Properties and dynamics of mesoscale-eddies in the Fram Strait from a comparison between two high-resolution ocean-sea ice models

. The Fram Strait, the deepest gateway to the Arctic Ocean, is strongly inﬂuenced by eddy dynamics. Here we analyse the output from two eddy-resolving models (ROMS and FESOM) with around 1 km mesh resolution in the Fram Strait, with focus on their representation of eddy properties and dynamics. A comparison with mooring observations shows that both models reasonably simulate hydrography and eddy kinetic energy. Despite differences in model formulation, they show relatively similar eddy properties. The eddies have a mean radius of 4.9 km and 5.6 km in ROMS and FESOM, respectively, 5 with slightly more cyclones than anticyclones (ROMS: 54%, FESOM: 55%). The lifetime of detected eddies is relatively short in both simulations (ROMS: 10 days, FESOM: 11 days), and the mean travel distance is 35 km in both models. More anticyclones are trapped in deep depressions or move toward deep locations. The two models show comparable patterns of baroclinic and barotropic instability. However, ROMS has relatively stronger eddy intensity and baroclinic instability, possibly due to its smaller grid size and higher effective resolution. Overall, the relatively good agreement between the two models 10 strengthens our conﬁdence in their ability to realistically represent the Fram Strait ocean dynamics, and also highlights the need for very high mesh resolution.

Given the possible sensitivity of simulations to model numerics, to the complex bottom topography and ocean currents in the Fram Strait, it is not known whether the above cited models have a broad agreement on the representation of eddy dynamics in terms of eddy generation and propagation. Answering this question will not only add credence to our understanding of eddy dynamics, but also create a reference for developing parameterisations required by coarse resolution ocean models. The aim of 60 this study is two-fold. First, we compare the output of two high-resolution, eddy-resolving ocean-sea ice models to answer the above question. We will show that there is good agreement in energy conversion that maintains eddy dynamics and in simulated eddy statistics as well, despite the fact that these models, namely ROMS  and FESOM (Wekerle et al., 2017), differ in many aspects such as numerical discretisation, horizontal and vertical mesh resolution, parameterisations, global vs. regional configurations. Second, we explore and describe the properties of eddies in the Fram Strait. We use an eddy-65 following approach to generate regional statistics focusing on the following questions: How are eddies spatially distributed?
Are anticyclones or cyclones dominating? What is their typical size, lifetime and what are their main travel pathways?

Model description FESOM
Model output from the Finite-Element Sea-ice Ocean Model (FESOM) version 1.4 (Wang et al., 2014;Danilov et al., 2015) is 70 used for eddy detection and tracking in this study. FESOM is an ocean-sea ice model which solves the hydrostatic primitive equations in the Boussinesq approximation and is discretised with the finite element method (Wang et al., 2008). In the vertical, z-levels are used. We use a global FESOM configuration that was optimised for the Fram Strait with regional resolution (grid size) refined to 1 km in this area, and a coarser resolution elsewhere (1 • resolution throughout most of the world's oceans, 24 km resolution north of 40 • N and 4.5 km resolution in the Nordic Seas and Arctic Ocean; Wekerle et al. (2017)). By comparing 75 with the local Rossby radius of deformation (around 3-6 km in the Fram Strait, see above), this configuration can be considered as "eddy-resolving". It is forced with atmospheric reanalysis data from COREv.2 (Large and Yeager, 2008), and river runoff is taken from the interannual monthly data set provided by Dai et al. (2009). Tides are not taken into account in the FESOM configuration used here. The simulation covers the time period 2000-2009, and has daily output. In this study, we analyse model output for the years 2006-2009. 80

Model description ROMS
The second high-resolution model simulation used in this study is based on the Regional Ocean Modeling System (ROMS) (Budgell, 2005;Haidvogel et al., 2008;McWilliams, 2005, 2009) with a configuration optimised for Fram Strait and the waters around Svalbard (called S800). With 800 m x 800 m horizontal resolution, S800 is eddy resolving in Fram Strait. S800 was initialised with and forced at the ocean boundaries with daily ocean and sea ice data from a 4 km 85 resolution pan-Arctic model called A4, together with tidal elevations from global TPXO tidal model (Egbert and Erofeeva, 2002). A4's initial state and boundary conditions were taken from monthly-averaged global reanalyses (Storkey et al., 2010).
Atmospheric forcing in A4 and S800 used 6-hourly ERA-Interim reanalysis (Dee et al., 2011). A4 was initialised in 1993, and following A4 spin-up S800 was initialised in January 2005. Analyses in this paper are done for the period of 2006-2009. Model characteristics of ROMS, and also of FESOM, are summarised in Table 1. Additional information about S800, including 90 discussions of its ability to reproduce boundary current observations in Fram Strait and along the continental slope north of Svalbard, is given in Hattermann et al. (2016), Sundfjord et al. (2017), Crews et al. (2018) and Crews et al. (2019).

