For this study, we use version 4.2 of NEMO (NEMO System Team, 2022). NEMO is a three-dimensional, free-surface, hydrostatic, primitive-equation global ocean general circulation model. Our configuration uses the eORCA1 global grid, with a nominal horizontal resolution of 1∘ at the Equator and a reduction in meridional grid spacing towards higher latitudes to match the accompanying shrinking of the zonal dimension of the grid cells. In the Southern Hemisphere, the model grid has been extended to reach 85∘ S to allow for the representation of the sub-ice shelf seas according to the procedure described in Mathiot et al. (2017). The average horizontal resolution of the grid under RIS, FRIS and LCIS is 20, 22 and 42 km respectively. To account for the decrease in the horizontal size of grid cells at high latitudes, we decide to linearly scale the Laplacian eddy viscosity south of 65∘ S according to grid cell size. In the vertical, the configuration possesses 75 levels, with thickness increasing from 1 m at the surface to 200 m at depth (Storkey et al., 2018). We use the z* vertical coordinate adapted to the ice shelf so that all cells between the surface and the ice shelf base are masked at initialization and the effect of the ice shelf on friction and pressure gradient is calculated (Madec and NEMO System Team, 2019; Mathiot et al., 2017). The bathymetry used is derived from the Earth TOPOgraphy version 2 dataset (ETOPO2v2; NOAA, 2006) with information for the extension under the ice shelves based on the International Bathymetric Chart of the Southern Ocean (IBSCO; Arndt et al., 2013). For the calculation of the thermodynamic properties of seawater, NEMO uses the Thermodynamic Equation Of Seawater – 2010 (TEOS-10), giving results in conservative temperature and absolute salinity, which, for the purposes of this study, were converted to potential temperature and practical salinity in order to facilitate comparison of the model results with observations and known signatures of water masses. For more information regarding the choices of advection and diffusion schemes, mixing coefficients and eddy parameterizations, please refer to the copy of the namelists provided in the accompanying data repository. A note explaining the nomenclature of the namelists and the differences between the open and “closed” cavity simulations can be found in the Supplement Sect. S1.

The effect of tides on vertical mixing (through breaking of internal waves) is taken into account in NEMO using the energy constrained parameterization of de Lavergne et al. (2020). This mixing parameterization does not, however, represent trapped waves at high latitudes or any tide-induced internal-wave mixing below ice shelves and does not include the effect of tides on basal friction and thus melting of the ice shelves. To address this, by default there is a parameter (rn_ke0) representing the background kinetic energy associated with tides which is set to a constant of 2.5×10-3 m2 s-2 everywhere. We tested another methodology of parameterizing the impact of tides on melting according to Jourdain et al. (2019) using CATS2008 two-dimensional tidal velocities; as summarized in the Supplement Sect. S2 and Fig. S1, this alternative parameterization brings marginal changes in the simulated melt patterns and bulk melt rates (< 10 %). The explicit representation of tides is not advisable in a configuration designed for climate applications due to the high levels of numerical mixing induced.

The ocean dynamics component, NEMO OCE, is coupled with SI3, the dynamic and thermodynamic sea ice model of NEMO (Rousset et al., 2015; Vancoppenolle et al., 2023). SI3 is directly resolved on the ocean grid, based on an energy- and salt-conserving approach for sea ice thermodynamics (Vancoppenolle et al., 2023), multiple categories to resolve subgrid-scale variations in ice thickness (Bitz et al., 2001; Lipscomb, 2001), a second-order-moment-conserving scheme for horizontal advection (Prather, 1986), and the adaptive elastic–viscous–plastic formulation for the rheology term of the momentum equation (Kimmritz et al., 2016).

