A weakened AMOC could cause Southern Ocean temperature and sea-ice change on multidecadal timescales

manuscript submitted to JGR: Oceans A weakened AMOC could cause Southern Ocean1 temperature and sea-ice change on multidecadal2 timescales3 Rachel Diamond1,2, Louise C. Sime1, David Schroeder3, Laura C. Jackson4,4 Paul R. Holland1, Eduardo Alastru´e de Asenjo5,6, Katinka Bellomo7,8, Gokhan5 Danabasoglu9, Aixue Hu9, Johann Jungclaus10, Marisa Montoya11,12, Virna L.6 Meccia13, Oleg A. Saenko14, Didier Swingedouw15 7 1British Antarctic Survey, Cambridge, UK8 2Department of Earth Sciences, University of Cambridge, Cambridge, UK9 3Centre for Polar Observation and Modelling, Department of Meteorology, University of Reading,10 Reading, UK11 4Met Office, Exeter, UK12 5Institute of Oceanography, Center for Earth System Research and Sustainability (CEN), Universit¨at13 Hamburg, Hamburg, Germany14 6International Max Planck Research School on Earth System Modelling, Max Planck Institute for15 Meteorology, Hamburg, Germany16 7National Research Council of Italy, Institute of Atmospheric Sciences and Climate, Turin, Italy17 8Department of Environment, Land and Infrastructure Engineering, Polytechnic University of Turin,18 Turin, Italy19 9Climate and Global Dynamics Lab, US National Science Foundation, National Center for Atmospheric20 Research, Boulder, CO, USA21 10Climate Variability Department, Max Planck Institute for Meteorology, Hamburg, Germany22 11Departamento de F´ısica de la Tierra y Astrof´ısica, Universidad Complutense de Madrid, Facultad de23 Ciencias F´ısicas, 28040 Madrid, Spain24 12Instituto de Geociencias, Consejo Superior de Investigaciones Cient´ıficas-Universidad Complutense de25 Madrid, 28040 Madrid, Spain26 13National Research Council of Italy, Institute of Atmospheric Sciences and Climate, Bologna, Italy27 14SEOS, University of Victoria, Victoria, BC, Canada28 15Univ. Bordeaux, CNRS, Bordeaux INP, EPOC, UMR 5805, 33600 Pessac, France29 Key Points:30 •We present the first CMIP6 multi-model intercomparison of AMOC weakening im-31 pacts on Southern Ocean temperatures and sea ice.32 •Increased southwards heat transport into the Southern Ocean causes warming and33 sea-ice loss, smaller than direct climate forcing effects.34 •We find a novel, tropical-Antarctic atmospheric teleconnection, which causes re-35 gional temperature and sea-ice change.36 Corresponding author: Rachel Diamond, [email protected] –1– manuscript submitted to JGR: Oceans Abstract37 We present the first CMIP6-era multi-model intercomparison of the Southern Ocean tem-38 perature and sea-ice response to substantial AMOC weakening. Results are based on anal-39 ysis of the North Atlantic Hosing Model Intercomparison Project (NAHosMIP), involv-40 ing eight CMIP6 models under identical North Atlantic freshwater hosing. On multidecadal41 timescales, we find that southwards ocean heat transport into the Southern Ocean in-42 creases, causing surface warming and sea-ice loss. Additionally, an atmospheric tropical-43 Antarctic teleconnection, identified here for the first time, causes regional temperature44 and sea-ice changes in the Southern Ocean. Unlike previous studies, we find that the Amund-45 sen Sea Low deepens for only some models. Overall, in the multi-model ensemble mean46 (multi-model range in brackets), over years 50-100 after AMOC weakening: Southern47 Ocean surface air temperature warms by 0.3 (0.1–0.7)◦C, sea level pressure decreases by48 30 (10–70)Pa, and sea-ice area decreases by 0.4 (-0.2–1.3)Mkm2. The teleconnection leads49 to regional differences between the response in the Indian sector and the Weddell Sea,50 of 180 (80–320)Pa in sea level pressure, 0.6 (0.5–1.4)◦C in surface air temperature, and51 0.1 (0.1–0.2)Mkm2in sea-ice area. These Southern Ocean heat transport, temperature,52 pressure, and sea-ice changes are small relative to the changes expected under future an-53 thropogenic warming, despite the large and idealised 0.3 Sv hosing used to weaken the54 AMOC.55 Plain Language Summary56 The Atlantic Meridional Overturning Circulation (AMOC) could substantially weaken57 over the next century due to climate change. The Southern Ocean is a key control of global58 ocean circulation and climate. Here, we use the latest generation of climate models to59 assess the impacts of this potential AMOC weakening on the Southern Ocean and Antarc-60 tic sea ice, on timescales of less than a century. Following AMOC weakening, ocean trans-61 ports move heat southwards into the Southern Ocean, causing Southern Ocean surface62 warming and sea-ice loss. We also identify a new atmospheric connection, from the trop-63 ics to Antarctica: this connection enhances warming and sea-ice loss in one Southern Ocean64 region, but causes cooling and sea-ice growth in another. This shows that substantial65 AMOC weakening could impact the Southern Ocean on multidecadal timescales. How-66 ever, these Southern Ocean changes resulting from AMOC collapse are much smaller than67 the projected direct impacts of greenhouse-gas-induced warming.68 1 Introduction69 The crucial importance of the Southern Ocean (SO) for global climate has been70 highlighted in recent years. The SO accounts for most of the excess heat and carbon ab-71 sorbed by the global ocean under anthropogenic climate change, alleviating short-term72 atmospheric warming (Gruber et al., 2019; Sall´ee, 2018). Furthermore, SO sea ice ex-73 erts a strong control on the global ocean circulation (Ferrari et al., 2014; Carter et al.,74 2008), but processes governing its recent variations are not well understood (Eayrs et75 al., 2021; Meehl et al., 2019; Purich & Doddridge, 2023; Diamond et al., 2024; Jena et76 al., 2024).77 The Atlantic Meridional Overturning Circulation (AMOC) forms part of the global78 ocean circulation system. It is the Atlantic Ocean’s primary overturning current system,79 and transports warm water northwards in the upper ocean, and cold water southwards80 in the deeper ocean (Buckley & Marshall, 2016; Fox-Kemper et al., 2021). Within the81 21st century, AMOC weakening is ‘very likely’, due to Atlantic Ocean warming, along-82 side freshening from Northern Hemisphere meltwater (Fox-Kemper et al., 2021). There83 is ‘medium confidence’ that an abrupt collapse will not occur before 2100 (Fox-Kemper84 et al., 2021), but it is still a possibility (Sgubin et al., 2017; Liu et al., 2017). It is there-85 fore vital to understand the potential impacts of a substantial weakening or shutdown86 –2– manuscript submitted to JGR: Oceans of the AMOC (Orihuela-Pinto et al., 2022; Liu et al., 2020; Buckley & Marshall, 2016;87 Bellomo et al., 2021; Jackson et al., 2015).88 One of the largest impacts of the AMOC on climate is through its transport of heat89 from the Southern to Northern Hemisphere (Buckley & Marshall, 2016; Zhang et al., 2017).90 A weakened (strengthened) AMOC would weaken (strengthen) this northward heat trans-91 port (Chiang & Bitz, 2005; Krebs & Timmermann, 2007; Stocker et al., 2007; Laurian92 & Drijfhout, 2011; Zhang et al., 2017). This would cause Northern Hemisphere cooling93 (warming) (Krebs & Timmermann, 2007; Broccoli et