Modulation of western South Atlantic marine heatwaves by meridional ocean heat transport

,


Introduction
Marine heatwaves (MHWs) and cold spells (CSs) are sustained extreme warm and cold sea surface temperature (SST) anomalies, respectively.In particular, MHWs have received considerable attention in recent years, but both MHWs and CSs can have drastic impacts on marine ecosystems (e.g., Schlegel et al., 2021), such as coral bleaching (Couch et al., 2017;Dalton et al., 2020;Hsieh et al., 2008;Le Nohaïc et al., 2017;Zapata et al., 2011), changes in primary productivity and hypoxia (Sen Gupta et al., 2020;Wheeler et al., 2003), pelagic species mortality (McEachron et al., 1994;Smale et al., 2019;Woodhead, 1964), and closing of commercial and recreational fisheries (Cavole et al., 2016;Chang et al., 2013;Santos et al., 2016;Stuart-Smith et al., 2018).The compound effect of extreme temperatures with other stressors such as ocean acidification, tropical cyclones, algae blooms, and marine pollution, can have long-term impacts on marine ecosystems.In addition to these stressors, global warming affects the baseline of transient anomalies (Hobday et al., 2018) and, therefore, there has been an increase in the duration, frequency and intensity of MHWs relative to CSs in most regions of the globe (Costa & Rodrigues, 2021;Frölicher & Laufkötter, 2018;Oliver et al., 2018).
One of the most resilient MHW events was registered in the Northeast Pacific region during 2013-2016.This event, commonly known as the "Blob," has been linked to large scale atmospheric and oceanic patterns such as a reduction in wind-driven upper-ocean mixing and a shallow mixed layer depth (Amaya et al., 2020;Di Lorenzo & Mantua, 2016;Joh & Di Lorenzo, 2017).Other recent studies have shown the importance of the ocean state for similar types of extreme temperature anomalies in different basins.For instance, the occurrence, location and intensity of MHWs in the Tasman Sea have been linked to upper 2,000 m warm ocean heat content anomalies on interannual to decadal timescales (Behrens et al., 2019), and the likelihood of MHWs near the East Australian Current has been shown to be modulated by westward propagation of Rossby waves (Li et al., 2020).
In the western South Atlantic, Rodrigues et al. (2019); Li et al., 2020 linked the occurrence of MHWs to blocking events associated with atmospheric Rossby wave trains propagating from the South Pacific Ocean.These blocking events are characterized by increased atmospheric sea level pressure, suppressed formation of clouds and increased solar radiation into the ocean.For some MHW events, this mechanism can be the trigger.In another study, Manta et al. (2018) found that the year of 2017 was also characterized by an intense MHW over the continental shelf off the Rio da Plata in Uruguay, and this event was triggered by extreme atmospheric heat fluxes related to reduced winds.However, the large-scale ocean circulation may precondition or interact with these events and, thus, modulate their occurrence on interannual timescales.For example, Goes et al. (2019) showed that the South Atlantic MHW during the austral summer of 2009/2010 propagated zonally from the center of the basin to the east coast of South America near 22°S.Upon reaching the western boundary, this MHW propagated southwards along the Brazil Current and dissipated two months later near 30°S.The mechanisms by which largescale ocean processes may influence MHW events in the South Atlantic are mostly unknown.In other locations, such as the North Atlantic, the tripole pattern, which is the first mode of interannual variability of SST and sea level anomaly (SLA), has been linked to convergences and divergences of the integrated meridional heat transport, which impacts the large scale upper ocean heat content and coastal sea level in neighboring areas (e.g., Roberts et al., 2016;Volkov et al., 2019).
In the South Atlantic, evidence has been found that large scale SST patterns can serve as a fingerprint for South Atlantic Meridional Overturning Circulation (AMOC) variability (Dima & Lohmann, 2010;Lopez et al., 2016), which may change the baseline for the occurrence of MHW and CS events in the region.This paper analyzes the effect of basin-scale propagating ocean anomalies on the occurrence of MHWs and CSs in the western South Atlantic.Known methodologies (Section 2.2) are used to detect the propagating modes and the extreme temperature events.A mixed layer heat budget analysis is presented to investigate the role of ocean advection and heat fluxes in the western South Atlantic (Section 3.3).Finally, a reconstruction of the AMOC in the South Atlantic will be used to infer the role of large-scale volume and heat transport in triggering these anomalies (Section 3.4).The implications of the results and the potential to predict extreme temperature events in the western South Atlantic is also discussed (Section 4).

