Projected changes to South Atlantic boundary currents and confluence region in the CMIP5 models: the role of wind and deep ocean changes

The South Atlantic (SA) circulation plays an important role in the oceanic teleconnections from the Indian, Pacific and Southern oceans to the North Atlantic, with inter-hemispheric exchanges of heat and salt. Here, we show that the large-scale features of the SA circulation are projected to change significantly under ‘business as usual’ greenhouse gas increases. Based on 19 models from the Coupled Model Intercomparison Project phase 5 there is a projected weakening in the upper ocean interior transport (<1000 m) between 15° and ∼32°S, largely related to a weakening of the wind stress curl over this region. The reduction in ocean interior circulation is largely compensated by a decrease in the net deep southward ocean transport (>1000 m), mainly related to a decrease in the North Atlantic deep water transport. Between 30° and 40°S, there is a consistent projected intensification in the Brazil current strength of about 40% (30%–58% interquartile range) primarily compensated by an intensification of the upper interior circulation across the Indo-Atlantic basin. The Brazil–Malvinas confluence is projected to shift southwards, driven by a weakening of the Malvinas current. Such a change could have important implications for the distribution of marine species in the southwestern SA in the future.


Introduction
The southwestern South Atlantic (SA) is one of the world's most energetic oceanic regions. This is a result of the meeting of two intense opposing flows: the poleward Brazil current (BC) and the equatorward Malvinas current (MC), to form the most important oceanographic front in the SA: the Brazil-Malvinas confluence (BMC; figure S1).
The BMC strongly influences the local climate and biological environment. The large sea surface temperature (SST) gradient affects local surface winds and heat advection to adjacent coastal areas (De Camargo et al 2013). Frontal zones are characterized by high primary production that enhances the local food chain (Alemany et al 2014).
This study will examine projected changes and driving factors for the SA circulation. The large-scale ocean circulation can, to some extent, be broken down into a surface wind-driven circulation and a deep density-driven overturning circulation. The surface circulation can reach 1000-2000 m depending on the region. Of particular importance are the western boundary currents (WBCs) that transport large amounts of salt, heat, and other tracers poleward. The basin interior surface flow usually opposes the WBC, partially compensating its poleward mass flux. These flows are also affected by air-sea fluxes, inter-basin transport and vertical transport from below associated with the meridional overturning circulation.
In the SA, the BC extends between about 20 and 40°S (figure S1) with considerable transport variation Original content from this work may be used under the terms of the Creative Commons Attribution 3.0 licence.
Any further distribution of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI. along its length. Generally, its transport north of 25°S is less than 10 Sv (Peterson and Stramma 1991) increasing substantially south of 30°S. Depth integrated transport to 1400-1500 m around 38°S might reach 19 Sv based on geostrophic estimates (Gordon and Greengrove 1986). Table S1 summarizes previous transport estimates.
The BMC is situated at around 38°S and shows seasonal variability with meridional shifts of about 5° ( Wainer et al 2000, Goni et al 2011. A number of mechanisms have been proposed to explain the seasonal BMC movements, including variations in BC transport (Olson et al 1988), MC transport (Matano 1993), Antarctic Circumpolar current (ACC) transport (Gan et al 1998), and seasonal wind stress changes (Fetter and Matano 2008). There is also evidence of longer-term change in the confluence location with both observational data (Goni et al 2011) and numerical simulations (Combes and Matano 2014) indicating a southward shift at a rate of between 0.39°a nd 0.81°per decade over the last two decades. The primary drivers of this change remain a topic of research.
Many studies have used coupled atmosphereocean general circulation models (AOGCM) to investigate changes in oceanic and atmospheric circulation in Global Warming scenarios. One robust change evident in both observations (Hill et al 2008, Wu et al 2012 and climate simulations (Wainer et al 2004, Sen Gupta et al 2009 is an intensification and poleward shift of portions of the subtropical gyres associated with an intensification of the WBCs or their extensions. This has been related to long-term changes in the wind stress curl (WSC), associated with changes in ozone and CO 2 forcing (e.g. Saenko et al 2005, Polvani et al 2011. Robust changes in the large-scale atmosphere circulation are also projected by climate models including a widening and slowing down of the Hadley cell and poleward shift of the Southern Hemisphere westerlies (Lu et al 2007, Harvey et al 2014. The aim of this study is to examine the projected changes to SA ocean circulation by analyzing output from the Coupled Model Intercomparison Project phase 5 (CMIP5) models (Taylor et al 2012). In particular, we examine changes in the SA basin circulation in relation to projected changes in surface winds and the large-scale overturning circulation and their consequences for the WBCs and BMC.

