, we extend our analysis of cyclone impacts to dynamic and thermodynamic changes in SIT, which have been rarely studied. A comparison of the simulated mean SIT from February 9 to 22 February 2020 with observations based on CryoSat-2 and SMOS satellite data (Ricker et al., 2017b) demonstrates that the spatial distribution of regions with relatively thin ice and relatively thick ice is captured by the model (Supplementary Figures S4A, B). However, it should be noted that the simulated SIT field is generally too smooth, leading to an underestimation of SIT in the central Arctic and to an overestimation of SIT in the Kara Sea and some parts of the marginal ice zone.

Simulated SIT changes during the cyclone passages mainly consist of a decrease in SIT east of Novaya Zemlya and an increase in SIT northwest of Svalbard (Figure 7A). Generally, there are some differences between the simulated and observed SIT response to the cyclone passages (Supplementary Figures S4C, D), which should be kept in mind when interpreting the results. However, the previously described main features around Novaya Zemlya and Svalbard are consistent in both datasets, which confirms the applicability of the simulation for this case study.

To generalize the results from this case study and move towards more solid conclusions on dynamic and thermodynamic contributions to cyclone-driven sea ice changes, we extend our analysis period to the whole winter of our selected year, specifically to January-March 2020.

The time series of the daily SLP averaged over the BKS region (Figure 9, domain shown in Figure 4) is an indicator for cyclone activity in the BKS. The figure highlights the main cyclone cases during the period and their tracks are shown in the Supplementary Figure S9. The cyclone at the beginning of February affected the western BKS around Svalbard for several days starting at February 2, before moving southeastward through the Barents Sea while decaying. The mid-February case of three successive cyclones was discussed in detail in Section 3. The two March events affected the BKS for several days each, with one cyclone taking a similar path as the mid-February cases (through the southern Barents Sea to the Kara Sea into the central Arctic) and the other cyclone entering the BKS on a more northerly track close to Svalbard. They occurred one after the other at intervals of approximately 1 week.

To discuss the importance of dynamics and thermodynamics for the related SIC changes, we follow the approach of Section 3 and show in Figure 9 the time series of the difference between the absolute value of the temporally integrated dynamic SIC change and its thermodynamic counterpart, integrated over 5 days and averaged over the BKS region. Based on our previous analysis, 5 days are an appropriate time period to capture the SIC impacts of individual cyclones (see, e.g., Figure 5).

First of all, the SIC difference is positive throughout the 3 months with a mean value of approximately 4% (indicated by the black dashed line in Figure 9). This indicates that for the complete January-March 2020, dynamic SIC changes dominate the overall SIC changes in the BKS region. This is in accordance with our previous analysis (Figure 4). Importantly, there is a significant temporal variability in the SIC difference, and thus in the relative importance of dynamics and thermodynamics, often associated with changes in SLP as indicated by the red line in Figure 9.

In agreement with our previous analysis of three consecutive cyclones in mid-February (Figure 4; Figure 6), the SIC difference is shifted towards more positive values (compared to the mean value) during these cyclone passages. This indicates a dominance of dynamic over thermodynamic contributions for up to 2 weeks. After the last of the three cyclones however, a shift towards smaller values indicates an increasing relative importance of thermodynamic processes for SIC changes, which lasts for about 2 weeks with comparatively high air pressure. A similar phase of higher relative importance of thermodynamic SIC changes under high air pressure conditions can be found end of January.

The process of enhanced dynamic SIC changes during and enhanced thermodynamic SIC changes after a cyclone passage is also found for the two cyclone cases that affected the BKS in March 2020 (Figure 9). After both of these cyclones, it took approximately 1 week before the next cyclone arrived in the BKS. Consequently, there was time for thermodynamic SIC changes to take the lead. This supports the hypothesis raised in Section 3 that enhanced thermodynamic SIC changes can take place after cyclone passages if not another cyclone follows in quick succession. In accordance with this, Figure 9 does not show a period of enhanced thermodynamic SIC changes after the passage of the early February cyclone case. This can be related to the fact the series of the mid-February cyclones started shortly afterwards.