We use output from the quasi-equilibrium freshwater forcing simulation performed by using a strongly-eddying configuration of the Parallel Ocean Program (POP, version 2) , with a nominal horizontal resolution of 0.1° and 42 non-equidistant vertical levels. The model is forced using observed river run-off fields and a prescribed atmospheric state based on the repeat annual cycle (normal year) Coordinated Ocean Reference Experiment (CORE) forcing data set , with 6 hourly forcing averaged to a monthly resolution. Precipitation is also taken from the CORE forcing dataset. Wind stress is computed offline using the Hurrell SST climatology and standard bulk formulae, whereas evaporation and sensible heat flux are calculated online using the model prescribed SST and bulk formulae. A diagnosed freshwater flux, determined from an equilibrium spin-up, is also prescribed. Sea-ice cover is prescribed based on the -1.8°C isoline of the SST climatology, with both temperature and salinity restored on a timescale of 30 d under diagnosed climatological ice . Apart from this, there is no salinity restoring in the model. Details on model configuration and simulation procedure can be found in the Supplementary Material of .

The quasi-equilibrium freshwater flux forcing simulation in was initialised from model year 300 of the multi-century simulation performed by . Model drift is still present , but occurs on much longer timescales than the AMOC decline. From the start of the quasi-equilibrium simulation, a freshwater flux with strength FH is applied over the North Atlantic sector (20–50° N) with a constant rate of 3×10-4Svyr-1, similar to the hosing simulation performed in the Community Earth System Model (CESM) . The freshwater flux anomaly is globally compensated to conserve salinity. The AMOC strength is weakening under increasing FH values and collapses around FH=0.125Sv (model year 415); more details on this simulation can be found in .

In the simulations of , the temperature anomalies associated with the SOM (see their Fig. 2) propagate eastward along the ACC in the South Atlantic, where they are disrupted south of Africa around 30° E due to the interaction with the Agulhas Current Retroflection . The anomalies then split into two pathways, where one continues along the ACC and slowly dissipates, while the other enters the Weddell Gyre, likely due to enhanced cross-stream eddy diffusivity compared to the South Atlantic . The heat anomalies also propagate northward through the Atlantic basin, thereby inducing multidecadal variations in the AMOC strength up to 3 Sv. The SOM is associated with a peak-to-peak variability of approximately 60 ZJ (1ZJ≡1021J) in global ocean heat content (OHC), highlighting its potential significance for large-scale climate variability.

Although the spatial pattern of the SST anomalies associated with the SOM extends across the entire Southern Ocean, the largest anomalies can be found in the South Atlantic sector . To quantify this variability, it is measured using the SOM index, which is defined as the SST anomaly averaged over the region 50–35° S, 0–50° W (black outlined region in Fig. a).

Previous studies have analysed the mechanical energy budget of the SOM in the POP control simulation . It was shown that eddy-mean flow interactions are central to explain the SOM, similar to that of a mode of multidecadal variability identified in a three-layer, eddy-resolving, quasi-geostrophic model of a zonal channel flow . The oscillatory behaviour arises from phase lags between mechanical energy input by the wind, generation of eddies by baroclinic instability, and the zonality of the mean flow. Below, we also apply a similar mechanical energy budget analysis to the quasi-equilibrium POP simulation to investigate the effects of an AMOC weakening on the SOM.

The Lorenz Energy Cycle (LEC) framework has proven effective for analysing the multidecadal variability of the SOM . Further simplifications of the full LEC framework have been proposed by , and these have been successfully applied to the POP model output to explain the oscillatory behaviour of the SOM . In the following, we denote K, Km, Ke as the volume-integrated total kinetic energy, mean kinetic energy and eddy kinetic energy, respectively; and P, Pm and Pe as the volume-integrated available potential energy, mean available potential energy and eddy available potential energy, respectively.

