Ice‐Front Retreat Controls on Ocean Dynamics Under Larsen C Ice Shelf, Antarctica

Iceberg A‐68 separated from the Larsen C Ice Shelf in July 2017 and the impact of this event on the local ocean circulation has yet to be assessed. Here, we conduct numerical simulations of ocean dynamics near and below the ice shelf pre‐ and post‐calving. Results agree with in situ and remote observations of the area as they indicate that basal melt is primarily controlled by wintertime sea‐ice formation, which in turn produces High Salinity Shelf Water (HSSW). After the calving event, we simulate a 50% increase in HSSW intrusion under the ice shelf, enhancing ocean heat delivery by 30%. This results in doubling of the melt rate under Gipps Ice Rise, suggesting a positive feedback for further retreat that could destabilize the Larsen C Ice Shelf. Assessing the impact of ice‐front retreat on the heat delivery under the ice is crucial to better understand ice‐shelf dynamics in a warming environment.


Introduction
Most ice shelves in the eastern Antarctic Peninsula have either retreated significantly or disintegrated over the past few decades (Cook & Vaughan, 2010).The consequent loss of buttressing force resulted in up to an eightfold increase in freshwater discharge from glaciers north of the Larsen C Ice Shelf, with an estimated sea-level rise equivalent of approximately 0.9 mm since 1979 (Rignot et al., 2004(Rignot et al., , 2019;;Scambos et al., 2004).In July 2017, Larsen C calved one of the largest icebergs ever observed: A-68.This event did not completely destabilize the ice shelf but it brought the ice front into its furthest receded position observed so far.With further calving, the ice front may recede beyond the compressive arch, a mechanical threshold that could lead to irreversible retreat (Doake et al., 1998;Fuerst et al., 2016).Processes driving ice-front retreat require further understanding as consequences of climate warming become more evident.
After ice-shelf retreat events, the temporary ice front is often pinned by ice rises, rumples, islands, and shallow bathymetric features that buttress the ice flow (Thomas, 1979).Past studies suggest that ocean-induced ablation of thin ice around such critical pinning points is important for the ice-shelf stability.For the Larsen C Ice Shelf, basal melt surrounding pinning points such as the Gipps and Bawden Ice Rises (GIR, BIR, Figure 1) is thought to reactivate local rifts, potentially leading to calving events such as the event in 2017 (Borstad et al., 2013;P. R. Holland et al., 2009;Kulessa et al., 2014;Larour et al., 2021).However, pathways of ocean heat intrusion under Larsen C during and after the calving of iceberg A-68, along with the aftermath on basal melting around local pinning points, remain poorly understood.Long-term in situ data are sparse as persistent sea-ice makes these locations particularly inaccessible.
Peninsula (Harrison et al., 2022;P. R. Holland et al., 2009;Hutchinson et al., 2020;Mueller et al., 2012).Data from hot-water boreholes drilled near the grounding line (magenta stars, Figure 1) in December 2011 showed that sampled water was at the local freezing point (−1.9°C), which excludes direct on-shelf intrusion of a "pure" (hence warmer) version of mWDW (Nicholls et al., 2012).As a consequence, the average basal melt near the grounding line is relatively low, that is, approximately 0.7 m yr −1 (Davis & Nicholls, 2019).
It is unknown if and how the calving of iceberg A-68 affected these sub-shelf circulation patterns.The impact of ice-front retreat and ice rifts on the dynamics of Antarctic ice shelves is only beginning to be explored (Bradley et al., 2022;Poinelli et al., 2023;Yoon et al., 2022).In this work, we use a regional ocean simulation of the Larsen C ice cavity at high-resolution to study sub-shelf ocean circulation and its effects on basal melt rates.We then investigate the impact of ice-front retreat on pathways of heat intrusion under the ice.2012) and the shallow areas that coincide with the Bawden and Gipps Ice Rises, respectively.(f) and (e) also include monthly average of sea-ice concentration at the end of summer (March 2016), that is, when its concentration is the lowest.BIR: Bawden Ice Rise, GIR: Gipps Ice Rise, CA: Cape Alexander, CF: Cape Fairweather, CI: Cabinet Inlet, KP: Kenyon Peninsula, MI: Mobiloil Inlet.

