For this study, we use both the high- and low-resolution versions of E3SM v1 (E3SM-HR and E3SM-LR, respectively): these simulations are fully coupled and are run with greenhouse gas conditions in the atmosphere that are fixed at 1950 levels. E3SM v1 consists of atmosphere, land, river, ocean, and sea ice components communicating via a flux coupler (cpl7; ). The atmospheric component, the E3SM Atmospheric Model (EAM; ), uses a spectral-element atmospheric dynamical core . The land component is the E3SM Land Model (ELM), which is a slightly revised version of the Community Land Model version 4.5 (CLM4.5; ). The river component is the newly developed Model for Scale Adaptive River Transport (MOSART; ). The ocean and sea ice components of E3SM v1 are based on the Model for Prediction Across Scale (MPAS) modeling framework and share the same unstructured, horizontal grid. The vertical grid of the ocean model is a structured, z-star coordinate . While E3SM-HR does not use the Gent–McWilliams (GM; ) mesoscale eddy parameterization, E3SM-LR does, with a constant GM bolus eddy diffusivity of 1800 m2s-1. A more detailed description of the E3SM v1 ocean and sea ice components is available in and , respectively. To identify the effect of model resolution on simulation results, we use the same tuning parameters on both high- and low-resolution E3SM simulations, as described in .

Table  describes the horizontal and vertical resolution for each model component. The atmosphere and land models share the same horizontal grid, having 25 and 110 km horizontal spacing for E3SM-HR and E3SM-LR, respectively. The atmospheric model uses a hybrid, sigma-pressure coordinate with 72 vertical layers and a top of the atmosphere at approximately 60 km for both the high- and low-resolution configurations. The land model uses 15 vertical levels for both simulations. For E3SM-HR, the ocean–sea ice mesh features a horizontal resolution varying between 18 km at the Equator and 6 km near the poles and 80 ocean vertical levels with spacing ranging between 2 m at the surface and ∼ 150 m at depth, following , to resolve the first and second baroclinic modes in the open ocean. For E3SM-LR, the ocean–sea ice mesh has a resolution varying between 60 km in the midlatitudes and 30 km at the Equator and poles, as well as 60 ocean vertical levels, with a layer thickness varying between 10 m at the surface and 250 m in the deep ocean. The river component employs a latitude and longitude grid with uniform grid spacing in both directions of 0.125 and 0.5∘ for high and low resolution, respectively. For more detailed information on time steps and coupling frequency for each component model, refer to Table 2 in .

Before performing fully coupled simulations, we run the ocean model in stand-alone mode for 1 and 3 months for E3SM-HR and E3SM-LR, respectively, starting from rest and the Polar Hydrographic Climatology (PHC; ) temperature and salinity profiles to slightly spin up the velocity field and damp out high-velocity waves. The results of the stand-alone ocean simulation are used as the initial condition for an ocean- and sea-ice-only, 3-year simulation using the Common Ocean Reference Experiment version 2 (COREv2; ) protocol for interannually varying atmospheric forcing. The sea ice is initialized with a uniform thickness of 1 m at all locations south of 70∘ S and north of 70∘ N. Finally, the model state from the end of this simulation is used as the initial condition for the ocean and sea ice components in the fully coupled simulations. The atmospheric initial condition is obtained from an earlier high-resolution E3SM simulation run with fixed sea surface temperature (SST). The land initial condition is interpolated from the year 1950 of a low-resolution E3SM v1 CMIP simulation . With these initial conditions, we run E3SM in a fully coupled model for 50 years at both high and low resolution for comparison. We use results from the last 30 years of each simulation for the analysis presented here.