Here, we present results from two configurations: first a closed-cavity reference configuration, where ice shelf melt is prescribed in a way that mimics the ice-shelf overturning, and secondly an open-cavity configuration. For the reference closed-cavity configuration, a fixed freshwater flux corresponding to the volume of basal meltwater estimated by Depoorter et al. (2013) for each ice shelf is added into the ocean evenly between the ocean floor and the base of the ice shelf, horizontally uniform across the ice shelf front, and a vertical wall closes the cavity at this location (as in Mathiot et al., 2017). The fixed freshwater flux is based on Depoorter et al. (2013) melt estimates as this is the same file used for the IPSL climate model. Furthermore, the ice shelf area surveyed by Adusumilli et al. (2020) only extends to 81.5∘ S so that RIS and FRIS are not fully covered and therefore do not have the full melt flux. For the open-cavity configuration, the majority of ice shelves are kept closed using the same method as described above, and only three of the largest cold water ice shelves are opened. Circulation is simulated under RIS, FRIS and LCIS where the prescribed freshwater flux is turned off at the mouths of these cavities and interactive melt is activated. Ice shelf melt and freeze are calculated using the three-equation formulation (Hellmer and Olbers, 1989; Holland and Jenkins, 1999; Asay-Davis et al., 2016) in which the temperature, salinity and velocities are averaged over a fixed boundary layer thickness of 30 m chosen according to Losch (2008). The top drag coefficient used is 10-3, and the temperature and salinity transfer coefficients used are 1.4×10-2 and 4×10-4 respectively. Note that a fixed ice shelf geometry is maintained, thereby assuming a steady state where all ice melted by the ocean is replaced by the seaward advection of new ice (Schodlok et al., 2016; Mathiot et al., 2017).

By using this combination of explicit and parameterized ice shelf cavities, we provide an intermediate step between prescribed melt and explicit cavities or even ice sheet coupling and gain experience and a better understanding of the impact on ocean dynamics in order to better inform future choices. The advantage of this approach is that it allows us to specify the melt for small cavities which remain unresolved or insufficiently resolved at a 1∘ resolution and simultaneously utilize the model capability to resolve sub-ice shelf cavity circulation under the large, cold ice shelves, which allows for more realistic formation of the source waters of AABW. In terms of computing cost, the open-cavity configuration costs 11 % more than the closed-cavity simulation (mostly due to addition of cells as the model grid is extended further south; only 0.3 % of this is associated with the cost of the ice shelf routines themselves). Figure 1 shows the extended bathymetry of eORCA1 for the Weddell Sea and the Ross Sea, with the three ice shelf cavities of interest un-masked and important features labelled.

For both open and closed configurations, the model was run for 124 years using two cycles of interannual (1948–2009) CORE forcing (Coordinated Ocean-ice Reference Experiments version 2; Large and Yeager, 2004; Griffies et al., 2009). Sea surface salinity restoring is activated but not under sea ice as we have low confidence in the sea surface salinity climatology in this area due to limited observations. Freshwater discharge from iceberg melt is parameterized using a prescribed surface flux with realistic distribution (Merino et al., 2016), based on calving estimates from Depoorter et al. (2013).

For all simulations, global ocean properties were initialized using the 1981–2010 climatology of World Ocean Atlas 2013 (WOA2013; Locarnini et al., 2013; Zweng et al., 2013) as this dataset is used for the IPSL climate model and so was a convenient choice. This climatology does not, however, extend under the Antarctic ice shelves, and so in order to provide somewhat realistic initial conditions underneath FRIS, LCIS and RIS, we employed an idealized regional configuration of each ice shelf. For this we created a NEMO test case using a closed domain, with temperature and salinity restoring at the boundaries; 75 vertical layers; and a resolution, time step and bathymetry corresponding to those of eORCA1. The domain for each of the 3 configurations included just the ice shelf and adjacent continental shelf and slope and so were reasonably low-cost and fast to run in order to perform sensitivity experiments. The simulations were initialized with a constant and uniform temperature and salinity and restored at the boundaries using a mean profile from WOA2013 for that region. The choices for initial thermohaline properties inside the cavities were informed by calculating the mean values of detected ISW from CTD (conductivity, temperature, and depth) observations performed in the area adjacent to each ice shelf and converting these to conservative temperature and absolute salinity for input to the model (-2 ∘C and 34.76 for FRIS (Janout et al., 2021), -1.95 ∘C and 34.74 for LCIS (Nicholls et al., 2004; Hutchinson et al., 2020), and -1.94 ∘C and 34.76 for RIS (Bergamasco et al., 2003; Budillon et al., 2003)). Each simulation was run for 10 years, which was found to be sufficiently long to spin up the circulation within each cavity and reach a stable melt rate. The temperature–salinity distributions within the cavity were extracted and merged with WOA2013 data re-gridded to the NEMO eORCA1 grid, with a cubic spline used to smooth the data discontinuity across the ice shelf front. By following this method we have attempted to provide as realistic initial conditions for eORCA1 as possible, with the simulation starting with CORE forcing from the 1 January 1948.

