Highly stratified mid-Pliocene Southern Ocean in PlioMIP2

During the mid-Pliocene warm period (mPWP; 3.264–3.025 Ma), atmospheric CO₂ concentrations were approximately 400 ppm, and the Antarctic Ice Sheet was substantially reduced compared to today. Antarctica is surrounded by the Southern Ocean, which plays a crucial role in the global oceanic circulation and climate regulation. Using results from the Pliocene Model Intercomparison Project (PlioMIP2), we investigate Southern Ocean conditions during the mPWP with respect to the pre-industrial period. We find that the mean sea surface temperature (SST) warming in the Southern Ocean is 2.8 °C, while global mean SST warming is 2.4 °C. The enhanced warming is strongly tied to a dramatic decrease in sea ice cover over the mPWP Southern Ocean. We also see a freshening of the ocean (sub)surface, driven by an increase in precipitation over the Southern Ocean and Antarctica. The warmer and fresher surface leads to a highly stratified Southern Ocean that can be related to weakening of the deep abyssal overturning circulation. Sensitivity simulations show that the decrease in sea ice cover and enhanced warming is largely a consequence of the reduction in the Antarctic Ice Sheet. In addition, the mPWP geographic boundary conditions are responsible for approximately half of the increase in mPWP SST warming, sea ice loss, precipitation, and stratification increase over the Southern Ocean. From these results, we conclude that a strongly reduced Antarctic Ice Sheet during the mPWP has a substantial influence on the state of the Southern Ocean and exacerbates the changes that are induced by a higher CO₂ concentration alone. This is relevant for the long-term future of the Southern Ocean, as we expect melting of the western Antarctic Ice Sheet in the future, an effect that is not currently taken into account in future projections by Coupled Model Intercomparison Project (CMIP) ensembles.

1 Introduction

The Southern Ocean is a critical component of the Earth's climate system, as it acts as a major sink of heat and carbon dioxide (Khatiwala et al., 2009; Tjiputra et al., 2010; Iudicone et al., 2011) and affects global ocean circulation patterns (Talley, 2013; Rintoul, 2018). Close to the Antarctic margin, dense Antarctic Bottom Water (AABW) is formed that fills the abyssal ocean of the Atlantic and Indo-Pacific basins (Orsi et al., 1999; Johnson, 2008). Further north, strong westerly winds drive the upwelling of Circumpolar Deep Water (Marshall and Speer, 2012) and downwelling of Antarctic Intermediate Water (Sloyan and Rintoul, 2001). The Southern Ocean circulation plays a vital role in mitigating the current rise in atmospheric CO₂ levels through its uptake of CO₂, where it is estimated that over 1999–2007 the Southern Ocean has taken up 40 % of excess anthropogenic carbon present in seawater (Bopp et al., 2015). In addition, through its heat uptake, the Southern Ocean modulates rising global air temperatures and ocean heat content (Frölicher et al., 2015; Liu et al., 2018).

While there have been several model studies concerning the impact of rising CO₂ levels on the future Southern Ocean (e.g., Ito et al., 2015; Bracegirdle et al., 2020; Almeida et al., 2021), these studies consider transient simulations following Coupled Model Intercomparison Project (CMIP) scenarios. Due to the absence of slow feedbacks, they cannot not inform us of the long-term impacts of higher CO₂ levels on the global and regional climate and typically do not take into account Antarctic Ice Sheet adjustment in a warming climate. One way to consider these effects is by studying warm paleoclimates during geological periods in which the atmospheric CO₂ concentration was similar to that of a possible near-future warm climate. One of these periods is the mid-Pliocene warm period (mPWP; 3.264–3.025 Ma). Burke et al. (2018) have shown that the near-future climate stabilized at the Representative Concentration Pathway 4.5 (RCP4.5) scenario is similar to the climate during the mPWP, and this period can therefore be considered a good geological analog for long-term future climate under moderate CO₂ emissions. The mPWP is the most recent geological period during which the CO₂ concentration was around 400 ppm (e.g., Seki et al., 2010; Pagani et al., 2010; de la Vega et al., 2020), which is close to present-day levels. The mPWP global mean surface temperature was ∼3 °C higher than during the pre-industrial (PI) (Haywood et al., 2013, 2020; McClymont et al., 2020a). Geographic boundary conditions were also similar to present-day conditions, with some important differences including a substantially reduced Greenland Ice Sheet (Dolan et al., 2015), absence of the West Antarctic Ice Sheet, reduction in the East Antarctic Ice Sheet (Dowsett et al., 2010), and a different configuration of mostly high-latitude Northern Hemisphere (NH) ocean gateways, including the Bering Strait and Canadian Archipelago (Dowsett et al., 2016).

The Pliocene Model Intercomparison Project Phase 2 (PlioMIP2) encompasses an ensemble of 17 Earth system models and was initiated to investigate the mPWP climate and to study its potential as a future climate analog (Haywood et al., 2016). Each participating model has provided at least a mPWP simulation (Eoi⁴⁰⁰) and a PI simulation (E²⁸⁰). The mPWP Eoi⁴⁰⁰ simulations are centered on an interglacial peak, the KM5c time slice at 3.205 Ma, during which orbital forcing was similar to present day. The mPWP boundary conditions, including orography, ice sheet extent, and vegetation cover, are prescribed based on the reconstruction by the Pliocene Research, Interpretation and Synoptic Mapping (PRISM4) project (Dowsett et al., 2016). Previous studies using the PlioMIP2 ensemble have shown that surface temperature proxies generally agree well with the models (Haywood et al., 2020; McClymont et al., 2020a) but that the boundary conditions, including orography, vegetation, and ice sheet conditions, can have a strong regional influence on the simulated mPWP climate (Feng et al., 2022; Burton et al., 2023).

While the influence of orographic changes on the simulated mPWP climate can reduce the suitability of the mid-Pliocene warm period as a geological analog for a future climate, the smaller ice sheet cover over Antarctica and Greenland during the mPWP can be very relevant for informing us on the impact of higher CO₂ levels combined with reduced ice sheet cover in the long-term future. In the Southern Hemisphere (SH) polar region, there are few orographic differences between the Eoi⁴⁰⁰ and E²⁸⁰ cases, apart from those related to the reduced Antarctic Ice Sheet (Dowsett et al., 2016). It is therefore likely that the Southern Ocean in the mPWP simulations is primarily affected by a higher CO₂ level and the reduced Antarctic Ice Sheet cover. While the reconstructed ∼24 m sea level equivalent of combined ice mass loss from the Greenland Ice Sheet and Antarctic Ice Sheet in the mPWP (Haywood et al., 2016) seems unlikely for near-future scenarios, future projections on multi-centennial timescales do show substantial Antarctic Ice Sheet loss and potential West Antarctic Ice Sheet collapse under multiple climate scenarios (Cornford et al., 2015; Golledge et al., 2015; DeConto and Pollard, 2016; Bulthuis et al., 2019; Pattyn and Morlighem, 2020). Studying the state of the Southern Ocean in the mPWP simulations can therefore provide valuable insight into the impact of a reduced Antarctic Ice Sheet on both global and Southern Hemisphere ocean conditions in a warm equilibrium climate.