Data
Five main data sets are used in this work, consisting of SST and sea level height from remote sensing, reanalyses for the ocean (ORAS5) and air-sea interface (ERA5), and an AMOC reconstruction from satellite altimetry and in situ temperature.
The analysis of SST data is performed using the NOAA Optimum Interpolation Sea Surface Temperature data set (OISSTv.2; Reynolds et al., 2007), a blended global gridded product available daily at a 1/4°horizontal resolution since January 1982.Daily resolution is used to identify metrics of MHWs and CSs, and monthly resolution is used in basin-scale correlation maps.For the purpose of this work, the SST data are spatially regridded using a bilinear interpolation to a 1°× 1°horizontal resolution from 1993 to 2020.
The monthly AMOC strength and meridional heat transport timeseries at four different latitudes in the South Atlantic (35°S, 30°S, 25°S, and 20°S) for the period 1993-2021 were produced using synthetic temperature profiles based on statistical relationships between SLA and the depths of isotherms (Dong et al., 2015(Dong et al., , 2021)), salinity profiles from historical T/S relationships (Goes et al., 2018), and monthly surface wind stress data from ERA5 for the Ekman component.The methodology has been validated against the AMOC and meridional heat transport estimates based on cross-basin expendable bathythermograph (XBT) high-density transect data near 35°S (Dong et al., 2015).
We examine vertical temperature anomalies and calculate the heat budget in the western South Atlantic using the ORAS5 reanalysis (Zuo et al., 2018), which is based on an ensemble of global eddy-permitting (0.25°) ocean model runs, forced with air-sea fluxes from ERA-Interim (1979-2015) and ECMWF NWP thereafter.The monthly output of ORAS5 is interpolated to a 1 × 1°horizontal grid and 75 vertical levels.For this work, we use the following ORAS5 fields: temperature (T ), horizontal velocity (v), mixed layer depth (h), wind stress (τ) and net surface heat flux (Qnet) from 1993 to 2021.

Complex EOF
To define the interannual propagating patterns of SLA in the South Atlantic, first the monthly de-seasoned SLA fields are filtered using a bandpass wavelet filter of 0.8-16 years.Then we perform a Complex Empirical Orthogonal function (CEOF) analysis (e.g., Navarra & Simoncini, 2010;Rasmusson et al., 1981) to extract the main propagating patterns of variability between 15°S and 40°S.The CEOF uses a principal component analysis on a Hilbert (quadrature) transform of a field.Therefore, it produces real and imaginary parts of loadings (maps), here defined as CEOF j (x,y) for a particular mode j, and their associated expansion coefficients or principal components (PC j (t)).The CEOFs and PCs are used to compute spatial and temporal amplitudes and phases, necessary conditions to describe a propagating wave pattern (Majumder et al., 2019).Therefore, the advantage of using CEOFs is the ability to define a propagating mode in space and time, as opposed to conventional EOFs which only provide snapshots of two phases (0°or 180°) of the propagation (Rasmusson et al., 1981).The SLA, associated with a particular mode of the variability j (SLA j ), can be reconstructed as the product of its loadings and its associated expansion coefficients: The time evolution of the mode j can be examined by rotating in phase the CEOF and the PC by an angle θ using a rotation matrix R 2x2 (θ), which produces the same reconstruction: Cross-correlations and composites of SST, SLP, and 10-m wind anomalies are obtained for the rotated phases of the CEOF modes, using a 13-month Gaussian filter on these variables to remove sub-annual timescales.

Definition of Marine Heatwaves and Cold Spells
The MHWs and CSs are defined as the extreme SST anomaly (SSTA) events that are above the seasonally varying the 90th percentile and below the 10th percentile, respectively, and last for at least 5 days (Hobday et al., 2016).We follow the definitions of MHWs and CSs described in Hobday et al. (2016) using detrended daily values of SSTA at a 1°× 1°degree horizontal resolution.The K-means cluster analysis (Arthur & Vassilvitskii, 2007), which is a method that minimizes the Euclidean distance between groups of observations, is used to determine the large-scale SSTA patterns associated with MHWs or CSs events.The cluster analysis is a method of dimension reduction, therefore categorizing the sparse extreme SSTA events into a set of spatial and temporal modes.In this analysis, we identified the locations where the daily SST anomalies crossed the MHW and CS thresholds and divided them into four clusters.The method was applied to MHW and CS anomalies separately, and then we visually selected the cluster modes that were mostly related to the western subtropical South Atlantic region.

Mixed Layer Heat Budget
To examine the roles of atmospheric and oceanic contributions to the mixed layer temperature changes, we use a simplified temperature tendency equation: which states that the mixed layer temperature tendency (a) is driven by contributions from the net surface heat flux (b), horizontal advection (c), and vertical entrainment (d).The residual R (e) represents unresolved processes such as horizontal and vertical mixing, eddy covariances, and the accumulation of errors from the terms (a)-(d) (Vialard et al., 2001).The mixed layer depth (h) provided by the reanalysis is calculated using the criterion of an increase in potential density of 0.03 kg•m 3 from the surface.Mixed layer depth typically varies from 15 to 180 m in the western South Atlantic.Temperature (T ) and horizontal velocity (v) are averaged over the mixed layer.In (b), Qnet is reduced by the amount of shortwave radiation that penetrates through the base of the mixed layer (Q pen ), estimated from an exponential function of surface ocean color (Morel & Antoine, 1994;Sweeney et al., 2005).Shortwave radiation from ERA5 is used in the Q pen calculation, since ORAS5 does not provide shortwave radiation as a standard output.In addition, ocean color is derived from the SeaWiFS satellite climatological Chlorophyll-a concentration data, interpolated from its original 9 km resolution to the ORAS5 output grid.In the entrainment term (d), ΔT is the difference between the mixed layer temperature and the temperature 5 m below the mixed layer and w is the entrainment velocity at the base of the mixed layer (upward only, given by the Heaviside function: H(w > 0) = 1; H(w < 0) = 0), estimated as: which consists of a mass flux that crosses an isopycnal surface (Stevenson & Niiler, 1983).The vertical entrainment velocity is decomposed into vertical advection or the Ekman pumping contribution )), the tendency of the mixed layer (∂h/∂t), and the horizontal induction across the mixed layer (v • ∇h).The constants used are the density of seawater (ρ 0 = 1,025 kg m 3 ) and the specific heat of seawater (C p = 4,000 J kg 1 K 1 ).
The horizontal advection term (c) can be further decomposed into its mean (bar) and time-variable (prime) constituents: where the subscripts denote the gradient in the x and y directions, the mean variability is defined as a 28-year mean, and the time-variable component is the residual from the mean.In the analysis of interannual variability of heat budget and heat transport, a 19-month window is applied to better match the variability of the PC.