Climate models and methods
CMIP5 models include state-of-the-art AOGCM and Earth System Models that were used to inform the Intergovernmental Panel on Climate Change Fifth Assessment Report. Typically, ocean horizontal resolution is approximately 1.0°×1.0°although there is considerable inter-model differences (table S2). As a result, these models do not explicitly resolve mesoscale processes, although the effects of these processes are included through parameterisation. The vertical ocean resolution varies between 31 and 70 layers.
We used output from 19 CMIP5 models (table S2), which have the available variables and scenarios for the current study. To examine projected changes we have made comparisons between the historical simulations that covers 1850-2005 and a high emissions scenario (RCP8.5) that extends the historical simulation to 2100. Historical simulations are forced by observationally derived data of the major climate forcings (greenhouse gases, anthropogenic and volcanic aerosols, ozone and solar irradiance). In the RCP8.5 scenario greenhouse gas forcing increases throughout the 21st century reaching an additional radiative forcing of about 8.5 W m −2 , equivalent to ∼1370 CO 2 -equivalent at the end of the century (Taylor et al 2012). A single ensemble member is used from each model (i.e. the first member with all available variables; usually r1i1p1). Variables from both ocean and atmospheric components were used: mass transport (umo and vmo) and wind stress (tauu and tauv). When mass transport variables were not available, we calculated mass transport from velocity fields (uo and vo). Despite long spin-up integrations some model drift may remain in the subsurface fields, however this is likely small compared to forced trends, particularly when considering ensemble averages (Sen Gupta et al 2016) and so has been neglected.
To estimate the state of the ocean during the 20th century the data was averaged over the period 1900-2000 from the historical simulations. In order to examine the projected changes, the averaged 2050-2100 period from the RCP8.5 simulations was compared with the 20th century state.
The calculation of the WBC transports is problematic as there are differences in their representation across models due to different bathymetries, coastlines, and resolutions. As a result we have manually selected the eastern boundary of the WBC for each model as the longitude where meridional poleward flow and any associated weak offshore counter flow vanishes in the long term mean velocities (see example in figure S2; following Sen Gupta et al 2016). The counter current is included as we are interested in identifying the western limit of the interior flow (that extends across the remainder of the basin) that can be related to Sverdrup transport. In most of the models the BC does not extend deeper than 1000 m, as a result we use this depth to define upper ocean transports. This is consistent with observations reported by Gordon and Greengrove (1986) who showed that most of the transport occurs in the upper 800 m.
The statistical significance of projected changes has been determined by applying the Robust Rank Test (Fligner and Policello 1981) to examine if two distributions have equal medians and are indicated in table S5. Unless otherwise stated, all values presented are multi-model medians (MMM) and associated interquartile range (IQR).

Basin interior and WSC
The upper-ocean interior circulation in the SA is estimated as the meridional volume transport shallower than 1000 m integrated from the eastern edge of WBC to the African coast (or east to the next land boundary south of Africa). The zonally integrated 20th century MMM equatorward interior transport decreases almost linearly towards the north from 27.5 Sv (23.7 to 29.1 Sv) at 30°S to zero at about 16°S (figure 1(a); individual model tranports, inter-quartile ranges and MMM are collated in table S5 for all results). North of this, the interior flow is weakly southwards.
Between 17 and ∼30°S, all but one model project a decrease in the basin interior transport strength. The MMM decrease is statistically significant at the 95% level between 13°S and 32°S (figure 1(b)). The change in the interior transport is largest at 23°S (21°-25°S). In this region, there is a mean projected weakening in transport of ∼3 Sv (2.4-3.4 Sv) corresponding to a 22% decrease (16%-24%). In this same region (between 21°and 25°S), there is no significant change in the BC transport (figures 1(b) and 2(a); see section 3.3). Thus, the MMM weakening in the ocean interior transport is not compensated by an associated weakening of the WBC.
To help understand the cause of the circulation change we examine the zonally averaged WSC integrated over the SA basin (or east to the next land boundary south of Africa; figures 1(c) and (d)). The WSC can be related via Sverdrup theory to the basin interior transport (positive/negative WSC implies a net northward/southwards interior transport). As the CMIP5 models have coarse resolution, we have linearly interpolated the zonally averaged WSC to 0.1°r esolution and estimated the latitude of the maximum WSC by calculating the WSC weighted mean latitude based on latitudes where WSC was greater than 50% of its maximum. The WSC maximum is located around 38.4°S in the models (36.4°-39.4°S; figure 1(c)). Comparison with observations (ERA-Interim reanalysis from 1979 to 2013) shows that the latitude of the MMM WSC maximum is biased polewards by ∼2.9°( figure 1(c)). This is despite the fact that further south there is a distinct equatorwards bias in the maximum zonal winds, consistent with previous studies (i.e. Barnes andPolvani 2013, Bracegirdle et al 2013).