al., 2006) and Southern Hemisphere94 warming (cooling) (Crowley, 1992; Stocker & Johnsen, 2003; Stocker et al., 2007; Lau-95 rian & Drijfhout, 2011). The ‘bipolar seesaw’ mechanism describes this balancing of heat96 between hemispheres, controlled by AMOC strength (Stocker & Johnsen, 2003; Stocker97 et al., 2007). Specifically, a reduced AMOC, particularly the weakening of South Atlantic98 currents that transport warm waters equatorward, would lead to heat accumulation in99 the South Atlantic and north of the Antarctic Circumpolar Current (ACC) (Stocker et100 al., 2007; Timmermann et al., 2007). This has been linked to a weakened North Brazil101 Current and altered heat transport pathways (Timmermann et al., 2007; Laurian & Dri-102 jfhout, 2011), resulting in ocean warming north of the ACC (Timmermann et al., 2007;103 Laurian & Drijfhout, 2011; Orihuela-Pinto et al., 2022). Over timescales of centuries, these104 ocean heat anomalies are gradually advected southward and mix across the ACC (Pedro105 et al., 2018).106 Palaeoclimate records support this mechanism: in Arctic ice cores, a ‘Dansgaard-107 Oeschger (DO) event’ during the last glacial period is marked by a rapid transition from108 a cool ‘stadial’ weak-AMOC state, to a warm ‘interstadial’ strong-AMOC state, and sub-109 sequent slow cooling back to a stadial (Gottschalk et al., 2015; Menviel et al., 2020; Thomas110 et al., 2009; Dansgaard et al., 1993; Malmierca-Vallet & Sime, 2023). Each DO event is111 associated with an Antarctic Isotope Maximum (AIM) event: AMOC weakening dur-112 ing DO events is associated with Antarctic warming, lagged by one to two centuries (Svensson113 et al., 2020; WAIS-Divide-Members, 2015; Markle et al., 2017). The ‘thermal bipolar see-114 saw’ mechanism explains the lag in terms of a slow heat build-up in a Southern Hemi-115 sphere oceanic heat reservoir that modulates and delays the Antarctic temperature re-116 sponse (Stocker & Johnsen, 2003; WAIS-Divide-Members, 2015; Buizert et al., 2018; Landais117 et al., 2015). These linked DO and AIM events indicate how, on centennial timescales,118 AMOC changes imprint themselves on the SO and Antarctica, as supported by model119 studies, e.g. Brown and Galbraith (2016); Liu et al. (2017); Pedro et al. (2018).120 However, we note the SO response to substantial AMOC weakening may differ un-121 der present-day (rather than glacial) conditions (Timmermann et al., 2010). Further-122 more, the timing and magnitude of the SO response remains uncertain, particularly at123 shorter-than-centennial timescales. Some evidence exists for dynamic multidecadal-scale124 atmospheric responses. These may include a poleward shift of the Southern Hemisphere125 westerlies (Chen et al., 2019; Liu et al., 2017; Markle et al., 2017) linked to a more pos-126 itive Southern Annular Mode (SAM) (Yuan & Li, 2008; Rind et al., 2018, 2001; Tim-127 mermann et al., 2010). Furthermore, tropical-Antarctic teleconnections may lead to fast128 pressure changes over the SO, including deepening the Amundsen Sea Low (ASL) (Orihuela-129 Pinto et al., 2022; Timmermann et al., 2010).130 Thus far, no near-present-day studies of AMOC weakening show multidecadal-scale131 Antarctic sea-ice change. The expected sea-ice behaviour is unclear, given the combined132 potential influences on the sea ice of ocean and/or atmospheric warming (Liu et al., 2020;133 Pedro et al., 2018; Orihuela-Pinto et al., 2022; Timmermann et al., 2010), and atmospheric134 changes including a more positive phase SAM (Zhang et al., 2017; Guarino et al., 2023;135 Rind et al., 2001), and deepened ASL (Timmermann et al., 2010; Orihuela-Pinto et al.,136 2022).137 –3– manuscript submitted to JGR: Oceans Table 1. List of global climate models used in this study, and their associated atmospheric, ocean and sea-ice components. Model Atm model Ocean model Ice model Reference HadGEM3-GC3-1LL UM GA7 NEMO3.6 CICE GSI8.1 Williams et al. (2018) HadGEM3-GC3-1MM UM GA7 NEMO3.6 CICE GSI8.1 Williams et al. (2018) CanESM5 CanAM5 NEMO3.4.1 LIM2 Swart et al. (2019) EC-Earth3 IFS 36r4 NEMO3.6 LIM3 D¨oscher et al. (2021) CESM2 CAM6 POP2 CICE5 Danabasoglu et al. (2020) IPSL-CM6A-LR LMDZ6 NEMO3.6 LIM3 Boucher et al. (2020) MPI-ESM-LR ECHAM6.3 MPIOM MPIOM Mauritsen et al. (2019) MPI-ESM-HR ECHAM6.3 MPIOM MPIOM M¨uller et al. (2018) Given the importance of SO processes (Sall´ee, 2018; Shi et al., 2021; Li et al., 2023)138 and the concern over AMOC weakening, we must understand potential impacts of sub-139 stantial AMOC weakening on the SO and Antarctic sea ice under sub-centennial timescales140 and near-present-day conditions. To address this, here, we investigate results for the SO141 from the newly released North Atlantic Hosing Model Intercomparison Project, ‘NAHos-142 MIP’ (Jackson et al., 2023). An idealised freshwater forcing was applied identically in143 eight CMIP6-era models over the North Atlantic, significantly weakening the AMOC (Fig.144 S1). We focus on the SO for the first 100 years, to better understand mechanisms of rapid145 change, and the balance between oceanic and atmospheric responses.146 The structure of this study is as follows. In Sec. 3.1, we present a multi-model in-147 tercomparison of AMOC weakening impacts on global temperatures, and SO temper-148 atures and sea ice. We show the SO response is dominated by warming and sea-ice melt,149 but includes a temperature dipole associated with Weddell Sea cooling and sea-ice growth.150 In Sec. 3.2, we focus on ocean and atmosphere heat transport changes, and ocean cir-151 culation changes, to determine the drivers of SO warming. In Sec. 3.3, we investigate152 atmospheric changes, and determine the drivers of Weddell Sea cooling. Finally, in Sec.153 3.4, we draw together Secs. 3.2 and 3.3 to explain the patterns of temperature and sea-154 ice change identified in Sec. 3.1.155 2 Methods156 2.1 Model Simulations157 Details of the eight latest-generation climate models used in this study are shown158 in Tables 1 and S1. The two simulations for each model used are the pre-industrial con-159 trol (‘piControl’) experiment and the ‘u03hos’ experiment. piControl is a core experi-160 ment run by all groups that contributed to the sixth Coupled Model Intercomparison161 Project CMIP6 (Eyring et al., 2016). The piControl simulation uses invariant solar, green-162 house gases, ozone, tropospheric aerosol, volcanic and land-use forcing from the year 1850163 for the whole simulation. The u03hos experiment is part of NAHosMIP (Jackson et al.,164 2023): the simulation is initialised from piControl with identical boundary conditions.165 In the u03hos experiments, an additional uniform, idealised, virtual negative salinity flux166 (‘hosing’) is applied to the North Atlantic and Arctic surface ocean, from 50◦N to the167 Bering Strait, with hosing strength of 0.3 Sv (1 Sv=106m3s−1), for minimum 100 years,168 see Jackson and Wood (2018b); Jackson et al. (2023) for details. This results in signif-169 icant AMOC weakening for all models (Jackson et al., 2023): there is an abrupt weak-170 –4– manuscript submitted to JGR: Oceans ening over the first few decades, that transitions to a