Ocean Model
We use the Massachusetts Institute of Technology General Circulation Model (MITgcm) with thermodynamic ice shelf and dynamic/thermodynamic sea-ice (Losch, 2008;Losch et al., 2010).Basal ice-shelf melt is computed using the three-equation model (Hellmer & Olbers, 1989;D. M. Holland & Jenkins, 1999;Jenkins, 1991).Ocean, sea-ice, and ice-shelf parameters are nearly identical to those used by Nakayama et al. (2019).However, we calculate here coefficients of heat and salt transfer using drag laws and friction velocities based on a momentum-flux-derived drag coefficient C d = 0.0022, which was obtained from in situ observations (Davis & Nicholls, 2019).Bed topography (Figure 1a) and ice-shelf geometry are interpolated from the state-of-the-art compilation BedMachine-V3 (Morlighem et al., 2020).Nominal horizontal and vertical grid spacing are set to 500 and 10 m, respectively, which constrain the step size of integration time to 30 s.
Boundary conditions are applied at the eastern and southern domain boundaries, while the northern boundary is closed and coincides with the seaward extension of Cape Fairweather.Summertime (December-March) boundary conditions of potential temperature and salinity (θ-S) are derived from available conductivity-temperature-depth (CTD) observations from the expedition of 2019 (Figure S1 in Supporting Information S1; Hutchinson et al., 2020).During the wintertime period (April-November), temperatures at the boundaries of the domain are maintained at a constant value of −1.9°C.Meanwhile, the salinity profile remains the same as during the summer season.We also apply tidal forcing as Dirichlet boundary conditions.Tidal amplitudes and phases are obtained from the data-assimilative barotropic tidal model of Padman et al. (2002).Realistic monthly-averaged surface forcing is interpolated on the model domain from numerically-analyzed atmospheric data from the ERA5 land reanalysis (Hersbach et al., 2020).Sea-ice concentration and thickness along the eastern and southern boundaries are interpolated from the coarse resolution global state estimate ECCO-LLC270 optimization (Zhang et al., 2018).Due to computational constraints, sea-ice velocities at the boundaries are directly calculated by surface fluxes forcing in the interior.

Experiments
We conduct two distinct experiments running over the same temporal span: from 1 January 2010 to 1 January 2017.In the first, we simulate ocean dynamics using the ice-shelf front prior to the calving of iceberg A-68 (control run).In the second, we use the ice-shelf front after the calving event, in which iceberg A-68 is removed from the model domain (iceberg run).Both simulations spin up for a period of 5 years under identical atmospheric and boundary forcing.
The following analysis refers to model output after the spin-up, that is, between 1 January 2015 and 1 January 2017.
To investigate differences in sub-shelf circulation in the two experiments, we release two passive tracers on 1 January 2015, following the approach of Nakayama et al. (2017Nakayama et al. ( , 2021)).Tracer tr01 is restored to a unity value and released at the ocean surface as a proxy for HSSW concentration.The restoring time scale is one hour.A second tracer is released at the ice-shelf base at the same rate as melting occurs in order to track meltwater fraction (MWF).

Results
Model results of the control run are discussed in Section 3.1 and a brief comparison with available observations is in Section 3.2.Results of the iceberg run are in Section 3.3.