All . The latter identified a 2°southward shift in the circumpolarly averaged WSC, associated with a doubling of CO 2 in a single climate model. Figure 1(b) reveals a strong correspondance between interior and Sverdrup transport changes (north of the southern tip of Africa), associated with a WSC weakening over this region (figure 1(d)). A strong inter-model correlation (r =0.8, p<0.05) between Sverdrup transport at 25°S and the associated model interior transport changes indicates that intermodel differences in the projected changes are well explained by differences in the projected surface wind changes (table 1).

Deep circulation
The 20th century deep meridional transport (integrated net transport below 1000 m) is almost constant along the SA (figure 1(a)) with a mean transport of −15.9 Sv (−18.2 to −12.9 Sv). Given that there is a net projected change in the upper ocean circulation (i.e. the BC and interior changes do not compensate each other north of the southern tip of Africa) and the Atlantic Basin is closed to the north, there must be a compensating change in the deep ocean circulation. Indeed, all models project a weakening of the deep transport in the 21st century. The net MMM SA deep circulation decreases significantly by approximately 3.4 Sv (∼2-5 Sv; figure 1(b)) corresponding to a 19% weakening (14%-28%).
We examine whether this weakening can be related to changes in the southward flowing North Atlantic deep water (NADW) and northward flowing Antarctic bottom water (AABW) entering the Atlantic basin at 33°S. To quantify this we have defined NADW as the southward transport below 1000 m between the South American coast and 40°W ( figure 2(b)). The AABW transport is defined as any northward flux on the western side of the basin (west of 20°W) and below 3000 m ( figure 2(b)). Based on these definitions, the NADW transport is −20.8 Sv (−22.8 to −19.3 Sv; figure S3) for the historical period. This value is similar to the estimated by Garzoli et al (2015) of −19.5 Sv (5.4) based on observations at 35°S. All but one model (NorESM1-M) suggest that NADW transport will decrease by 2.4 Sv (1.7-3.5 Sv; table S5) corresponding to a 13% reduction (8%-14%).
The AABW transport for the 20th century is 5.5 Sv (3.8-8.5 Sv; figure S3). While there is a MMM projected decrease of 0.6 Sv (−1.4 to 0.1 Sv), the change is not statistically significant with 13 out of 19 models projecting a weakening.
As such, the NADW plays the dominant role in the net deep southward transport decrease and in balancing most of the mismatch between surface interior and BC transport changes. It is interesting that most models also show a net northward projected change at about 10°W and 25°W (figure 2(c); albeit considerably weaker than the WBC change). This may relate to a weakening in parts of the NADW that have separated away the South American coastline ( figure 2(b)) as a result of interaction with Vitoria-Trindade Ridge at 20°S (Garzoli et al 2015).

Western boundary currents
All models simulate the BC between approximately 20°and 40°S. The MMM maximum transport (based on each model's maximum transport which occurs at different latitudes) is -20.1 Sv (−23.0 to -13.2 Sv). Here, we have applied the same interpolation as for WSC (see section 3.1) for the integrated transport along the western boundary. The maximum transport latitude is 33.3°S (31.8-34.9°S, figure S4). Observational estimates around this region lie within the model range (19.2 Sv, Stramma 1989;16 Sv, Lentini et al 2006, 23 Sv, Garzoli 1993, see figure S4 and table S1). There is a projected increase in the strength of the BC along its length (figures 1(a) and (b)). Between 20°a nd 30°S, the BC intensification is weak and not statistically significant, even with 14 of 19 models projecting an increased transport. South of 30°S all models agree on the intensification although the magnitude varies considerably. Averaged between 30°and 40°S the increase ranges between 4.8 and 8.1 Sv (IQR) with a MMM increase of ∼6 Sv. This corresponds to a substantial 40% (30%-58%) increase in mean transport.  The projected change is greatest at around 39°S, poleward of the maximum mean state BC transport, with an intensification of 8.8 Sv (∼5-11 Sv; figure 1(b)). In addition, 18 out of 19 models show a southward shift in the latitude of the maximum BC transport of ∼1.2°( 0.7°-1.5°; figure S4). At 33°S, where the historical BC transport is strongest, the MMM projected intensification in the BC is largely compensated by a decrease in the NADW transport (figures 2(a) and (c)). However, even though the upper interior MMM change is not significant, inter model differences in BC transport are strongly related to differences in interior transport (r=0.81, p<0.05; table 1). At 38°S, close to where the BC intensification is strongest, the projected MMM BC intensification is associated with a complex set of changes to NADW, interior, other deep ocean transports and Indonesian throughflow (ITF) (figure 2(a)). However, the inter-model differences in the BC change is most strongly associated with changes in the upper interior transport across the Indo-Atlantic basin (r=0.95, p<0.05, table 1).
