We use the low-resolution configuration of the EC-Earth model, EC-Earth3-LR, an Earth system model (ESM) developed collaboratively by the European research consortium EC-Earth. The EC-Earth model has flexible configurations that allow the inclusion or exclusion of various climate processes, making it a versatile tool for a wide range of climate studies . EC-Earth3 integrates several key components, including the atmospheric model IFS cycle 36r4, the land surface module HTESSEL, the ocean model NEMO3.6 , and the sea ice module LIM3 , all coupled via the OASIS3-MCT coupler . IFS and HTESSEL have a horizontal linear resolution of TL159 (1.125°), and the ocean and sea ice components (NEMO and LIM) have a nominal resolution of 1° .

The low-resolution configuration was selected to significantly reduce computational costs and because it has been extensively validated in both modern and palaeoclimate studies, showing robust performance in simulating the climates of past warm periods such as mid-Holocene, Last Interglacial, and Late Pliocene . These simulations have provided valuable information that has been integrated to major model intercomparison projects, such as the Paleoclimate Model Intercomparison Project phase 4 (PMIP4) and the Pliocene Model Intercomparison Project phase 2 (PlioMIP2) . Our setup allows us to conduct multi-centennial simulations and various sensitivity experiments, being particularly suited for exploring slow processes in the deep ocean, which are central to the goals of this study. Such processes include changes in stratification, overturning circulation, and AABW formation in response to altered climate forcing. Overall, the EC-Earth3 model has consistently demonstrated its effectiveness in capturing key climate dynamics, including temperature variability, heat fluxes, and other essential aspects of the Earth's system. This capability facilitates a more comprehensive understanding of the impacts of natural and anthropogenic forcing on the global climate system .

To investigate the impacts of varying ice sheet extent and CO2 concentrations in the polar Southern Hemisphere (pSH), we performed a series of sensitivity experiments, displayed in Table . The experiment design is based on the Core and Tier 2/Extension experiments as outlined in the Pliocene for Future protocol of PlioMIP2 and PlioMIP3 . These experiments also follow the PlioMIP2 naming convention in which the experiments with modern ice sheet extent are labelled E and the Late Pliocene ice sheets are labelled Ei, followed by their atmospheric CO2 levels. In our simulations, the experiment E280 comprises the CO2 concentration reconstructed for the pre-industrial period (280 ppmv), while the experiments E400 and Ei400 employed the reconstructed CO2 concentration of the Late Pliocene (400 ppmv). All simulations were started in parallel after branching off from a quasi-equilibrated PI spinup spanning 800 years to ensure consistent baseline conditions and that any changes observed in the simulations are due to the perturbation and not model drift. We defined quasi-equilibrium as a global surface air temperature trend of less than 0.05 K per century. Thus, the E280 experiment represents our pre-industrial control simulation, which is a core experiment of PlioMIP2/3, while E400 and Ei400 represent our sensitivity experiments, being within the Tier 2 experimental design of PlioMIP2 and continuing as optional but pivotal experiments in PlioMIP3 . Specifically, the primary purpose of the E400 experiment is to clarify how an elevated CO2 level with respect to PI, without other boundary condition changes, affects climate, a process usually referred to as forcing factorisation, which isolates CO2-driven climate change from other palaeoclimate forcings. Conversely, Ei400 focuses on the combined impact of CO2 and ice sheet extent change. Here we define the polar Southern Hemisphere as our domain of study, which includes the entire Southern Hemisphere from 60–90° S. To ensure consistency across all simulations, modern vegetation, as simulated for the year 1850 CE, was held fixed by disabling the offline LPJ-GUESS dynamic vegetation model . The final 200 years of model output is used for analysis of the mean state, with the pre-industrial control (E280) simulation serving as a baseline for comparison with the sensitivity experiments.

The protocol for our pre-industrial (PI) simulation follows the framework of for the Coupled Model Intercomparison Project version 6 (CMIP6) piControl experiment. Ice sheets, land geography, topography, and vegetation are all unmodified from the model. GHG concentrations for CO2, CH4, and N2O are 284.3 ppmv, 808.2, and 273.0 ppbv, respectively. For orbital parameters, eccentricity is set at 0.016764, obliquity is set at 23.549, and perihelion – 180° is set at 100.33.