Ocean Circulation Prior to the Calving of Iceberg A-68
The control run simulates ocean conditions before the calving of iceberg A-68 (Figures 1,.High sea-ice production is simulated near the ice front during the austral winter season between March and October, with polynya modeled in the north of the domain.As a result of loss of buoyancy in response to brine rejection from sea-ice formation, tracer tr01 (as proxy for HSSW) convects to deeper layers and intrudes into the ice cavity (Movie S1).
Water in the ocean flows along lines of constant depth.In order to conserve potential vorticity, the flow entering the cavity is constrained by the presence of an ice front which acts as a barrier (Grosfeld et al., 1997).This abruptly reduces the water column thickness and curtails on-shelf intrusion.As a consequence, tracer tr01 is channeled along three distinct bathymetric pathways (red arrows in Figure 2a).The most intense intrusion occurs around latitude 68°S and along the "southern trough" (a deep bathymetric feature documented in Brisbourne et al., 2020).Along this pathway, tr01 sinks to the seabed in the open ocean and drains into the deep trough toward the Mobiloil inlet.Second and third pathways occur around the two pinning points.Near GIR, ocean currents interact with thin ice, and the relatively flat seabed favors intrusion of tr01.Water that eventually enters the cavity wraps around the Kenyon Peninsula and reaches the Mobiloil inlet at the end of the second winter season.Similarly, tr01 enters the cavity following the flat seabed near BIR.This water wraps around Cape Alexander and sinks toward the Cabinet inlet.
Due to the seasonal variability of sea-ice, the intrusion of tr01 into the cavity occurs in pulses (Figure S6 in Supporting Information S1).During the first summer months, the model simulates sea-ice melting with the consequent freshening of surface water and tr01 being confined in the mixed layer.At the beginning of the first winter, sea-ice production increases.Cumulative on-shelf intrusion of tr01 accelerates in June 2015, stabilizes at the end of the first winter season, and remains constant throughout the summer.At the beginning of the second winter, tr01 intrusion accelerates again and at the end of 2016, the cumulative on-shelf tr01 volume is twice as much as at the end of 2015.
The intrusion of HSSW drives melting when this saline water interacts with the base of the ice shelf (Figures 1e  and 2a).HSSW carries sufficient heat to melt ice as the freezing point decreases with increasing pressure.We find that the thermal driving pattern matches well the spatial distribution of melt rate, while changes in current velocity alone cannot explain the high simulated melt pattern at the grounding line (Figure S7 in Supporting Information S1).The model simulates three peaks of meltwater (Figure 2b) that correspond to the three intrusive pathways.The most significant peak, with MWF around ∼0.1%, is found north of the Kenyon Peninsula (box A).Meltwater then flows north-northeast along the ice draft toward shallower layers and exits the cavity between latitude 68°S and 67°S.Other sources of meltwater are the GIR and BIR (boxes B and C), where the MWF is twice as high as in box A. Seasonal variability of melt rates in box A (Figure S6f in Supporting Information S1) further shows a strong correlation (R = 0.84) with thermal driving (similar trends found under BIR and GIR, Figures S8 and S9 in Supporting Information S1).This confirms the dominance of ocean temperature on ice-shelf melting rather than ocean current speed (R = 0.42).

Model-Data Comparison
Simulated vertical profiles of potential temperature and salinity under the ice of the control run match well CTD observations (Figures 3c and 3d, 4c and 4d, Figures S1 and S5 in Supporting Information S1) accessed through the two drill sites in December 2011 (Nicholls et al., 2012).First, the model reproduces a layer of cold and fresh meltwater beneath the ice base around the local freezing temperature.Meltwater sits above denser and relatively warmer HSSW characterized by potential temperature around −1.9°C.Second, the modeled power spectrum shows   Average melt rate at the southern drill site (box A) is 0.18 m yr −1 and, in proximity of BIR and GIR (box B and C), it reaches 0.46 and 0.11 m yr −1 , respectively.The total ice-shelf melt rate average is in good agreement with 0.17 m yr −1 , estimated with altimetry observations spanning the same period (Adusumilli et al., 2018).At box A, in situ data (0.70 m yr −1 , Davis & Nicholls, 2019) and altimetry-derived measurements (0.36 m yr −1 , Adusumilli et al., 2018) are underestimated by the model.At boxes B and C, calculated melt is in reasonable alignment with altimetry-derived estimations of the two pinning points (1.11 and 0.36 m yr −1 , respectively).