Eddy detection and tracking
Eddy detection and tracking algorithms are important tools to understand eddy properties such as their size, strength, lifetime and travel pathways. For datasets as large as the output of ocean models, automated methods need to be used. Eddy detection 95 methods can be assigned to two categories, based either on (1) geometrical or on (2) physical characteristics of the flow field, or on a combination of both. In this study, we apply a method developed by Nencioli et al. (2010) to detect and track eddies simulated with ROMS and FESOM, which is based on the geometry of velocity vectors and thus belongs to the first category of methods. The eddy detection is based on four constraints derived from the general characteristics of velocity fields in the presence of eddies, e.g. the reversal of the flow field across the eddy centre. Two parameters, a and b, which determine the 100 minimum size of detectable vortices, have to be set in the algorithm. After some sensitivity tests, we set a = 4 and b = 3, which equals the values used in the test case of Nencioli et al. (2010). Note that our mesh resolutions (800 m and 1 km in ROMS and FESOM, respectively) are similar to theirs (1 km). Eddy boundaries around each detected centre are determined by the outermost closed contour of the stream function field.
To cross-validate our results, we also the Okubo-Weiss criterion, which belongs to the second category of methods (Okubo, 105 1970;Weiss, 1991). Eddies are identified as areas where vorticity dominates over strain. More precisely, the area where the Okubo-Weiss parameter is below a threshold of OW 0 = −0.2σ OW with same sign of vorticity, where σ OW is the spatial standard deviation of OW, is considered as an eddy (Isern-Fontanet et al., 2006). Here (u, v) is the horizontal velocity field, and f is the Coriolis parameter.
After eddies are detected, eddy tracks are computed by comparing eddy centres in successive time steps. More precisely, if 110 two eddies at successive time steps lie within a search radius and have the same sense of rotation, they form a track. The eddy tracking scheme is thus sensitive to the prescribed search radius. A too small value might lead to a false splitting of the track, whereas a too large value would lead to more than one eddy within the searching area. As a first approximation, eddies are advected with the mean current. Considering a mean velocity of around 0.2 m/s (see e.g. Figure 5 in Wekerle et al. (2017)) and a daily mean model velocity field, a possible choice would be a search radius of 17 km. After performing sensitivity tests with 115 different radii, we chose a radius of 14 km. This value reduced the number of occasions when several eddies were detected in the searching area. Furthermore, eddies with a lifetime shorter than 3 days were discarded. (2)

Energy budget
An energy budget can be obtained by expressing velocity as u = u + u as described in the previous section, inserting it in the momentum equation in the Boussinesq approximation, multiplying the equation with u , and time-averaging it. This leads to a conservation equation for EKE. The change of EKE in time is governed by the advection of eddies, energy transfer from mean kinetic energy (MKE) and eddy available potential energy (EPE) to EKE, and energy dissipation (vertical mixing and 140 horizontal diffusion) (e.g. Olbers et al., 2012, chapter 12): where b = − g ρ ρ0 is the buoyancy and D i and V i are horizontal and vertical dissipation terms. Cartesian tensor notation with summation convention has been used, with i = 1, 2 and j = 1, 2, 3. u i is thus the horizontal component of the velocity vector u j , and u 3 = w is the vertical velocity. In this study, we diagnose the first two terms on the right hand side of the equation.
They are the main source terms of EKE, and indicate barotropic and baroclinic instability.

