Multidecadal trends in the large-scale ocean circulation are influenced by changes in radiative forcings such as long-lived greenhouse gases, volcanic aerosols, and solar irradiance. Model simulations suggest that anthropogenic aerosols can also force circulation changes, including delaying the weakening of the North Atlantic thermohaline circulation, altering the interhemispheric sea surface temperature gradient, and inducing a pan-oceanic heat redistribution. The extent to which aerosols from different regions contribute to these oceanic changes is currently unknown. Using specifically designed 20th century coupled climate model experiments that separate Asian and non-Asian aerosol impacts, it is shown that the non-Asian aerosol component, predominantly sulfate aerosols, accounts for much of the simulated aerosol-induced oceanic changes. These include delaying the weakening of the global meridional circulation, increasing the northward heat transport across the equatorial Atlantic, and inducing a subsurface cooling in the subtropical southern Indian Ocean. As global sulfate aerosol levels peaked in the 1980s, these trends may be starting to reverse. This study highlights the importance of Northern Hemisphere non-Asian anthropogenic aerosols in driving remote changes in Southern Hemisphere subtropical and extratropical oceans.
 Since the advent of coupled climate models, modeling groups have sought to use individual radiative forcing experiments for detecting and attributing long-term changes in the atmosphere and oceans [e.g., Huber and Knutti, 2012; Levitus et al., 2012]. While changes due to greenhouse gases are well understood, the atmosphere and ocean circulation response to anthropogenic aerosols is less well quantified. This is due to the fact that model simulations of aerosols and their radiative impacts are, by and large, poorly constrained by observations, making their radiative impact quite uncertain [Forster et al., 2007]. Despite this, many studies have shown that anthropogenic aerosols can induce changes to regional monsoonal rainfall. For example, black carbon can induce an intensification of pre-monsoonal rainfall across India [e.g., Lau et al., 2006], while sulfate aerosols are known to affect the land-sea temperature contrast, leading to shifts in convection systems across East Asia, and possibly Australia [e.g., Rotstayn et al., 2007].
 Volcanic and anthropogenic aerosols (predominantly sulfates) are associated with a direct oceanic cooling, due to the reflection of incoming shortwave radiation [Delworth et al., 2005]. As most anthropogenic emissions originate from the Northern Hemisphere (NH) [Stern, 2006], aerosols also induce an interhemispheric asymmetry in surface ocean temperatures [Cai et al., 2006; Chang et al., 2011]. A direct effect of this hemispheric surface imbalance is a pan-oceanic heat transport from the Southern Hemisphere (SH) to the NH, via a relatively “stronger” global Conveyor [Cai et al., 2006], compared to the case without increasing anthropogenic aerosols. The redistribution of heat from the midlatitude band 10–35°S northward leads to a major cooling in the subtropical and midlatitude South Atlantic [Cai et al., 2006]. This, in turn, draws heat along the pathway of the Agulhas outflow (south of South Africa) and the Indonesian Throughflow, inducing the cooling in the SH subtropical Indian Ocean [Cai et al., 2007], seen in observations [Alory et al., 2007; Levitus et al., 2012]. Through this process, anthropogenic aerosols delay the greenhouse-gas induced weakening in the Meridional Overturning Circulation (MOC) [Delworth and Dixon, 2006; Cai et al., 2006; Collier et al., 2013], as well as delaying the total warming of the ocean by reducing the net shortwave radiation reaching the surface [Delworth et al., 2005]. Scattering aerosols such as sulfates have also induced changes in the interhemispheric sea surface temperature (SST) gradient across the tropical Atlantic [Chang et al., 2011], shifting southward of the tropical Atlantic Intertropical Convergence Zone. This is somewhat consistent with modeling results from an atmosphere/mixed layer ocean model, in which aerosols induce a northward flow of atmospheric energy flux from the SH to the NH tropics [Ming and Ramaswamy, 2011], to compensate for the interhemispheric radiative imbalance caused by NH aerosols. Anthropogenic and volcanic aerosols are also strong drivers of North Atlantic SSTs on multidecadal timescales [Booth et al., 2012].
 However, many questions remain. For example, the extent to which large-scale oceanic changes can be attributed to aerosol emissions from particular regions (e.g., Asia, Europe, North America) is currently unknown. This is important as aerosol emissions across the NH have slowed and possibly peaked during the late 20th century [e.g., Streets et al., 2006, Philipona et al., 2009]. We utilize targeted experiments from a coupled general circulation model (GCM) to assess the relative importance of anthropogenic aerosols from Asian and non-Asian sources on historical large-scale ocean circulation trends.