more gradual decrease over the sub-171 sequent decades (Fig. S1)172 2.1.1 Model selection173 Of the eight CMIP6-era models that ran the u03hos simulation, three models (HadGEM3-174 GC3.1-MM, MPI-ESM1.2-LR and -HR) have especially frequent spurious deep convec-175 tion in the SO Weddell Sea, a model artifact common in CMIP5-era and some CMIP6-176 era models (Heuz´e, 2021; Mohrmann et al., 2021). This leads to very frequent forma-177 tion of large open-water polynyas in the simulated Weddell Sea sea ice, impacting SO178 conditions, particularly in the Weddell Sea (Heuz´e, 2021; Mohrmann et al., 2021). The179 focus of our work is on the SO under near-present-day conditions, and including all the180 models in the multi-model ensemble mean would induce spurious signals in the regions181 of deep convection of some of the models. Therefore, we calculate multi-model ensem-182 ble results (shown in Sec. 3.1) using the five models without this artifact (CanESM5,183 CESM2, EC-Earth3, HadGEM3-GC3.1-LL, and IPSL-CM6A-LR). However, we found184 that the global-scale response to hosing is similar whether considering the five non-convecting,185 or all eight, models. Therefore, for interest, we show results including the additional three186 convecting models from Sec 3.2.1 onwards.187 2.2 Analysis188 All model output variables were aggregated to a regular 1◦latitude-longitude grid189 using the Climate Data Operators library (Schulzweida, 2023) for inter-model compar-190 ison. We determine the impact of freshwater hosing in the u03hos simulations by com-191 paring them to the piControl simulations as follows. For all maps, unless otherwise in-192 dicated, we present the time-mean of the u03hos simulation over years 50-100, with the193 time-mean of the piControl simulation over the same period subtracted. As the simu-194 lations are all between 100 and 200 years long, year 50-100 is the final 50-year period that195 can be compared across models. By year 50-100, the AMOC is substantially weakened196 for all models (Fig. S1). For timeseries, unless otherwise indicated, we show the u03hos197 simulation result with the piControl simulation result subtracted, year by year, as in Andrews198 et al. (2019), to account for unforced model drift in the simulations without assuming199 the drift is linear. We hereafter refer to the difference between the u03hos and piCon-200 trol simulations as the ‘u03hos–piControl anomaly’. For a given model, the u03hos and201 piControl simulation outputs are compared using a t-test. Unhatched regions on maps202 show results statistically significant at the p<0.05 level.203 In Sec. 3.1 we identify in the u03hos simulations, relative to the piControl simu-204 lations, SO dipoles in SLP, SAT, and SIC. We define the ‘dipole magnitude’ as the dif-205 ference between the high and low centres in the dipoles as follows. For all variables, we206 begin with the multi-model ensemble mean of the u03hos–piControl anomaly, for all lat-207 itudes and longitudes over the SO region (south of 30◦S), using annual means for years208 0–100. For SLP and SAT, to identify the locations of the high and low centres, we first209 identify the two grid-cells with the maximum and minimum values over the full 100 years.210 We then take the means over the 3×3 matrix of 1◦grid-cells around each of these two211 grid-cell centres, and then find the difference between the two means to yield the ‘dipole212 magnitude’ for each variable. For SIC, we could use the same method as is used for SLP213 and SAT, but the following method yields a more physically meaningful quantity: for214 SIC, the ‘dipole magnitude’ is simply the difference in SIA between the Weddell Sea (60◦W–215 0◦E) and west Indian (0◦E–60◦E) sectors (both methods yield similar results after nor-216 malisation).217 –5– manuscript submitted to JGR: Oceans 2.2.1 OHT and AHT218 Meridional heat transport (MHT) across a given latitude band is the sum of net219 ocean heat transport (OHT) and atmospheric heat transport (AHT) across the band.220 We use the CMIP6 diagnostic ‘hfbasin’ for OHT, and calculate AHT from the divergence221 of the vertical fluxes integrated over the atmosphere (equal to the sum of the AHT and222 atmospheric heat storage, but annual and longer-term-mean atmospheric heat storage223 is negligible) (Liu et al., 2018; Donohoe et al., 2020).224 2.2.2 Ocean circulation and heat transport225 In Sec. 3.2.2, we consider meridional ocean heat transport and circulation. The net226 (or ‘residual’) meridional overturning circulation is composed of the Deacon (Eulerian-227 mean) cell, compensated by the eddy-induced circulation (Danabasoglu et al., 1994; Mar-228 shall & Speer, 2012). Their respective streamfunctions are given by ψres,ψ,ψ* such that229 ψres =ψ+ψ∗(1) The meridional OHT is the sum of contributions from this Eulerian-mean flow OHTeul,230 eddies OHT∗ ed and diffusion OHTdiff (Yang et al., 2015; Li et al., 2022):231 OHT =OHTeul +OHT∗ ed +OHTdiff (2) For parametrisations of mesoscale advection and diffusion, all models use the Gent-232 McWilliams scheme, with formulations from Gent and McWilliams (1990) and Redi (1982)233 or the formulation from Griffies (1998). For CESM2 and HadGEM3-GC3.1-LL, relevant234 for Sec. 3.2.2, Table S2 shows parametrisations of mesoscale advection and diffusion, sub-235 mesoscale eddy processes (Fox-Kemper et al., 2011), and background vertical diffusion236 (Large et al., 1994; Madec et al., 2013, 2017). Table 3, Jackson et al. (2023), details these237 parametrisations for all models. For all models, the parametrised eddy-induced advec-238 tive transport ψ∗follows (Gent & McWilliams, 1990):239 ψ∗=K∗S(3) where K is the eddy-induced transfer coefficient and S is the slope of the isopycnals.240 2.2.3 SAM and ASL index241 The dimensionless SAM index is (Gong & Wang, 1999):242 SAM =P∗ 40 −P∗ 65 (4) where P∗ lat is the normalised zonally averaged sea level pressure (SLP) at this latitude,243 computed from the zonally averaged SLP timeseries Plat:244 P∗ lat(t) = Plat(t)−µPlat σPlat (5) where µPlat is the time-mean of the timeseries. To compute σPlat , the standard devia-245 tion of the timeseries, we apply a high-pass second-order forward-backward Butterworth246 filter with cut-off period of 50 years to remove any long-term trend, and calculate the247 standard deviation from this filtered timeseries (this step changes the standard devia-248 tion value by <5% in most and <15% in all cases).249 The ASL index is the minimum Pacific sector SLP (60◦–75◦S and 180◦–310◦E; Turner250 et al., 2013). For easier inter-model comparison, the ASL timeseries is normalised (as251 described above for SLP: using the ASL timeseries long-term mean and standard devi-252 ation).253 –6– manuscript submitted to JGR: Oceans Figure 1. Multi-model mean of u03hos–piControl anomaly, taken over years 50-100 of run. Unhatched regions: agreement for at least 4 of the 5 models on sign change. (a) global SAT (b) global SLP (with zonal mean subtracted at each latitude, to show meridional variation) (c) SO SLP, with zonal mean subtracted at each latitude (contours), overlaid with surface wind vectors (d) SO SAT (e) SO sea ice