The datasets used here for model evaluation purposes are summarized in Table ; they include direct observations, reanalyses, and interpolated climatologies for the atmosphere, ocean, and sea ice in the Southern Ocean. For the evaluation of sea ice, we use sea ice concentration from the NOAA/NSIDC Climate Data Record of Passive Microwave Sea Ice Concentration version 3 NCDR; and sea ice production from , which is derived from data from the Advanced Microwave Scanning Radiometer for Earth Observing System (AMSR-E). For the evaluation of oceanic currents and sea ice concentration, we utilize the Southern Ocean State Estimate SOSE;, a state-of-the-art, data assimilation product that incorporates millions of ocean and sea ice observations while maintaining dynamically consistent ocean state variables. Given the sparsity of observations in many regions around Antarctica, SOSE offers a comprehensive, physically based estimate of ocean properties that would otherwise be entirely uncharacterized . Specifically, we use the SOSE dataset with 1/6∘ horizontal resolution, spanning from 2005 to 2010. We also use the World Ocean Atlas 2018 (WOA18; ), which provides a global data product of ocean temperature, salinity, and density. Considering that both SOSE and WOA18 are based on offshore observational data, we utilize in situ conductivity–temperature–depth (CTD) observations to better characterize the model subsurface temperature and density on the continental shelf.

For evaluating the atmospheric state over the Southern Ocean, we use zonal and meridional wind velocities at 10 m from the European Centre for Medium-Range Weather Forecasts (ECMWF) ERA5 reanalysis product . It should be noted here that the ocean, sea ice, and atmosphere datasets described above represent present-day conditions, whereas the E3SM simulations are representative of model conditions for the 1950s. We compare monthly averaged products from both present-day observations and E3SM simulations, noting that the horizontal resolution of observational datasets – varying between 16 and 25 km – differs from E3SM simulation output.

Coastal polynyas are areas covered by “thin” sea ice, as defined by having a thickness of less than 20 cm, and surrounded by coastline and/or consolidated thick sea ice during the freezing season (e.g., ). We apply this definition of coastal polynyas to our model results. Our definition of OOPs follows a methodology similar to previous studies (e.g., ); i.e., we consider areas within the ice pack where the open-ocean sea ice concentration is less than 15 % during the freezing season.

In this section, we investigate the fidelity of Antarctic coastal polynyas simulated in E3SM (a list of the 13 polynyas is included in Table ) by comparing model output with available observational data. Cold katabatic winds blowing off Antarctica in E3SM-HR (Fig. a) push the thick sea ice offshore, leading to open water. Consequently, intensive latent heat flux exchanges from the ocean to the atmosphere favor sea ice production in the Antarctic coastal polynyas (Fig. b; note enhanced sea ice production near the Antarctic coast compared to further offshore). Sea ice volume accumulated in coastal polynyas from March to October is shown in Fig. c as a function of longitude for E3SM-HR and the AMSR-E satellite estimate. The AMSR-E data show that the highest rates of sea ice production occur along the East Antarctic coast, while the Ross Sea (Ross Ice Shelf polynya – RISP, around 180∘ longitude) has the largest total sea ice production. E3SM-HR generally does well at representing the accumulated sea ice volume in coastal Antarctic polynyas, especially in high sea ice production areas along the East Antarctic coast. Furthermore, over the course of the simulation, E3SM-HR displays interannual variability in total sea ice production from Antarctic coastal polynyas of up to 43 % (Fig. d). This suggests that E3SM-HR may be useful for analyzing Antarctic coastal polynya variability.

The mean coastal polynya area and sea ice volume accumulated during the March to October freezing season are presented in Fig.  for each of the 13 polynyas indicated in Fig. b and Table ; results from both the high- and low-resolution versions of E3SM are included and compared against observations. As expected, E3SM-LR shows very little polynya area or sea ice volume production over the defined coastal polynya region, whereas E3SM-HR has significant improvements in both polynya area and sea ice production compared to the low-resolution simulation. Yet, sea ice volume production in E3SM-HR is smaller than observed in several coastal polynyas such as the Dalton, Dibble, and Amundsen polynyas (DaP, DiP, AP, respectively). In reality, these coastal polynyas are strongly affected by the presence of landfast ice, a feature not yet represented in E3SM. Landfast ice is stationary sea ice attached to coastal features such as the shoreline and grounded icebergs . Several previous studies suggested that landfast ice and glacier tongues play an important role in the formation of some coastal polynyas by blocking sea ice advection of upstream sea ice into the polynya and thereby facilitating divergent motion . Therefore, a representation of landfast ice may be needed to more accurately capture coastal polynya characteristics. When we compare the sum of polynya area and sea ice volume production for polynyas that are not associated with landfast ice (Fig. o), it can be seen that E3SM-HR does reasonably well at reproducing sea ice volume production compared to AMSR-E observations, despite generally underestimating polynya area.