The North Brazil current (NBC) is the northward flowing branch of the South Equatorial current. In the historical simulations it shows an approximately linear increase from ∼20°S to ∼11°S reaching up to 20.5 Sv towards the north (17.5-26.7 Sv; figure 1(a)). This range lies within observational estimates at 11°S of 23-26 Sv (Hummels et al 2015; table S1). In the region north of 15°S the 21st century simulation shows a significant (>95%) decrease on NBC strength ( figure 1(b)). The mean transport for 15°S-20°S is ∼19 Sv (16-24.3 Sv; table S5). All but two models project a decrease in transport with a MMM of ∼2.5 Sv (−4.7 to −1.4 Sv; table S5). At 10°S, the projected NBC decrease is largely compensated by changes in the NADW (∼3 Sv; figure 2(a)) with a strong inter-model correlation (r=0.95, p<0.05; table 1).
At the southern extent of the Atlantic basin the powerful Malvinas current flows equatorwards fed by part of the ACC that migrates northwards as a result of topographic steering after transiting the Drake Passage (DP; figure S1). In the historical simulations the MC has a transport of 37.4 Sv (30-50.4 Sv) averaged between 50°S and 42°S. All but two models project a decrease in transport with a reduction of 3.7 Sv (6.7-1 Sv). This corresponds to a weakening of 8% (−12% to −3%). At 45°S the MC is projected to decrease by ∼5 Sv, largely associated with a weakening of the upper ocean interior transport integrated across all basins (r=0.95, p<0.05; figure 2(a); table 1).

Brazil-Malvinas confluence (BMC)
Next we quantify the projected changes to the BMC and examine drivers that may give rise to changes in its position. We define the BMC position as the latitude where the meridional transport vanishes along the western boundary of the Atlantic basin. In order to better determine this latitude, the integrated western boundary meridional transport data was linearly interpolated to 0.1°. The simulated BMC is located at 39.6°S (38.8°-41.6°S; figure 3(a); equivalent to the latitude of zero transport in figure 1(a)-red solid line). By comparison the observed location is 38°S (from 1993 to 2008- Goni et al 2011, from 1992to 2007-Lumpkin and Garzoli 2011. As such the BMC is systematically simulated too far to the south. In the warming scenario, all models project a southward shift in the mean BMC position of 1.2°(0.7°-1.6°; figure 3(a)). To help understand the cause of the BMC latitudinal change we examine the various drivers that have been proposed in the literature: basin wide WSC (Lumpkin and Garzoli 2011), BC transport (Olson et al 1988), MC transport (Matano 1993, Combes andMatano 2014), and DP transport (Gan et al 1998). To examine the influence of the WSC, we look at the projected change in the latitude of the WSC zero line. Sverdrup theory suggests an interior divergence and an associated zonal flow fed from the WBCs at this latitude. To analyze the influence due to BC and MC transport, the meridional transport at latitudes 5°n orth and south, respectively, of each model's 20th century BMC position was related to the BMC position. For latitudes closer than this there may be aliasing of the two current as they can both exist at the same latitude (see schematic figure S1). The DP transport is calculated at 68°W.
Changes in the WSC zero line (which shifts southwards in all models), BC transport (which increases in all models) and MC transport (which decreases in 16 out of 19 models) are all consistent with the robust southward shift in the confluence position. However, based on a multiple linear regression (MLR) analysis using these potential drivers as predictors of mean BMC latitude (table S3) we find that the MC transport plays a primary role in explaining intermodel differences in the mean BMC position (r=0.79, p<0.05), followed by BC transport (r=0.5, p<0.05). This is in agreement with Matano (1993) who suggested that the BMC position is mainly related to the MC strength.