concentration. To estimate the sign of the response of the SAM and ASL indices in the u03hos254 simulations, we fit a linear trend, using a least-squares linear regression model (Virtanen255 et al., 2020). This returns the slope of the regression line, the standard error of this es-256 timated slope, and the p-value that no trend exists (calculated using a hypothesis test257 with null hypothesis that the slope is zero, using a Wald Test with t-distribution of the258 test statistic).259 –7– manuscript submitted to JGR: Oceans Figure 2. u03hos–piControl anomaly, for each model, of (a-e) global SAT, mean over final 50 years (see Table S1 for simulation lengths); and SO, mean over years 50-100 for (f-j) SAT and (k-o) SIC. Blue: piControl and red: u03hos sea-ice edge (threshold where SIC>0.15) –8– manuscript submitted to JGR: Oceans 3 Results260 3.1 Southern Ocean temperature and sea-ice response261 For all models, the AMOC weakens significantly over the first 50 years of the u03hos262 simulation, to a new, weaker, state (Fig. S1 for timeseries of AMOC strength; see also263 Jackson et al. (2023)). We consider first the response in the multi-model mean. Fig. 1(a)264 shows strong Northern Hemisphere cooling and weaker Southern Hemisphere warming:265 the well-established ‘bipolar seesaw’ pattern (e.g. Stocker and Johnsen (2003); Stocker266 et al. (2007); Timmermann et al. (2007)). From Fig. 1(b,c), we note in the Southern Hemi-267 sphere a ‘Rossby-wave-train’-like pattern, originating from central America and prop-268 agating southand eastwards to the Southern Ocean (SO), investigated in Sec. 3.3.2. This269 pattern includes low/high sea level pressure (SLP) centres over the Atlantic/Indian SO270 sectors. Considering the SO response: overall, SO SAT increases (Fig. 1(d)), leading to271 reduced SLP (not shown) and reduced sea-ice concentration (Fig. 1(e)). The previously-272 identified low/high SLP centres (Fig. 1(b,c)) are associated with cyclonic/anticyclonic273 winds and regional temperature and sea-ice change (Figs 1(d,e)) further detailed in Sec.274 3.3). Over years 50-100, in the multi-model ensemble mean (multi-model range), the over-275 all SO SLP is reduced by 31 (6 – 72)Pa, SAT is increased by 0.29 (0.13 – 0.69)◦C, and276 SIA is reduced by 0.40 (-0.15 – 1.32)Mkm2. We define the ‘dipole magnitude’ as the dif-277 ference between the high and low centres of SLP, SAT, and SIA visible in Figs 1(c-e) (de-278 tailed in Sec. 2.2). The SLP dipole magnitude is 180 (80 – 320)Pa, the SAT dipole mag-279 nitude is 0.56 (0.46 – 1.35)◦C, and the SIA dipole magnitude is 0.10 (0.05 – 0.18)Mkm2.280 Timeseries of all these quantities over years 0–100 are shown in Fig. S2. In the Ross and281 Amundsen Sea regions, SLP decreases by up to 30Pa (Fig. 1(b,c)), and Amundsen Sea282 SIC decreases by 2-8%.283 We now consider the Southern Hemisphere response by individual model (see also284 Figs S3–S6). In the Southern Hemisphere, northwards of the SO (0◦to 30◦S), all mod-285 els eventually warm over all sectors, by 0.1 to 1.5◦C (Figs 2(a-e)). The magnitude of the286 warming differs between models, with strongest warming for CESM2, and weakest for287 EC-Earth3.288 We move on to the response in the SO region (30◦to 90◦S), shown in Figs 2(f-j).289 The Antarctic Circumpolar Current (ACC) warms for all models, over all sectors, par-290 ticularly in the Indian sector at 200m–1km depth (see also Figs S7–S9). Southwards of291 the ACC, there is regional warming for all models, particularly in the Indian sector. We292 note IPSL-CM6A-LR has some Pacific sector cooling, similar to some CMIP3 and CMIP5293 models under freshwater hosing (Timmermann et al., 2010; Buiron et al., 2012).294 Fig. 2(k-o) shows the u03hos–piControl sea-ice concentration (SIC) anomaly; this295 broadly matches the SAT anomaly where sea-ice exists. All models have statistically sig-296 nificant ice loss, with maximum regional SIC reductions ranging from 5 to 25% across297 models. Regions of greater sea-ice loss (∼10 to 20%) have especially high SATs; see e.g.298 CESM2 and HadGEM3-GC3.1-LL, with SAT increases of 1.5◦–3◦C in the regions of great-299 est sea-ice loss, likely linked to ice-albedo feedbacks (Curry et al., 1995). Despite ice loss300 in other regions, Weddell Sea SIC (and ice thickness, not shown) either increases, or does301 not change, for 4/5 models.302 To summarise, all models have Southern Hemisphere warming (and Northern Hemi-303 sphere cooling). By years 50-100, warming extends to the SO Indian sector for all mod-304 els, especially along the ACC at ∼200-500m depth. For most models, sea ice is reduced305 particularly in the Indian sector, but in the Weddell Sea there is regional cooling and306 sea-ice growth. Over years 50-100, in the multi-model ensemble mean, the SO SLP de-307 creases by 31 Pa, SAT increases by 0.29◦C, and SIA decreases by 0.40Mkm2. The dif-308 ference between the Indian sector and Weddell Sea response in SLP is 180 Pa, in SAT309 is 0.56◦C, and in SIA is 0.1Mkm2.310 –9– manuscript submitted to JGR: Oceans Figure 7. u03hos–piControl anomaly SO sea level pressure mean over years 50-100 (contours). Mean wind vector anomalies over years 50-100 are overplotted as vectors; black arrows show changes significant at the p=0.05 level. Note different wind vector scale for CESM2. –16– manuscript submitted to JGR: Oceans Figure 8. (a-f) u03hos–piControl anomaly of global mean geopotential height at 200hPa (zonal mean subtracted) over years 50-100, for the six of eight models where data accessible. Unhatched regions: statistically significant at p=0.05 level. (g) Multi-model mean. Unhatched regions show where all models agree on change of sign. –17– manuscript submitted to JGR: Oceans Figure 9. (a-f) HadGEM3-GC3.1-LL u03hos–piControl anomaly of mean over years 50-100. Unhatched regions: statistically significant at p=0.05 level. (a) SAT [◦C] (b) sea-ice thickness [m] (c) net heat flux at ocean surface; positive downwards [W/m2](d) longwave downwelling radiation at surface; positive downwards [W/m2](e) Ice volume tendency (due to thermodynamics) [cm/day] (f) Ice volume tendency (due to dynamics) [cm/day]. For (e-f): overplotted arrows indicate surface wind vector anomalies. Note also, to guide the eye, only ice volume tendency anomalies that match the sign of the sea-ice thickness anomaly, and thus could explain it, are shown. (g) attribution of sea-ice thermodynamic thickness change in (e) to heat fluxes from the ocean and/or atmosphere (shown in (c) and (d) respectively), based on the sign and magnitude of the change. For example, if a region of sea-ice melt has a more positive atmospheric downwelling longwave anomaly but the upwards ocean heat flux anomaly is negative (or negligible, defined as less than 1/10 magnitude of the longwave anomaly) the sea-ice melt is attributed to the atmospheric flux change only. –18– manuscript submitted to JGR: Oceans 3.4 Causes of Southern Ocean temperature and sea-ice change462 Here, we further explain the previously identified patterns of SO temperature and463 sea-ice change, in terms of the SO warming discussed in Sec. 3.2, and the SO temper-464 ature dipole discussed