We employ a modified version of the CM2Mc Earth System Model of coupled to the Parallel Ice Sheet Model (PISM) v1.0 via the offline coupling framework described in . CM2Mc consists of the atmosphere model AM2.1, the land model LandLAD, the Modular Ocean Model version 5 (MOM5), and the dynamical Sea-Ice Simulator (SIS). They are coupled by the Flexible Modeling System (FMS) . The atmosphere grid has a latitudinal resolution of 3° and a longitudinal resolution of 3.75°, with 24 vertical levels. MOM5 utilises an Arakawa B-grid in a tri-polar configuration . Its lateral grid resolution is nominally 3°, varying latitudinally to a minimum of 0.6° at the equator to better resolve equatorial dynamics. The vertical grid has 28 layers implemented in the rescaled pressure (p*) coordinate. Layer thickness varies between 10dbar at the surface and 506dbar in the deep ocean .

PISM is used to simulate the AIS on a cartesian grid with a horizontal resolution of 16km×16km. In the vertical, the grid spacing ranges from 20m at the ice base to 100m for the thickest ice domes. Ice velocities are calculated by a superposition of the shallow-ice approximation and shallow-shelf approximation of the Stokes flow. We use an adapted version of PISM v1.0, which includes a precipitation scaling as introduced by . The scaling increases precipitation with decreasing ice elevation in order to account for the moisture holding capacity of air dependent on its temperature. The surface mass balance (surface melt and runoff) is computed using a positive degree-day (PDD) scheme. Melt coefficients are set to 3mm per PDD for snow and 9mm per PDD for ice. The Glen–Paterson–Budd–Lliboutry–Duval flow law describes the ice rheology in the thermomechanically coupled model with a freely evolving three-dimensional enthalpy field . Basal shear stress is parameterized dependent on the basal velocity and the yield stress that results from the Mohr–Coulomb criterion . Calving is implemented by the eigencalving approach by . Additionally, a subgrid scheme that captures calving fronts for different shelf geometries is used and a minimum thickness criterion of 50m at the calving front is applied. The adaptive time-stepping scheme reduces computational costs by choosing the maximum possible time step based on the internal dynamic state of the system. The source code of PISM is identical to the one used in , except for one bug fix on the approximation of the driving stress at floating ice margins that was committed later in PISM version 2.0.

The coupling of CM2Mc and PISM is done offline using the framework of . It exchanges mass and energy fluxes between ocean and ice sheet through the Potsdam Ice-shelf Cavity model (PICO) , which is implemented as a sub-module in PISM. PICO calculates sub-shelf melt rates by parametrising the vertical overturning circulation inside ice-shelf cavities. For that purpose it uses a box model based on , but extended to two horizontal dimensions, that divides the ocean ice-shelf boundary into 19 Antarctic basins see Fig. 2 in. In each basin, melting and freezing below the ice shelves is calculated based on ocean temperature and salinity at the depth of the continental shelf. Figure 3 in visualises the conceptual idea of the offline variable exchange between MOM5 and PISM. MOM5 and PISM are run sequentially with a fixed coupling time step of 10 years. After each step, the coupling framework processes the model outputs to make it compatible between the different model grids and then restarts the models. In the time domain, model outputs are averaged over the 10 years, i.e. seasonal ocean changes are not seen by the AIS. Spatially, the oceanic fluxes to PISM are based on regridded temperature and salinity fields at the AIS margin of MOM5. The horizontal resolution of MOM5 does not resolve the continental shelf of the Antarctic continent. Nevertheless, values for temperature and salinity at the depth corresponding to the mean continental shelf topography are extracted from MOM5 and horizontally averaged, resulting in scalar values for T and S for each basin . As MOM5 shows warm biases around the Antarctic continent, anomalies relative to the last 500 years of the climate spinup (see Sect. ) are calculated and applied to the forcing data of PISM standalone runs the basin mean values of these data are included in Fig. 2 in. PISM provides basal mass, calving, and surface mass fluxes aggregated per basin as well as the enthalpy required to melt the ice. These are regridded and inserted into the ocean as freshwater and enthalpy fluxes. In contrast to , basal mass fluxes are inserted not at the ocean surface but at the calving front ice-draft depth (which is determined as the mean depth at the outermost PICO box) as this represents the vertical insertion of ice-shelf cavity meltwater more realistically.