The Main Propagating Mode in the South Atlantic
The main propagating mode of SLA at interannual timescales in the South Atlantic (CEOF1) is an east-west pattern centered between 25°S and 35°S (e.g., Majumder et al., 2019).This mode has a main periodicity of 3-5 years and explains ∼28% of the interannual variance of SLA in the South Atlantic (Figure 1c).The correlation between the reconstructed SLA and the band-pass filtered SLA reaches r = 0.75, and the correlation with SSTA reaches r = 0.7, averaged over 33°S-27°S and 46°W-35°W in the western subtropical South Atlantic (Figure 1c).The evolution of this pattern for half-cycle (0-180°phase) snapshots every 45°(Figure 1a) shows that it has a signature in the South Atlantic SST field (Figure 1b), in that the correlation of SSTA with this mode shows an inphase relationship pattern, where positive (negative) SLA are associated with high (low) SSTA.These co-located in-phase SLA and SSTA are observed mostly in the eastern and western parts of the basin.In the center of the basin, the anomalies in sea level and SST do not overlap; instead the SST correlation map shows a dipole pattern centered at ∼30°S, which hints at an exchange of waters from north and south of the region.This SST dipole pattern is similar to the South Atlantic Subtropical Dipole mode (SASD), which is the leading EOF mode of interannual SST variability in the South Atlantic, explaining about 24% of the total variance (e.g., Morioka et al., 2011;Wainer et al., 2014).The SASD mode has been previously correlated to EOFs of the east-west SLA mode (Grodsky & Carton, 2006), suggesting a connection between the tropics and subtropics via Kelvin and Rossby waves.The SASD variability has been associated with changes in the strength of the subtropical high, which interacts with SST by changing the wind speed pattern and latent heat flux (Sterl & Hazeleger, 2003).
The vertical structure of temperature anomalies and their association with the propagating SLA features are shown in Figure 2. The longitude-time diagram of the upper 500 m temperature anomalies (T500) from ORAS5 reanalysis and SLA (Figure 2a), averaged between 32°S and 28°S, shows that T500 propagates along with the SLA from east to west in approximately 5 years.The vertical structure of these temperature anomalies in two different longitudinal areas, [15-20]°W and [35-40]°W (Figures 2b and 2c), shows that they are generally subsurface intensified and located mostly in the upper 200 m near the base of the mixed layer.This association suggests that SST anomalies are linked to ocean dynamics and heat transport.

Detection of Marine Heatwaves and Cold Spells
Given the in-phase relationship found between the propagating SLA and SST signals in the South Atlantic, in this section we investigate whether this mechanism can serve as a precursor for the frequency and duration of extreme daily near-surface temperature.We apply a cluster analysis to the SSTA during the MHW/CS events, calculated on a 1°× 1°grid, to define the four main clusters of MHWs and CSs in the South Atlantic (Figure S1 in Supporting Information S1).The clusters in which centroids are located in the western South Atlantic are selected to be used as the MHW and CS modes (Figure 3).These modes show similar patterns: large-scale cooling (Figure 3a) and warming (Figure 3b) centered near [30°S, 30°W] with a maximum intensity of approximately 1.5°C . The temporal evolution of their occurrence and intensity shows year-to-year variability with a slight trend toward more frequent MHW events in recent years relative to CS, suggesting that non-linear SST trends may exist in the South Atlantic during the analyzed period.To understand the co-variability of the extreme SST events and the propagating SLA features, the joint variability of MHW and CS cluster modes of Figure 3 is compared with the reconstructed SLA from the CEOF1 mode in the western South Atlantic (Figure 4a).A clear correspondence between the extreme SST events and the propagating SLA mode is observed, in which daily warm SST extremes are associated with high SLA anomalies, and viceversa.To quantify this relationship, a conditional probability is calculated.The probability of MHW and CS events conditional to the CEOF phase is calculated using the Bayes' theorem (Supporting Information S1).Four probabilities are examined, which are calculated from the combination of two binary variables, one associated with the sign of the CEOF1 SLA reconstruction (CEOF+, CEOF-) and the other to the occurrence of extreme daily SST events (MHW, CS).The estimated values of conditional probabilities (Figure S2 in Supporting  (1998( , 1999( , 2004( , 2008( , and 2011( ) and MHWs (2001( , 2002( , 2005( , 2014( , and 2015)), respectively.
Information S1) suggest that the occurrence of extreme SST events and SLA with the same phase, that is, P(MHW | CEOF+) = 82.2% and P(CS | CEOF-) = 81.0%,are much more likely than the events with opposite phase, that is, P(MHW | CEOF-) = 19.0%and P(CS | CEOF+) = 17.8%.The total conditional probability that both MHW and CS events are in phase with the SLA reconstructed from the CEOF1 mode is calculated simply by averaging their two conditional probabilities, P(MHW | CEOF+) and P(CS | CEOF-), which results in ∼82% of the occurrences, suggesting a strong potential for using the propagating mode to predict such extreme events.
In further analysis, the annual metrics of the MHWs and CSs are analyzed (Figures 4b-4d).The time evolutions of the duration, frequency and intensity of the MHW and CS events averaged in the western South Atlantic often show out-of-phase relationships.The years with more intense, frequent and longer MHWs show weaker, less frequent and shorter CS events.Particular years that show such out-of-phase relationships are 2001, 2002, 2009, 2014, and 2015 for more positive (MHW) events, and 1998, 1999, 2004, 2008, and 2011 for more negative (CS) events.During those years with increased MHW occurrence there are approximately five to six events, with an intensity of 1.5-2°C and duration of 10-20 days per event, resulting in more than 60 MHW days each year.A composite analysis of the three MHW metrics for the selected years with increased and decreased MHWs (Figure 5) shows that both events are characterized by a dipole between the eastern and western sides of the basin.Similar patterns are observed for CSs (not shown).Such a pattern suggests an out-of-phase occurrence of SST extreme events between east and west, which can be related to the propagating SLA mode for those years.