in Sec. 3.3. To do this, we present results for HadGEM3-GC3.1-465 LL, which represents the mean ensemble behaviour well. Figs 9(a-f) respectively show466 the u03hos–piControl anomaly of SAT, SIC, net downwards heat flux into the ocean sur-467 face (calculated at the air-sea interface, or ice-sea interface in ice-covered regions, using468 CMIP6 variable ‘hfds’, from Griffies et al. (2016)), downwelling longwave flux from the469 atmosphere at the surface (the ocean, ice, or land surface, as applicable), and the sea-470 ice volume tendencies due to thermodynamics and dynamics. In addition to the surface471 downwelling longwave flux, we also considered all other atmospheric surface heat flux472 terms (including shortwave and latent and sensible heat fluxes), but found that only the473 longwave term had a statistically significant u03hos–piControl anomaly relevant for the474 following analysis.475 We begin with the SO warming. In Sec. 3.2, we found that AMOC weakening led476 to heat accumulation in the SH upper ocean, north of the ACC. This ocean heat anomaly477 is predominantly transported southwards through the ACC via ocean eddies and diffu-478 sion, warming the upper ocean. The ocean surface releases this excess heat to the at-479 mosphere (Fig. 9(c)), which warms (Fig. S9). The warmed atmosphere responds by re-480 leasing heat, both upwards (not shown) and back downwards (Fig. 9(d)). Overall, air481 temperatures increase in most regions (Fig. 9(a)). South of the ice edge: sea-ice melts482 in most regions (Fig. 9(b)). This is due partly to direct ocean warming (Fig. 9(c)), and483 partly to the ocean warming the atmosphere north of the ice edge, so the ice surface is484 warmed by downwelling longwave radiation (Fig. 9(d)).485 We now explain the SO temperature and sea-ice dipole in terms of the atmospheric486 response to AMOC weakening. From Sec. 3.3, we found that a Rossby-wave-like pat-487 tern leads to a SO pressure dipole, associated with cyclonic/anticyclonic winds (arrows488 in Figs 9(e,f)). Due to a combination of dynamic and thermodynamic processes (Fig.489 9(e,f)), the cyclonic anomaly leads to sea-ice thickness increase in the northern Weddell490 Sea near the ice edge, as in Holland and Kwok (2012); Vernet et al. (2019). Near the ice491 edge, in the sector 0-30°W, ice thickness increases due to dynamics (Fig. 9(f)). Slightly492 polewards, in the ice interior, the south-westwards wind anomaly drives convergent ice493 drift (ridging); near the sea-ice edge, the north-westwards wind anomaly drives diver-494 gent ice drift, advecting the sea ice northwards. In the sector 30-60°W, thermodynamic495 processes lead to ice growth (Fig. 9(e)): at the sea-ice edge, the north-westwards wind496 anomaly advects relatively cool, dry air, cooling the ocean surface and resulting in ad-497 ditional thermodynamic ice growth; additionally we note thermodynamic ice growth in498 the ice interior. Conversely, south of Africa, warm, moist air is advected southwards, lead-499 ing to warming, sea-ice loss, and inhibiting ice growth at the ice edge. Conversely, south500 of Africa, over the King Hakon low (Holland & Kwok, 2012), warm, moist air is advected501 southwards, leading to warming, sea-ice loss, and inhibiting ice growth at the ice edge.502 Furthermore, the warming/cooling pattern in the Amundsen-Bellinghausen/Ross seas503 and associated sea-ice loss/increase (Fig. 9(a,b)) is likely explained by the deepened ASL504 (Fig. S15) and associated wind anomalies (Fig. 9(c,d)).505 Fig. 9(g) shows an attribution of sea-ice thermodynamic change in Fig. 9(e) to heat506 fluxes from the ocean or atmosphere (Figs 9(c) and (d) respectively), based on the sign507 and magnitude of the change. In a few areas, ice melt can be attributed primarily to only508 one of the atmosphere or ocean (in particular, some regions of melt are attributed to at-509 mosphere only, in light orange). However, in most regions, melt or growth is attributed510 to a combination of both atmosphere and ocean change (red for melt and blue for growth)511 – this is especially clear in the Pacific and west Indian Sectors, and in the Weddell Sea.512 –19– manuscript submitted to JGR: Oceans This analysis demonstrates that the SAT and sea-ice response is determined by both513 ocean and atmosphere. Specifically, the response is governed both by increased south-514 wards OHT across the ACC and associated ocean and atmospheric warming, and by the515 strongly zonally asymmetric pressure dipole. Furthermore, the slightly deepened ASL516 likely contributes to sea-ice reduction in the Amundsen sector.517 4 Discussion518 Here, we investigated the multidecadal-timescale SO response to AMOC weaken-519 ing, using a CMIP6-era model ensemble. In particular, we focused on the SO temper-520 ature and sea-ice responses, which have not previously been investigated in detail. We521 explained the overall SO response in terms of two ‘groups’ of processes. We find an ap-522 proximately linear inter-model relationship between the magnitude of the AMOC weak-523 ening, and a key process linked to each of these two groups (Fig. 4): models with greater524 AMOC weakening tend to have both greater southwards OHT transport increase into525 the SO, and a greater SO temperature dipole magnitude.526 The first group involves coupled ocean/atmospheric processes (and has some zonal527 asymmetry). AMOC weakening reduces northwards ocean heat transport across the equa-528 tor. Heat accumulates in the SH upper ocean, north of the ACC. This leads to the well-529 established ‘bipolar seesaw’ pattern of NH cooling and SH warming (Crowley, 1992; Stocker530 & Johnsen, 2003; Stocker et al., 2007). Meanwhile, atmospheric adjustments shift the531 SH westerlies southwards (also shown by e.g. Liu et al. (2017); Markle et al. (2017)). This532 southwards shift of the westerlies shifts the SO Deacon cell southwards. The Deacon cell533 change is partially compensated by a strengthened SO eddy circulation, in agreement534 with e.g. Chen et al. (2019). Some of the accummulated upper ocean heat is transported535 southwards across the ACC, predominantly by parameterised (eddy and diffusive) heat536 transports, in agreement with similar experiments under glacial conditions (Pedro et al.,537 2018). This anomalous heat transport into the SO region leads to ocean and atmospheric538 warming from the ACC polewards, eventually causing sea-ice melt.539 The second group of processes is atmosphere-dominated (and has much stronger540 zonal asymmetry). A Rossby-wave-like pattern originates from Central America and prop-541 agates eastwards into the SO. Due to its origin over Central America, this teleconnec-542 tion is novel: previous studies have identified in response to AMOC weakening a ‘La Nina’-543 like teleconnection originating from a different location, the central Pacific Ocean (Timmermann544 et al., 2010; Orihuela-Pinto et al., 2022).545 A full investigation of the tropical processes generating the teleconnection are out-546 side this study’s scope, but we suggest a possible source here. In the u03hos relative to547 the piControl simulations, we identify a strong positive