In this study, we will use model output from the PlioMIP2 ensemble to investigate Southern Ocean conditions during the mPWP. Our aim is to answer the following questions:

1. What are the differences between mPWP and PI states of the Southern Ocean, and how do these impact the abyssal global ocean circulation?

2. How much do mPWP boundary conditions, primarily the reduced Antarctic Ice Sheet, contribute to the difference in Southern Ocean conditions between the mPWP and the PI, and what are the resulting implications for a possible future climate?

To answer the first question, we will look at changes in Southern Ocean conditions in the mPWP compared to the PI in the PlioMIP2 models. This analysis includes sea ice extent, temperature, and salinity changes in the ocean, as well as atmospheric variables such as precipitation and wind that are first-order controls on the oceanography of the Southern Ocean. We will then link changes in these variables to altered deep-ocean circulation and ocean heat and freshwater transport in the Southern Ocean. For the second question, we consider seven additional E⁴⁰⁰ sensitivity simulations of the ensemble. These simulations are in equilibrium at a mid-Pliocene CO₂ level, 400 ppm, but do not feature mPWP boundary condition changes. Using these sensitivity simulations, we assess to what extent sea ice, temperature, and salinity changes are influenced by a reduced Antarctic Ice Sheet.

2 Methods

2.1 PlioMIP2 experimental design

The participating models of PlioMIP2 performed two core simulations (Haywood et al., 2016). Each model performed a PI control simulation (E²⁸⁰) at approximately 280 ppm CO₂ and a mPWP simulation (Eoi⁴⁰⁰) at 400 ppm CO₂, both with a simulation length set to be as long as possible or at least 500 model years. The mPWP Eoi⁴⁰⁰ simulation is performed using the PRISM4 boundary conditions (Dowsett et al., 2016). These boundary conditions prescribe a mPWP land–sea mask, topography, bathymetry, vegetation, soils, lakes, and land ice cover. These boundary conditions include the closure of the Bering Strait and Canadian Archipelago in the mPWP, as well as modifications to other seaways. The Greenland Ice Sheet was substantially reduced to approximately 25 % of its modern-day cover, where land ice remains only in the eastern Greenland mountains. The Antarctic Ice Sheet prescribed by PRISM4 has no ice over western Antarctica and reduced ice cover over eastern Antarctica. The total reduction in the global ice sheet volume is equivalent to a sea level increase less than ∼24 m with respect to the present day (Dowsett et al., 2016). Trace gases, orbital parameters, and the solar constant in the mPWP Eoi⁴⁰⁰ simulation are set to the same value as the model's PI E²⁸⁰ simulation. For the full details of the experimental design of PlioMIP2, the reader is referred to Haywood et al. (2016).

2.2 Data

We use datasets from simulations performed with 15 out of 17 models participating in PlioMIP2. NorESM-L and MRI-CGCM2.3 are excluded from this study because not all necessary data are available. Table 1 lists the 15 models used in this study, along with their resolution and reference. It is important to note that 5 out of the 15 models in the ensemble are Community Earth System Models (CESMs), of which 3 are CCSM4 models, creating an imbalance in the ensemble. However, due to differences in model versions, as well as model settings of individual groups, earlier PlioMIP2 studies show that the climate response varies significantly among the CESM members (e.g., Haywood et al., 2020; Z. Zhang et al., 2021), and we therefore consider it unlikely that this imbalance skews our results.

Feng et al. (2020)Peltier and Vettoretti (2014)Chandan and Peltier (2017)Baatsen et al. (2022)Feng et al. (2020)Feng et al. (2020)Feng et al. (2022)Stepanek et al. (2020)Q. Zhang et al. (2021)Hunter et al. (2019)Williams et al. (2021)Tan et al. (2020)Tan et al. (2020)Lurton et al. (2020)Chan and Abe-Ouchi (2020)Li et al. (2020)

Table 1 Overview of PlioMIP2 models used in this study. E⁴⁰⁰ sensitivity experiments have been provided by models typeset in bold.

^* Pre-industrial land–sea mask in both simulations.

Each of the models listed in Table 1 has completed the PI E²⁸⁰ simulation and the mPWP Eoi⁴⁰⁰ simulation with the mPWP boundary conditions that are described in Haywood et al. (2016). All models that are analyzed in this study, except HadGEM3, use the mPWP boundary conditions that include changes to the land–sea mask and ocean bathymetry. HadGEM3 uses a pre-industrial land–sea mask in both simulations. Additionally, seven model groups have performed a sensitivity study in which a PI simulation is performed at 400 ppm CO₂ (E⁴⁰⁰). We use this simulation to separate the effect of increased CO₂ levels from that of other mPWP boundary conditions.

2.3 Analysis

We have used 100-year averages for all PlioMIP2 model fields, except for IPSL-CM5A and IPSL-CM5A2. For the latter two models, we have used the available 50-year averages for E²⁸⁰ and Eoi⁴⁰⁰ experiments and a 30-year average for the E⁴⁰⁰ IPSL-CM5A2 experiment. Ocean freshwater and heat transport have been calculated on native model grids, except for COSMOS, where computation is done on a regular interpolated 1×1° grid, as the native curvilinear grid is unsuitable for this calculation. All other variables are horizontally interpolated to a regular 1×1° grid, and any spatial averages presented in this study are calculated from these fields. Unless indicated otherwise, an anomaly is defined as Eoi⁴⁰⁰–E²⁸⁰, and any percentage anomalies are defined with respect to E²⁸⁰.

The multi-model mean (MMM) is calculated by averaging over all the models in Table 1. The special model mean (SMM) is calculated by averaging over the seven models that have E⁴⁰⁰ data. In Fig. S1 in the Supplement, we compare the E²⁸⁰, Eoi⁴⁰⁰, and Eoi⁴⁰⁰–E²⁸⁰ SMM and MMM of surface variables used in this study. For the Southern Ocean, the SMM–MMM difference in the spatially averaged Eoi⁴⁰⁰–E²⁸⁰ sea surface temperature (SST) anomaly is approximately −0.5 °C. The models included in the SMM therefore show substantially colder Southern Ocean SSTs than those included in the MMM. The difference in average sea surface salinity (SSS) anomaly is less than 0.1 psu. The location of the MMM and SMM sea ice edge (15 % sea ice cover) is almost identical, although there are local differences in sea ice cover between the SMM and MMM.