Ocean Circulation After the Calving of Iceberg A-68
The iceberg run simulates ocean conditions after the calving of iceberg A-68 (Figures 1,.The removal of iceberg A-68 from the ice-shelf geometry substantially alters the pathways through which ocean water intrudes and distributes heat beneath the ice by means of two processes.First, the retreat event results in the ice front migrating westward.This leads to polynya near the ice shelf also in the southern section of the model domain, while, in the control run, sea-ice presence in this area was more stable.Second, the removal of a substantial portion of ice, combined with the deeper bathymetry, increases the thickness of the water column along the new ice front with respect to the pre-calving conditions. The aforementioned processes result in a two-fold effect.(a) An enhanced HSSW production along the new margin, which is a direct results of an increase in the number of polynya days during the two years of simulations.(b) A thicker water column along the ice front, which weakens the potential vorticity barrier caused by the ice shelf.As a consequence, these processes favor a widespread and more intense on-shelf intrusion of tr01 with respect to the control run.
In the iceberg run, there are notable changes in the sinking behavior of tr01 and the intrusion pattern along the southern trough compared to the control run.Instead of sinking primarily along the southern trough as observed in the latter, tr01 sinks along the entire western edge of iceberg A-68 in the former.This sinking and intrusion are most pronounced south of latitude 67°S (Figure 2d), aligning with the new peaks of sea-ice production in the southern section of the model domain as well as the thicker water column along the ice front in the south.As a consequence, the overall intrusion of tr01 below GIR is much stronger in the iceberg run, and the tracer reaches the western-most extensions of the ice-shelf cavity near the Mobiloil inlet after only a single winter season since the initial release (Movie S1).
Similar to the control run, the intrusion of HSSW occurs in pulses.tr01 intrusion accelerates at the beginning of each winter season and stabilizes during the summer months when the model simulate sea-ice melting.However, in comparison to the control run, the iceberg run shows that the total on-shelf intrusion of tr01 increases by 50% while heat transport toward the grounding line increases by 30% (Figures S6c and S6d in Supporting Information S1), due to the processes explained above.Consequently, the total integrated melt rate is 73% higher in the iceberg run, while the largest increase is concentrated south of latitude 68°S (Figures S6, S8, and S9 in Supporting Information S1).In particular, basal melt below GIR (calculated over box C) increases by 113% and the area north of Kenyon Peninsula, in proximity of the south drill site (calculated over box A) experiences a 54% increase.In contrast, basal melt around BIR (calculated over box B) decreases by 12%.While the basal melt increase south of latitude 68°S is a direct result of a stronger HSSW intrusion after ice-front retreat, the decrease of melt under BIR is a consequence of an enhanced northward flow of meltwater from the Mobiloil inlet.In comparison to the control run, meltwater flows more northward and along the new ice front and cools the ambient water under the ice in proximity of BIR (Figures 2e and 2f), curtailing local melt.