Mixed Layer Heat Budget
Here, we explore the potential role of surface heat fluxes and ocean advection in modulating MHWs on interannual timescales, using data from the ORAS5 and ERA5 reanalyses.The mixed layer heat budget is performed in a box within the region of maximum warming in the western subtropical South Atlantic, defined as 35°-46°W/ 27°-33°S (Figure 6a).The seasonal cycle of the mixed layer heat budget (Figure S3a in Supporting Information S1) is dominated by the Qnet term.Qnet warms the ocean during summer (November-March), and dampens the ocean's cooling during winter (May-September).The advection term is small, and vertical advection cools the ocean during fall (March-June), a period in which the mixed layer depth increases from 20 to 180 m.To focus on interannual timescales, monthly anomalies of the heat budget terms (Equation 3) are calculated, and a low-pass filter of 19 months is applied.To understand the spatial contributions of the terms in the mixed layer heat budget, a correlation analysis was performed between ∂T/∂t and the three analyzed components (Figures 6a-6c) by adding the components sequentially at each grid point.The correlation between ∂T/∂t and Qnet (Figure 6a) is high (r > 0.7) particularly in the tropical region north of 25°S, and more modest (0.7 > r > 0.4) in the subtropical region.Adding the vertical entrainment to Qnet does not increase the correlation values considerably, but some improvement is observed in the subtropical region.When horizontal advection is included, the correlations in the subtropical region north of 33°S increase almost everywhere, reaching values of r = 0.9 and above, which shows the importance of the oceanic contribution to changes in mixed layer temperature in the western South Atlantic.
The time evolution of the heat budget terms averaged in the western subtropical South Atlantic (box in Figure 6c) estimated in ORAS5 shows that the atmospheric (Qnet) and oceanic (horizontal advection and vertical entrainment) terms contribute similarly to the SST tendency (Figure 6d).The variability of the residual between the SST tendency and the sum of the atmospheric and oceanic contributions is smaller than the individual components, meaning that the residual does not have a major impact on the heat budget variability in the selected region.Other regions such as the western boundary and south of 35°S show lower correlations between changes in mixed layer temperature and the sum of atmospheric and oceanic terms (Figure 6c), probably due to eddy covariance and mixing terms.
On interannual timescales, the net surface heat flux from ERA5 is mostly dominated by the latent heat flux (Figure 6f) with a regression coefficient of 0.71, compared to much lower values for sensible (0.13), shortwave (0.07), and longwave (0.08) radiation components, and there is a good degree of compensation between shortwave and longwave + sensible radiative fluxes.Most of the contribution of the vertical entrainment term comes from the tendency of the mixed layer term (∂h/∂t), with smaller contributions from Ekman vertical advection and lateral induction terms (not shown).The decomposition of the horizontal advection term into meridional and zonal mean and time-variable components shows that v′Ty is by far the dominant component (Figure 6e).This suggests that meridional velocity is driving the oceanic exchanges in the region.The dominance of the meridional component of temperature advection also holds for the seasonal variability (Figure S3b in Supporting Information S1) and emphasizes the importance of the Brazil Current transport for the mixed layer heat budget.The results above suggest that the relationship between SLA and SST in the interior of the basin is mainly due to variations in advection and latent heat fluxes in the mixed layer (Figures 6d-6f).Using the SLA field for each phase of the propagating feature, we calculated the associated geostrophic velocities from the thermal wind relationship.For better visualization, the SLA was smoothed using a 5 × 5 degree Gaussian filter.According to geostrophy in the Southern Hemisphere, flow is cyclonic (clockwise) around a negative SLA, and anticyclonic (counterclockwise) circulation flows around a positive SLA (see Figure S4 in Supporting Information S1 for full propagation cycle).Therefore, a positive SLA anomaly would favor advection of warm tropical waters ahead of the westward propagating SLA, and cooler subpolar waters behind the SLA anomaly (Figures 7a-7d).Warm anomalies are then generally associated with southward advection and cold anomalies with northward advection.Consequently, the dipolar SST circulation centered over the SLA monopole anomaly (Figures 7e and g) is generated because of the exchange of waters between the subpolar and tropical regions that takes place in the subtropics.In addition, the CEOF1 mode is related to sea level pressure (SLP) anomalies in the center and south of the basin (Figures 7i-7l).SLP anomalies are collocated with SLA from the CEOF1 mode pattern at the center of the basin (90°and 270°phase angles) and have the same sign.A positive SLP anomaly (Figure 7k) may warm the surface locally due to a reduction of winds and cloud cover.It also generates an anticyclonic wind pattern around it, reinforcing the subtropical circulation and increasing the latent heat flux, particularly north of 30°S, thus contributing significantly to the dipole SST pattern.The opposite happens for the reverse phase (90°, Figure 7i).In the western side of the basin, strengthened winds may dampen the positive SSTA (Figures 7k and 7l), similar to what was shown in Sterl and Hazeleger (2003), leading to the transition to the next phase of the oscillation.Composites of monthly SST, SLA, Wind speed, SLP, and MLD are also shown for four selected MHW (2002, 2014) and CS (1999, 2008) events (Figures S5 in Supporting Information S1).During the warm and cold events (Figure S5 in Supporting Information S1), the maximum intensity of SST anomalies is above 1°C.SSH anomalies of the same sign and magnitudes of up to 4 cm are observed in the western basin during these events.In general there are anomalies of the mixed layer of opposite sign relative to the SST, that is, shallower during warm events and deeper during cold events.The shallowing of the mixed layer is generally concomitant with a decrease in near-surface wind speed, and an anticyclonic (counterclockwise) wind anomaly pattern around a high SLP anomaly in the center-west of the basin.Therefore, the shoaling of the mixed layer under warm SST/SSH is a response to the reduction in wind speed and increases in stratification and buoyancy, which are mostly driven by horizontal advection and surface heat fluxes (Figure 6).Conversely, deepening of the mixed layer during cold events is associated with stronger, cyclonic winds, and weaker stratification.Volkov et al. (2019) showed that the tripole SLA pattern in the subtropical North Atlantic is partly driven by heat and freshwater divergence associated with the AMOC.Similarly, we investigate the link between the South Atlantic MOC at four different latitudes and the dominant SLA propagating feature.For this, we use the same methodology that was applied for generating Figure 1, in which the CEOF1 mode is rotated in phase (Equation 2).Similar to a lagged correlation, we calculate the correlation between the real part of PC1 and AMOC strength for each rotation angle θ, and select the angle at which the best correlation is achieved (Figure 8).This is performed with the AMOC time series at the four latitudes (20°S, 25°S, 30°S, and 34.5°S) computed by Dong et al. (2021).The AMOC time series agree reasonably well with PC1 at all analyzed latitudes, with correlations above 0.4.At 34.5°S, the phase relationship was not as clear in the beginning of the timeseries but improved after 2000.The weaker relationship may be due to the fact that the spatial pattern of the CEOF1 mode is more dominant north of 33°S.Indeed, Dong et al. (2021) showed that the AMOC at 34.5°S does not agree well with the AMOC at the other 3 latitudes further north.In addition, we can observe a slight shift toward the combination of higher angles and SLA patterns moving westward as the latitude at which the AMOC is calculated moves farther north (Figure 8), suggesting that coherent anomalies are advected from the south.