The stratification of the Southern Ocean is analyzed using the potential density ρ, which is calculated from potential temperature and salinity fields using the TEOS-10 equation of state (Roquet et al., 2015). To provide a measure for the stratification, we use the following stratification index (SI), following Bourgeois et al. (2022), based on Sgubin et al. (2017):

$$\begin{array}{}\text{(1)}& \mathrm{SI}=\sum _{i=\mathrm{1}}^{\mathrm{10}}\mathit{\rho }\left(z_i\right)-\mathit{\rho }\left(z_{\mathrm{0}}\right),\end{array}$$

where z₀ is the sea surface and $z_i=z_{i-\mathrm{1}}+\mathrm{200}$ for i=1, …, 10 (units in m).

Averages for the Southern Ocean are calculated over all grid cells between 45–90° S, unless indicated otherwise. We calculate the Antarctic sea ice area as the sum of the product of the sea ice cover fraction and the grid cell area over all SH ocean grid cells. For Eoi⁴⁰⁰, only grid cells that were also ocean grid cells in E²⁸⁰ were included in this analysis. Thus, any additional sea ice that is formed in the mPWP in areas where the land–sea mask has changed from land to sea is not taken into account. This is to avoid bias in the sea ice area anomaly due to ocean cells being created in locations where the Antarctic Ice Sheet has disappeared.

Furthermore, we use the following definitions for the Atlantic Meridional Overturning Circulation (AMOC) and the abyssal deep-cell circulation: the AMOC strength is the maximum value of the Atlantic meridional overturning streamfunction at latitudes north of 0° N and below a depth of 500 m, and the abyssal deep-cell circulation strength is the absolute value of the most negative global meridional overturning streamfunction below a depth of 1000 m.

Figure 1 (a) Multi-model mean (MMM) Eoi⁴⁰⁰–E²⁸⁰ sea surface temperature (SST) anomaly. (b) MMM Eoi⁴⁰⁰–E²⁸⁰ SST anomaly minus the MMM global mean SST anomaly. (c) Same as panel (b) but on an orthographic projection centered on Antarctica. Dashed white and gray lines show the MMM annual mean sea ice edge (15 % sea ice cover) in E²⁸⁰ and Eoi⁴⁰⁰, respectively. (d) Scatterplot of the individual model sea ice area anomaly relative to E²⁸⁰ versus the SST anomaly in the Southern Ocean (45–90° S). Dashed line shows a linear least squares fit, with the indicated correlation (R value) and p value. E²⁸⁰ and Eoi⁴⁰⁰ are the respective PI and mPWP experiments.

2.4 Results

2.4.1 Mid-Pliocene Southern Ocean in PlioMIP2

Figure 1a shows the MMM Eoi⁴⁰⁰-E²⁸⁰ SST anomaly. The MMM global average SST is 2.4 °C warmer in the mPWP than in the PI, while the Southern Ocean (45–90° S) is 2.8 °C warmer. The ratio of SST warming in the Southern Ocean to global average warming, the Southern Ocean SST amplification, is therefore 1.2. We can see regions of amplified warming more clearly in Fig. 1b, which shows the MMM SST anomaly minus the MMM global average SST anomaly. Warming amplification occurs in both the Northern Hemisphere and Southern Hemisphere at latitudes higher than approximately 40° in regions where sea ice cover is absent in the Eoi⁴⁰⁰ simulations.

From Fig. 1c, which focuses on the southern polar region, it is clear that the largest SST warming amplification in the Eoi⁴⁰⁰ simulations occurs around the pre-industrial sea ice edge. In locations where sea ice cover persists in the mPWP, the warming is at or below the global average. There is a strong correlation between the Southern Ocean temperature anomaly and sea ice area decrease within the models ($R=-\mathrm{0.90}$ and p<0.05; Fig. 1d). Models with a relatively large decrease in sea ice area generally also experience the highest SST warming in the Southern Ocean. Sea ice reduction is therefore correlated to mPWP polar amplification in the Southern Hemisphere, although the analysis performed does not exclude the possibility that other mechanisms may be at work. Across the study ensemble, the MMM decrease in sea ice area in the mPWP is 71 % with respect to the pre-industrial sea ice area.

Burton et al. (2023) show that higher mPWP CO₂ levels combined with the reduced albedo of the SH polar region are the main drivers behind the warming in this region. The 60—90° S average annual mean MMM SAT is 7.6 °C higher than in the PI. When normalized by the MMM global average SAT warming of 3.4 °C, we find a SH polar amplification index of 2.2. This is substantially higher than both the CMIP6 abrupt-4 × CO₂ SH polar amplification of 1.1 (centered around year 100) and the CMIP6 historical (1979–2014) SH polar amplification of 0.9 (Hahn et al., 2021), in which simulations are purely driven by fast feedbacks and ice sheet adjustments are not taken into account (Eyring et al., 2016). The amplified SAT warming over the majority of Antarctica and the high-latitude Southern Ocean (Fig. S2) likely contributes to the increase in precipitation over many parts of the SH polar region in the mPWP with respect to the PI (Fig. 2a). The MMM increase in precipitation over the Southern Ocean and Antarctica is 0.35 mm d⁻¹ (+14 %) and 0.36 mm d⁻¹ (+50 %), respectively. Part of this increase in precipitation, particularly over the Southern Ocean, is offset by an equal or even larger increase in evaporation, as illustrated by the net surface freshwater flux anomaly in Fig. 2b. Nevertheless, the net change in surface freshwater flux is still positive over the majority of the Southern Ocean, with a MMM increase in surface freshwater flux of 0.13 mm d⁻¹ (+10 %). We find that, as a result, the majority of the Southern Ocean becomes less saline at the surface, especially close to the Antarctic continent and that this salinity decrease is robust across the PlioMIP2 ensemble (Fig. 2c). The MMM change in Southern Ocean sea surface salinity (SSS) is −0.17 psu, with local minima down to −3.0 psu.

Figure 2 (a) Multi-model mean (MMM) Eoi⁴⁰⁰–E²⁸⁰ precipitation anomaly. (b) MMM Eoi⁴⁰⁰–E²⁸⁰ surface freshwater flux (precipitation minus evaporation) anomaly. (c) MMM Eoi⁴⁰⁰–E²⁸⁰ sea surface salinity (SSS) anomaly. Stippling indicates that fewer than 12 out of 15 (<80 %) models agree on the sign of change. Dashed white and black lines show the MMM annual mean sea ice edge (15 % sea ice cover) in E²⁸⁰ and Eoi⁴⁰⁰, respectively. E²⁸⁰ and Eoi⁴⁰⁰ are the respective PI and mPWP experiments.