The aim of these sensitivity experiments is to unveil the isolated impact of LP ice sheet extent to the climate of the polar Southern Hemisphere, without introducing confounding factors. To achieve this, we modify only the ice sheet mask to represent LP ice sheet extent while retaining pre-industrial albedo values and topography in the model. The LP AIS reconstruction was originally developed using the high-resolution British Antarctic Survey Ice Sheet Model, integrated with climatologies from the Hadley Centre Global Climate Model , utilising PRISM2 boundary conditions . The reliability of the AIS extent is further supported by the results of PLISMIP, which evaluated the dependencies of the ice sheet model for the warm period of the Late Pliocene using 30 different models . Figure  provides a visual comparison of the modern and LP ice sheet extent. LP GrIS reconstruction is provided for PlioMIP2 and based on 30 modelling results from PLISMIP . provide more detail, including spatial configuration of the LP GrIS and associated climatic impacts of modifying the GrIS in the polar Northern Hemisphere.

In our sensitivity experiments, the LP ice sheet masks were interpolated to the EC-Earth IFS grid and substituted into the initial condition referred to as the snow-depth field. In IFS, ice sheet presence is defined as grid cells with snow depth >10 m. By altering this field, we therefore reclassify those cells as exposed land or ocean in the LP experiments. The snow scheme in EC-Earth3 (based on ECMWF's HTESSEL land surface model) treats snow as perennial when snow depth exceeds a threshold 0.5 m of water equivalent, assigning a fixed high albedo and preventing further accumulation, effectively acting as a proxy for ice sheets. When snow depth is below this threshold, snow is considered seasonal and surface albedo is calculated as a weighted average between snow albedo and underlying surface albedo, reflecting seasonal snow cover variability . In our experiments, the AIS orography is retained, but, in regions where snow depth falls below the perennial threshold, seasonal snow processes dominate, allowing accumulation and melt with seasonally varying albedo. All other initial and boundary conditions were held fixed at PI values, including the prescribed surface albedo fields, orography/elevation, GHG concentrations, aerosol fields, ocean boundaries, and soil and vegetation distribution, together with their modified albedo properties. Additionally, no freshwater hosing was applied.

This experimental design therefore isolates the climatic response to the geometric removal of ice cover under fixed pre-industrial albedo and topography. Areas that are ice-covered in PI but ice-free in LP retain the model's default PI surface properties for that grid cell type (bare land or ocean), rather than adopting LP-specific albedo or vegetation reconstructions. This design allows us to focus on the first-order radiative effect. By doing so, we isolate the radiative effect of land ice loss, how the change in surface reflectivity influences the local and regional energy balance, and the resulting dynamical response, including how these changes affect atmospheric circulation, wind patterns, and ocean feedbacks within the model framework. This controlled experimental design avoids confounding influences from additional forcings, such as changes in vegetation, soil moisture, or orography, which may otherwise obscure the direct climatic impact of ice retreat. In this way, the experiment acts as a valuable idealised sensitivity test that serves as a baseline for understanding the isolated role of ice sheet retreat on Southern Hemisphere climate dynamics.

In the E400 scenario (400 ppm CO2, relative to E280), the average Antarctic surface air temperature rises by 2.51 °C. The warming is most pronounced in two specific regions that we refer to as “hotspots”, the Weddell (75° S, 50° W) and Ross (73° S, 160° W) seas, with temperatures increasing by up to 6 °C in the Weddell Sea and by 5 °C in the Ross Sea (Fig. a). Changes to albedo (Fig. c) are primarily confined to these two regions, with the most significant decrease (up to 20 %) occurring in the Weddell Sea, extending between the coastline and 60° S and clustered to the Weddell gyre. A smaller area of albedo decline (10 %) is observed west of the Ross Sea. Sea ice loss replicates these patterns surrounding the hotspots. The largest sea ice decline occurs to the east of the Weddell Sea (Fig. e) and is clustered to the coastline moving eastward, and a smaller area of sea ice loss is found west of the Ross Sea. More moderate warming occurs across the majority of the remaining area, with generally less than 2 °C increase over the interior of Antarctica and less than 1 °C at the periphery of East Antarctica. This is accompanied by virtually no changes in albedo. A localised cooling of 1–2 °C is observed in the Southern Ocean between 62° S, 160° W–160° E, where a small loss in albedo is also displayed (Fig. c).