145
For more than two decades, mooring measurements have been conducted across Fram Strait at around 79 • N to monitor the exchange of water masses through this gateway (e.g. Beszczynska-Möller et al., 2012;von Appen et al., 2016;von Appen et al., 2019). To assess the overall model performance in reproducing the mean state and resolving the flow variability, we use the observed hydrography as well as the velocity field and compare the latter in terms of power density spectra (PDS) and EKE 150 to the model results.
The two models simulated very similar spatial distributions of water masses. The simulated mean temperature and salinity at 100 m depth reveal that the warm (>5 • C) and narrow WSC closely follows the 1000 m isobath along the Svalbard shelf break correctly represent the main water masses. ROMS shows a slightly cold bias which is not present in FESOM (ROMS: root mean square (rms) error of 1.28 • C, FESOM: rms error of 0.49 • C), and has earlier been identified to be associated with a cold bias in the A4 model that provides the inflow boundary conditions for S800 . The simulated water mass in FESOM, particularly in central and eastern Fram Strait, is slightly too saline, whereas it is slightly too fresh in ROMS.

160
The overall rms error in salinity is 0.26 and 0.31 in ROMS and FESOM, respectively.
For the comparison of velocity time series, current meter data from three moorings located in the WSC and three moorings between individual deployments) and from the models were estimated via the Thomson multitaper method (Figure 4). For the estimated PDS, the median in log 10 (0.05/day) frequency steps was calculated for frequencies between -1.2/day and -0.35/day, and the slopes were fitted to those binned values. The slopes of the observations are ∼-1.7 and ∼-1.5 for WSC and EGC 170 moorings, respectively, while ROMS/FESOM showed slopes of ∼-1.7/∼-2.1 and ∼-2.2/∼-2.7 respectively. The difference between the models is larger at high frequency, which might be related to the fact that tides were simulated in ROMS.
A seasonal cycle of EKE in 75 m depth computed from current meter data of moorings deployed across Fram Strait is shown in Figure 5a. The highest level of EKE is reached in the winter months (January-March), and lowest values are reached in early autumn (September-November). There is a strong lateral gradient from west to east, with a much higher level of EKE in 175 the eastern part of Fram Strait, the WSC region. Both models well reproduce the observed seasonal and spatial variations of EKE (Figure 5b and c), except that the observation shows a higher EKE level in the central Fram Strait than the models.

Eddy size
In this study we compute the eddy radius as average distance from the eddy centre to the eddy boundary, which is defined by the outermost closed contour of the stream function field. Eddies detected in both models are relatively small, with 95%/92% of cyclones and 92%/87% of anticyclones in ROMS/FESOM having a radius below 10 km (Figure 7a). The mean/median radius for ROMS and FESOM is 4.9/4.1 and 5.6/4.7 km, respectively (Table 2). Eddies simulated in FESOM are thus slightly larger 195 than in ROMS. The eddy radius compares well with the Rossby radius of deformation (∼4-6 km in summer and smaller values in winter ). This suggests that baroclinic instability is likely the main mechanism of eddy generation, which will be further investigated in Section 5. In both simulations, cyclones are slightly smaller than anticyclones (Table 2).

Eddy intensity
The mean/median intensity of eddies (expressed as the absolute value of relative vorticity normalised by f , averaged over all 200 detected eddies) simulated by ROMS and FESOM is 0.4/0.36 and 0.28/0.24, respectively. Eddies simulated by FESOM are thus weaker than eddies simulated by ROMS (see also Figure 7b and Table 2). The proportion of eddies with absolute values below 0.3 is larger for FESOM (63%) than for ROMS (38%). Cyclones are slightly more intensive and have a larger standard deviation than anticyclones in both models (Table 2).

205
The duration over which eddies are continuously detected by the employed method is on average 10 and 11 days in ROMS and FESOM, respectively (Figure 7d). 85%/82% of eddies detected in ROMS/FESOM have lifetimes below 15 days, whereas only 4%/6% of eddies detected in ROMS/FESOM have lifetimes above 30 days. Pathways of these long-living eddies will be analysed in the next section. Note that the eddy lifetime may be longer if one considers that eddies likely can exist for some time before and after being detected as an eddy by the tracking method. Also, a false splitting of the track could occur if the eddy 210 moved relatively fast in combination with a too small searching area. In both simulations, there is no significant difference in lifetime regarding polarisation. They are very similar regarding travel distance. On average, eddies travel around 34 and 35 km in ROMS and FESOM, respectively (Table 2). Again, there is no significant difference in travel distance regarding polarisation ( Figure 7e). Compared to eddies generated e.g. in the Gulf Stream region, the lifetime of Fram Strait eddies is rather short (Kang and Curchitser, 2013).