Propagating SLA and AMOC
The lagged correlation between the AMOC and SLA reconstructed from CEOF1 mode in the eastern part of the basin (25°S-33°S/10°W-10°E) (Figure 9) shows that the magnitude of the correlation is largest at 30°S (r = 0.63, lag = 4; significant at 90%), followed by 25°S (r = 0.56, lag = 3), 20°S (r = 0.43, lag = 5), and 35°S (r = 0.4, lag = 9).The negative correlations indicate that when the AMOC is stronger, there is a negative sea level anomaly in the east and therefore a positive SLA in the western side of the basin.These correlations are mostly driven by the geostrophic component of the AMOC, with correlations of about 0.55 for 25°S and 30°S, with the Ekman component playing a smaller role.The typical lags are between 3 and 9 months, similar to the timescales of northward propagation defined in Dong et al. (2021).The negative lags suggest that the AMOC leads the propagating pattern and therefore could be a precursor to the propagation.
To further explore links between the propagating mode, the AMOC, and meridional heat transport (MHT), we calculated MHT using the reconstruction of Dong et al. (2021) across 30°S, the latitude with the highest correlation between CEOF1 and the AMOC.We divided the MHT across 30°S into western (west of 43°W), eastern (east of 3°E), and interior contributions, similar to what was performed in Dong et al. (2009) across 35°S.To focus on the upper layer of the ocean, we integrated the heat transport contributions from the surface to 500 m, instead of through the full ocean depth.The mean transport contribution in the west is southward ( 0.59 ± 0.12 PW), and it is northward in the interior (0.74 ± 0.19 PW) and east (0.58 ± 0.16 PW).Following the same procedure described above, the phase of the CEOF1 mode with the highest correlation with MHT anomalies is calculated for each of the three areas (Figure 10).The optimal propagation phase in each area confirms the role of the CEOF1 mode in regulating the amount of heat exchanged between the tropics and extratropics.As such, stronger heat transport southward in the west and northward in the interior correspond to the CEOF1 phase with a positive SLA in the west (Figures 10a and 10b), and an increased northward transport in the east corresponds to a CEOF1 phase with high SLA in the east (Figure 10c).
These results suggest that the 3-5 years propagation period of this mode from east to west could be modulated by and also modulate the AMOC on interannual timescales.