The average ice shelf melt rate pattern for FRIS, LCIS and RIS is shown in Fig. 4 for the model simulation equivalent years 1995 to 2009, where orange indicates melt and purple shows refreezing. The average total melt flux for this time period is shown in Table 1 and compared to Depoorter et al. (2013) from which the volumes for the prescribed melt were taken for the reference configuration (closed). Opening the cavities results in at least double the melt reported from Depoorter et al. (2013). This discrepancy reflects both a warm bias on the continental shelf in NEMO (Sect. 4.4) and a possible bias in Depoorter's estimates which are lower than all other satellite estimates (Table 1). The total melt fluxes of each ice shelf from various other observational and model studies are also listed in the table, showing the wide spread in basal melt estimates both within values calculated from observations and between observations and models (Table 1). The model studies of Mathiot et al. (2017) and Bull et al. (2021), which are both regional NEMO 1/4∘ configurations, and the NEMO 1/12∘ configuration of the southwestern Weddell Sea of Hausmann et al. (2020) are particularly relevant to compare eORCA1 with, as here we see the possible impact of lowering the resolution in NEMO. For the Weddell Sea, our global 1∘ (eORCA1) compares well with these regional high-resolution studies, producing a net basal melt within 12 Gt yr-1 of the other estimates for FRIS and LCIS. The eORCA1 melt rate for RIS, while higher than observational studies, is in the middle of other model estimates and is especially well aligned with that of NEMO 1/4∘ from Mathiot et al. (2017). Overall, eORCA1's total melt fluxes correspond well with the average from all other estimates and are well within the standard deviations (last line of Table 1).

The patterns of melt shown in Fig. 4 also compare well with those of observational estimates like Rignot et al. (2013; their Fig. 1) and high-resolution model results like Hausmann et al. (2020; their Fig. 3), whose colour bar we replicated for ease of cross-comparison. If we look at the melt pattern of FRIS and compare it with these two aforementioned studies, we see that eORCA1 captures the high melt rates at the western portion of the ice shelf front, at the southern edge of Berkner Island and along the grounding line at the back of the cavity. The model also correctly simulates the region of refreezing along the western boundary of the circulation cell within the cavity, in both the Ronne and Filchner depressions and the re-freezing in the shallow region between the Korff and Henry Ice Rises (Fig. 4a, see bathymetry location in Fig. 1a). For LCIS, the entire shelf shows a positive melt (Fig. 4b). Observations from Rignot et al. (2013) and simulations from Harrison et al. (2022) indicate some re-freezing under this ice shelf, but the regional high-resolution model studies of Mathiot et al. (2017) and Hausmann et al. (2020) similarly show melting only. The pattern for RIS generally compares well with that reported from observations, but the magnitude of melt at the ice shelf front, especially to the east, is elevated (Fig. 4c).

Opening the sub-ice shelf cavities in eORCA1 allows for the establishment of a horizontal gyre circulation within the cavity and on the continental shelf of the Weddell Sea and the Ross Sea, in line with previous studies (Losch, 2008; Mathiot et al., 2017).

The mean state of circulation from the last 10 years of simulation within the FRIS cavity, along with the associated bottom thermohaline properties, can be seen in Fig. 5a–d. The circulation patterns shown here are in good agreement with Bull et al. (2021) at 1/4∘ and with Hausmann et al. (2020) at 1/12∘, with the exception of higher bottom salinities in eORCA1 and a slightly weaker barotropic circulation strength. Note that here we use potential temperature and practical salinity so as to be in line with the other figures of this paper, so approximately 0.17 psu must be added when juxtaposing with absolute salinity plots. The depth-averaged velocity and barotropic circulation pattern in Fig. 5a and b both indicate an anticlockwise circulation under the ice shelf. Comparatively warm and salty HSSW enters via the Ronne Depression, circulates from west to east, melts the base of the ice shelf mostly along the grounding line (cold, fresh signatures in Fig. 5c and d) and exits via the Filchner Trough as ISW. This pattern is consistent with observations (Nicholls et al., 2001; Janout et al., 2021). Two pathways of Modified Circumpolar Deep Water (MCDW) towards the ice shelf front can be seen, both in the circulation pattern (Fig. 5a) and via the bottom temperature (Fig. 5c): one located in the middle of the continental shelf (Central Trough) and the other on the shelf to the east of Filchner Trough. These pathways provide a conduit for heat towards the ice shelf and facilitate the mixing of shelf water masses with MCDW. It is therefore encouraging that eORCA1 (with an effective horizontal resolution under FRIS of 22 km) captures these, as they could play an important role in the evolution of shelf circulation in future climate scenarios (Naughten et al., 2021).