In contrast, with LP ice sheet extent (Ei400 relative to E280), warming is much greater than in E400, with the near-surface air temperature over Antarctica increasing by an average of 9.49 °C. The warming hotspots shift further inland, with temperatures rising by over 17 °C inland from the Ross Sea (81–83° S, 180–155° W) and up to 16 °C inland from the Weddell Sea (81–83° S, 20–35° W) (Fig. b). The Ross and Weddell seas themselves experience warming of up to 12 and 13 °C, respectively. The most substantial albedo declines also occur at these inland hotspots, with a decline of more than 50 % inland of the Ross Sea, whilst the Ross Sea itself experiences a 30 % decrease. There is an albedo reduction of 40 %–50 % inland of the Weddell Sea, which extends into the Weddell Sea itself (Fig. d). Sea ice losses are consequently the most drastic in these locations, with a decline of over 65 % in the Weddell Sea and 60 % in the Ross Sea. Moreover, extensive areas of sea ice loss are observed extending eastward from the Weddell Sea and westward from the Ross Sea.

Over the interior of Antarctica, warming reaches 11–12 °C, decreasing towards the eastern coastline where temperatures increase by 6–7 °C. In Ei400, albedo changes are not confined to regions affected by ice sheet change, and there is an overall albedo decline of 20 % across the Antarctic interior, with decreasing severity toward the eastern coastline. There is a small hotspot showing a pronounced loss of 30 %–40 % on the east coast (75° S, 60–70° E). Additionally, albedo decreases of up to 30 % are observed along the coastline at 0–10° E and 140–160° E. This widespread albedo reduction is a result of the interplay of climate feedbacks that likely include changes in cloud cover, atmospheric temperature, and moisture transport influencing the radiation balance and surface reflectivity.

As the surface temperature rises due to the abrupt change in radiative forcing applied through our experiments, the subsequent shifts in climate create significant feedbacks that can influence large-scale atmospheric circulation, particularly the Southern Annular Mode (SAM). SAM is the leading mode of atmospheric variability in the Southern Hemisphere, characterised by fluctuations in the strength and position of the westerly winds encircling Antarctica . It has a large influence on pSH climate, as the wind regimes over this region modulate sea ice and deep-water formation, along with other climate patterns . Therefore, understanding how it responds to increased CO2 concentrations and ice sheet changes is vital. Here we derive the SAM mean state of the sensitivity experiments (Fig. a, b, and c) by applying empirical orthogonal functions (EOFs) to the sea level pressure (SLP) field and extracting its first mode. SAM variability in the form of a time series spanning the last 200 simulation years (Fig. ) was extracted through the first principal component of the EOF (PC1) and standardised.

In E280, Fig. a reveals an atmospheric structure characteristic of a positive SAM phase, with lower-than-normal SLP over Antarctica (90–60° S) and higher-than-normal SLP in the mid-latitudes. Regression of austral summer (December–January–February, DJF) 850 hPa zonal wind onto the leading SAM principal component (PC1) shows the strengthening of mid-latitude westerlies, whilst easterlies strengthen to the north of the Antarctic Polar Front (APF; Fig. d), reflecting the poleward shift in the mid-latitude westerly jet during positive SAM phases . The PC1 reiterates (Fig. a) the overall positive phase, with a slight positive central tendency (median 0.04) and marginally more positive months than negative (Table ). A more positive SAM phase is typically associated with cooler temperatures over Antarctica in summer, such as those established in the E280 simulation, as stronger westerly winds act as a barrier to warm air transport from lower latitudes .

Under increased CO2 forcing (E400), there is a reversal in EOF polarity. Higher-than-normal SLP is present over Antarctica, and lower-than-normal SLP is present in the mid-latitudes (Fig. b), indicative of a negative SAM phase. Mid-latitude Southern Ocean winds show strong negative anomalies (Fig. e), indicating an equatorial shift in the westerlies, thereby reaffirming the negative SAM phase. The PC1 time series (Fig. b) has a slight negative central tendency (median -0.07) and more negative months than positive. A negative SAM phase is associated with higher air temperatures across Antarctica and often leads to reduced sea ice formation that is triggered by katabatic winds in the pSH . This, in turn, reduces the upwelling of cold deep-ocean water onto the Antarctic continental shelf, further reinforcing this negative feedback.

Combining CO2 with LP ice sheet extent results in the same large-scale atmospheric structure as in E400, characterised by a negative SAM phase (Fig. c). The PC1 time series (Fig. c), however, demonstrates a behaviour closer to neutral in Ei400 than in E400, with a very small negative median (-0.03) and the most even split of negative to positive months out of the three experiments (Table ). Additionally, Ei400 displays the largest raw variability (in units of the original PC1), potentially indicating a more chaotic and less stable SAM pattern.