The total kinetic energy K is computed according to: 1K=ρ02∫V(u2+v2‾)dV, where u=(u,v) denotes the horizontal velocity vector, ρ0 the global average density of sea water, and V the volume over which the energetics are evaluated. The overbar represents a 5 year time average used to perform the eddy-mean decomposition, consistent with previous studies on the energetics of the SOM . The results are not sensitive to small changes in the averaging window (tested for 3–10 year time averages). The mean kinetic energy is given by: 2Km=ρ02∫V(u‾2+v‾2)dV, and the eddy kinetic energy is computed as the difference between K and Km per gridpoint. The available potential energy is expressed as: 3P=-g2∫V1n0ρ*2‾dV, where the potential density anomalies are defined as ρ*(x,y,z,t)=ρ(x,y,z,t)-ρref(z), with ρref(z)=〈ρ(x,y,z,t)〉av representing the global reference state . In the expression for ρref, the angled brackets indicate a global area average, while the subscript denotes a time average over one SOM cycle. Furthermore, g denotes the gravitational acceleration and n0(z) the vertical gradient of the reference potential density. An eddy-mean decomposition can be performed to determine the mean and eddy potential energies (Pm and Pe, respectively). The analysis is conducted over the entire Southern Ocean south of 30° S (SO30 region), thereby capturing the complete SOM variability and avoiding problems of boundary terms .

Under the approximations outlined in detail by , the evolution equations for the volume integrated eddy kinetic energy Ke and available potential energy P are given by: 4dPdt≈G(Km)-C(Pe,Ke)-D(Km),5dKedt≈C(Pe,Ke)-D(Ke). Here, the generation of mean kinetic energy by the wind forcing exerting a stress on the ocean surface S is expressed as: 6G(Km)=∫S(τx‾u‾+τy‾v‾)dS, with τ=(τx,τy) the wind stress. The exchange of potential to eddy kinetic energy associated with baroclinic instability, C(Pe,Ke), is determined by: 7C(Pe,Ke)=-g∫Vρ′w′‾dV≈C(P,K)-C(Pm,Km), where w denotes the vertical velocity and the prime indicates anomalies with respect to the 5 year time average. The terms in the mechanical energy budget are calculated using 5 year moving averages, while eddy contributions are determined from monthly mean data. The terms D(X) above are dissipation terms, but are not crucial to study the SOM cycle . The degree of non-zonality of the mean flow, which is a proxy for the baroclinic generation of eddies, is quantified by ζ=(∫v2‾dV)/(∫u2‾dV), where the squared meridional and zonal velocities are computed first and then volume integrated over the SO30 region and the top 300 m.

In the quasi-equilibrium simulation of the POP, a gradual increase in surface freshwater forcing leads to a weakening of the AMOC (Fig. a). The AMOC strength at 26° N decreases from a mean of 19.4 Sv during the first 50 model years to a mean of 4.8 Sv during the last 50 model years. The onset of the collapse occurs around model year 415 . The SOM index (Fig. b, black curve) exhibits distinct multidecadal variability over the first 200 model years, with a dominant period of approximately 45 years that is statistically significant at the 95 % confidence level based on a Morlet wavelet power spectrum (Fig. b). As the freshwater forcing increases, this period becomes longer, reaching approximately 50 years between model years 300–500. After the AMOC collapse and over the last 100 model years, the variability in the SOM index disappears entirely (Fig. b). The Drake Passage transport (Fig. c) displays the same multidecadal variability as the SOM index over the first 300 years (Fig. a), with a peak-to-peak amplitude of about 17 Sv. Subsequently, the mean transport and the amplitude of multidecadal variability slightly decrease, the former by about 10 Sv and the latter being minimal just before the onset of the AMOC collapse. Interestingly, this amplitude increases again after the AMOC collapse, while the frequency of variability slightly decreases (Fig. a).

To understand the disappearing multidecadal variability in the AMOC strength time series, we conduct an empirical orthogonal function (EOF) analysis on annual mean SSTs south of 30° S for the first and last 100 model years of the simulation. All time series are first linearly detrended and normalised by their standard deviations, and subsequently weighted to their surface area prior to conducting the EOF analysis. The dominant EOF pattern and associated principle component (PC) are shown in Fig. . During the first 100 model years (Fig. a), the EOF exhibits relatively strong (negative) amplitudes over the Atlantic sector in the Southern Ocean. This motivates the choice of the region used for the SOM index , as indicated by the black outlined region (50–35° S, 0–50° W). The corresponding PC (Fig. c) exhibits a similar period to that of the SOM index, highlighting the dominance of this multidecadal variability in the SO30 region.