Discussion
Past studies delineated two future scenarios that would destabilize the Larsen C Ice Shelf and precondition its collapse (e.g., Doake et al., 1998;P. R. Holland et al., 2015;Jansen et al., 2013): (a) ungrounding of ice near BIR and GIR pinning points and (b) ice-front retreat beyond the compressive arch.Our simulations suggest that the calving event of 2017 may have redistributed ocean heat delivery under the ice base with an estimated doubling of basal melt around GIR.This is particularly important as sustained thinning in this area has a two-fold effect.First, it may lead to ungrounding the ice shelf from the shallow bathymetry, with the consequent acceleration of the ice flow (Borstad et al., 2013).Second, it may re-activate dormant rifts, which could then lead to further ice-front retreat, similar to the event of 2017 (P.R. Holland et al., 2009;Larour et al., 2021).This positive feedback for further ice-shelf retreat after calving events is worthy of further investigation in order to accurately estimate the evolution and the impact of global sea level of the largest remaining ice shelf in the Antarctic Peninsula.
Our results build understanding on the need to consider updated ice-shelf margin geometry to accurately project ocean circulation below the ice.A recent study has demonstrated that ice-front retreat can have significant implications for the intrusion of warm Circumpolar Deep Water into the Pine Island Glacier cavity in the Amundsen Sea (Yoon et al., 2022).Here, our results show that calving events can also impact the mechanisms of HSSW intrusion under cold water ice shelves, which is the dominant driver of their basal ablation.In order to extend these results to other ice shelves, it is important to note that, rather than the physical retreat of the ice front alone, it is the overall hydrography of the post-calving condition that plays a crucial role in delivering heat toward the grounding line.For instance, the presence of bathymetric ridges (absent in the case of Larsen C) and the ice-front retreat beyond their location was shown to determine the actual extent of off-shelf water intrusion in the Pine Island Glacier cavity (Bradley et al., 2022).
A limitation of our simulations is the lack of large-scale ocean circulation in the Weddell Sea, as our lateral ocean boundary conditions are only based on barotropic tidal forcing.For example, our model is unable to account for the presence of relatively warm mWDW documented near BIR (Hutchinson et al., 2020;Nicholls et al., 2004), which could substantially increase the simulated basal melt.Furthermore, we do not consider the impact of frazil ice crystals, which accumulate under Antarctic ice shelves and form marine ice bodies.Under Larsen C, the thickness of marine ice is estimated to be up to a third to half of the ice-shelf depth, in agreement with numerical models (P.R. Holland et al., 2009;McGrath et al., 2014;Harrison et al., 2022).The role of frazil accretion in the form of such thick layers of marine ice is well documented for the ice flow, but the impact of accreted ice on the on-shelf heat intrusion is not fully understood (Jansen et al., 2013;Kulessa et al., 2014).Further numerical simulations need to take into account realistic ocean forcing and to investigate its impact on frazil ice accretion under weakly-stratified ice shelves such as Larsen C, where marine ice is abundant.
Finally, we find that ocean temperature has a stronger impact on ice-shelf melting compared to current speed.This is in contrast of what has been modeled for warm water ice shelves (e.g., Pine Island and Thwaites Ice Shelves, Nakayama et al., 2019) and suggests a distinctive characteristic specific to cold water ice shelves.As a response to ongoing and future climate change, warm water could be transported progressively closer to large cold water ice shelves in the Weddell Sea (Darelius et al., 2016;Hellmer et al., 2012), most likely in the form of mWDW.Calving events could facilitate the intrusion of relatively warm mWDW under the ice and therefore increase their basal melt.Although there have been no direct observations of mWDW intrusion under the Larsen C Ice Shelf to date, our simulations emphasize the importance of additional observations and modeling efforts to monitor the arrival and impact of warm coastal currents beneath this and other cold water ice shelves.

Conclusions
We present high-resolution ocean simulations nearby and below the Larsen C Ice Shelf before and after the calving of iceberg A-68.Results from the post-calving run suggest that the removal of iceberg A-68 yields High Salinity Shelf Water and ocean heat intrusion to increase by 50% and 30%, respectively.Particularly, this results in basal melt around the Gipps Ice Rise to double.Enhanced thinning around this pinning point can lead to further rifting and calving, ultimately compromising the mechanical stability of the ice shelf.Consequently, our results suggest a positive feedback loop for further ice-front retreat.Our findings underscore the critical importance of considering updated ice-shelf margin geometry when aiming to understand the processes controlling ocean heat intrusion beneath ice shelves.By extension, this knowledge is crucial for accurately quantifying basal ablation and thus the contribution of Antarctica to global sea-level rise, as the effects of climate warming become more evident.