Figure 3 (a) Multi-model mean (MMM) zonal mean Eoi⁴⁰⁰–E²⁸⁰ potential temperature anomaly. (b) MMM zonal mean Eoi⁴⁰⁰–E²⁸⁰ salinity anomaly. (c) MMM zonal mean Eoi⁴⁰⁰–E²⁸⁰ potential density anomaly. Stippling indicates that fewer than 12 out of 15 (<80 %) models agree on the sign of change. Vertical dotted and dashed lines indicate the MMM annual mean zonal mean sea ice edge (15 % sea ice cover) in E²⁸⁰ and Eoi⁴⁰⁰, respectively. (d) Scatterplot of the high-latitude (>60° S) Southern Ocean stratification index in E²⁸⁰ PI and mPWP experiments. E²⁸⁰ and Eoi⁴⁰⁰ are the respective PI and mPWP experiments.

Figure 4 (a) Individual model Eoi⁴⁰⁰–E²⁸⁰ Atlantic Meridional Overturning Circulation (AMOC) and abyssal overturning strength anomalies. Horizontal lines indicate the multi-model mean (MMM) overturning strength anomalies. (b) Scatterplot of individual model Eoi⁴⁰⁰–E²⁸⁰ Southern Ocean stratification index anomalies against the abyssal cell strength anomalies. (c) Scatterplot of individual model Eoi⁴⁰⁰–E²⁸⁰ AMOC strength anomalies against the abyssal cell strength anomalies. Dashed lines show a linear least squares fit to the individual model data, with indicated correlation (R) and p values. E²⁸⁰ and Eoi⁴⁰⁰ are the respective PI and mPWP experiments.

As Figs. 1 and 2 show, the majority of the surface of the Southern Ocean both freshens and warms in the Eoi⁴⁰⁰ simulations with respect to E²⁸⁰. It is inevitable that these changes affect the density stratification in the Southern Ocean. Figure 3a and b show the MMM Southern Ocean zonal mean potential temperature and salinity anomalies, respectively. The warming is quite uniform throughout the water column, apart from relatively low surface warming close to Antarctica and an increase in surface warming between approximately 45–65° S. In contrast, in the MMM zonal mean salinity anomaly, we see a clear pattern emerging that features a fresher ocean surface and saltier deeper ocean across all latitudes in the mid-Pliocene Southern Ocean. The same pattern is reflected in the MMM zonal mean potential density anomaly in Fig. 3c. In most models, we see a stronger stratification in the mid-Pliocene Southern Ocean than in the pre-industrial (Fig. S3). At high latitudes (>60° S), the increase in stratification is consistent across all PlioMIP2 ensemble members, as illustrated by their increase in high-latitude Southern Ocean stratification index (Fig. 3d; Table S1 in the Supplement). The high-latitude Southern Ocean stratification index is defined as the average stratification index over all latitudes higher than 60° S, which is where abyssal deep-water formation takes place. The gridded individual model stratification index anomalies are shown in Fig. S4. The MMM Eoi⁴⁰⁰–E²⁸⁰ high-latitude Southern Ocean stratification index anomaly is 1.9 kg m⁻³, corresponding to a relative increase of 37 % with respect to the E²⁸⁰ MMM.

AABW formation is driven by buoyancy loss at the ocean surface. Consequently, the substantial increase in density stratification in high-latitude, non-circumpolar waters of the Southern Ocean is likely to have an impact on abyssal ocean circulation. Figure 4a shows the individual model and MMM Eoi⁴⁰⁰–E²⁸⁰ anomalies of the AMOC and abyssal overturning strengths. Overall, we see a decrease in abyssal cell strength in 9 out of 15 models, with a MMM decrease of 2.5 Sv (−22 %). The MMM AMOC strength increases by 3.1 Sv (+16 %) in the mPWP simulations with respect to the PI, which agrees with Z. Zhang et al. (2021), who show a consistently stronger mPWP AMOC in PlioMIP2. As 13 out of 15 models show a stronger mid-Pliocene AMOC, the response of the AMOC cell is more consistent than that of the abyssal cell.

The abyssal cell circulation strength should be related to the volume of deep-water formation in the Southern Ocean and therefore to surface water buoyancy in the high-latitude Southern Ocean. We find that the high-latitude (>60° S) Southern Ocean stratification index anomaly shows a significant anti-correlation ($R=-\mathrm{0.76}$; p=0.001) with the abyssal cell strength anomaly (Fig. 4b). The increased Southern Ocean stratification in the mid-Pliocene simulations therefore indeed appears to significantly impact the abyssal cell circulation and thereby the global ocean circulation. It is interesting, however, that the six models that do not feature a decrease in the abyssal cell strength still have a more stratified high-latitude Southern Ocean (Fig. 4b). Figure 4c shows that four out of these six models do have the highest AMOC anomaly of the ensemble, which points to a possible interaction between the abyssal cell and the AMOC cell.

When correlating the abyssal cell strength and the AMOC cell strength anomalies, there is no significant correlation (Fig. 4d). However, when the outlier IPSL-CM6A is removed, the correlation does become significant (R=0.57; p=0.03). A possible explanation for this correlation is that when the abyssal waters return to the Southern Ocean, both by diapycnal upwelling to intermediate depths and by transport to the surface along isopycnals (Lumpkin and Speer, 2007), they interact with North Atlantic Deep Water (NADW) from the AMOC cell. In the Southern Ocean, wind-driven Ekman transport upwells NADW to the surface to ultimately return northward (Toggweiler and Samuels, 1995), but the majority of NADW is transformed to AABW (Talley, 2013; Rousselet et al., 2021). The stronger AMOC in the Eoi⁴⁰⁰ simulations is driven by more vigorous NADW formation due to increased salinity in the high-latitude North Atlantic (Weiffenbach et al., 2023). In this way, the higher volume of relatively saline NADW results in more AABW transformation, thereby partly compensating the effects of higher density stratification in the high-latitude Southern Ocean.

Figure 5 (a) Multi-model mean (MMM) E²⁸⁰ 1000 hPa zonal wind, where positive values indicate eastward wind. (b) MMM Eoi⁴⁰⁰ 1000 hPa zonal wind. (c) MMM Eoi⁴⁰⁰–E²⁸⁰ 1000 hPa zonal wind anomaly. Stippling indicates that fewer than 12 out of 15 (<80 %) models agree on the sign of change. Dashed white and black lines show the MMM annual mean sea ice edge (15 % sea ice cover) in E²⁸⁰ and Eoi⁴⁰⁰, respectively. (d) MMM zonal mean 1000 hPa zonal wind in Eoi⁴⁰⁰, E²⁸⁰, and Eoi⁴⁰⁰–E²⁸⁰. Shading indicates 1 standard deviation around the MMM Eoi⁴⁰⁰–E²⁸⁰ anomaly. E²⁸⁰ and Eoi⁴⁰⁰ are the respective PI and mPWP experiments.