After the AMOC collapse (last 100 model years), the pattern of the EOF changes significantly, with the largest amplitudes now located in the Pacific sector of the Southern Ocean (Fig. b). The period of the associated PC increases to approximately 75 years (Fig. d), matching the period of the Drake Passage transport over the last 100 years (Fig. a). This motivates us to also study the variability in the Pacific sector, where we define the SOM-P index as the spatially-averaged SSTs over the blue outlined region (60–45° S, 170° E–150° W), which coincides with the region of largest EOF amplitudes (Fig. b). Interestingly, a pronounced mode of multi-decacal variability emerges in the SOM-P index only after the onset of the AMOC collapse around model year 415 (blue curve in Fig. b). The dominant period during the last 150 model years is approximately 75 years, and is statistically significant at the 95 % confidence level based on a Morlet wavelet power spectrum (Fig. c). The dominant period of the SOM-P index is consistent with that of the Drake Passage transport after the AMOC has collapsed (Fig. a). This suggests that Pacific variability is associated with the multidecadal behaviour of the Drake Passage transport after the AMOC has collapsed, as will be further explored below in Sect. .

Density anomalies associated with the SOM propagate northward and submerge around 40° S , generating multidecadal variability in the subsurface temperature and salinity fields (Fig. ). These subsurface anomalies continue to propagate northward within the Atlantic basin, thereby influencing the strength of the upper branch of the AMOC . As a result, the AMOC strength at 26° N exhibits a similar 45 year oscillation period as the SOM index during the first 300 model years (Fig. a). As the AMOC weakens, the distinct temperature and salinity patterns associated with the positive and negative phases of the SOM disappear (Fig. ), and the AMOC loses its multidecadal variability (Fig. b).

The substantially weakened AMOC leads to a pronounced reduction in meridional heat and salinity transport , resulting in an accumulation of heat and convergence of salt in the Southern Hemisphere ocean interior, extending to depths up to 1000 m north of the Antarctic subpolar front (50° S). This causes both warming (Fig. a) and salinification (Fig. b) of the subsurface waters, consistent with previous studies on a weakened AMOC . In contrast, the deep waters experience a freshening and slight cooling related to the reduced formation of North Atlantic Deep Water (NADW), and a reorganization of the water masses in the Southern Ocean. The resulting density changes show an increase in the upper 1000 m between 50 and 20° S, and a decrease further north (Fig. c). At greater depths, the waters become less dense, primarily driven by salinity changes (Fig. ).

Between 50–40° S and over the upper 1000 m, the isopycnals slope downward and this is indicative of the Antarctic subpolar front (solid curves in Fig. c–f). This meridional density gradient becomes less negative after the AMOC collapse, resulting in a decrease of the isopycnal slope (shading and dashed curves in Fig. c). By contrast, a negative meridional density gradient difference is found at subsurface depths (1000–2000 m) and near 50° S, corresponding to a steepening of the isopycnals. Changes in the isopycnal slopes modify the baroclinicity, thereby influencing the strength of the ACC through thermal wind balance (Fig. c). The changes in the meridional density gradient are not uniform across the ACC latitude band (50° S - 40° S), and may therefore lead to spatially heterogeneous changes in the ACC. showed that an increased meridional slope of the isopycnals near 50° S corresponds to a reduction of the SOM period in the CESM model. This is consistent with the results here, where an increase in the SOM period occurs simultaneously with the meridional isopycnal slopes becoming less negative (near 50° S and upper 1000 m).

The displacement of the isopycnals leads to changes in the stratification of the water masses (Fig. e), with clear bands of increased and decreased stratification. Stratification decreases in the water mass north of 50° S between 1000 and 1500 m, but, in contrast, waters below 1500 m and those that upwell south of 50° S show an overall increase in stratification. The latter increases are primarily driven by salinity-controlled changes in N2 (Fig. f), while stratification changes north of 50° S and in the upper 1500 m result from a combination of temperature- and salinity-driven effects.

Similar to the Atlantic sector, the Indian and Pacific sectors of the Southern Ocean also exhibit warming and salinification over the upper 1000 m and north of 45° S (Figs.  and ). These changes are accompanied by a decrease of the meridional density gradient over the near-surface layer, and an increase of this gradient in the subsurface layers across the ACC latitude band (between 50 and 40° S). Stratification weakens north of 60° S in the upper 1500 m, while deeper waters show an overall increase in stratification. The AMOC weakening therefore leads to a basin-wide reorganization of the Southern Ocean density structure, which appears to suppress SOM-related variability in the Atlantic sector while enabling the emergence of a new mode of multidecadal variability in the Pacific sector.

To understand the reduced SOM variability after the AMOC collapse, we analysed the mechanical energy changes over the SO30 region. The dominant terms of the mechanical energy budget are evaluated over the full simulation period, and over three different SOM cycles: one early in the simulation (SOM cycle 1, model years 63–114), one during the AMOC collapse (SOM cycle 2, model years 410–480), and one during the weak AMOC state (SOM cycle 3, model years 500–600). The difference in cycle length reflects the increasing SOM (SOM-P) period under increasing freshwater flux forcing. The mechanical energy budget analysis is displayed in Fig. .