Our model setup only couples the ocean component (MOM5) to the AIS model PISM , with no coupling between atmosphere and Antarctic ice sheet components. The surface of PISM is forced with climatological means of surface air temperature and precipitation which are described in the next section. As PISM is not coupled to the atmosphere component of CM2Mc, we do not strictly conserve water there. However, discrepancies are negligible, as we put artificial freshwater to the system in our experiments (i.e. hosing). In the Southern Hemisphere, the removal of river runoff fluxes in the Antarctic domain was implemented as a minor modification to , because precipitation runoff into the Southern Ocean is represented by the PISM ice-sheet instance instead.

Most climate models use prescribed fluxes for Antarctic freshwater discharge, whereas our approach is able to capture changes in the dynamics of ice sheet and shelves in Antarctica and their interaction with the ocean. As a result, the different components of discharge from the Antarctic Ice Sheet into the surrounding ocean (surface runoff, basal melt, and calving) evolve dynamically in correspondence to the applied ocean-to-ice forcing.

The atmospheric boundary conditions for PISM are based on a multiple regression analysis of ERA-Interim data resulting in a parameterisation of mean annual and mean summer surface air temperature as a function of latitude and surface elevation, using an atmospheric lapse rate of Γ=-8.2°Ckm-1 . The mean precipitation field is calculated as the average between 1986 and 2005 from the output of the Regional Atmospheric Climate MOdel (RACMOv2.3) published by . For the PISM spinup, ocean forcing is provided by observational temperature and salinity data at the sea floor on the continental shelf of Antarctica averaged over the time period from 1975–2012 . Anomalies in the coupled framework are applied to these oceanic fields.

CM2Mc runs with pre-industrial atmospheric conditions as well as land cover of the year 1860. For the radiative forcing this implies a solar irradiance of 1364.67Wm-2 and greenhouse gas concentrations as given by . Further conditions provided by monthly mean climatologies derived from reanalysis are described by . The parameter set in the example configuration CM2M_coarse_BLING as distributed with the MOM5 code does not entirely reproduce the results shown by . Therefore, we modified several parameters (see Appendix ) to represent the pre-industrial climate state. Additionally, the CM2Mc time steps are reduced from 3 down to 1.5 h for the ocean and from 1.5 down to 0.75 h for the atmosphere.

The coupled system was spun-up in three stages. The first makes use of the ocean and sea-ice components of CM2Mc in standalone mode with prescribed atmospheric boundary conditions. It allows for an equilibration of the ocean with respect to a changed freshwater input of the Antarctic Ice Sheet. In this, the default river runoff values at the southernmost cells of the ocean were replaced by static freshwater fluxes calculated as the 1000 year mean of a PISM standalone spinup. This setup was run for 8000 years to avoid abrupt changes when coupling it to the interactive PISM. In the second spinup stage, CM2Mc was initialized using ocean and sea-ice restarts from stage 1 with the same PISM forcing and integrated for further 1000 years, now including a dynamic atmospheric component. Finally, once stage 2 was complete, PISM was coupled interactively and integrated for 500 years. All following experiments use the coupled model configuration and are extended from the state of this last spinup. In Appendix  we show model evaluation plots for the stage 3 spinup (compared to observational data), however we note that the climate state is broadly the same as in the uncoupled version of .

Throughout our analysis, we define the AMOC as the maximum zonally integrated meridional stream function strength in the Atlantic between 20 and 60°N below 500m and the AABW circulation cell strength is defined as the absolute value of the minimum global meridional overturning circulation (GMOC) below 2000m. The GMOC is defined as the aggregation of AMOC and the Indo-Pacific Meridional Overturning Circulation (PMOC) strength (maximum between 20 and 90°N below 500m in the Pacific basin). The stage 3 climate state has a maximum AMOC strength of ≈21Sv (1Sverdrup=1Sv=106m3s-1) and a GMOC strength of ≈23Sv. The mean temperature and salinity values that are used as input for PISM oscillate around -0.95±0.3°C and 34.575±0.025gkg-1 during the stage 3 spinup.