Data Availability Statement
The Massachusetts Institute of Technology general circulation model is an open source numerical model freely available on Github at https://github.com/MITgcm/MITgcm(checkpoint 66j).Model code, input and results are available at https://ecco.jpl.nasa.gov/drive/files/ECCO2/High_res_LarsenC(registration is required to access the

Figure 1 .
Figure 1.Model domain (dotted black box), integrated barotropic streamfunction, melt rate and sea-ice concentration.Magenta stars are the locations of drill sites after Nicholls et al. (2012).Yellow and red squares are the locations of CTD observations after Nicholls et al. (2004) and Hutchinson et al. (2020), respectively.(a) Bathymetry, both background colormap and contour lines.Magenta dashed lines are vertical cross-sections X-X' and Y-Y'.(b) Barotropic streamfunction in Sverdrup (Sv), control run.(c) Barotropic streamfunction, iceberg run.(d) Integrated melt anomaly as difference between integrated melt rate for iceberg (f) and control (e) runs.Streamfunction and melt patterns are calculated as temporal average between 1 January 2015 and 1 January 2017.Magenta boxes A, B, and C are key locations investigated in this paper, covering the survey area of Nicholls et al. (2012) and the shallow areas that coincide with the Bawden and Gipps Ice Rises, respectively.(f) and (e) also include monthly average of sea-ice concentration at the end of summer (March 2016), that is, when its concentration is the lowest.BIR: Bawden Ice Rise, GIR: Gipps Ice Rise, CA: Cape Alexander, CF: Cape Fairweather, CI: Cabinet Inlet, KP: Kenyon Peninsula, MI: Mobiloil Inlet.

Figure 2 .
Figure 2. Simulated ocean conditions along the horizontal cross-section corresponding to isopycnal surface 27.90.(a) Concentration of passive tracer tr01 as a snapshot of 1 January 2017.Red arrows show direction of the pathways of tr01 intrusion.(b) Distribution of meltwater fraction (MWF) calculated on the same day.(c) Distribution of potential temperature as an average for December 2016.(d)-(f) are same as (a)-(c) but for the iceberg run.

Figure 3 .
Figure 3. Simulated ocean conditions along vertical cross-sections X-X', which is aligned with the latitude of the south drill site.Black contour lines are isopycnal levels.(a), (b) Concentration of passive tracer tr01 and meltwater fraction (MWF), respectively, as a snapshot for 1 January 2017 in the control run.(c), (d) Distribution of potential temperature and salinity in the control run as a monthly average for December 2016.Dotted markers are observation profiles of potential temperature (c) and salinity (d) after CTD data of Nicholls et al. (2012) (under the ice base) and Hutchinson et al. (2020) (open ocean).(e)-(h) are same as (a)-(d) but for the iceberg run.

Figure 4 .
Figure 4. Simulated ocean conditions along vertical cross-section Y-Y'.Black contour lines are isopycnal levels.(a), (b) Concentration of passive tracer tr01 and meltwater fraction (MWF), respectively, as a snapshot for 1 January 2017 in the control run.(c), (d) Distribution of potential temperature and salinity in the control run as a monthly average for December 2016.(e)-(h) are same as (a)-(d) but for the iceberg run.
good agreement with in situ data acquired by a Nobska Modular Acoustic Velocity Sensor at the southern drill site (FigureS10in Supporting Information S1)(Davis & Nicholls, 2019).Moreover, modeled sea-ice results for 2015 are in fairly good agreement (R = 0.82) with average sea-ice concentration in the Weddell Sea obtained via DMSP SSM/I-SSMIS passive microwave data (FigureS6in Supporting Information S1)(DiGirolamo et al., 2022).Simulated melt rate of the entire Larsen C Ice Shelf as an average of the two years of simulations is 0.11 m yr −1 .