The available potential energy, P, is dependent on the reference density ρref used (see Eq. ). We compute P over the full simulation using a fixed ρref derived from the first 50 model years (dark blue curve in Fig. a), such that all changes in P reflect variations in the density structure rather than changes in the reference state. In contrast, P for the individual SOM cycles is computed using ρref averaged over the corresponding SOM cycle period (cyan curves in Fig. a). The different choices of ρref do influence the magnitude of P, but not their overall variability and tendency.

The available potential energy starts to increase in the SO30 region following the AMOC collapse, with the Atlantic, Indian, and Pacific sectors contributing approximately 29 %, 30 %, and 41 %, respectively, to the total P in this region. An increase in P indicates that the water column is further displaced from its stable reference state. This is also consistent with the decrease in vertical stratification in the upper 1500 m in the Atlantic, Indian and Pacific sectors south of 30° S, as shown in Fig. . The available potential energy, however, has not yet reached an equilibrium state at the end of the simulation, as P continues to increase.

The tendency of P (Fig. ) is affected by the terms G(Km), representing the mean wind energy input, and by C(Pe,Ke), representing conversion of eddy potential to eddy kinetic energy. An increase in G(Km), e.g. due to a more zonal flow with steeper isopycnals, increases P. In contrast, an increase in C(Pe,Ke) reflects enhanced eddy generation through baroclinic instability, transferring more energy from Pe to Ke, and thus reduces P over time. Figure  shows both a mean reduction of G(Km) and C(Pe,Ke), although the reduction in G(Km) is larger (Fig. c and d). The increase in P at the end of the simulation is therefore not driven by enhanced mean wind energy input, but instead is a consequence of changes in the density field associated with the AMOC collapse. Additionally, K and Ke primarily exhibit a reduced amplitude of variability rather than a shift in mean magnitude (Fig. b).

The relatively zonal background ACC flow starts to meander more after the AMOC collapse, which is reflected in the increase of the non-zonality parameter (ζ, Fig. e). This increase is consistent with the reduction in mean wind energy input over time, as the correlation between the zonal wind forcing and the zonal mean flow weakens. While an increased non-zonal flow is typically associated with enhanced eddy activity, this is not reflected in an overall increase of C(Pe,Ke). The non-zonality parameter loses its multidecadal variability entirely after the AMOC collapse, suggesting that its increase is not primarily driven by variability in the eddy generation and mean wind energy input. Mean flow changes likely also play a role, e.g. a reduction in the mean ACC strength or a shift in its position could lead to increased meandering of the zonal flow.

Following the framework established in earlier studies , the SOM cycle can be divided into four distinct regimes (Fig. a), where the vertical stratification plays a key role in setting the timescale of this SOM cycle . Note that the quantities are now centered (zero mean) and normalised by their standard deviation, to more clearly see the phase differences between the different terms. Regime A corresponds to a low-energy state with a relatively zonal ACC, starting at the minimum of the total potential energy P and ending at the minimum of the total kinetic energy K. In regime B, P increases as the zonal flow is accelerated by the wind work, leading to a maximum of P. Regime C represents the high-energy state, spanning from the maximum of P to the maximum of K. During this period, the flow undergoes enhanced baroclinic instability, leading to an increase in the generation of eddies by the mean flow. The P accumulated in regimes A and B is now converted to eddy kinetic energy. This enhanced eddying flow rearranges the flow field, making it less zonal, thereby disrupting the correlation between the surface ocean velocity and the wind stress. As a result, the wind energy input quickly decreases. Finally, regime D is characterised by declining P and K. As the storage of P becomes exhausted, the conversion of P to K begins to diminish as well. This, combined with the reduction in wind work, drives the flow back to its low-energy state, thereby completing the cycle.