The pattern and strength of the zonal wind stress at the Southern Ocean surface also exert a strong effect on the global meridional overturning circulation through the wind-driven equatorward Ekman transport. Over the past few decades, the westerly winds over the mid-latitude Southern Ocean have been strengthening and shifting polewards (Swart and Fyfe, 2012; Deng et al., 2022). An increase in strength and poleward shift in the mid-latitude Southern Hemisphere westerlies has also been shown under future warming scenarios (e.g., Kidston and Gerber, 2010; Bracegirdle et al., 2013). In PlioMIP2, we do not see any systematic changes in surface westerly winds over the Southern Ocean (Fig. 5). Fewer than half of the models show a poleward shift in the zonal mean surface wind, with a minimal increase in maximum strength (Fig. S5), and there is little agreement over the sign of change, apart from a decrease in westerly wind strength south of Australia and increase in strength over the Pacific sector (Fig. 5c).

The Southern Ocean is the only location where the meridional ocean heat transport (OHT) in the mid-Pliocene simulations differs significantly from the pre-industrial (Fig. 6a; see Fig. S6 for individual model plots). Between 35–65° S, the MMM poleward heat transport reduces by 0.040 PW (−19 %). We show in Fig. 6b that the decrease in Southern Ocean ocean heat transport is significantly correlated with the decrease in abyssal cell strength ($R=-\mathrm{0.73}$; p=0.002). The four models that show an increase in poleward OHT in the Southern Ocean feature a stronger rather than weaker abyssal cell. This is consistent with Menviel et al. (2015), who show that a stronger AABW circulation leads to an increase in poleward heat transport. These results suggest that mid-Pliocene changes in abyssal cell strength affect the Southern Ocean heat transport.

Figure 6 (a) Multi-model mean (MMM) meridional ocean heat transport (OHT) in Eoi⁴⁰⁰, E²⁸⁰, and Eoi⁴⁰⁰–E²⁸⁰; positive values indicate northward transport. Shading indicates 1 standard deviation around the MMM Eoi⁴⁰⁰–E²⁸⁰ anomaly. (b) Scatterplot of the individual model Eoi⁴⁰⁰–E²⁸⁰ abyssal cell strength anomaly against the 35–65° S average OHT anomaly. Dashed lines show a linear least squares fit to the individual model data, with indicated correlation (R) and p values. (c, d) Same as panels (a) and (b) for the northward meridional freshwater transport (FWT), respectively. E²⁸⁰ and Eoi⁴⁰⁰ are the respective PI and mPWP experiments.

In the Eoi⁴⁰⁰ simulations, we also find a significant increase in northward freshwater transport in the Southern Ocean with respect to the E²⁸⁰ (Fig. 6c; see Fig. S7 for individual model plots). The northward freshwater transport increases between 35–65° S by 0.19 Sv (+31 %). Figure 6d shows a correlation between the Southern Ocean freshwater transport anomaly and the abyssal cell strength that is just outside the 95 % significance interval ($R=-\mathrm{0.49}$; p=0.07). The weakest increase or even decrease in northward freshwater transport is again shown by a subset of models that does not have a weaker abyssal cell circulation in the mid-Pliocene. It is also important to consider that the OHT and freshwater transport anomalies will likely be influenced by the stronger AMOC in the mid-Pliocene simulations. We do not, however, find a significant correlation between Southern Ocean heat or freshwater transport and the AMOC strength (Fig. S8).

2.4.2 Influence of boundary conditions and CO₂

The results in the previous section indicate that the density stratification of the mid-Pliocene Southern Ocean increases with respect to the pre-industrial in the PlioMIP2 ensemble, illustrating a significant impact on the global abyssal cell overturning circulation and its associated heat and freshwater transport. It is imperative to further investigate the cause of the changes in temperature and salinity in the Southern Ocean that is driving increased density stratification in the mPWP. To this end, we employ the seven available E⁴⁰⁰ sensitivity simulations to separate the effects of CO₂ concentration and geographic boundary conditions, in particular the reduction in the Antarctic Ice Sheet in the Eoi⁴⁰⁰ simulations. In this section, for all the comparisons between Eoi⁴⁰⁰, E⁴⁰⁰, and E²⁸⁰ cases, we use the SMM, i.e., the mean of the seven models that have an E⁴⁰⁰ simulation.

Figure 7a–c illustrate that the effect of the mid-Pliocene boundary conditions on the sea ice area is substantial. Where an increase in CO₂ from 280 to 400 ppm causes a decrease in sea ice area of 28 % relative to the pre-industrial, implementing mid-Pliocene boundary conditions causes an additional 31 % of sea ice area loss. Figure 7d–f reinforce the earlier link that we found between enhanced warming in the Southern Ocean and sea ice area decrease (in which we do not find an amplification of Southern Ocean warming in E⁴⁰⁰). The SMM Southern Ocean SST amplification is 0.9 in E⁴⁰⁰, while the SMM Southern Ocean amplification index is 1.1 in Eoi⁴⁰⁰, which is 0.1 lower than the MMM Eoi⁴⁰⁰ Southern Ocean amplification index. The Southern Ocean SST amplification observed in Eoi⁴⁰⁰ simulations is therefore mainly a consequence of the mPWP geographic boundary conditions. The SMM Southern Ocean average E⁴⁰⁰–E²⁸⁰ SST anomaly is 1.1 °C, while the Eoi⁴⁰⁰–E²⁸⁰ anomaly is 2.3 °C. This means that the CO₂ forcing and mPWP boundary conditions both contribute approximately equally to mPWP Southern Ocean warming.

Figure 7 (a) Special model mean (SMM; see Sect. 2) E²⁸⁰ annual mean sea ice cover. Dashed white line indicates the SMM annual mean sea ice edge (15 % sea ice cover). (b, c) Same as panel (a) for E⁴⁰⁰ and Eoi⁴⁰⁰. (d) SMM E⁴⁰⁰–E²⁸⁰ sea surface temperature (SST) anomaly minus the SMM global mean SST anomaly. (e, f) Same as panel (d) for Eoi⁴⁰⁰–E⁴⁰⁰ and Eoi⁴⁰⁰–E²⁸⁰. Dashed white, blue, and black lines show the SMM annual mean sea ice edge (15 % sea ice cover) in E²⁸⁰, E⁴⁰⁰, and Eoi⁴⁰⁰, respectively. Note that panel (f) is the same as Fig. 1c, but here the SMM is shown. E²⁸⁰ and Eoi⁴⁰⁰ are the respective PI and mPWP experiments, and E⁴⁰⁰ is the PI experiment at 400 ppm CO₂.