As evidenced in the previous section, the modified boundary conditions that were imposed in our experiments have significant implications for processes occurring in the near-surface atmosphere. Consequently, the sea surface and the ocean interior of the pSH are also affected. In E400, where we solely increase the atmospheric CO2 concentration, the surface ocean exhibits similar warming patterns to the atmosphere (Fig. a), albeit at a much lower magnitude. Figure a shows an overall warming along the path of the Antarctic Circumpolar Current (ACC), particularly within 45–55° S and through the Brazil–Malvinas Confluence (BMC), along with warming hotspots in the Weddell and Ross seas that are advected eastward through the ACC. Additionally, the Pacific upper-ocean cooling exhibited in Fig. a is in close agreement in both magnitude and location with the region where atmospheric cooling occurs. In Ei400, sea surface warming (Fig. b) agrees even more consistently with the change in near-surface atmospheric temperature (Fig. b). The warming hotspots confined to the Ross and Weddell seas and the Adélie Coast also remain, with the same pattern of eastward advection of warm waters as in E400, although with SST increasing up to 5 °C in the Weddell Sea.

With warming of the Earth's surface leading to an extensive sea ice melt (Fig. e and f), the surface layer of the Southern Ocean undergoes substantial freshening around the Sea Ice Zone (SIZ), which is highly sectorised. In E400 (Fig. a), the Bellinghausen and Davis seas exhibit the highest freshening, whereas the region encircling the APF (north of 55° S), the wind-driven outcrop of the Circumpolar Deep Water (CDW) in the Weddell gyre, and some parts of the Ross Sea exhibit upper-ocean salinisation. Conversely, with the stronger reduction in sea ice concentration that is imposed by reducing the ice sheet extent (Ei400), the surface freshening of the Southern Ocean is amplified (Fig. b), while the salinisation in the outcrop region of the CDW, in the Ross Sea, and outside the APF remains, albeit at a much reduced magnitude.

The combined effect of upper-ocean warming and freshening with weaker wind regimes within the APF (Fig. e and f) also affects the distribution of these thermohaline properties in the water column. Within the limit of the polar Southern Hemisphere, at 60° S, both E400 and Ei400 experiments exhibit the development of a fresh cap in the upper ocean that by itself increases the stratification of the water column (Fig. a and b) with respect to the pre-industrial climatology. As the westerlies weaken concomitantly, the stratification is sustained throughout the simulation, which further isolates the surface ocean. Moving down from the upper ocean, in E400 (Fig. a), the entire water column exhibits a uniform warming and salinisation during the runtime that is consistent with the higher sectorisation of areas that experience cooling (warming) and freshening (salinisation).

In Ei400 (Fig. b), however, the upper ocean experiences warming and freshening, whereas the subsurface undergoes warming and salinisation down to the intermediate layer (∼1000 m). This indicates an increase in stratification and a subsequent isolation of the ocean interior that allows its stronger salinisation in comparison to E400, especially at deep and abyssal depths. This likely occurs as a combined effect of the stronger stabilisation of the upper ocean, further reinforced by even weaker westerlies (Fig. f) and the entrainment of salty water masses through the intermediate layers of the Southern Ocean. Additionally, the abrupt forcing that is introduced by reducing the ice sheet extent in Ei400 results in an initial cooling of the ocean interior of about 3 °C, which offsets the degree of deep ocean warming in this simulation, when compared to E400. Therefore, in comparison to the pattern revealed in Fig. a, in Ei400, the Southern Ocean interior (surface) undergoes more salinisation (freshening) and less (more) warming.

As the changes in the sea surface temperature and salinity, which are amplified in Ei400 with respect to E400, impose contrasting effects in the ocean interior, the Southern Ocean Meridional Overturning Circulation (SMOC) undergoes a major shift (Fig. ), particularly with respect to the strength of the AABW, which is essential for the ventilation of the global ocean . Firstly, the SMOC reveals two major cells in E280: a clockwise cell that represents the northward flow of surface waters and the return flow of deep and intermediate waters that upwell at around 60° S and an anticlockwise cell that represents the northward flow of the AABW towards Indo-Pacific and Atlantic basins . In E400, the clockwise cell deepens and the AABW is significantly weakened from 8–2 Sv. In Ei400, on the other hand, the clockwise cell weakens and shoals, while the AABW exhibits a slight strengthening of about 2 Sv.