The temporal evolution for SOM cycle 2 and 3 are shown in Fig. b and c, respectively. Note that we normalised the quantities as (X-Xmean)/σX1, where we use the standard deviation of the first cycle as reference. Equation () shows that the rate of change of Ke, and therefore K, is mainly influenced by C(Pe,Ke). Large transfers of potential to kinetic energy feed almost directly into the eddy kinetic energy field. Consistent with this, in all three SOM cycles, K and C(Pe,Ke) are positively correlated with a lag varying between 1–5 years. The non-zonality parameter ζ loses its multidecadal oscillatory behaviour in SOM cycle 3, in contrast to the pronounced variability found during SOM cycles 1 and 2 (Fig. a and b). During SOM cycle 1 and 2, the minima and maxima of P lead the minima and maxima of K with an approximate phase difference of 90°. The phase offset between P and K vanishes completely in SOM cycle 3 due to the altered behaviour of P as a consequence of stratification changes in the water column. Furthermore, a clear phase difference between G(Km) and C(Pe,Ke) is found in SOM cycle 1, which is essential to sustain the SOM . In SOM cycle 3, however, G(Km) and C(Pe,Ke) begin to co-vary, with reduced phase differences between their minima and maxima.

Figure  shows the energetics integrated over the entire Southern Ocean. Performing the same analysis for the Atlantic sector alone yields a similar behaviour: during SOM cycles 1 and 2, the phase differences between the energy terms resemble those found in the SO30 region, while these phase relationships disappear during SOM cycle 3 (Fig. ). In contrast, in the Pacific sector the emergence of the SOM-P cycle is not accompanied by clear phase differences during SOM cycle 3 that are consistent with the established framework of (Fig. ). This suggests that, unlike in the Atlantic sector, the SOM-P may not be strongly influenced by eddy–mean flow interactions. In conclusion, the reorganization of the Southern Ocean density field leads to a fundamental change in the mechanical energy budget, particularly affecting the phase relationships between P and K, and between G(Km) and C(Pe,Ke). As these phase relationships are essential for sustaining the SOM, their alteration due to the AMOC collapse leads to the disappearance of the SOM cycle in the Atlantic sector.

Changes in stratification likely influence deep convection across the Southern Ocean and the exact role of deep convection in the SOM variability is not completely clear . The reason is that convection is parameterised (using the KPP mixing scheme in POP), which provides enhanced mixing without explicitly resolving vertical velocities. Hence, the contribution by deep convection cannot be assessed using the mechanical energy pathways . A recent study by , based on the analysis of a high-resolution CESM simulation, found only weak evidence for atmosphere–ocean feedbacks contributing to Southern Ocean multidecadal variability and instead attributed this variability primarily to oceanic processes. They propose that the SOM mechanism operates as part of a coupled oscillator involving Southern Ocean deep convection events, with salinity upwelling east of Maud Rise playing a crucial role. These findings support the conclusion of that deep convection can also play a substantial role in sustaining the SOM.

Based on the largest differences in the maximum and variance of the mixed layer depth (MLD) between model years 500–600 and 1–100 (Fig. a and b), four regions are identified for further analysis. Similar to , we define the Weddell Gyre to Kerguelen Plateau (WGKP) region as 80–50° S and 35° W–80° E (green outlined region in Fig. b). The brown-outlined region in the eastern Indian sector south of Australia, extending from 70–50° S and from 80–150° E, is hereafter referred to as AU. The convective region highlighted in magenta, located at longitudes aligned with New Zealand and spanning 70–62° S and 160° E–170° W, is hereafter denoted as NZ. Finally, the blue-outlined western Pacific region, extending from 70–50° S and 110–60° W, is hereafter referred to as PA.

The area-averaged potential density (PD) profiles of the four regions are shown in Fig. c and d for model year 1–100 and model year 500–600, respectively. The stratification over the upper 2000 m is relatively weak for the NZ and WGKP regions, whereas the AU and PA regions are stronger stratified (Fig. c). Following the AMOC collapse, stratification increases in the NZ and WGKP regions, while the opposite is true for the AU and PA regions (Fig. d). A stronger (weaker) stratification reduces (increases) the MLD, which is found for the NZ and WGKP (AU and PA) regions.

To make the latter more explicit, we present the maximum MLD and Brunt-Väisälä frequency differences (relative to the first 100 model years) over the four regions in Fig. . The stratification in the WGKP region strongly increases in the upper 100 m, decreases in the layer just below (down to 250–500 m), and increases again at depths down to 2000 m (Fig. b). The NZ region exhibits an overall increase in stratification, with the strongest anomalies occurring in the upper 100 m. In contrast, the AU and PA regions show an overall decrease in stratification in the upper 1000–1500 m, and an increase at greater depths.