MLR is also used to examine inter-model differences in projected BMC latitude changes (table S4). Only the MC transport change has a significant relationship with the BMC position change (r=0.49, p<0.05; figure 3(b)). As such, intermodel differences in the projected shift in BMC can be partly associated with differences in projected MC transport changes. Adding the other variables as predictors of BMC change provides little additional benefit for predicting inter-model differences in the confluence position change (table S4). We would note that south of the BMC at 45°S there is no relationship between MC transport and the strength of the DP transport (although there is a weak relationship further south r=0.44, p=0.05 at 50°S). As noted, there is a strong relationship between projected MC change and the change in interior transport across all basins (r=0.95; table 1; figure 2(a)), however the interior transport changes are unrelated to WSC changes via Sverdrup dynamics (r<0.1).

Discussion and conclusions
Sverdrup theory relates WSC changes to changes in the basin interior meridional transport, in the absence of deep density driven circulation changes. Here we find that a projected decrease in the WSC between 15°S and 30°S indeed explains the ∼20% reduction in northward upper ocean interior transport. However, there is no significant reduction in the WBC transport at this region, and the reduced interior transport is largely compensated by a reduction in the deep (>1000 m) southward transport. This is consistent with a weakening of NADW transport resulting from high latitude North Atlantic warming and freshening (Dong et al 2014), and with the projected northward anomalies in the abyssal circulation ( figure 2(c)).
The southward shift in the WSC zero line and the weakening in the upper-ocean basin interior transport are likely to have important consequences to the world's ocean dynamics. The change in the WSC pattern drives a polewards shift of the frontal zone between the subtropical gyre circulation and the ACC, that leads an increased Agulhas leakage (Biastoch et al 2009). However, this change in the Agulhas leakage may not reflect an increased inter-hemispheric heat transport, as there is a projected decrease in both basin interior and North Brazil current volume transports: the primary pathways of this heat exchange.
There is a consistent and substantial intensification of the BC of almost 40% between 30°and 40°S. North of the southern tip of Africa the relatively weak BC increase is largely compensated by the NADW reduction with inter-model differences mainly associated with large differences in upper ocean interior transport related to changes in WSC. South of Africa, despite a decrease in the water entering the combined Indo-Atlantic region via ITF (figure 2(a); Sen Gupta et al 2016) there is a very large increase in the BC. This increased BC transport is primarily compensated by increases in the upper ocean interior Indo-Atlantic transport. However, changes in the Indo-Atlantic WSC and associated Sverdrup transport cannot explain these interior changes (intermodel correlation is not significant, table 1). This is unsurprising as this latitude intersects the location of the Agulhas current retroflection.
The location of the BMC is projected to shift southward by ∼1.2°. The relationship between the ACC and MC transport, which can modulate the BMC position, and surface winds is not straightforward (Fetter andMatano 2008, Lumpkin andGarzoli 2011), especially due to the existence of a complex topography in the Drake Passage region. Despite southward shifts in the WSC pattern, we found that intermodel differences in the projected BMC southward shift location only showed a significant relationship with MC strength. This is consistent with findings of Combes and Matano (2014), who reported an observed southward shift of 0.62°decade −1 between 1993 and 2008 and related this to a weakening of the northern branch of the ACC and a consequent reduction of the MC transport. Moreover, results are consistent with Matano (1993) who used idealised experiments to show that the BMC is displaced northwards of the WSC zero line by a strong MC, with BMC location modulated by the strength of the MC transport.
A key uncertainty in this analysis lies on the coarse resolution of the CMIP5 models. The models are unable to simulate the mesoscale processes that are important in this highly energetic region. Despite this, the gross changes described here should be constrained by basin wide wind changes. Indeed a similar analysis for the East Australian current (Oliver and Holbrook 2014) found very little difference between CMIP5 projections and projections from an eddy permitting model with regard to changes in the long-term transport of the WBCs. There is also considerable uncertainty with regards to the future evolution of the winds over the Southern ocean. This is related to compensating effects of greenhouse gases and the ozone forcing (e.g. Gillett and Fyfe 2013).
Our findings may have important consequences for regional climate and ecosystems. BC intensification can potentially increase heat and salt transport along the BC pathway, modifying the SST gradient that can affect regional weather in coastal areas (Pezzi et al 2005, De Camargo et al 2013. Additionally, Prado et al (2016) observed a decrease in the occurrence of temperate marine mammals between 30°and 34°S since 1970s, where we project a robust change in the BC. Thusly, the changes can potentially modify the distribution of marine species and affect economically local fishery production (e.g. Andrade 2003, Alemany et al 2014.