The setup of the North Atlantic freshwater hosing experiment which we apply to the model follows the protocols of the North Atlantic Hosing Model Intercomparison Project (NAHos-MIP) of . We use a uniform distribution of freshwater with hosing flux of 0.3Sv that is applied to the river runoff fields of the model in the North Atlantic and Arctic Oceans, distributing it in the regions above 50°N in the Atlantic and above the Bering Strait in the Pacific see Fig. 1a in. Other than described in the protocol, we do not compensate the artificial freshwater input, as MOM5 does not have an option for a flux adjustment over the total volume of the ocean. Surface compensation was tested, however in this setup had direct implications for Southern Ocean convection by affecting surface stratification (as shown in previous studies; ). To prevent a rapid rebound of the AMOC, we maintain the hosing for the entire duration of the experiment, causing a continuous sea level rise and freshening of the global ocean. However, the resulting forcing of sea level is not applied to PISM. We performed both uniform hosing experiments using 0.1 (not shown here) and 0.3Sv hosing strength, but show only results of the 0.3Sv uniform hosing experiment here, as the weaker forcing does not lead to a full AMOC shutdown in our model and therefore is not suitable to investigate our research question. A control experiment is run in parallel with the hosing run. Both runs are integrated for 1500 years. These two runs, hereafter called HOSING and CONTROL, and their difference are the focus of this study. All comparisons are presented as the difference between the 100 year means of HOSING and CONTROL to avoid the evaluation of short term climate variability. To estimate the significance of these anomalies, we test for each grid cell if the 100 year means of HOSING are within the range of internal variability of CONTROL. This is done by comparing the mean of the target time period in HOSING with the distribution of 100 year running averages in CONTROL. HOSING results are considered significant if they lie outside of the 95 % quantile of this CONTROL distribution.

The North Atlantic freshwater forcing weakens the AMOC from 21.5–5 Sv (that we refer to as AMOC collapse) after 100 years in the HOSING experiment (Fig. a). This reduction and its impacts are in line with the results of the NaHosMIP models in the first 100 years after the AMOC collapse. With the weakening of the AMOC, the surface air temperature in the Northern Hemisphere decreases (Fig. Appendix a), particularly over the Atlantic north of 60°N (up to -9°C), as the northward oceanic heat transport is reduced. In the SH, this results in increased surface air temperature (Fig. a) similar to , as well as in a southward shift of the ITCZ (Fig. c), and decreased sea level pressure in the SH (Fig. d), especially in the subtropical high regions . Due to the collapsed ocean circulation, heat accumulates in the subsurface South Atlantic north of 40°S (Fig. b) as also found in . This increases the temperature gradient across the ACC, which in combination with strengthened westerly winds leads to an intensification of ACC by approx 20Sv (Fig. c) as suggested by . In the ninth century of the HOSING simulation, there is a climate regime shift, as the rate of Antarctic Bottom Water (AABW) formation increases by more than 10Sv (Fig. b). A similar shift is present in the ACC strength (Fig. c) and maximum sea ice extent in the SH abruptly decreases by around 40% (Fig. d). We analyse these changes in detail in Sect. .