Discussion
In two recent studies, Rodrigues et al. (2019) analyzed the MHW event of 2014/2015, and Manta et al. (2018) analyzed the MHW event of 2017.Using a mixed layer heat budget methodology, they showed that surface heat flux anomalies, including increased shortwave radiation from reduced cloud cover and reduced latent heat loss from weaker winds, were responsible for the onset of marine heatwaves in the region.Both the onsets of the 2014/ 2015 and the 2017 MHWs were associated with the Madden-Julian oscillation (MJO) (Barreiro et al., 2019), which is the dominant mode of atmospheric variability on intraseasonal timescales in the tropics (mainly between 40 and 60 days; Madden & Julian, 1972).As the MJO propagates from the tropics, it affects convection, tropospheric winds, rainfall and surface fluxes (e.g., Matthews et al., 2004).
The role of the ocean has received much less attention than atmospheric teleconnections.Here we show that the large-scale ocean circulation (through the AMOC) is a pacemaker for the interannual occurrence of marine heatwaves in the western subtropical South Atlantic.This relationship was hypothesized here from the strong relationship between western South Atlantic temperature anomalies and sea surface height anomalies on interannual timescales.The SLA variability in the western South Atlantic is highly correlated (r = 0.77) with the main east-west propagating mode of SLA.The propagation of this mode agrees with the first baroclinic Rossby wave mode (Majumder et al., 2019), which is linked to the large-scale ocean adjustment.In the East, our results suggest that this mode is triggered by anomalies in the strength and location of the South Atlantic subtropical height, observed in SLP and wind composites, which can induce anomalies in coastal upwelling in the Benguela Current System region (Sun et al., 2018).SLP anomalies associated with the subtropical anticyclone have also been linked to the South Atlantic's north-south dipole of SST (Sterl & Hazeleger, 2003;Venegas et al., 1997), and our results suggest both SASD and the east-west propagating mode are related to the same forcing.The mechanism for SST propagation seems to be related to the one described in Colin de Verdière and Huck (1999) and te Raa and Dijkstra (2002), in which by crossing the South Atlantic, a warm anomaly induces southward velocity perturbations to the west and northward to the east of the center of the anomaly.This leads to a phase difference between the temperature and velocity anomalies.The perturbed velocities advect warm water southward to the west and cold water northward to the east of the initial anomaly, thereby moving the warm anomaly westward.Our results suggest that, as the mode propagates, it influences the meridional heat transport of eastern and western boundary currents and the ocean interior in different phases (Figure 10).A similar mechanism was also suggested for other western boundary currents (Elzahaby et al., 2021;Zhang et al., 2021).Because the AMOC consists of both boundary flows and ocean adjustment, this mechanism may modulate but also be influenced by AMOC variability.This relationship is supported by the significant correlation between the AMOC and the cross-basin SSH pattern (Figure 10).Therefore, this work highlights the importance of sustained AMOC monitoring for regional climate in the South Atlantic.The AMOC in the South Atlantic has been monitored for more than a decade by in situ observations (e.g., Dong et al., 2009;Meinen et al., 2013), and for almost three decades using satellite observations (e.g., Dong et al., 2015;Majumder et al., 2016).Previous studies have linked the AMOC to SST fingerprints in the South Atlantic (Dima & Lohmann, 2010;Lopez et al., 2016), which allows extending these time series back more than a century.Recently, Bodnariuk et al. (2021) found a link between the propagating modes in the South Atlantic and Indian and South Pacific basins, suggesting that there could be coherence of oceanic features throughout the Southern Hemisphere.Further investigations of links between the propagating SSH modes, the AMOC, and atmospheric teleconnections should be performed using numerical and simplified model studies.

Conclusions
The leading mode of the interannual variability of SLA in the South Atlantic is characterized by westward propagating anomalies centered at 30°S, with a periodicity of 3-5 years.The propagating SLA signals associated with this mode are positively correlated with SSTA in the western subtropical South Atlantic.The temporal phase of the propagating mode is well correlated with interannual modulations of the MHW and CS mode centered in the western South Atlantic.We estimated that there is an 82% probability that the extreme daily sea surface temperature events occur when the SLA associated with the propagating mode bears the same sign.Our analysis shows that SST modulation by SLA in the subtropical region is driven mainly by latent heat flux anomalies and by oceanic horizontal advection.The latent heat flux appears to be related to wind speed anomalies associated with the subtropical high and large-scale atmospheric waves.However, the present analysis cannot address the atmospheric mechanisms and teleconnections that influence the oceanic propagations and the large-scale heat fluxes patterns in the South Atlantic.The oceanic advection term mediates the exchange of tropical (warm) and subpolar (cold) waters in the region (Figure 11): clockwise circulation around a negative SLA anomaly favors advection of cold subpolar waters toward the subtropics, while counterclockwise circulation is generated around a positive SLA anomaly, advecting warm tropical waters into the subtropical region.The analyzed westwardpropagating mode is significantly (at 90%) correlated (r = 0.63) with the AMOC at 30°S, and its origin in the eastern part of the basin lags the AMOC by approximately 3-9 months.Therefore, the sustained AMOC observations in the South Atlantic can provide enhanced multi-year predictability for extreme temperature events in the western side of the basin.