Moving now to the Ross Sea, the time mean circulation pattern under RIS along with the bottom temperature and salinity can be seen in Fig. 5e–h. Here, we notice a strong anticlockwise circulation concentrated at the western boundary, with reduced magnitude currents towards the back and east of the cavity. The west of the cavity is overall warmer and saltier and the east shows signatures of ISW. Bottom temperature indicates the presence of a cold ISW plume exiting the cavity to the far east (Fig. 5g), which is not seen in the time-averaged velocities or barotropic streamfunction, likely because the associated speeds are slow. Instead, the simulated circulation indicates an offshore advection of sub-ice shelf water following the Challenger Trough (see location marked in Fig. 1b). This water mass is likely recirculated HSSW as its temperature remains at surface freezing point (-1.9 ∘C). A strong clockwise circulation cell offshore of RIS (red in Fig. 5f) brings warm CDW into contact with the ice shelf front to the east, mixing out the signature of ISW further offshore (Fig. 5g). While this simulated circulation pattern agrees well with that described by observations (Fig. 1; Bergamasco et al., 2003; Budillon et al., 2003), it is likely too strong, resulting in an exaggerated net melt flux compared to the observational estimates (Table 1; anomalously high melt at the eastern portion of the ice shelf front in Fig. 4c).

To highlight the impact of opening the FRIS, LCIS and RIS sub-ice shelf cavities on the offshore properties, Figs. 2g and 3g show the temperature versus salinity distribution excluding the data under the ice shelves. The differences in bottom temperature and salinity can be seen in Fig. 2h and i for the Weddell Sea and Fig. 3h and i for the Ross Sea.

A significant improvement in the representation of Weddell shelf water properties is evident as now HSSW is within the observed range and ISW is detected on the continental shelf (see HSSW and ISW red boxes in Fig. 2g). Opening the sub-ice-shelf cavity of FRIS has allowed the HSSW that previously built up in the Ronne Depression to advect under the ice shelf, become modified through basal interactions, and exit the cavity as colder and fresher ISW. Consequently, the temperature and salinity differences are polarized west and east, with warmer fresher conditions along the entire western boundary of the Weddell Gyre and cooler, saltier conditions on the eastern continental shelf (Fig. 2h and i). These results agree well with those of Mathiot et al. (2017). The impact of opening LCIS can be seen via the maintenance of cold bottom properties immediately to the north (despite the fact that the shelf circulation has changed so that HSSW no-longer floods this region), along with the presence of a large negative salinity anomaly indicative of ice shelf melt (Fig. 2i). As the simulation is only 124 years long, the impact of opening the cavities on AABW cannot be fully assessed due to the slow renewal of this water mass at the bottom of the global ocean. A small change in signature of AABW can, however, be seen in the volumetric T–S plot (Supplement Fig. S3a), where explicit ocean–ice shelf interaction results in a shift in volume towards cooler, fresher AABW (open-closed weighted average shift in AABW volume by -0.008 ∘C and -0.003 psu). This shift is accompanied by a small increase in volume of the water mass by 0.23 % (AABW limits delineated in green in Fig. S3a).

The impact of opening the RIS cavity on offshore properties can be seen in Fig. 3h and i. Similar to the Weddell Sea, conditions in the west, where in the reference run HSSW was built up, now become warmer and fresher as the path under the ice shelf is open. The signature of the cold plumes of dense shelf water (Fig. 5g) on either side of Roosevelt Island can clearly be seen in the temperature difference plot (Fig. 3h), but curiously they do not possess the same salinity anomaly (Fig. 3i). The positive salinity difference of the western plume indicates that this water is a variety of HSSW which has circulated under the ice shelf and was previously not present in this area. The small negative anomaly to the east indicates that this cold plume is, as previously hypothesized, outflowing ISW. Small temperature differences on the continental slope and further offshore indicate that there has been some communication of the changes in shelf waters further afield. The volumetric T–S plot for the Ross Sea found in the Supplement (Fig. S3b) indicates that opening the RIS cavity has moved the core of AABW towards slightly cooler fresher values, accompanied by a 0.34 % decrease in volume of AABW as defined by the original water mass limits (delineated in green in Fig. S3b; open-closed weighted average shift in AABW volume by -0.001 ∘C and -0.005 psu).