Figure  shows the transient response of the AIS to an AMOC collapse. The total change in the ice sheet volume is comparably small, as it barely exceeds the range of internal variability during the coupled spinup. Sea-level-rise potential (SLRP) with respect to the CONTROL experiment is decreased by around 2.5cm after 1500 years (Fig. a), without showing any abrupt deviations during the whole integration. Because PISM is only exposed to oceanic changes in our configuration, the main drivers of change in the ice sheet are basal melting and calving fluxes. Both fluxes are subject to oceanic temperature and salinity anomalies, which the model extracts from ocean depths between 500 and 1000m, depending on the glacial basin. Basal mass and calving fluxes in HOSING are within the range of variability of CONTROL in the first 800 years of the simulation (Fig. c and d), because during this period mean subsurface ocean temperature forcing in HOSING is in the range of values in CONTROL (-0.85±0.4°C) (Fig. b). There is less variability in the ocean temperature, explaining the decrease of basal mass variability (Fig. b and c). Calving, on average, is 3% larger than in CONTROL, driving the slight decrease in Antarctic mass (Fig. a and d). With the shift of AABW formation at around 830 years, the average ocean temperature across all basins decreases to -1.5°C in HOSING (Fig. b). This drives a subsequent decrease in basal mass flux (mean decrease rate of 160Gtyr-1) (Fig. c). In the same period, an increase in calving flux (on average 170Gtyr-1) (Fig. d) balances the basal mass decline, leading to similar rates of SLRP changes in HOSING and CONTROL. The reduction in basal melting leads to growing ice shelves that calve more frequently, which explains the increase in calving fluxes. The average ocean salinity across all basins shows a constant negative (freshening) trend (Fig. b). This constant decrease of salinity reflects the global freshening due to artificial freshwater forcing, and leads to subsurface conditions supporting basal melt reduction, though the linear trend of decreasing salinity is not reflected in the basal mass fluxes. Antarctic ice mass changes (Fig. a) are relatively small in response to the strongly reduced AMOC strength. Overall, in our simulations there is no destabilizing effect of an AMOC shutdown on the WAIS or other parts of the AIS, as hypothesized based on the SO surface warming after an AMOC collapse.

To explain this result, the following sections investigate SO conditions during the HOSING simulation, providing insights into the changes in SO subsurface temperature. We analyse the mean climate states of three time intervals indicated in Figs.  and . The first period (mean of model years 100–200) is chosen to focus on the time interval that previous studies have investigated . The second period (mean of years 600–700) represents the climate state before a reduction in maximum Southern Hemisphere (SH) sea ice extent (Fig. d) and the onset of deep convection in the Ross Sea. The last time period (mean of years 1400–1500) shows the climate state at the end of our simulation. The climate state of each time period is discussed in the following Sections with a focus on the changes in the SO and adjoint Antarctic ice sheet basins.

In the first period, the SO is characterized by positive sea-surface temperature (SST) anomalies up to 1.5°C south of 50°S (Fig. a). This warming, which is robust on the eastern side of the Antarctic continent (see stippling in Fig. a), leads to a decrease in maximum sea ice extent and decreasing sea-ice thickness by 5–10 cm in all coastal regions around Antarctica (Figs. d and a). Sea-ice melt dominates the change in total surface freshwater flux over changes in the precipitation-minus-evaporation balance (Fig. d and g) which can regionally decrease surface density. This is outweighed, however, by positive sea-surface salinity (SSS) anomalies originating in the South Atlantic (Fig. b), in line with results shown by , which diffuse across the ACC leading to a net increase in surface density in much of the SO (Fig. c), particularly the Weddell Sea region. In CONTROL, the Weddell Sea is the only SO region where convection occurs similar results of. This vertical mixing changes little in HOSING (Fig. e), shifting southwards due to reduced sea ice and more negative wind stress curl (Figs. f and d).

The atmospheric warming (Fig. a) south of 65°S leads, on average, to an ocean warming from the surface down to around 250m (Fig. a–c). However, temperature anomalies are negative in subsurface depths (500–1000 m) in this region (Fig. d) between years 100 and 200. This cooling originates in the North Atlantic, as residual North Atlantic Deep Water (NADW) transports the cooling signal of the collapsed AMOC to the SH. The inflowing NADW is upwelled mainly in the Atlantic sector and reaches the waters close to the Antarctic continent approximately in model year 100 (Fig. b). The ACC transports this signal along the coast, spreading it to all sectors of the SO (Fig. a–c). In regions without convection, we find that the cooling strengthens with time and represents a significant signal in the east of the Antarctic continent (see stippling in Fig. d). Inflowing water masses also drive a freshening trend (decrease of 0.1% per year) in the same subsurface depths (Fig. d). These changing subsurface conditions are decisive for the evolution of the AIS.