Introduction
The Text S1 of this Supplementary Material describes the concept of the Bayes' Theorem and the methodology used to calculate probabilities of MHWs and CSs occurrences conditional to the western South Atlantic SLA reconstructed by the CEOF1 mode. Figure S1 shows the four clusters calculated for MHW and CSs, with the first column shown the clusters analyzed in this work.Figure S2 shows the result of the posterior conditional probabilities described in Text S1, and Table S1 shows the values of the 1 individual probabilities used to calculate the posterior probabilities.Figure S3 shows the seasonal cycle of the mixed layer heat budget (in W/m 2 ) and the mixed layer depth (in meters) averaged in the western subtropical South Atlantic (46°W-35°W/33°S-27°S).The mixed Layer heat budget is calculated using the monthly ORAS5 reanalysis.Figure S4 shows the full cycle of the main westward propagating SSH mode in the subtropical South Atlantic and associated geostrophic velocity anomalies calculated from the SSH patterns.Figure S5 shows the composites of SST, SLA, wind/SLP and MLD during two MHW events (2002( , 2014( ) and two CS events (1999( , 2008)).

Text S1. Conditional Probability
The probability of an event occurring based on prior knowledge of the conditions that might be relevant to the event is described by the Bayes' theorem.The Bayes' theorem is expressed in the following formula:

P ( A∨B)= P (B∨ A) . P ( A ) P( B)
Where P(A|B) is the posterior probability that event A occurs given that event B has occurred, P(B|A) is the likelihood that both events A and B occur, and P(A) and P(B) are the observed independent probabilities of the events A and B occurring, respectively.
In our particular case, we assume that the variables A and B are binary, in which A represents the MHW (A+) and CS (A-) events, and B represents the CEOF+ (B+) and CEOF-(B-) events.The occurrences of these events are treated as mutually exclusive, thus P(B) can be defined as the sum of all factors affecting B: Therefore, P(B) can be regarded as a normalization factor for the posterior probabilities.Thus, in agreement with the main mechanism examined here, we are interested in analyzing if the alternative hypothesis [P(MHW,CEOF+), P(CS,CEOF-)], in which the CEOF and temperature extremes are in phase, is significantly higher than the null hypothesis of an out-of-phase relationship [P(MHW,CEOF-), P(CS,CEOF+)], and in this case we can reject the null hypothesis.
The events are analyzed over 9681 days, from which MHW events occurred during 651 days and CS events occurred during 635 days.The CEOF+ events occurred during 4922 days, from which 533 MHW events and 118 CS events occurred, and CEOF-events occurred during 4759 days, from which 517 CS and 118 MHW events occurred.The individual probabilities are given in Table 1.

2
Text S2.Seasonal mixed layer heat budget in the box of Figure 6.In Figure S3b, the temperature advection (ADV) across the boundaries of a box is calculated according to this equation:

Volume
Where T h is the averaged mixed layer temperature, T m is the temperature averaged over the mixed layer volume inside the closed domain S, h is the mixed layer depth, and V .n is the velocity vector across each wall of the box.S1.

Figure 1 .
Figure 1.(a) SLA spatial phase (cm) of the first interannual CEOF mode at 45°phase snapshots.(b) Correlation of SST anomalies from 1993 to 2020 with the PC1 timeseries for the respective temporal phases shown on top.SST anomalies were previously low-passed with a 13-month Gaussian filter.Hatched regions show the statistically significant regions according to a double tailed Student t-test method.(c) Timeseries of SST anomalies (red) and SLA reconstructed by the CEOF1 (black) averaged between 33°S and 27°S/46°W and 35°W (black box in top left panel).Dots are the temporal phase (in degrees) from the CEOF mode.

Figure 2 .
Figure 2. (a) Longitude-time evolution of the monthly temperature anomalies from ORAS5 reanalysis averaged between 28 and 32°S and over the upper 500 m (shaded) overlaid by monthly SLA in meters (black contours).Data was detrended, the zonal means were subtracted to show the propagating features, and a 13-month triangular filter was applied to highlight the interannual variability.Time-depth evolution of detrended ORAS5 temperature anomalies averaged between 28 and 32°S and between the longitudinal ranges (highlighted by green shades in (a)) of (b) 15 and 20°W, and (c) 35 and 40°W.The green line in (b) and (c) represents the MLD.Dashed arrows between (b) and (c) highlight the inferred timescale of propagating features between the two longitudinal ranges.

Figure 3 .
Figure 3. Marine (a, c) cold spell and (b, d) heatwave SST anomaly modes related to the western subtropical South Atlantic derived from a cluster analysis.Top panels (a, b) show the spatial maps of the anomalies and bottom panels (c, d) are the respective daily time series calculated from the averaged patterns on panels (a, b).