The first observed OOP, the Maud Rise polynya (MRP), was detected during the period June to October 1973, and it propagated westward from the Maud Rise seamount into the open ocean at an average velocity of 0.013 ms-1 . In the winters of 1974 and 1975, the MRP extended westward into the central Weddell Sea, initiating a Weddell Sea polynya (WSP; ). More recently, the MRP was observed intermittently in August 2016 and more consistently during the winter of 2017 (Fig. a–c; note that these observations are shown for reference only and not for the purpose of model evaluation). Similarly to a previous study comparing the difference between high- and low-resolution climate model results (e.g., ), E3SM-LR does not exhibit OOPs at any point in the simulation. E3SM-HR does produce MRPs, an example of which is shown in Fig. d–f for model year 54, at which relatively low sea ice concentration can be seen around 0∘ longitude. The presence of OOPs allows the ocean and atmosphere to interact directly through enhanced exchanges of sensible heat fluxes (results not shown), potentially affecting the dense water formation in E3SM-HR.

OOPs appear to occur more frequently in high-resolution model simulations than in the last 50-year record of satellite sea ice observations (e.g., ). Even in high-resolution ocean reanalysis products, there are several cases of OOPs (for instance, in 2005 in SOSE and in 2004, 2007, and 2010 in ECCO2) that do not appear in satellite observations . The convective activity accompanying OOPs is associated with decreased stability within the water column, which can be caused by buoyancy changes in surface or deep waters . OOP formation mechanisms have been investigated extensively in the past. Most studies recognize the existence of an interplay mechanism between atmospheric wind forcing and reduced upper-ocean stratification due to anomalously positive sea surface salinities that strengthens the Taylor column near the Maud Rise seamount . In addition, a strengthening of the subpolar cyclonic gyre in the Weddell Sea may be a factor in preconditioning the oceanic convective process, since it leads to shoaling of the pycnocline and circulation that strongly interacts with the bathymetry near Maud Rise . Possibly because of the presence of both an overly strong cyclonic gyre related to a deep subpolar low (see Sect.  below) and a reduced stratification in the Weddell Sea interior compared with WOA18 climatology (not shown), E3SM-HR produces OOPs in 18 of the 30 total simulation years (see Fig. ), far more frequently than observed.

For the purposes of this paper, we define the simulated OOP years as those in which either or both MRP and WSP occur or in which embayment-like features in sea ice concentration occur in the Weddell Sea (Fig. c–f). During non-OOP years, most of the Weddell Sea is covered with sea ice from September to November (Fig. a). As reported elsewhere, once an OOP occurs, it has a tendency to re-occur in subsequent years (Fig. g) due to the ventilation of salty and high-density waters within the deep convection area that induces a positive feedback on subsequent convections .

In this section, we apply a water mass transformation (WMT) analysis to the last 30 years of the E3SM-HR simulation in the Weddell Sea region over both the continental shelf and the open ocean to find which polynya type, coastal or OOP, predominantly produces dense water masses in E3SM-HR. The WMT analysis, first introduced by , quantifies the relationship between the thermodynamic transformation of water mass properties within an ocean basin and the net transport of those same properties into or out of the basin. This relationship has been used to characterize the thermodynamic processes that sustain the Southern Ocean overturning in models . A more detailed description of WMT analysis can be found in . discuss previous applications of WMT to the analysis of low-resolution E3SM simulations.

In Fig. a, we show the mean WMT rate over the Southern Ocean produced in E3SM-HR for all surface fluxes combined (positive WMT rates indicate that water masses become denser, implying buoyancy loss; negative WMT rates indicate that water masses become lighter, implying buoyancy gain). Because we only focus on the WMT rate during the freezing season from March to October, positive WMT rates dominate over most of the Southern Ocean. Over the continental shelf, the region delimited by the 1000 m isobath in Fig. a, we find strong positive WMT rates, which are the direct result of high sea ice production there (Fig. b). These results compare well in magnitude with observation-based estimates .