Eddy pathways
Eddy pathways are investigated by focusing only on long-living eddies, e.g. eddies with lifetime of more than 30 days, and by classifying them by generation areas (Figures 8 and 9). In both simulations, eddies generated on the Svalbard shelf have very distinct travel pathways for cyclones and anticyclones, which is consistent with their distribution (Figure 6e and f). Cyclones tend to stay on the shelf, and populate the narrow Svalbard fjords. Anticyclones in contrast leave the shallow shelf area and 220 tend to travel westward into the deep basin. As shown in Figure 7c, more cyclones (31% and 25% in ROMS and FESOM, respectively) are detected in shallow areas with water depths less than 500 m than anticyclones (21% and 19% in ROMS and FESOM, respectively). Note that as the number of detected eddies on the East Greenland shelf is relatively small in both simulations, most eddies detected in shallow areas are located on the Svalbard shelf.
Anticyclones generated in the WSC core region, here defined approximately as the area between the 500 m and 2000 m 225 isobaths, show longer travel pathways than cyclones. In both simulations, most of them travel westward along the recirculation pathway north of the Molloy Deep , and some even continue southward along the East Greenland continental shelf break. Some eddies travel northward along the western rim of the Yermak plateau or recirculate around the Molly Deep, while only few trajectories deviate westward south of 79 • N in both models.
The asymmetric pathways of eddies generated on the Svalbard shelf and in the WSC core region can have dynamical reasons.

230
As described by Cushman-Roisin (1994, Chapter 17), fluid parcels surrounding a rotating eddy are stretched when they move to deeper waters and thus acquire relative vorticity. In contrast, when moving to shallower waters, on the flank of the eddy the surrounding fluid is squeezed and thus relative vorticity is decreased. This results in a secondary drift of the vortices, with cyclones moving towards shallower regions and anticyclones moving to deeper regions. Morrow et al. (2004), based on satellite altimetry, showed that this dynamical reasoning can explain the diverging pathways of cyclones and anticyclones in different 235 ocean basins.
Tracks of long-living eddies generated in the southern central Fram Strait, in particular those simulated in ROMS, show a high density of anticyclones in the Boreas Basin, the region between 0 • EW-5 • E, 76 • N-77 • N. More anticyclones appear to be trapped in this depression, a similar situation as occurring in the Lofoten Basin (Raj et al., 2016;Volkov et al., 2015). As in the case of eddies generated along the Svalbard shelf break, the clustering of anticyclones can be explained by the dynamical reason Regarding eddies present in northern Fram Strait, both ROMS and FESOM show a high density along the western flank of the Yermak Plateau. Additionally, ROMS shows more long-living (>30 days lifetime) eddies west of the Plateau ( Figure   245 6a-d) than FESOM (Figures 9 and 8). Eddies in this region have earlier been identified to occur with a different seasonality than would be expected from changes in baroclinic instability of the boundary current that explains the seasonality in eddy occurrence along other parts of the shelf break (Crews et al., 2019). A difference between the two models is the inclusion of tidal forcing in ROMS. The circulation and water mass transformations above the Yermak plateau are known to be strongly influenced by barotropic to baroclinic tidal conversion and mixing at the semi-diurnal critical latitude (Fer et al., 2015), that 250 may also explain the enhanced eddy generation in this region in ROMS. As revealed by FESOM, more cyclones tend to follow the Svalbard Branch, whereas more anticyclones tend to follow the Yermak Branch.

Vertical extent and hydrographic properties
We determined the vertical extent of eddies detected in 100 m depth with lifetime above 30 days by calculating relative vorticity/f in the eddy centres in the water column ( Figure 10). In addition, temperature and salinity anomalies were calculated 255 in the same way to study the hydrographic properties of eddies, with anomalies computed relative to the mean value for the month. This was done for eddies generated in the five different regions shown in Figure 1

Energetics in the eastern Fram Strait
We now analyse the source of EKE as simulated in ROMS and FESOM. We focus here on the eastern side of Fram Strait, which is the most energetic region ( Figure 5). As described in section 2.5, the change of EKE in time is governed by the advection of eddies, energy transfer from mean kinetic energy (MKE) and eddy available potential energy (EPE) to EKE, and energy dissipation. In this study we analyse only the first two terms on the right hand side of the EKE conservation equation (Eq. 3), 270 which are the main source terms for EKE and are related to barotropic and baroclinic instability. 9 https://doi.org/10.5194/os-2020-24 Preprint. Discussion started: 14 April 2020 c Author(s) 2020. CC BY 4.0 License.