The SMM precipitation over the Southern Ocean and Antarctica is generally higher in Eoi⁴⁰⁰ than in E⁴⁰⁰ with respect to E²⁸⁰ (Fig. 8a–c). Over the 45–90° S Southern Ocean, the average SMM precipitation increases by 0.18 mm d⁻¹ (+7 %) in E⁴⁰⁰ and 0.31 mm d⁻¹ (+12 %) in Eoi⁴⁰⁰. The increase in precipitation over Antarctica is 0.10 mm d⁻¹ in E⁴⁰⁰ (+14 %) and 0.30 mm d⁻¹ (+43 %) in Eoi⁴⁰⁰. The larger increase in precipitation over the Southern Ocean in Eoi⁴⁰⁰ results in a larger decrease in SMM sea surface salinity than in E⁴⁰⁰, even though the mPWP boundary conditions also cause a substantial increase in evaporation over parts of the Southern Ocean and Antarctica (Fig. S9). On average, the sea surface salinity of the Southern Ocean decreases more in Eoi⁴⁰⁰ (−0.12 psu) than in E⁴⁰⁰ (−0.09 psu; Fig. 8d–f). This difference is larger in the areas close to the Antarctic continent that are covered by sea ice in Eoi⁴⁰⁰. In these areas, the sea surface salinity decreases by 0.54 psu in Eoi⁴⁰⁰ and 0.24 psu in E⁴⁰⁰.

Figure 8 (a) Special model mean (SMM; see Sect. 2) E⁴⁰⁰–E²⁸⁰ precipitation anomaly. (b, c) Same as panel (a) for Eoi⁴⁰⁰–E⁴⁰⁰ and Eoi⁴⁰⁰–E²⁸⁰. Dashed white, blue, and black lines show the SMM annual mean sea ice edge (15 % sea ice cover) in E²⁸⁰, E⁴⁰⁰, and Eoi⁴⁰⁰, respectively. Stippling indicates that fewer than five out of seven (<70 %) models included in the SMM agree on the sign of change. (d–f) Same as (a)–(c) but for sea surface salinity (SSS) anomaly. E²⁸⁰ and Eoi⁴⁰⁰ are the respective PI and mPWP experiments, and E⁴⁰⁰ is the PI experiment at 400 ppm CO₂.

In Fig. 9a–c, we show the SMM zonal mean potential density anomalies in the Southern Ocean. As expected, the warmer and fresher Southern Ocean surface in E⁴⁰⁰ also results in increased Southern Ocean stratification with respect to E²⁸⁰. The SMM high-latitude (60–90° S) Southern Ocean stratification index increases by 0.7 kg m⁻³ in E⁴⁰⁰ (+14 %) and 1.2 kg m⁻³ (+24 %) in Eoi⁴⁰⁰ (Table S1). Figure 9e shows that there are substantial regional variations in the SMM E⁴⁰⁰ and Eoi⁴⁰⁰ stratification indices. The Eoi⁴⁰⁰ SMM is less stratified in the Atlantic sector, possibly related to the stronger AMOC, while it is more stratified in the Indo-Pacific sector. At the mid-latitudes, the difference between the SMM 45–60° S average Southern Ocean stratification index anomalies is larger, with 0.8 kg m⁻³ in E⁴⁰⁰ (+10 %) and 1.9 kg m⁻³ (+24 %) in Eoi⁴⁰⁰. These results suggest that across most of the Southern Ocean, mPWP boundary conditions enhance the increase in Southern Ocean stratification induced by a higher CO₂ level.

Figure 9 (a) Full-depth special model mean (SMM; see Sect. 2) zonal mean E⁴⁰⁰–E²⁸⁰ potential density anomaly. (b, c) Same as panel (a) for Eoi⁴⁰⁰–E⁴⁰⁰ and Eoi⁴⁰⁰–E²⁸⁰. (d) SMM E⁴⁰⁰–E²⁸⁰ stratification index anomaly. (e, f) Same as panel (d) for Eoi⁴⁰⁰–E⁴⁰⁰ and Eoi⁴⁰⁰–E²⁸⁰. Stippling indicates that fewer than five out of seven (<70 %) models included in the SMM agree on the sign of change. Dotted, dashed, and dash-dotted red lines show the SMM annual mean sea ice edge (15 % sea ice cover) in E²⁸⁰, E⁴⁰⁰, and Eoi⁴⁰⁰, respectively. E²⁸⁰ and Eoi⁴⁰⁰ are the respective PI and mPWP experiments, and E⁴⁰⁰ is the PI experiment at 400 ppm CO₂.

Despite the greater increase in stratification index in Eoi⁴⁰⁰ than in E⁴⁰⁰, the abyssal cell shows a larger and more consistent decrease in the E⁴⁰⁰ simulations (Fig. 10a). The SMM E⁴⁰⁰–E²⁸⁰ abyssal cell strength anomaly, −1.9 Sv (−15 %), is nearly double the Eoi⁴⁰⁰–E²⁸⁰ anomaly of −1.0 Sv (+8 %). There is a significant negative correlation between the stratification index anomaly and the abyssal cell strength anomaly for both E⁴⁰⁰–E²⁸⁰ and Eoi⁴⁰⁰–E²⁸⁰, as shown in Fig. 10b. The difference between the slope and y intercept of the E⁴⁰⁰–E²⁸⁰ and Eoi⁴⁰⁰–E²⁸⁰ regressions is likely to be related to the interaction of the abyssal circulation with the AMOC. The results suggest that the stronger AMOC in Eoi⁴⁰⁰ may be partly compensating weakening of the abyssal cell. The AMOC cell is stronger in all Eoi⁴⁰⁰ simulations with respect to E²⁸⁰ (Fig. 10a), with an SMM increase of 3.4 Sv (+18 %), while the E⁴⁰⁰ SMM AMOC strength increase of 0.2 (+1 %) is substantially lower, and only three out of seven models simulate a stronger AMOC in E⁴⁰⁰.

Figure 10 (a) Eoi⁴⁰⁰–E²⁸⁰ and E⁴⁰⁰–E²⁸⁰ Atlantic Meridional Overturning Circulation (AMOC; green) and abyssal overturning (red) strength anomalies of models included in the special model mean (SMM; see Sect. 2). Horizontal lines indicate the SMM overturning strength anomalies. (b) Scatterplot of SMM ensemble individual model Eoi⁴⁰⁰–E²⁸⁰ and E⁴⁰⁰–E²⁸⁰ Southern Ocean stratification index anomalies against the abyssal cell strength anomalies. Dashed black and red lines show a linear least squares fit to the individual model data of the SMM ensemble, with indicated correlation (R) and p values. The multi-model mean (MMM) ensemble Eoi⁴⁰⁰–E²⁸⁰ regression from Fig. 4b and the Eoi⁴⁰⁰–E²⁸⁰ anomalies of the models not included in the SMM are shown in light gray. E²⁸⁰ and Eoi⁴⁰⁰ are the respective PI and mPWP experiments, and E⁴⁰⁰ is the PI experiment at 400 ppm CO₂.