Considering the mean state of the last 200 simulation years, we therefore see that the AABW undergoes extensive weakening in both sensitivity simulations. Such sustained weakening is consistent with the picture of freshening and warming of the upper ocean displayed in Fig.  but not completely consistent with the picture of salinisation of the deep ocean. To evaluate the strength and variability in AABW formation in our experiments during runtime and gain more insights into the overall state of abyssal overturning in the Southern Ocean, we derived an AABW index that consists of the absolute value of the minimum overturning south of 60° S and below 500 m depth (adapted from ). Note that we exclude the upper 500 m from the index to isolate the water column comprising the subsurface to abyssal layers from the upper ocean and avoid capturing its high-frequency variability. The AABW index (Fig. ) shows that, even though Fig. b and c indicate a similar pattern of AABW weakening in both sensitivity experiments, the evolution of the AABW strength shows a sustained weakening in E400 and a partial recovery in Ei400, particularly after simulation year 700. Additionally, in both sensitivity experiments, the variability in the AABW is significantly reduced with respect to E280. Such behaviour indicates that enhancing the atmospheric CO2 concentrations, accompanied by massively reducing the extent of ice sheets, triggers compensatory mechanisms to the initial AABW suppression, likely involving salinisation of the deep and bottom ocean, as suggested in Fig. .

Furthermore, as the AABW formation is a complex process that is not yet fully understood and receives contribution from other masses formed in the Southern Ocean, particularly in the deep ocean , the insights gained from Figs. –, indicate that the underlying mechanism for the partial recovery of the AABW in Ei400 is a combination of the changes in salinity and temperature that occur in the water column, isolated through the weakened wind regime (Fig. f), and the interplay between the water masses in the Southern Ocean that are directly affected by these changes. The overall change in water mass density and thermohaline properties can easily be visualised through a temperature–salinity (T–S) diagram (Fig. ). In the figure, we highlight major water masses that are formed in the Southern Ocean and directly contribute to, or are the main product of, deep-water formation that is exported to the global ocean, including Antarctic Surface Water (AASW), Antarctic Intermediate Water (AAIW), Circumpolar Deep Water (CDW), High Salinity Shelf Water (HSSW), and Antarctic Bottom Water (AABW). In this sense, AASW, CDW, and HSSW are contributors to AABW formation , whereas AAIW does not play a direct role in AABW formation but is directly impacted by the nature of our experiments . A detailed description of their thermohaline properties in each simulation, along with density levels where they are formed, is detailed in Table .

Upon comparing E400 and Ei400 with E280, all water masses displayed in Fig.  are formed at warmer temperatures. This reflects the overall warming of the water column that is observed in Fig. b and d. Conversely, the evolution of salinity during runtime in Fig. a and c displays a contrasting pattern within the entire water column, which is also reflected in the overall lighter densities that are occupied by these water masses. Specifically, the AASW is the lightest water mass (γθ<27kgm-3) displayed in Fig. , which loses buoyancy during winter through brine rejection and is transformed into the HSSW (γθ>27kgm-3) and which subsequently descends through the water column and feeds the AABW. On the other hand, the CDW, brought through the Meridional Overturning Circulation (MOC), is an important water mass that, via diapycnal mixing with AASW and HSSW, directly contributes to AABW formation . In E400 (Fig. b), the AABW is contracted, and the AASW and HSSW become substantially fresher and lighter, with respect to E280. This indicates a reduced dense shelf overflow that ultimately weakens the AABW formation. As CDW becomes saltier and warmer to a degree that does not modify its density and the upwelling induced by the westerlies is not increased during runtime, its entrainment into the Southern Ocean is not able to destabilise the stratification towards promoting more deep-water formation.

Conversely, in Ei400, the AABW is expanded, becoming denser and saltier, while maintaining its temperature in comparison to E280. Additionally, the change in thermohaline properties and the density of the AASW and HSSW are amplified with respect to E400, but the CDW becomes saltier, which suggests enhanced entrainment of saltier waters into the Antarctic shelf and a subsequent increase in abyssal salinity. These processes combined justify the partial recovery of the AABW towards the end of the Ei400 simulation, even with relatively stable conditions in the upper ocean. In summary, freshening of the upper ocean induces increased stratification in both experiments that reduces the export of dense shelf water to the bottom of the ocean and results in an overall AABW suppression in the first years of the E400 and Ei400 experiments. However, reducing the ice sheets imposes an extra freshening and warming of the upper ocean that further isolates the ocean interior, while the subsurface and deeper ocean undergo a more intense salinisation that occurs in increased AABW formation, not necessarily induced by the change in ice sheet itself but by possible teleconnections with other ocean basins further north.