meridional gradient in the at-548 mospheric relative vorticity at 200hPa (not shown). This gradient is strongest at ∼1◦S,549 80◦W, on the east coast of Central America, and indicates favourable conditions for Rossby550 wave generation (Lachlan-Cope & Connolley, 2006; Jin & Kirtman, 2009). The gradi-551 ent most likely arises from a combination of zonally asymmetric changes including (a)552 the net southwards shift of the ITCZ (Chiang & Bitz, 2005; Broccoli et al., 2006), lead-553 ing to warmer, wetter conditions particularly over the southern tropical ocean basins,554 and (b) a strengthened Walker Circulation and enhanced subsidence over the eastern Pa-555 cific and coast of Central America, which has been identified in a similar simulation (Orihuela-556 Pinto et al., 2022).557 Previous studies of DO events have suggested distinct timescales for the evolution558 of a ‘slower’ bipolar mode, and a ‘fast’ atmospheric teleconnection (Stocker & Johnsen,559 2003; Markle et al., 2017). We cannot distinguish a clear time-separation between the560 evolution of the pressure dipole, and the circumpolar upper-ocean warming (Fig. S16;561 see warming from approximately 30–60°S); both begin within decades of AMOC weak-562 –20– manuscript submitted to JGR: Oceans ening, and progressively strengthen throughout the simulations. Much longer simulations563 might be required to identify two distinct timescales of evolution. We suggest that on564 multi-centennial timescales, the tropical changes that generate the teleconnection would565 settle more quickly than the longer-timescale heat accumulation in the SO from the bipo-566 lar seesaw process, which could take multiple centuries to mature (Stocker & Johnsen,567 2003; Landais et al., 2015; WAIS-Divide-Members, 2015; Pedro et al., 2018).568 The magnitudes of SO temperature and sea-ice change in this study seem small rel-569 ative to some past events (WAIS-Divide-Members, 2015; Markle et al., 2017; Diamond570 et al., 2024; Holloway et al., 2018; Chadwick et al., 2023). SO mean SAT and SIA changes571 for most models are of a similar order to interannual variability (e.g. Figs S5-6). This572 is in some part explained by regional asymmetry (e.g. Fig. 2). Nevertheless, we show573 that these impacts are statistically significant and the processes that yield them are ro-574 bust across models (e.g. Figs 1 and 4). Furthermore, the centennial-scale DO-AIM ‘de-575 lay’ observed in ice cores seems consistent with our study: the relatively small changes576 we detect here on timescales of up to a century likely could not be resolved using ice core577 data (Pedro et al., 2018; WAIS-Divide-Members, 2015; Landais et al., 2015; Svensson578 et al., 2020).579 Through this study, we used existing model output from the NAHosMIP CMIP6580 project, detailed and analysed in Jackson et al. (2023), wherein a large (0.3 Sv) fresh-581 water hosing was applied to the North Atlantic. Applying similar large freshwater hos-582 ing is a standard method of weakening the AMOC in GCMs, to investigate hysteresis583 and global AMOC weakening impacts, see e.g. Krebs and Timmermann (2007); Stouf-584 fer et al. (2006); Vellinga et al. (2002); Timmermann et al. (2007); Liu and Liu (2014);585 Jackson and Wood (2018a); Jackson et al. (2015); Kageyama et al. (2013). The NAHos-586 MIP experiments allow us to investigate multidecadal scale impacts of substantial AMOC587 weakening on the SO, comparing systematically across 8 CMIP6 models. We note that588 the forcing applied is much larger than North Atlantic freshwater forcing that could be589 expected under climate change (Van den Berk & Drijfhout, 2014; Jackson et al., 2023;590 Lenaerts et al., 2015). Despite this large North Atlantic forcing and substantial AMOC591 weakening, the SO responses we report are smaller than what might be expected over592 the next century due to future anthropogenic warming. For example, under moderate593 climate change scenario SSP1-2.6, northwards OHT is expected to increase by ∼200TW594 at 40°S and ∼300TW at 50°S (Mecking & Drijfhout, 2023). By comparison in Fig. 3(a),595 considering the multimodel range, southwards OHT increases by 50-150TW at 40°S and596 0-50TW at 30°S. Furthermore, mean SO temperature under SSP1-2.6 is expected to in-597 crease by 0.5-1.5C by the 2050s, relative to the present-day (Fox-Kemper et al., 2021);598 this is also larger than the SO 0.3°C increase in the ensemble mean that we report here.599 We note that the larger regional differences due to the atmospheric teleconnection we600 identify (e.g. multimodel mean SAT difference of 0.6°C, and greater for some models,601 see Fig. 2) are more comparable to the magnitude of greenhouse-gas-related changes.602 Overall, this tells us that the multidecadal-scale SO sensitivity to substantial AMOC weak-603 ening is much smaller than the multidecadal-scale SO sensitivity to more direct green-604 house gas-induced warming.605 Although the two ‘groups’ of processes are similar across models, there are signif-606 icant inter-model differences in the magnitude of the SO temperature and sea-ice response607 (e.g SO SAT warming is 0.2◦C in EC-Earth3, versus over 1◦C for CESM2). This is likely608 explained by inter-model differences in ocean representation and AMOC weakening. Whilst609 each of the models has a similar timescale and pattern of AMOC weakening, there are610 significant inter-model differences in AMOC initial and final states (Fig. S1 and Jackson611 et al. (2023)), so the models have different magnitudes of southwards ocean heat trans-612 port increase. This is supported by the relationship shown in Fig. 4: models with greater613 AMOC reduction tend to have a greater increase in southwards OHT into the SO. There614 are also inter-model differences in ACC strength and position (Mohrmann et al., 2021;615 –21– manuscript submitted to JGR: Oceans Beadling et al., 2020), and the representation of mesoscale ocean eddies (Jackson et al.,616 2023), which largely control additional heat transport across the ACC (Beadling et al.,617 2020; Shi et al., 2021; Dufour et al., 2015). All models parameterise mesoscale eddies based618 on the Gent-McWilliams scheme. Here, the eddy-induced (advective) transport is a prod-619 uct of the eddy-induced transfer coefficient K and the slope of isopycnals S (Sec. 2.2.2).620 In these models, differences in magnitude of the additional eddy-induced heat transport621 across the ACC (identified in Sec. 3.2.2) may originate from inter-model differences in622 K, as well as from changes in S, and from the temperature contrast between the flows623 carried by the eddies southward and northward. It is outside this study’s scope to iden-624 tify which factors contributed the most, but we list them for a possible future study. Fur-625 thermore, models which instead have eddy-permitting ocean components tend to more626 effectively compensate a strengthened Euler circulation (Farneti et al., 2015). Therefore,627 eddy-permitting models would likely yield stronger eddy compensation under AMOC628 weakening, and even greater southwards OHT across the ACC.629 In summary, future work could include: continuing