The differences in circulation patterns and in thermohaline properties that result from opening the RIS and FRIS cavities documented above do not elucidate whether or not we have reduced biases and improved the realism of shelf waters in eORCA1. For this, a direct comparison with in situ observations is necessary. Due to the remote location of these ice shelves and the harsh conditions associated with obtaining hydrographic samples in these areas, there are limited observations, and so optimally interpolated atlases such as WOA or ocean reanalysis products like SOSE miss important local features or seasonal variability. For comparison purposes, we have consequently selected CTD data from research cruises that have sampled transects across the front of the ice shelves and extracted the model data corresponding to the approximate ship's track using PAGO, a tool to analyse gridded ocean datasets (Deshayes et al., 2014).

For FRIS we use two CTD sections across the ice shelf front undertaken in 1980 and 1995 on board the RV Polarstern by the Alfred Wegener Institute (Rohardt et al., 2016; Janout et al., 2021). The location of the section selected in NEMO to approximately overlay the CTD transects can be seen as a dotted magenta line in Fig. 1a. The output from NEMO corresponding to the same months and same equivalent year (for the second cycle of CORE forcing) in the simulation was selected for both closed-cavity (prescribed freshwater flux) and open-cavity (FRIS, LCIS and RIS) runs. A comparison between the CTD data and NEMO can be seen in Fig. 6a to f for January 1980 and Fig. 6g to l for February to March 1995. In terms of surface waters, NEMO does not capture the fine-scale horizontal variability and overestimates the subsurface salinity. For both observational years, evidence of warm, fresh, MCDW intrusions can be seen in the middle of the CTD sections (Central Trough; Fig. 6a and g). While the model struggles to capture the coherence of this subsurface temperature maximum, the anticlockwise circulation cell set up on the central continental shelf in the open-cavity simulation does aid the advection of MCDW towards the ice shelf, thereby producing a slightly better representation of this warm intrusion in Fig. 6c and i. The presence of cold ISW in Filchner Trough is clearer in the 1995 CTD data than in 1980, where the sampling frequency was sparser and this region not well covered. The 2018 Polarstern sampling of the Jason Trough was the highest resolution yet, and while we cannot directly compare with the simulation output as the CORE forcing ends in 2009, the presence of a tongue of ISW focused on the western bank of Filchner Trough is evident in Fig. 3 of Janout et al. (2021) and so should be kept in mind for comparison. Opening the FRIS cavity overall improves the thermohaline properties at the ice shelf front, most notably by spreading out the pool of HSSW from the Ronne Depression (e.g. Fig. 6k) across the continental shelf (e.g. Fig. 6l) and by facilitating the production and thus outflow of ISW within Filchner Trough (Fig. 6c and i).

The CTD sections used for comparison along the front of RIS were obtained through the World Ocean Circulation Experiment Database (Boyer et al., 2018) and correspond to cruises undertaken on board the RVIB Nathaniel B. Palmer in 2000 (cruise id: US010404; Smethie and Jacobs, 2005) and in 2007 (cruise id: US034357). Data were extracted from the eORCA1 simulation corresponding to the dates of these cruises and the approximate ship track across the ice shelf front (dotted magenta line in Fig. 1b). Similar to the Weddell Sea, the model tends to overestimate the subsurface temperature and salinity (Fig. 7b, e, h and k), suggestive of biases in the representation of coastal processes, including vertical mixing. This effect is somewhat reduced by allowing for circulation under RIS, especially by decreasing subsurface salinities (Fig. 7f and l). At depth, NEMO captures the east–west distribution of haline properties such as the HSSW pool located within Joides Basin, albeit with somewhat amplified salinities. In terms of temperature, the model has a clear bias to the east, especially in the closed-cavity run, where CDW is detected at the ice shelf front. Both the temperature and salinity biases are reduced in the open-cavity run (e.g. Fig. 7c and f). In particular, the significant reduction in the extent and magnitude of the subsurface warm water intrusions brings the model more in line with observations.