The mean WMT rates in latitude and longitude space can be converted into neutral density space and are shown in Fig. b for the whole Southern Ocean (gray line), the continental shelf only (green line), and the Weddell Sea (pink line). The maximum WMT rates over the whole Southern Ocean (≈ 20 Sv) are found at a neutral density of 27.6 kgm-3 and are larger than the observation-based estimate (≈ 13 Sv) in . In the high density ranges above 28.0 kgm-3, a local maximum WMT rate of 3.0 Sv is seen at a neutral density of 28.2 kgm-3. and define AABW as having neutral densities higher than 28.27 kgm-3. We consider 28.0 kgm-3 to be a suitable minimum threshold of AABW in the E3SM-HR simulation. As this high-density transformation rate in E3SM-HR is entirely due to WMT over the Weddell Sea rather than on the continental shelf (compare green and pink lines in Fig. b), the simulated bottom water is produced through open-ocean convection in the Weddell Sea only, similarly to what has been found in previous studies . Over the continental shelf, we also see positive transformation rates, but these occur at relatively lighter densities, with an average neutral density of 27.5 kgm-3, whereas no transformation occurs at densities higher than 28.0 kgm-3. These results are confirmed by the T–S diagrams for the Weddell Sea water column, presented in Fig. S1 in the Supplement and computed from both the E3SM-HR annual climatology and the WOA18 climatology. Compared to WOA18 and also considering Fig. 1a in , E3SM-HR exhibits a reasonably well-reproduced CDW and Antarctic surface water, but it misses shelf waters completely and produces AABW that is less dense (mostly because it is less cold) than observed.

When we consider the WMT rate contributed by several types of surface fluxes separately over the Antarctic continental shelf and over the Weddell Sea (Fig. c and d), the positive transformation rate over the continental shelf results almost entirely from brine rejection by sea ice production. The positive WMT rate over the deep Weddell Sea, meanwhile, is mostly due to the surface heat release from the ocean to the atmosphere, especially at relatively high neutral densities above 28.0 kgm-3, which is consistent with the strong ocean-to-atmosphere exchanges that occur in OOPs. We confirm, therefore, that the coastal polynyas and OOPs simulated in E3SM-HR have different WMT mechanisms. However, as described above, the simulation produces water masses at relatively low density levels on the continental shelf, despite the fact that the model does reasonably well at reproducing a large amount of sea ice formation in coastal polynyas during the freezing season. To investigate this apparent inconsistency, in the next section we examine E3SM-HR's hydrographic characteristics in the Southern Ocean.

We start by comparing E3SM-HR's atmospheric surface (10 m) winds with the ERA5 reanalysis data (Fig. a and b). Compared to ERA5, E3SM-HR has stronger atmospheric surface winds, not only along the Antarctic continental shelf but also in the interior Southern Ocean, which is consistent with an overly deep subpolar low-pressure system south of 60∘S (Fig. S2 in the Supplement). The overly strong atmospheric winds along the continental shelf may impact the strength of the ASC. Indeed, in Fig. c and d, we find that a very strong ASC is simulated in E3SM-HR, with speeds more than double those in SOSE (e.g., 0.34 ms-1 in E3SM-HR compared with 0.16 ms-1 in SOSE at 0∘ longitude). Therefore, we hypothesize that the strong winds over the continental shelf may be one of the drivers of the strong ASC in E3SM-HR.