2.5 Discussion

2.5.1 Model–data comparison

There are no data on the Southern Ocean sea ice cover during the KM5c time slice, but there are a number of studies on Pliocene sea ice and ocean temperatures that suggest there was reduced ice cover during this period (e.g., Barron, 1996; Whitehead and Bohaty, 2003; Whitehead et al., 2005; Dowsett et al., 2010). Whitehead et al. (2005) reconstructed winter sea ice cover at Site 1165 (64.380° S, 67.219° E) and Site 1166 (67.696° S, 74.787° E) across the Pliocene. They found a maximum of 78 % less sea ice relative to modern at Site 1165, and a 61 % reduction at Site 1166 during the Pliocene. Our MMM shows a respective relative sea ice concentration decrease of 72 % and 39 % at these locations, respectively, which matches reconstructions reasonably well, considering the range of uncertainty in the reconstructions of at least 30 % (Whitehead et al., 2005).

We also compare Southern Ocean PlioMIP2 SSTs to SST reconstructions from the KM5c time slice. The SST reconstruction by McClymont et al. (2020a) includes six proxy locations in (close proximity to) the Southern Ocean. We compare absolute mPWP SSTs and mPWP–PI SST anomalies at these locations for both the models and data in Fig. S10, using the observed pre-industrial SSTs from the ERSSTv5 1870–1900 dataset (Huang et al., 2017a) and reconstructed mPWP SSTs from McClymont et al. (2020a). While absolute mPWP SST reconstructions generally fall within the range of the PlioMIP2 ensemble SSTs, the average spread of modeled mPWP SSTs is 7.9 °C. This intermodel spread is reduced to 4.1 °C for the mPWP–PI SST anomalies. However, at three out of six SST proxy locations, the reconstructed SST anomaly falls outside the range of the modeled SST anomaly. The models are neither consistently warmer nor colder than the reconstructions, meaning we cannot detect a consistent bias in the modeled or reconstructed Southern Ocean SSTs. The models appear to match absolute mPWP Southern Ocean SSTs reconstructions better than reconstructed mPWP–PI SST anomalies, possibly due to discrepancies between reconstructed and modeled Southern Ocean SSTs in the PI.

2.5.2 Mid-Pliocene Southern Ocean as a future analog

Burton et al. (2023) have shown that the influence of mPWP boundary conditions on the simulated global average mPWP–PI SAT, SST, and precipitation anomalies in PlioMIP2 is substantial. They found that 44 % of the SAT and SST anomaly and 49 % of the precipitation anomaly were driven by non-CO₂ forcing. Our results align with this study and show that the mPWP boundary conditions are responsible for approximately half of the average mPWP–PI anomaly in sea ice cover, SST, precipitation, and stratification index in the Southern Ocean. While a separation of effects due to orographic and land ice cover is not possible, it is plausible that this influence is primarily due to the decreased Antarctic Ice Sheet as there are, apart from a closed Indonesian Throughflow, few other changes in SH mPWP boundary conditions with respect to the PI. This may make the mid-Pliocene warm period a good analog for our long-term future climate in which substantial melting of the Antarctic Ice Sheet is expected. We show that mPWP boundary conditions, primarily the reduction in the Antarctic Ice Sheet in the Southern Hemisphere, lead to additional precipitation, sea ice loss, and sea surface warming and are thus responsible for reinforcing increased stratification caused by a rise in CO₂ levels. This potential effect of the Antarctic Ice Sheet reduction is not taken into account in the CMIP Earth system model (ESM) projections of future climate but may therefore play an important role in Southern Ocean conditions on multi-centennial timescales. Moreover, a slowly changing ice sheet will contribute to changes in equilibrium climate sensitivity due to state dependence of fast feedback processes (von der Heydt et al., 2016), which is not accounted for in CMIP ESM simulations and multi-century ice sheet projections. It should be taken into account, however, that Antarctic Ice Sheet projections do not show a ∼21 m sea level equivalent decrease in ice volume, which has been reconstructed for the mPWP Antarctic Ice Sheet (Dowsett et al., 2010). Under unabated emission scenarios, multi-centennial projections show a maximum increase in sea level equivalent melt between approximately 5 m (Golledge et al., 2015) and 15 m (DeConto and Pollard, 2016) by 2500 CE, which makes the mPWP ice sheet extent unrealistic, even for longer-term future scenarios. This means that the Southern Ocean conditions from this study are not likely to be analogous to a long-term future, although a partial reduction that includes the collapse of the West Antarctic Ice Sheet would still greatly decrease the ice sheet extent and therefore possibly induce effects similar to those found in this study.

Our results suggest a relationship between increased stratification in the Southern Ocean and a weakened abyssal cell circulation during the mid-Pliocene warm period. CMIP5 studies investigating RCP4.5 and RCP8.5 scenarios show a decline in Antarctic Bottom Water formation due to fresher and warmer Southern Ocean surface conditions and consequentially a slowdown of deep-ocean circulation by the end of the 21st century (Meijers, 2014; Ito et al., 2015). This is consistent with de Lavergne et al. (2014), who show weakening of Southern Ocean deep convection due to surface freshening under the RCP8.5 scenario, as well as recent work by Chen et al. (2023) demonstrating that freshening of the Southern Ocean surface due to Antarctic meltwater induces weakening of deep convection. CMIP6 models also show future Southern Ocean surface warming and sea ice decrease, as well as mixed layer shoaling that suggests the slowdown of AABW formation (Purich and England, 2021).

While the majority of the PlioMIP2 models show a weaker abyssal cell circulation in the mPWP, there are 6 out of 15 ensemble members for which the Southern Ocean stratification increases in the mPWP simulations without showing weakening of the abyssal cell circulation. Interestingly, four of these six models also exhibit the largest AMOC strength increase during the mPWP with respect to the PI. This implies a potential linkage between the AMOC and abyssal cell circulation, where a stronger AMOC during the mid-Pliocene may partially compensate for the reduced formation and circulation of Southern Ocean bottom water. The stronger AMOC in the mPWP PlioMIP2 simulations has been attributed to changes in geographical boundary conditions, specifically the closure of the Bering Strait and Canadian Archipelago (Z. Zhang et al., 2021; Weiffenbach et al., 2023). Since these geographical changes are not relevant for the future, it is uncertain whether the strength of decline in abyssal cell circulation, or the lack of decline exhibited in some models, resembles a plausible future scenario. Given the substantial influence of the intensified AMOC in the mPWP PlioMIP2 simulations on global ocean circulation, it is reasonable to expect that it also impacts Southern Ocean conditions. To disentangle effects of orographic and ice sheet changes, additional sensitivity studies are necessary. Nevertheless, our results highlight the tight interplay between AMOC and abyssal cell, connecting Southern Hemisphere and Northern Hemisphere high latitudes, and their potential to impact fast feedback processes relevant for equilibrium climate sensitivity in various regions of the world.