A recurring theme throughout the results presented here is that the model is overall too salty, driven by what appears to be an over-production of HSSW in the Ronne Depression and Joides Basin. One driver for this could be the overestimated polynya activity which forms the totality of parent waters of AABW in the absence of ice shelves in eORCA1. This can be seen in Fig. 8 where the mean winter (July–August–September) mixed-layer depths (MLDs) in the reference run for the years 1971–2009 are compared to the climatology from Sallée et al. (2021a) for the same time period and using the same criteria for calculation (Fig. 8a and b; MLD defined as the depth at which density exceeds the 10 m density by 0.03 kg m-3). The model greatly overestimates winter MLDs in the Weddell Sea, both on the continental shelf adjacent to FRIS, where the depth of the base of the mixed layer aligns with bathymetric features indicating deep convection right to the ocean floor, and offshore of the continental slope, where a large region of MLD greater than 1000 m is present (Fig. 8c). This level of open-ocean deep convection has in reality only once been observed, during the 1974–1976 Weddell Polynya event near Maud Rise (3∘ E, 66∘ S), indicating a gross overestimation of winter MLDs in the model (Heuzé, 2021; Killworth, 1983). Ross Sea MLDs (Fig. 8d) compare better with observations but show values indicating a full water-column-depth convection in Terra Nova Bay which is not reported in Sallée at al. (2021a). Curiously, NEMO actually underestimates winter mixed layers in the eastern portion of the Ross continental shelf showing mean MLDs of under 100 m where the observational climatology indicates values of around 400 m (Fig. 8d compared to Fig. 8b). This too strong a stratification could be one of the factors facilitating the intrusion of CDW to the ice shelf front seen in Fig. 7b and h.

The authors note that the biased MLDs could be one of a number of factors contributing to the overly saline conditions; wrong sea ice parameters and biases in the atmospheric forcing could also play an important (and related) role. High sea ice production is seen on the southwest continental shelves of the Weddell Sea and Ross Sea in the Supplement Fig. S2a and b. Opening the cavities slightly reduces the magnitude of ice production in the Ronne Depression (Fig. S2c) and at the location of the Terra Nova Bay polynya (Fig. S2d) and increases the production of ice further east. There is no overall change in the principal location of polynya activity, and the slight west/east decrease/increase in sea ice is presumed to have a negligible effect on the total amount of HSSW generated. As such, the reduction of the highly saline HSSW signature seen in Figs. 2g and 3g when cavities are opened is likely due to a conversion to ISW (and not from a decrease in HSSW production itself). Please see the Supplement Sect. S3 for an evaluation of simulated polynyas near the studied ice shelves and a diagnosis of the effect of opening the cavities on ice production.

Once a model is able to explicitly form the parent waters of AABW in the right locations on the continental shelf (and export this dense water), it will become necessary for modellers to tone down open-ocean deep convection as this workaround will be longer relied upon to form the totality of AABW. Here we explore the impact that opening the cavities has on MLD to diagnose the extent of vertical convection in the model. Some reduction in MLD is seen on the continental shelf and slope in the Filchner (Fig. 8e) and Challenger troughs (Fig. 8f) due to the increase in stratification as a result of the greater bottom densities associated with outflowing ISW (Fig. S4a and c). The presence of ISW appears to promote slightly increased ice production in these areas, as discussed earlier. In this case, it is therefore the ocean properties that drive sea ice, and the brine rejection associated with elevated ice production is found to have a minor effect on water properties. Within the region of exaggerated MLDs off the Weddell continental slope, the MLDs deepen in the open-cavity experiment (positive anomalies Fig. 8e). We hypothesize that this deepening is associated with an overall cooling of the subsurface layers due to a horizontal mixing of ISW, unimpeded by a relatively weak and diffuse Antarctic Slope Current (ASC; discussed in the following section). Overall, in wintertime, mixed layers are on average 19 m deeper over the whole Weddell Sea region in the open-cavity experiment compared to the reference closed simulation. This reinforcement of the high MLD bias highlights the need for work to be done on reducing wintertime deep convection, together with better representing dense water overflows.