these simulations for several cen-630 turies, or running longer lower-resolution model simulations. This would help identify631 whether the two ‘groups’ of processes have distinct timescales of evolution. Running en-632 sembles of simulations with each model would also help isolate these signals. The timescales633 and magnitudes of the SO response could also be constrained using observations of high634 temporal resolution, for past periods following significant AMOC weakening, e.g. DO635 or Heinrich events (Marcott et al., 2011; Broecker et al., 1992). The atmospheric adjust-636 ment processes that lead to the novel atmospheric teleconnection could also be further637 investigated. We showed eddy and diffusive ocean processes transport accummulated heat638 southwards across the ACC to high SO latitudes, but the model-dependence of these pro-639 cesses could be investigated in terms of model resolution and parametrisation – although640 the timescale of heat propagation likely also depends on other factors, e.g., initial and641 final states of the AMOC, and ACC strength (Jackson et al., 2023; Pedro et al., 2018;642 Markle et al., 2017).643 5 Conclusions644 Over the next century, significant AMOC weakening is likely (Fox-Kemper et al.,645 2021). It is vital to understand multidecadal-timescale impacts of AMOC weakening on646 the Southern Ocean (SO). However, there is no current consensus on these impacts (WAIS-647 Divide-Members, 2015; Markle et al., 2017; Orihuela-Pinto et al., 2022; Timmermann648 et al., 2010; Liu et al., 2017; Chen et al., 2019) and no studies have investigated the SO649 sea-ice response in detail.650 We addressed this issue using the NAHosMIP CMIP6 model ensemble (Jackson651 et al., 2023). For the first time, we systematically assess, across a CMIP6 multi-model652 ensemble, oceanic and atmospheric impacts of substantial AMOC weakening on SO tem-653 peratures and sea ice. We explain the SO response in terms of two ‘groups’ of processes,654 which respectively cause more zonally uniform warming, and a temperature dipole with655 regional warming/cooling in the Indian Sector/Weddell Sea. The multidecadal temper-656 ature and sea-ice evolution is determined by a combination of these two groups of pro-657 cesses. Overall, the models share a similar multidecadal timescale for SO warming to be-658 gin. Over years 50–100, in the multi-model ensemble mean (multi-model range in brack-659 ets), the strongest ocean warming is in the Indian sector, at 200m to 500m depth along660 the ACC. The Southern Hemisphere westerlies weaken on the equatorward flank by 0.1661 to 1m/s, particularly in the Atlantic sector. Overall SO SLP decreases by 30 (10–70)Pa,662 SAT increases by 0.3 (0.1–0.7)◦C, and SIA decreases by 0.4 (-0.2–1.3)Mkm2. The dif-663 ference between the Indian sector and Weddell Sea responses in SLP is 180 (80–320) Pa,664 in SAT is 0.6 (0.5–1.4)◦C, and in SIA is 0.1 (0.1–0.2)Mkm2.665 –22– manuscript submitted to JGR: Oceans These impacts seem relatively modest: comparable for many models to the mag-666 nitude of interannual variability, and small relative to the changes expected under fu-667 ture anthropogenic warming. Nevertheless, these impacts, and the processes that yield668 them, are statistically significant and robust across models. Paleoclimate records and669 model studies suggest that on timescales of centuries to millenia, the weakened AMOC670 would lead to continued SO warming (of several degrees) and Antarctic sea-ice loss (on671 the order of a million km2) (WAIS-Divide-Members, 2015; Svensson et al., 2020; Liu &672 Fedorov, 2019; Holloway et al., 2018). A substantial future AMOC weakening would have673 global consequences (Chiang & Bitz, 2005; Krebs & Timmermann, 2007; Broccoli et al.,674 2006), and we show through this study that these consequences extend to the SO and675 Antarctic sea ice on multidecadal timescales.676 Open Research Section677 The piControl CMIP6 DECK model outputs are in the ESGF archive: https://678 aims2.llnl.gov/search/. Information on obtaining this data available from https://679 pcmdi.llnl.gov/CMIP6/Guide/dataUsers.html. For the piControl experiments, data680 is available at Danabasoglu et al. (2019); (EC-Earth) (2019); Boucher et al. (2018); Ri-681 dley et al. (2018, 2019); Wieners et al. (2019); Jungclaus et al. (2019); all last accessed682 July 2024. Processed model outputs used in this study are available at Diamond (2025).683 Acknowledgments684 RD acknowledges support from NERC training grant NE/S007164/1. RD thanks Rahul685 Sivankutty for useful conversations and guidance. Thanks to Warren Lee for help with686 accessing and processing data. LCS has received support from the NERC National Ca-687 pability International grant SURFEIT: NE/X009319/1, and acknowledges additional sup-688 port from ANTSIE: EU-H2020 G.N.864637. This work was supported by MBIE NZ: Antarc-689 tic Sea-Ice Switch (ASIS) – Preparing for New Threats, and by TiPES: EU-H2020 G.N.820970.690 D Schroeder acknowledges support from the NERC-UKESM program. This work was691 supported by NERC through National Capability funding, undertaken by a partnership692 between the Centre for Polar Observation Modelling and the British Antarctic Survey.693 This work used the JASMIN data analysis platform http://jasmin.ac.uk/. LCJ was694 supported by the Met Office Hadley Centre Climate Programme funded by DSIT. D Swinge-695 douw acknowledges the support from the TipESM project funded by the European Union’s696 Horizon Europe research and innovation programme under grant agreement No 101137673.697 AH is supported by the Regional and Global Model Analysis (RGMA) component of the698 Earth and Environmental System Modeling Program of the U.S. Department of Energy’s699 Office of Biological & Environmental Research (BER) via National Science Foundation700 (NSF) IA 1947282 (DE-SC0022070). This work is supported by the US National Science701 Foundation (NSF) National Center for Atmospheric Research (NCAR), a major facil-702 ity sponsored by the US NSF under Cooperative Agreement 1852977. This research has703 been supported by the Spanish Ministry of Science and Innovation (project MARINE,704 grant no. PID2020-117768RB-I00). KB has received funding from the European Union’s705 Horizon 2020 research and innovation programme under the Marie Sk lodowska-Curie Ac-706 tions (grant no. 101026907 (CliMOC)). The EC-Earth3 simulations shown in this work707 were carried out at ECMWF under the special projects SPITBELL and SPITMEC2. Thanks708 to the editor Martin Vancoppenolle and to two anonymous reviewers for their detailed709 and constructive comments.710 The authors declare no conflicts of interest.711 References712 Andrews, T., Andrews, M. B., Bodas-Salcedo, A., Jones, G. S., Kuhlbrodt, T., Man-713 –23– manuscript submitted to JGR: Oceans ners, J., . . . others (2019). Forcings, feedbacks, and climate sensitivity in714 HadGEM3-GC3. 