2.5.3 Southern Ocean biases

From historical CMIP5 and CMIP6 simulations, it is known that low-resolution Earth system models show systematic biases and inaccuracies concerning Southern Ocean surface conditions and deep-ocean circulation. Southern Ocean SSTs are among the most (warm) biased characteristics of Earth system model simulations (e.g., Wang et al., 2022; Zhang et al., 2023)). CMIP5 simulations have been shown to have a consistent warm and low-density bias over the entire water column and show a large spread in the volume and characteristics of Antarctic Bottom Water (Sallée et al., 2013). As most CMIP5 models create deep water by deep convection in the open ocean, which only rarely occurs in reality, bottom water formation processes are not well represented in these low-resolution ESMs (Heuzé et al., 2013). This remains a problem in the newer CMIP6 generation models (Mohrmann et al., 2021). The majority of the PlioMIP2 models are models also used for CMIP5 or CMIP6 and therefore have the same issues representing deep-water formation. As such, changes to deep-cell circulation in the mid-Pliocene warm period may also be biased due to the inaccurate representation of deep-water formation in the Southern Ocean. In addition, it has been shown that the CMIP5 ensemble shows biases in Southern Ocean SSTs due to errors in cloud-related parameterizations (Hyder et al., 2018). These biases have been correlated to simulated Last Glacial Maximum AMOC depth anomalies (Sherriff-Tadano et al., 2023) and could potentially also affect the AMOC and deep-cell circulation in the mPWP.

Another important bias in the PlioMIP2 models concerns the sea ice area. It has been shown that changes in temperature, sea ice area, and precipitation in future projections are strongly tied to the simulated historical mean sea ice area in CMIP5 (Bracegirdle et al., 2015; Kajtar et al., 2021) and the models strongly overestimate the historical variance in the sea ice extent (Zunz et al., 2013). The PlioMIP2 ensemble also features large ranges of uncertainty in sea ice cover, which we have shown to be crucial when it comes to polar amplification and Southern Ocean precipitation. Figure S11 shows that the MMM sea ice cover and extent in the PI E²⁸⁰ simulations is spatially similar to historical observations (1979–2004 NOAA/NSIDC Climate Data Record of Passive Microwave Sea Ice Concentration; Meier et al., 2021), with a MMM E²⁸⁰ sea ice area of 10.4 million km² and historical observed sea ice area of 9.5 million km². While the historical observed sea ice area should ideally be compared to historical simulations, Fig. S11 gives some confidence that the simulated MMM E²⁸⁰ sea ice cover is realistic for the PI. However, the sea ice area variation among models is 5.2–15.6 million km², which is substantially larger than the discrepancy between the MMM E²⁸⁰ and historical observations. Furthermore, the absolute Eoi⁴⁰⁰–E²⁸⁰ sea ice area anomaly varies between −4.0 and −11.7 million km², and relative sea ice area anomaly varies between −39 % and −96 %. This large variance in sea ice extent leads to uncertainty about the representativeness of the Southern Ocean conditions in response to both the higher CO₂ and the geographic boundary conditions in the mPWP, especially for those models that have a substantially lower or higher sea ice cover than the MMM.

3 Conclusions

The Southern Ocean simulated by PlioMIP2 is characterized by a large reduction in sea ice area in the mid-Pliocene warm period with respect to the pre-industrial. Due to the increase in CO₂ and the Antarctic Ice Sheet reduction, the 60–90° S MMM SAT is 7.6 °C warmer in the mPWP. This results in a polar amplification factor of 2.2 with respect to the global average MMM SAT anomaly of 3.4 °C, which is substantially higher than in future projections. The MMM relative sea ice area decrease is 71 %, with a substantial variation in sea ice area decrease among the 15 models that ranges from −39 % to −96 %. The decrease in Southern Ocean sea ice area is strongly tied to the SST increase in the mPWP, ranging from 1.1 to 5.1 °C, with a MMM increase of 2.8 °C. There is also a robust increase in precipitation, with a MMM increase of 0.35 mm d⁻¹ over the mPWP Southern Ocean and Antarctic continent, resulting in freshening of the ocean surface.

The warm and fresh mPWP Southern Ocean surface leads to an increase in density stratification with respect to the pre-industrial. This is correlated with a slowdown of the global abyssal cell circulation during the mid-Pliocene warm period, affecting both heat and freshwater transport in the Southern Ocean. The CO₂ concentration and Antarctic Ice Sheet reduction are primary drivers of warmer and fresher Southern Ocean surface conditions and of deep-ocean circulation in the mid-Pliocene warm period with respect to the pre-industrial. However, we do find a potential interaction of deep abyssal circulation with the stronger AMOC in the mPWP simulations, where the strengthened AMOC partly compensates weakening of the abyssal cell. Increased AMOC strength has been linked to high North Atlantic salinity due to closed Arctic gateways during the mid-Pliocene warm period and therefore does not appear to be directly related to the CO₂ increase or Antarctic Ice Sheet reduction.

Sensitivity simulations at the mid-Pliocene CO₂ level without any geographic, ice sheet, or vegetation changes also show decreased sea ice area, increased precipitation, and warming and freshening of the ocean surface. This effect is enhanced by mid-Pliocene boundary conditions, in particular by the Antarctic Ice Sheet reduction, where these boundary conditions contribute approximately one-half to the total sea ice area loss, SST warming, precipitation increase, and density stratification increase in the mPWP simulations. This illustrates that it is important to consider the effects of a smaller Antarctic Ice Sheet when projecting for possible long-term future climates. The multi-century reduction in the Antarctic Ice Sheet will drive changes in fast feedback processes (e.g., clouds and sea ice cover) affecting equilibrium climate sensitivity. However, uncertainties are present via the large model spread and biases in sea ice area and Southern Ocean deep-water formation, as well as via the effect of orographic changes on ocean circulation and Southern Ocean conditions. Additional sensitivity studies, where the land–sea mask and ice sheet size are implemented separately, are necessary to be able to separate the effect of orography and ice sheet changes in the mPWP simulations and further investigate the mechanisms behind the mid-Pliocene Southern Ocean's effect on global ocean circulation.