A closer look at the shelf hydrography supports this hypothesis. In particular, we select three vertical meridional and zonal cross-sections at 18∘W (eastern Weddell Sea), 63∘ S (western Weddell Sea), and 90∘ W (Bellingshausen Sea), which are representative of the fresh, dense, and warm shelf types, respectively (the locations of the cross-sections are indicated in Fig. d, and they coincide with the sections described in . The results are presented in Fig. , comparing the model temperature and density profiles with CTD observational data . In general, E3SM-HR features down-sloping isopycnals approaching the continental slope in all cross-sections. For the fresh shelf type (Fig. a, d, and g), the model isopycnal surfaces are strongly tilted downwards in the onshore direction, more so than in the observational results (Fig. d). The strong Antarctic Slope Front (ASF) in the fresh shelf may be induced by the higher easterly winds (with respect to ERA5 winds) along this shelf (Fig. a). At the dense shelf type (Fig. b, e, and h), the isopycnals associated with warm CDW tilt down toward the seafloor over most of the continental slope. However, as the CDW approaches the shelf break in the dense shelf type, the observed isopycnals shoal again, indicating a V-shaped isopycnal surface (Fig. e). Unlike observations, E3SM-HR exhibits no V-shaped isopycnals, but rather steep isopycnal surfaces tilted downward toward the shelf break from offshore. This is also likely related to overly strong model winds (Fig. b). Lastly, in the warm shelf type (Fig. c, f, and i), the observations show isopycnal surfaces that tilt upward towards Antarctica, resulting in warm CDW reaching the continental shelf (Fig. f). Again, E3SM-HR simulates hydrographic conditions that are more similar to the fresh shelf type, with downward-tilted isopycnals (Fig. i). This bias could similarly be related to the overly strong model winds (Fig. c).

To further support our hypothesis that strong winds are responsible for the strong ASF, we compare vertical cross-sections of salinity, neutral density, and ocean current speed for the dense shelf case and for a strong wind and a weak wind composite (Fig. ). The strong wind (weak wind) composite was computed by selecting all years with the top (bottom) 25 % winter wind speeds over the dense shelf (63∘ S). Compared to weak wind years, strong wind years are characterized by a pronounced subsurface salinity cross-section gradient between the relatively salty waters in the CDW layer and the fresher waters over the continental shelf (Fig. e). In response to the lateral gradient in salinity, the lateral density gradient is also strong in the strong wind years, which causes an intensification of the ASC and ASF at or close to the shelf break (Fig. f). These results further indicate that the wind bias is the main contributor to the strong ASC in the E3SM-HR simulation.

In summary, as a result of ubiquitous, strong polar easterly winds, we find that E3SM-HR consistently displays only the fresh shelf type. These overly strong winds cause enhanced onshore Ekman transport (see Fig. S3 in the Supplement, particularly the shelf results in panel c), which in turn may affect the advection of sea ice formed during the winter season. As shown in Fig. , E3SM-HR's simulated sea ice extent in the Southern Ocean is much less than that from SOSE in all seasons except for JAS (July–August–September), but sea ice volume is generally larger than SOSE's, and the largest differences are found in the coastal regions. We hypothesize that this is the result of sea ice build-up on the continental shelf in E3SM-HR, wherein sea ice thickness often exceeds 2 m, due to enhanced poleward surface Ekman transport, which prevents sea ice from moving to lower latitudes where it would potentially melt.

Modeled and observed SST and sea surface salinity (SSS) averaged over the continental shelf provide additional information regarding the impacts of these biases. In Fig.  we compare shelf SST and SSS from E3SM-HR with those from SOSE and WOA18. E3SM-HR has a good representation of seasonal SST and SSS variation, but with warmer and fresher as well as colder and fresher water properties in the summer and winter seasons, respectively, compared to SOSE and WOA18. In summer, the warm, fresh surface water has a neutral density of 26.0 kgm-3 in E3SM-HR compared with more than 27.2 kgm-3 in WOA18 and 26.8 kgm-3 in SOSE. The possible external sources of the relatively low summer density in E3SM-HR can be excessive precipitation (E-P), runoff, or sea ice melting. Consistent with the sea ice thickness bias discussed above (Fig. ), we find that the freshwater flux into the ocean from melting sea ice in E3SM-HR is larger than that of SOSE, especially over the continental shelf and shelf break (Fig. b), while precipitation and runoff differences (not shown) are not significant compared to SOSE. Because of extensive and excessive freshwater inputs from sea ice in E3SM-HR, the annual mean sea surface neutral density is lighter than WOA18 over the continental shelf (Fig. d and e). We thus conclude that the overly strong easterly winds are also (indirectly) responsible for the relatively lighter water masses forming on the continental shelf in E3SM-HR.