1 and UKESM1. Journal of Advances in Modeling Earth715 Systems,11(12), 4377–4394.716 Beadling, R. L., Russell, J., Stouffer, R., Mazloff, M., Talley, L., Goodman, P., . . .717 Pandde, A. (2020). Representation of Southern Ocean properties across cou-718 pled model intercomparison project generations: CMIP3 to CMIP6. Journal of719 Climate,33(15), 6555–6581.720 Bellomo, K., Angeloni, M., Corti, S., & von Hardenberg, J. (2021). Future climate721 change shaped by inter-model differences in Atlantic meridional overturning722 circulation response. Nature Communications,12(1), 1–10.723 Bjerknes, J. (1964). Atlantic air-sea interaction. In Advances in Geophysics (Vol. 10,724 pp. 1–82). Elsevier.725 Boucher, O., Denvil, S., Levavasseur, G., Cozic, A., Caubel, A., Foujols, M.-726 A., . . . Cheruy, F. (2018). IPSL IPSL-CM6A-LR model output pre-727 pared for CMIP6 CMIP piControl [dataset]. Earth System Grid Federa-728 tion. Retrieved from https://doi.org/10.22033/ESGF/CMIP6.5251 doi:729 10.22033/ESGF/CMIP6.5251730 Boucher, O., Servonnat, J., Albright, A. L., Aumont, O., Balkanski, Y., Bastrikov,731 V., . . . others (2020). Presentation and evaluation of the IPSL-CM6A-732 LR climate model. Journal of Advances in Modeling Earth Systems,12(7),733 e2019MS002010.734 Bouttes, N., Gregory, J. M., Kuhlbrodt, T., & Suzuki, T. (2012). The effect of wind-735 stress change on future sea level change in the Southern Ocean. Geophysical re-736 search letters,39(23).737 Broccoli, A. J., Dahl, K. A., & Stouffer, R. J. (2006). Response of the ITCZ to738 Northern Hemisphere cooling. Geophysical Research Letters,33(1).739 Broecker, W., Bond, G., Klas, M., Clark, E., & McManus, J. (1992). Origin of the740 northern Atlantic’s Heinrich events. Climate Dynamics,6, 265–273.741 Brown, N., & Galbraith, E. D. (2016). Hosed vs. unhosed: interruptions of the At-742 lantic Meridional Overturning Circulation in a global coupled model, with and743 without freshwater forcing. Climate of the Past,12(8), 1663–1679.744 Buckley, M. W., & Marshall, J. (2016). Observations, inferences, and mechanisms745 of the Atlantic Meridional Overturning Circulation: A review. Reviews of Geo-746 physics,54(1), 5–63.747 Buiron, D., Stenni, B., Chappellaz, J., Landais, A., Baumgartner, M., Bonazza, M.,748 . . . others (2012). Regional imprints of millennial variability during the MIS 3749 period around Antarctica. Quaternary Science Reviews,48, 99–112.750 Buizert, C., Sigl, M., Severi, M., Markle, B. R., Wettstein, J. J., McConnell, J. R.,751 . . . others (2018). Abrupt ice-age shifts in southern westerly winds and752 Antarctic climate forced from the north. Nature,563(7733), 681–685.753 Carter, L., McCave, I., & Williams, M. J. (2008). Circulation and water masses of754 the Southern Ocean: a review. Developments in Earth and Environmental Sci-755 ences,8, 85–114.756 Chadwick, M., Sime, L., Allen, C., & Guarino, M.-V. (2023). Model-data compari-757 son of Antarctic winter sea-ice extent and Southern Ocean sea-surface temper-758 atures during Marine Isotope Stage 5e. Paleoceanography and Paleoclimatology,759 38(6), e2022PA004600.760 Chen, C., Liu, W., & Wang, G. (2019). Understanding the uncertainty in the 21st761 century dynamic sea level projections: The role of the AMOC. Geophysical Re-762 search Letters,46(1), 210–217.763 Chiang, J. C., & Bitz, C. M. (2005). Influence of high latitude ice cover on the ma-764 rine Intertropical Convergence Zone. Climate Dynamics,25(5), 477–496.765 Crowley, T. J. (1992). North Atlantic deep water cools the Southern Hemisphere.766 Paleoceanography,7(4), 489–497.767 Curry, J. A., Schramm, J. L., & Ebert, E. E. (1995). Sea ice-albedo climate feedback768 –24– manuscript submitted to JGR: Oceans mechanism. Journal of Climate,8(2), 240–247.769 Danabasoglu, G., Lamarque, J.-F., Bacmeister, J., Bailey, D., DuVivier, A., Ed-770 wards, J., . . . others (2020). The community earth system model ver-771 sion 2 (CESM2). Journal of Advances in Modeling Earth Systems,12 (2),772 e2019MS001916.773 Danabasoglu, G., Lawrence, D., Lindsay, K., Lipscomb, W., & Strand, G. (2019).774 NCAR CESM2 model output prepared for CMIP6 CMIP piControl [dataset].775 Earth System Grid Federation. Retrieved from https://doi.org/10.22033/776 ESGF/CMIP6.7733 doi: 10.22033/ESGF/CMIP6.7733777 Danabasoglu, G., McWilliams, J. C., & Gent, P. R. (1994). The role of mesoscale778 tracer transports in the global ocean circulation. Science,264(5162), 1123–779 1126.780 Dansgaard, W., Johnsen, S. J., Clausen, H. B., Dahl-Jensen, D., Gundestrup, N. S.,781 Hammer, C. U., . . . others (1993). Evidence for general instability of past782 climate from a 250-kyr ice-core record. Nature,364(6434), 218–220.783 Diamond, R. (2025, May). Supporting dataset for Diamond et al. (2025) JGR784 Oceans [dataset]. Zenodo. Retrieved from https://doi.org/10.5281/785 zenodo.15441431 doi: 10.5281/zenodo.15441431786 Diamond, R., Sime, L. C., Holmes, C. R., & Schroeder, D. (2024). CMIP6 models787 rarely simulate Antarctic winter sea-ice anomalies as large as observed in 2023.788 Geophysical Research Letters,51(10), e2024GL109265.789 Donohoe, A., Armour, K. C., Roe, G. H., Battisti, D. S., & Hahn, L. (2020). The790 partitioning of meridional heat transport from the last glacial maximum to791 CO2 quadrupling in coupled climate models. Journal of Climate,33(10),792 4141–4165.793 D¨oscher, R., Acosta, M., Alessandri, A., Anthoni, P., Arneth, A., Arsouze, T., . . .794 others (2021). The EC-Earth3 Earth system model for the climate model in-795 tercomparison project 6. Geoscientific Model Development Discussions,2021,796 1–90.797 Dufour, C. O., Griffies, S. M., de Souza, G. F., Frenger, I., Morrison, A. K., Palter,798 J. B., . . . others (2015). Role of mesoscale eddies in cross-frontal transport of799 heat and biogeochemical tracers in the Southern Ocean. Journal of Physical800 Oceanography,45 (12), 3057–3081.801 Eayrs, C., Li, X., Raphael, M. N., & Holland, D. M. (2021). Rapid decline in802 Antarctic sea ice in recent years hints at future change. Nature Geoscience,803 14(7), 460–464.804 (EC-Earth), E.-E. C. (2019). EC-Earth-Consortium EC-Earth3 model output805 prepared for CMIP6 CMIP piControl [dataset]. Earth System Grid Federa-806 tion. Retrieved from https://doi.org/10.22033/ESGF/CMIP6.4842 doi:807 10.22033/ESGF/CMIP6.4842808 Eyring, V., Bony, S., Meehl, G. A., Senior, C. A., Stevens, B., Stouffer, R. J., &809 Taylor, K. E. (2016). Overview of the Coupled Model Intercomparison Project810 Phase 6 (CMIP6) experimental design and organization. Geoscientific Model811 Development (Online),9.812 Farneti, R., Downes, S. M., Griffies, S. M., Marsland, S. J., Behrens, E., Bentsen,813 M., . . . others (2015). An assessment of Antarctic Circumpolar Current and814 Southern Ocean meridional overturning circulation during 1958–2007 in a suite815 of interannual CORE-II simulations. Ocean Modelling,93, 84–120.816 Ferrari, R., Jansen, M. F., Adkins, J. F., Burke, A., Stewart, A. L., & Thompson,817 A. F. (2014). Antarctic sea ice control on ocean circulation in present and818 glacial climates. Proceedings of the National Academy of Sciences,111(24),819 8753–8758.820 Fox-Kemper, B., Danabasoglu, G., Ferrari, R., Griffies, S., Hallberg, R., Holland, M.,821 . . . Samuels, B. (2011). Parameterization of mixed layer eddies. III: Imple-822 mentation and impact in global ocean climate simulations. Ocean Modelling,823 –25–