Barotropic instability
The transfer of MKE to EKE is related to barotropic instability. It can be expressed as the sum of two terms, the product of horizontal eddy Reynolds stress and horizontal mean shear, and the product of vertical eddy Reynolds stress and vertical mean shear. Strong velocity shear thus support barotropic instability. Here we consider only terms that contain horizontal derivatives, 275 and assume that the terms with vertical derivatives play a minor role (as shown for the Gulf Stream region by Gula et al. (2015)).
In the two models, the energy conversion between MKE and EKE is directed in both ways: it shows an alternating pattern, with positive values indicating conversion from MKE to EKE and negative values indicating conversion from EKE to MKE (Figure 11a,b) 1 . The alternating pattern is very similar between the both models, with consistent locations and magnitude of positive and negative energy transfer. The energy transfer occurs mainly along the pathway of the WSC core, which is located 280 approximately along the 500-1000 m isobaths (see also Figure 3).
The relatively similar pattern in both models suggests that there is a strong influence of bathymetry, which determines positive and negative spots of energy conversion. A necessary condition for barotropic instability is that β − ∂ yyū vanishes within the domain, where u =ū(y) is a zonal current with arbitrary meridional profile (e.g. Cushman-Roisin, 1994). The planetary potential vorticity is weak and can be ignored in polar regions, so we only consider the topographic β, with β = 285 − f H ∇H quantifying the change in potential vorticity across the bathymetry and H and ∇H being the water depth and its horizontal gradient. A map of the topographic β west of Svalbard reveals large values along the Svalbard shelf break ( Figure   12a). We take the monthly mean meridional velocity v from FESOM as an approximation of the along-stream velocity, and compute β − ∂ xx v (Figure 12b). In many places along the Svalbard shelf break, β is much larger than ∂ xx v. However, in some places, e.g. at the entrance of Kongsfjorden (79 • N) and Isfjorden (78 • 10'N) and along the 250 m isobath at around 80 • N, 290 β−∂ xx v changes sign. These regions are characterised by positive values of energy conversion in both models, indicating active barotropic instability there. This is comparable to the Norwegian continental slope off the Lofoten islands, where barotropic instability becomes particularly important in regions with steep bottom slope (Ghaffari et al., 2018).

Baroclinic instability
For baroclinic instability to be active, a horizontal density gradient must be present to provide available potential energy 295 which can be converted to EKE. This transfer from EPE to EKE can be expressed as the mean vertical eddy buoyancy flux (Eq. 3). In contrast to barotropic instability, the energy conversion between EPE and EKE in FESOM and ROMS is directed mostly one way, with mainly positive values revealing conversion from EPE to EKE (Figure 11c,d). As the eastern Fram Strait is temperature-stratified, it is mainly the vertical eddy temperature flux that contributes to vertical eddy buoyancy flux    (Tverberg and Nøst, 2009), with upward sloping isopycnals near the seafloor that locally enhance the APE of the mean 305 field.
A necessary condition for baroclinic instability is that the cross-stream gradient of Ertel potential vorticity (PV) changes sign with depth (e.g. Spall and Pedlosky, 2008). Ertel PV Π is defined as Here we compute Π from simulated long-term mean velocity u = (u, v, w) and buoyancy b, and neglect the small terms containing derivatives of vertical velocity w. Figure 13 shows the Ertel PV and its gradient in zonal direction for two sections across the Svalbard shelf break (78 • N and 78 • 50'N) for the FESOM simulation. The dominant term is the vertical stretching term, with a smaller contribution from the relative vorticity terms. The tilting terms are one order of magnitude smaller ( Figure   not shown). At both sections, the cross-stream gradient reveals a change in sign with depth, indicating that the mean current is 315 baroclinically unstable. This is in agreement with studies by Teigen et al. (2011) andvon Appen et al. (2016), and our simulated energy conversions (Figure 11c,d).

Discussion
Despite their very fine resolution, ROMS and FESOM simulate a weaker variability in velocity than the observed in terms of the power density spectrum (Figure 4). This might indicate that the resolution used is still insufficient to well resolve all 320 the mesoscale eddies in the presence of numerical dissipation. Part of the variability revealed by the power density spectrum can also be attributed to the atmospheric forcing. Although the forcing datasets are different in the two cases, both of them are derived from relatively coarse reanalysis products (in particular, COREv.2 used in the FESOM simulation has a zonal resolution of approximately 1.875 • ) and may miss part of small-scale variability. A topic for further research is to clarify the importance of these factors.