The temporal evolution of basal mass flux is strongly aligned with the mean subsurface ocean temperatures (between 500 and 1000m depth), which decrease by around 0.15°C (Fig. b). Therefore, between years 100 and 200, changes of basal mass flux show on average a decrease up to 100Gtyr-1 after 200 years in HOSING (Fig. c). Resulting AIS thickness changes are limited to the coastal regions and mostly smaller than 5m (Fig. d). Despite the decrease of basal melt in most regions, there is a peak of mean basal mass flux in the Atlantic sector around year 90 of our HOSING simulation (Fig. , basin 1). It results due to several years with less convection in which subsurface temperature anomalies in the Weddell Sea are positive (maximum positive anomaly of 0.4°C between year 80 and 130) (Fig. b). The reduced vertical mixing, which does not yet represent a regime shift compared to CONTROL (see stippling in Fig. e), stops the heat release from subsurface waters to the atmosphere (Fig. d), driving a briefly increase of melt rate up to 200Gtyr-1 that is reflected in decreasing AIS thickness of the Filchner–Ronne ice shelf. These thickness changes dominate the 100 year mean AIS thickness anomaly (Fig. d). Local ice-sheet changes are therefore directly linked to the changing frequency of convection in the Weddell Sea, which impacts interior temperature and salinity changes of coastal ocean waters. Similar fluctuations also exist in the spinup and CONTROL, which explains why results in this period are mostly not significant in the Weddell Sea region (see stippling in Fig. ).

The intermediate part of our simulation (years 600–700) is analysed to present ocean and SH climate conditions before the climatic shift, ie. the onset of deep convection in the SO. In this period, SSTs in most regions south of 60°S other than over Prydz Bay in East Antarctica show negative anomalies (Fig. a). This is caused by an increase in sea ice, as after approximately 250 model years, sea ice in the SH regrows and its maximum extent exceeds that of the CONTROL run (see contour lines in Fig. a–c). Sea-ice thickness increases in the Weddell and Ross Sea region (Fig. b) by 5–20 cm and variability of the averaged maximum sea-ice extent decreases compared to CONTROL (Fig. d). Thereby less deep warm water is upwelled and sea surface temperatures as well as surface–air temperatures over and close to Antarctica show a cooling signal (Figs. a and b). The maximum cooling in this time period is -3°C over parts of the Weddell Sea and over the Amundsen and the Ross Sea. The wind stress curl at the ocean surface intensifies along the maximum sea-ice edge, especially in the Bellingshausen and Amundsen Sea (Fig. f). Thereby, fresher circumpolar deep water is upwelled, conveying the low salinity signal of the North Atlantic hosing (Figs. c, d, and e). We therefore see negative SSS anomalies (Fig. b), despite less freshwater flux due to sea-ice melt south of 70°S (Fig. e), and surface densities south of 60°S decrease on average by 0.25kgm-3 (Fig. c) in all sectors except for the eastern part of the Pacific sector and latitudes north of 70°S in the Pacific. Positive sea-surface salinity anomalies (up to +0.5psu) in the eastern Bellingshausen Sea (Fig. b) are probably transported there from the South Atlantic, as the Ross Sea gyre is strengthened over time (Fig. e) and increases the water transport from the ACC towards the margin of the Antarctic continent .

Decreased densities of upwelled waters increase surface stratification, leading to a strong decrease in vertical mixing (maximum mixed layer depth (MLD) reduction of around 1250m) in the Weddell Sea (Fig. e). Therefore, variability of AABW formation (Fig. b) decreases in this time period. The reduced convection drives an increase of subsurface temperatures in the Atlantic sector by a maximum of 0.5°C at depths of 500–1000 m (Fig. d), as sensible and latent heat loss to the atmosphere is reduced (Fig. e). As a result, basal mass fluxes slightly increase in this region compared to CONTROL (Fig. , basin 1), however AIS thickness anomalies have the same magnitude as in the previous time period (see Sect. ), i.e. ice loss does not accelerate (Fig. d). With the reduction in Weddell Sea convection, heat begins accumulating in the deep SO below 2000m in all ocean basins (Figs. a, b, and a, b). Salinity values below 2000m show increasing negative anomalies, which originate from subsurface depths and over time diffuse to the deep SO (Figs. c and e).