Figure 4 .
Figure 4. (a) Averaged daily SST anomalies (°C) associated to the western South Atlantic cluster mode of MHW (red bars) and CS (blue bars) overlaid by the time series of the SLA (gray line) in the western South Atlantic reconstructed from the CEOF1 mode.Annual averages of (b) Frequency, (c) Maximum Intensity (MaxInt), (d) Duration, and (e) Days per year of MHW (red) and CS (blue) in the western South Atlantic.The blue and shaded areas are selected years with strong CSs(1998( , 1999( , 2004( , 2008( , and 2011( ) and MHWs (2001( , 2002( , 2005( , 2014( , and 2015)), respectively.

Figure 6 .
Figure 6.(a-c) Maps of correlation between the anomalies of temperature tendency (dT/dt) and the anomaly terms of the mixed heat budget: (a) Qnet, (b) Qnet + ADVW, and (c) Qnet + ADVW + ADV.The black box in (c) represents the region where the heat budget timeseries were calculated (27-33°S/35-46°W).Hatched areas are the locations above 95% significance.(d) Mixed layer heat budget anomalies averaged in the box.Red/blue shading represent the residuals (dT/dt minus Sum).(e) ADV (black) and its decomposition into meridional (VTy) and zonal (UTx) directions divided into eddy and mean contributions (see Equation 5 in Section 2.2.3).(f) Timeseries of the components of the anomalous surface heat flux (Qnet) from ERA5 reanalysis: latent (LHF), sensible (SHF), shortwave (SWR) and longwave (LWR) fluxes.The mixed layer budget terms are labeled mixed layer temperature tendency (dT/dt), surface fluxes (Qnet), horizontal advection (ADV), vertical advection (ADVW) and Sum = QNET + ADV + ADVW.Anomalies are calculated by subtracting the monthly climatology of the monthly averages, and Gaussian low-pass filter of 19 months is applied to the timeseries.

Figure 7 .
Figure 7. Composites for different phases of the propagating CEOF1 mode of SLA.Top panels: CEOF1 spatial pattern at 90°phases shown at the top left of panels a-d.The associated geostrophic current vectors are overlaid.The box of Figure 6 is shown in panel (a).Middle panels: Composites of SST anomalies (°C) with the geostrophic current vectors overlaid.Bottom panels: Composites of surface wind anomalies with the direction given by the arrows and speed (m/s) given by the shades.Overlaid on panels (i-l) are the contours of SLP anomalies (Pa).

Figure 8 .
Figure 8. CEOF1 pattern of SLA in cm (maps) and PC1 and AMOC anomaly (timeseries) associated with the maximum correlation between PC1 mode and the AMOC anomalies at four different latitudes: (a) 20°S, (b) 25°S, (c) 30°S, and (d) 35°S.The maximum correlation between the AMOC and the CEOF1 is estimated by rotating the CEOF1 phase.The AMOC anomalies timeseries are given in Sv and smoothed with a 19-month Gaussian filter.The black box in (a) shows the region where the CEOF1 reconstructed SLA is used to calculate lagged correlations with the AMOC strength (Figure 9).

Figure 9 .
Figure 9. Lagged correlation of the reconstructed SLA in the eastern side of the basin (10°W-10°E/34°S-25°S; box in panel (a)) with the AMOC at four different latitudes (20°S, 25°S, 30°S, and 35°S) in the South Atlantic.For negative lags, the AMOC leads SLA.The dashed lines are the 90% confidence intervals given the adjusted degrees of freedom.

Figure 10 .
Figure 10.Same as Figure 8 but for the (a) West, (b) Interior, and (c) East components of the meridional heat transport (PW) across 30°S integrated between 0 and 500 m.The black arrows on maps show the direction of the increased geostrophic transport in each phase of the CEOF1 mode.

Figure 11 .
Figure 11.Schematic of the effect of the propagating oceanic heat transport on the warming and cooling of the western and eastern South Atlantic regions at two opposite phases.The contours (colored shading) are for SLA and the black horizontal arrows indicate westward propagation of the oceanic features.The red and blue arrows indicate the tendency for warming and cooling of the regions represented by the polygonal regions.

Figure S1 .
Figure S1.Averaged SSTA for the defined four cluster modes for MHWs (top) and CSs (bottom).The number of days and percentage of total days in the analyzed period are shown on top of each panel.The left column shows the clusters selected for the western South Atlantic.

Figure S2 .
Figure S2.Conditional probabilities (%) of MHW and CS events given a particular phase of the SLA reconstructed in the western South Atlantic (box of Figure 6c) using the CEOF1 mode.Conditional probabilities are calculated using the Bayes' theorem (see Text S1), and values are shown in TableS1.

Figure S3 .
Figure S3.Seasonal cycle of the mixed layer heat budget terms in the box of Figure 6c.(a) Deepening of the MLD during fall (Feb-Jun) provides cooling to the mixed layer due to vertical entrainment.A rapid warming due to net heat heating occurs when the mixed layer is shallow during late spring and summer.(b) Seasonal cycle of the horizontal advection directional components into the box of Figure 6c.

Figure S4 .
Figure S4.Full cycle (360° every 45°) of the East-west propagating SSH mode, filtered spatially using a 5° horizontal Gaussian filter, overlaid by the statistically significant calculated geostrophic velocities (black arrows).

Table S1 .
Individual probabilities used in the Bayes' theorem estimation of conditional probabilities.