325
Snapshots of simulated relative vorticity ( Figure 2) and the baroclinic energy transfer (Figure 11c,d) suggest that the model effective resolution (Soufflet et al., 2016) in FESOM might be slightly lower than in ROMS. First, the grid size is slightly larger for FESOM (1 km vs 800 m for ROMS). This small difference in the grid size (20%) might matter as both numerical dissipation and explicit viscosity decrease with the grid size. In both models, biharmonic viscosity which scales with grid size cubed is applied. Second, FESOM1.4 is based on a collocated discretisation (an analog of Arakawa A-grid), whereas a stag-330 gered Arakawa C-grid is employed by ROMS. Because of pressure gradient averaging required by collocated discretisations, the effective resolution could be reduced. The collocated discretisation of FESOM also requires to use the no-slip boundary condition, which implies more dissipation along the boundary as well. Third, FESOM relies on implicit time stepping for external mode whereas ROMS uses a specially selected split-explicit method (see, e.g. Soufflet et al. (2016)) which is less dissipative. There might be other reasons for the difference in the simulated EKE in certain regions between the two models.
In particular, a higher EKE level in western Fram Strait in ROMS might be related to the difference in the simulated sea ice.
Sea ice could damp eddies through the ocean-ice stress.
In addition to the high similarity in the eddy properties, both models exhibit a very similar pattern in barotropic energy conversion in eastern Fram Strait (Figure 11a,b). The degree of similarity is quite surprising, given that FESOM uses z-levels in the vertical whereas ROMS relies on terrain-following coordinates, which might lead to differences in topographic steering.

340
Topography is bi-linearly interpolated to grid points and only smoothed over the 2d-stencil of nearest vertices in FESOM. In contrast, ROMS requires a smoother bathymetry.
Our instability analysis provides information that can be used to develop parameterisations for coarse resolution climate models. A commonly used eddy parameterisation is the one developed by Gent and McWilliams (1990), which mixes tracers along local potential density, and thus flatten isopycnals. However, our analysis shows that there are areas with conversion 345 from EKE to MKE (Figure 11a,b) associated with a steepening of isopycnals, which is not taken into account by the Gent- McWilliams parameterisation.

Conclusions
Several studies (e.g. Hattermann et al., 2016;Wekerle et al., 2017) indicate that mesoscale eddies are an important ingredient of ocean dynamics in the Fram Strait, determining the recirculation of AW and also the amount of oceanic heat that enters the 350 Arctic Ocean. However, the eddy properties have not been thoroughly investigated before.
Based on the results of two eddy-resolving ocean-sea ice models, ROMS and FESOM, we examined the properties and generation mechanisms of mesoscale eddies in Fram Strait. We found that the models agree with each other with respect to the modelled circulation, hydrography and eddy characteristics. They simulate rather short-living eddies (lifetime is on average 10-11 days), with a very slight dominance of cyclones (ROMS: 54%, FESOM: 55%). Cyclones and anticyclones show very 355 distinct travel pathways, e.g., cyclones generated on the shallow Svalbard shelf tend to stay there, whereas anticyclones tend to travel offshore into the deep basin. More anticyclones tend to be trapped in deep depressions. Mean eddy radius is 5.0-6.0 km, which compares well with the first baroclinic Rossby radius of deformation in this region. On average, eddies travel around 35 km in both models. Eddy cores are located at about 100 m depth on average. Cyclones are predominantly cold eddies, while anticyclones are predominantly warm eddies.

360
The models also agree on mechanisms driving eddy generation, with consistent patterns of conversions to EKE from the mean kinetic and eddy available potential energies. The small size of eddies explains why a very high (1 km or finer) resolution is needed to simulate them. The good agreement on eddy generation and properties despite the very different numerics of FESOM (unstructured horizontal grid with vertical z-levels) and ROMS (regular horizontal grid with a terrain following vertical coordinate) gives us confidence in their ability to realistically simulate eddy processes. The similarities of the simulated eddy 365 fields despite the different sea ice components, surface and boundary conditions of the two models further imply that the observed eddy characteristics are likely a result of the fundamental dynamics associated with the ocean mean state in this