The stratification changes in the convective regions are related to a salinity-dominated reorganization of the water-column structure following the AMOC collapse (Fig. ). The increased stratification in the deeper layers of the AU and PA regions, and in the intermediate layers of the NZ and WGKP regions, can be linked to a reduced poleward advection of warm and saline Circumpolar Deep Water (CDW) after the AMOC has weakened. Furthermore, a downward displacement or reorganization of Antarctic Intermediate Water (AAIW), identified by its salinity minimum, leads to reduced vertical salinity gradients causing weakening of the stratification. This mechanism explains the reduced stratification in the upper ∼1500m of the AU and PA regions, and the weakening of stratification in the WGKP region between roughly 100 and 250–500 m. Although the reduced stratification barely penetrates into the WGKP region (Fig. ), it dominates over the otherwise increasing stratification in the same depth range. The stratification changes north of the NZ convective region show a similar structure to those in the WGKP region.

Although the stratification weakening over the AU and PA regions induces a slightly deeper mixed-layer (Fig. c and d), the stratification remains sufficiently strong such that the MLD responses are limited. A weak multidecadal oscillation arises after the collapse of the AMOC with comparable periods to that of the SOM-P index (Fig. c and d). In contrast, the stratification over the NZ region remains relatively weak compared to the other three regions (Fig. c and d), making this region the most prone to the deepest MLD events after the AMOC collapse. The onset of deep convection with strong multidecadal variability (Fig. a) in this region is closely linked to the stratification anomalies in the upper 100 m after the AMOC collapse (Figs. a).

Up to the AMOC collapse, strong MLD changes occur in the WGKP region with a period closely following that of the SOM index (Fig. b). Over these 400 years, the MLD in the NZ region is deep (∼800m) but its variability is relatively small (Fig. a). Deep convection events in the WGKP convective region follow a convection–restratification mechanism . Heat originating from the inflow of relatively warm North Atlantic Deep Water (NADW) is transported into the Weddell Sea by the westward return flow in the southern branch of the Weddell Gyre, where it remains effectively trapped within the gyre circulation. The accumulation of heat at mid-depth levels destabilizes the water column and can eventually trigger deep convection. Once the mid-depth heat reservoir is depleted, convection shuts down (Fig. b). The multidecadal recharge of this heat reservoir depends on both the AMOC and the Weddell Gyre circulation , and is closely related to the SOM variability as generated by the eddy-mean flow interaction mechanism described in the previous section. The ACC is modulated by these convective episodes, with the meridional pressure gradient weakening during non-convective phases due to gradual mid-depth warming of waters south of the ACC, and strengthening during convective phases as these waters cool. This variability in the pressure gradient leads to a corresponding weakening or strengthening of the ACC, typically with a lag of a few years (Fig. ).

The stratification over the WGKP region is relatively weak compared to the AU and PA regions (Fig. ), but increases due to the reduced inflow of NADW. As a result, the water column cannot support any deep convection (>1000m) events anymore after the AMOC collapse. At the same time, stratification anomalies in the upper 100 m start to appear again over the NZ region, predominantly salinity-driven (not shown), and are apparently sufficient to initiate convection and support relatively strong MLD changes (Fig. a), which in turn affect the Drake Passage transport. After deep convection ceases in the WGKP region and the SOM cycle vanishes, a pronounced oscillatory signal still persists in the Drake Passage transport with a period similar to that of the SOM-P index (Fig. ). Deep convection in the NZ region now leads the multidecadal oscillations in the Drake Passage transport (Fig. ).

The occurrence of deep convection events have been linked to an overall strengthening of the ACC, as they facilitate the conversion of potential to kinetic energy, thereby energizing the ocean circulation . Indeed, we find an overall decrease in mean strength of the ACC (Fig. c), consistent with a decrease in C(Pe,Ke) (Fig. d) and the termination of deep convection events in the WGKP region.

This shift in Southern Ocean deep convection from the Atlantic to the Pacific sector is closely linked to the behaviour of the SOM-P index, which starts to exhibit pronounced multidecadal variability around the same time deep convection in the NZ region starts (Fig. b). The convective episodes in the NZ region now lead the oscillations in the SOM-P index, in a similar way the minima and maxima of the WGKP convective episodes led the minima and maxima of the SOM-index (Fig. a). Whereas the SOM mechanism involves a combination of eddy–mean flow interactions and deep convection in the WGKP region, the oscillations emerging after the AMOC collapse cannot be explained by the first mechanism, as its signature is not detectable in the mechanical energy budget over the SO30 region. Instead, the SOM-P variability has a purely convective origin and in this way, the Pacific sector becomes the primary source region for multidecadal variability of the Southern Ocean origin when the AMOC has collapsed.