In all other ocean sectors, the density of intermediate waters continue to decrease, driven by the freshening of upwelled NADW in the SO that is transported around the Antarctic continent by the ACC, as discussed for the previous time period in Sect. . Therefore, in the Indian and Pacific sector we find negative subsurface temperature anomalies (up to -1°C) along the coast (Fig. d), as well as decreasing salinity in subsurface depths in all regions south of 50°S, up to -0.45psu (Fig. e). These changes in subsurface ocean forcing (temperature and salinity) between years 200 and 800 therefore show the continuation of the trends seen in the previous time period. Consequently, the response of the AIS is very similar, showing an averaged change of basal mass flux decrease of 100Gtyr-1 with respect to CONTROL (Fig. c) and an increase in calving flux of the same rate (Fig. d). Both signals are most evident in ice-sheet basins in the Ross Sea region as well as basins close to and east of the Amery ice shelf (Fig. ), where AIS thickness anomalies are also positive (Fig. d).

From year 700 onwards, vertical mixing in the Weddell Sea progressively increases, down to 2000m depth (Fig. a). Convection in the Ross Sea starts to ventilate the deep SO (Fig. e) from year 830 onwards, mixing water columns in this region continuously every year (Fig. b). Convection starts due to the growing water column instability driven by the accumulation of heat in the deep SO in previous centuries (Fig. a) as well as decreased salinity of subsurface waters (Fig. c). As a result, AABW formation in HOSING increases by 15Sv with respect to CONTROL at the end of the simulation (Fig. b), exhibiting an anti-phased behaviour of NADW and AABW formation . Another contributor to the destabilisation of the SO water column might be changes in temperature of NADW, which is upwelled in the SO (Fig. b). As in the study by , heat accumulates in the South Atlantic, where over time it gradually diffuses to the deep ocean (see Video S1 in the Supplement). This diffusion is a continuous process during the whole simulation, warming over time the NADW that flows to the SH. Consequently, the negative subsurface temperature anomalies in the SO diminish over time and switch to positive anomalies (see Video S1). Around one century after the convection onset in the Ross Sea, water columns also reach instability in other SO sectors, inducing convection in coastal regions eastward from the Weddell Sea to around 70°E (the ocean in front of the Amery Ice Shelf) (Fig. e).

Along with the strengthening of convection in the SO there is a resulting increase of global mean SST (Fig. a) driven by the release of subsurface heat (Figs. f and e). The SST increase around the Antarctic continent is around 4 times higher in the climate state after enhanced AABW onset (Fig. a) than after the atmospheric warming due to the AMOC collapse (Fig. a). In line with this warming, there is a shift in SO sea-ice extent, decreasing to around 55% of the extent in CONTROL (Fig. d). Sea ice is thinning in all regions around Antarctica by up to 50cm (Fig. f).

Furthermore, the increase of deep convection leads to a cooling of Antarctic Bottom waters as heat is released to the surface. This temperature decrease affects all depths, except for the surface layer above ≈500m (Fig. a) and the maximum cooling averaged along the continental slope of Antarctica up to ≈65°S is 1°C with respect to CONTROL. Due to the climate regime shift in the SO in the last time period the Ross Sea gyre shifts its location again, so that with respect to CONTROL it expands to the east Indian sector in front of Wilkes Land (Fig. f). Thereby, cool water masses are transported westward from the Ross Sea and consequently, subsurface temperatures in the East Antarctic also decrease by up to 2°C (Fig. d). In the adjacent ice-sheet basins, the basal mass flux reduces to values close to zero. Especially in the East Antarctic, the calving rates increase by up to 100Gtyr-1 compared to CONTROL as ice shelves grow larger with less basal melt. In the Weddell Sea subsurface temperatures partly increase along the coast (Fig. d), which can be explained by shallower mixed layer depth in the eastern Weddell Sea (Fig. e). Reduced heat loss to the atmosphere (Fig. f) leads to less pronounced SST warming compared to most other regions around Antarctica (Fig. a) and temperature anomalies at 500–1000 m depth are positive (0.05–0.3 °C) along the coastal margin (Fig. d). In basins located nearby the Weddell Sea, basal melt increases in years of positive temperature forcing, i.e. only for certain years when no convection site opens up. Basins with ice shelves in Bellingshausen Sea show no significant change in basal mass or calving fluxes (Figs.  and ).