We have seen how opening the large, cold ice shelf cavities in eORCA1 leads to a better representation of continental shelf circulation and thermohaline property distributions. But the question remains regarding the transfer of these now more realistic dense shelf waters downslope and offshore, to feed the globally important AABW. While the simulation period of 124 years (two CORE forcing cycles) is too short to explore the impact of these changes far afield, it is sufficient to investigate the changes on the continental shelf and slope adjacent to the large ice shelves. To do this, we use PAGO (Deshayes et al., 2014) to select a cross section of data following the bathymetric troughs of the Weddell Sea and the Ross Sea, which are thought to be important for dense water export (Foldvik et al., 1985; Jacobs, 1991), namely the Filchner and Challenger troughs (sections shown in green in Fig. 1).

The thermohaline cross sections of Filchner Trough and a continuation down the continental slope can be seen in Fig. 9a and b for the open-cavity run, and the difference between these values and the reference run (open-closed) is shown in Fig. 9c and d. By opening the sub-ice shelf cavity, the properties within Filchner Trough have decreased in temperature and increased in salinity as the candidate parent waters of AABW build up on the continental shelf. This results in a net increase in density at the bottom of the trough (Fig. S4b), but there is very little indication of a coherent cascading of this water down the continental slope.

A cross section of the Challenger Trough (Fig. 10) reveals depth-varying thermohaline changes. Opening the sub-ice shelf cavity has allowed for the water adjacent to the ice shelf to advect into the cavity leaving the bottom properties here slightly warmer. The layer immediately above conversely experiences cooling and salinification due to the outflow of ISW driven by the “ice pump” (Fig. 10c). Here we see some evidence indicating the translation of this dense cold water tongue over the continental shelf break and downslope (Figs. 10c and S4d). The overflow of this water results in the pulling in of warmer offshore water at intermediate depth (Fig. 10c). A horizontal redistribution of surface waters simultaneously takes place due to the anti-clockwise circulation pattern (Fig. 5e), which in turn produces a cooling and freshening in the surface layer (Fig. 10c and d).

For both the Filchner and Challenger troughs, the downslope export of the ISW tongue is limited due to the commonly known and acknowledged problem of correctly capturing this overflow in a coarse z-coordinate model (Heuzé, 2021). The aptitude of representing dense water overflows is thought to increase with models of higher resolution, but this is difficult to achieve in a global model for climate coupling purposes without a nested zoom (Storkey et al., 2018; Colombo et al., 2020; Solodoch et al., 2022).

Another important dynamic for Antarctic shelf water realism is the ASC (red arrows in Fig. 1) and related Antarctic Slope Front, which together restrict the lateral mixing of offshore and shelf water masses, acting as an effective barrier protecting the large, cold ice shelves from warm water masses of circumpolar origin (Thompson et al., 2018). Some CDW, or a modified version thereof, is carried within the ASC and occasionally fluxes onshore to mix with dense shelf waters (Beadling et al., 2020; Bull et al., 2021). This interaction between dense shelf water and CDW is important for the formation of AABW, as the onshore flux of water replaces the dense water transported offshore and thus sustains formation of shelf water (Thompson et al., 2018). Figure 9e shows a velocity cross section for the Weddell Sea shelf and slope where westward velocities are positive so as to correspond with the direction of the ASC, and the net westward transport across the section is 9.8 Sv (1 Sv is 106 m3 s-1). This can be compared to Fig. 9f, which is a cross section from SOSE output (same time periods used) where the net transport is 3 times higher at 32.8 Sv. Similarly for the Ross Sea, Fig. 10e shows a cross section of westward velocities in eORCA1 where the volume transport is 13.3 Sv, which is less than half of that estimated from SOSE in Fig. 10f of 20.9 Sv. As can be seen from both SOSE cross sections, the ASC flows eastward as a narrow jet, closely following the shelf break in the Weddell Sea and slightly further offshore in the Ross Sea. It is well known that coarse-resolution models are unable to correctly represent the ASC as a resolution of at least 0.5∘ is needed to capture the dynamics and net transport (Mathiot et al., 2011). The absence of realistic ASC in NEMO eORCA1 (Figs. 9e and 10e) has important consequences, as a weaker and more diffuse ASC allows for a greater level of onshore–offshore exchange of water masses. This is one important restriction that needs to be kept in mind when using this coarse-resolution configuration for process studies in the area.