2 Model Set-up and Experiment Description
 The utilized GCM is a low-resolution (spectral R21) coupled atmospheric-ocean model (CSIRO Mk3 R21, referred to as Mk3 R21), with a horizontal resolution of 5.6° (longitude) × 3.2° (latitude) for both the atmosphere and ocean. The atmosphere contains 18 hybrid vertical levels, while the ocean has 21 vertical levels. It includes schemes such as dynamic sea-ice and interactive aerosols [Rotstayn et al., 2007], with reflective anthropogenic aerosol species such as sulfates, and absorbing aerosol species such as black carbon and particulate organic matter, as well as natural aerosols such as mineral dust, sea salt, and stratospheric aerosols from volcanoes. The Mk3 R21 includes both the direct and indirect aerosol effects and is coupled to a temporally evolving ocean. As a comparison to the Mk3 R21, we also examine the CSIRO Mk3.6 coupled GCM, which has an ocean spatial resolution of 1.875° (longitude) × 0.9375° (latitude) and includes an updated radiation scheme, upgraded ocean model, and an interactive aerosol scheme with up-to-date aerosol forcings (more details are given in Rotstayn et al. ). To test the robustness of the impact of aerosols on the subsurface temperature evolution, we also analyze experiments from the IPSL CM5A-LR (low resolution) coupled GCM with an ocean spatial resolution of 2° [Dufresne et al., 2013]. To isolate the role of aerosols, three ensembles of eight (five) simulations from the Mk3 R21 (Mk3.6) are examined for the period 1901–2000 (1901–2005): the first set contains all temporally evolving global radiative forcings, including both natural (volcanic aerosols and solar irradiance) and anthropogenic (greenhouse gases, stratospheric ozone depletion), except for global anthropogenic aerosols (all forcings except aerosols; referred to as AXA). The second ensemble set includes all radiative forcings included in AXA with the inclusion of time-evolving anthropogenic aerosol emissions across Asia (defined as 0–45°N and 70–160°E; referred to as AS). The third ensemble contains all forcings included in AXA with the inclusion of both Asian and non-Asian (global) anthropogenic aerosols (all forcings; referred to as ALL). The spatial distribution of anthropogenic aerosols between the different experiments can be viewed through the net shortwave radiation at the surface (Figure S1, supporting information), which shows Europe, eastern USA, central Africa, East Asia, and India as anthropogenic aerosol hot spots. For this study, we focus on the annual mean changes of the ensemble of each set of eight simulations (five for the Mk3.6), calculated as 100 year linear trends (105 years for the Mk3.6). To assess the significance of these trends, we use a t test, based on the standard error of the eight individual simulations.
 The Mk3 R21 has certain deficiencies in its aerosol scheme, including emissions of black carbon that are too low relative to other models used in a model intercomparison (subsequently, emissions of black carbon were increased by 50% in the Mk3.6) [Rotstayn et al., 2010]; despite this the Mk3 R21 performs well at simulating the 20th century global mean temperature change [Cai et al., 2006; Rotstayn et al., 2007]. In the Mk3 R21, the estimate of the net aerosol radiative forcing is −1.2 W m−2 (between the preindustrial period to 1990): the direct effect is estimated to be −0.39 W m−2 while the indirect effect is −0.8 W m−2. This is well within the range of −1.8 to −0.3 W m−2 given in the Fourth Assessment Report of the Intergovernmental Panel on Climate [Forster et al., 2007]. Some of the major features of global aerosols in the Mk3 R21 include strong increases in sulfur and black carbon emissions after 1940 during a period of industrialization [Rotstayn et al., 2007, Figure 1]. During the late 1970s and early 1980s, there was a sharp decline in the global sulfur emissions, predominantly over Europe and North America [Stern, 2006, Figure 3]. Across Asia, there was a continued growth in sulfur emissions up to the late 1990s, followed by small decline since that decade [Streets et al., 2006]. This has led to a hypothesis that declining emissions of non-Asian aerosols may have led to the recent regional warming of the North Atlantic since the 1980s [Chang et al., 2011; Booth et al., 2012]. As such, this study focuses on the relative importance of Asian versus non-Asian aerosols in driving long-term trends in the ocean circulation.
3 Aerosol-Induced Changes in the MOC
 The Atlantic Ocean is a crucial component of the global Conveyor, in which heat is transported via near-surface water from the South to the North Atlantic, where it sinks in the process of North Atlantic Deep Water Formation (NADW) [Cai et al., 2006]. Over time, this water upwells across the Pacific and Indian Oceans, with the Pacific water flowing through the Indonesian Throughflow and across the tropical Indian Ocean, through to the Atlantic via the Agulhas outflow, thus completing the global loop. While the global Conveyor is a conceptual model of a more complicated system [Lozier, 2010], it has been shown, using the coarse Mk3 R21, that aerosols induce an intensification of the Conveyor, leading to an aerosol-induced increase in the Atlantic MOC [Cai et al., 2006]. How much of that increase is due to Asian aerosols? Figure 1 compares MOC trends in the Atlantic and Indo-Pacific basins between the three ensembles. In the AXA ensemble, there is a broad weakening of the Atlantic MOC of 1.5 Sv over the 100 year period (Figure 1a), which is larger in the Mk3.6 (Figure S2, supporting information); across the Indo-Pacific, there is a decreased northward flow (blue) in the upper 200 m from 0–20°N, offset by weak and broad increased northward flows in the subsurface layers (Figure 1b). The addition of Asian aerosols (AS; Figures 1c and 1d) only slightly changes these trends, and the differences in the MOC in the Atlantic and Indo-Pacific between the AS and AXA ensemble are not significant. Only the addition of non-Asian aerosols (ALL ensemble) suppresses the Atlantic weakening seen in the AXA ensemble (cf. Figures 1a and 1e) and subsurface Indo-Pacific northward flow (Figure 1f). Similar results are seen in the Mk3.6 (Figures S2e and S2f), where aerosols induce an oceanic circulation response that leads to local increases in SSTs and sea surface salinity north of 45N [Collier et al., 2013].
 To evaluate the long-term evolution of the meridional flow in the Atlantic, we compute a thermohaline circulation (THC) index, based on the maximum stream function of the north Atlantic MOC (between 20–80°N; [Delworth and Dixon, 2006]). The simulated Atlantic THC index displays strong decadal variability in all three experiment ensembles (Figure 2a, thick lines) and commences to decline in the AS ensemble prior to 1900 and at around 1920 in AXA (figure not shown). In contrast, the ALL ensemble shows very little change in the MOC index throughout the entire 20th century, supporting the notion that non-Asian aerosols are crucial for delaying any weakening of the MOC that may be generated by increased greenhouse gases [Delworth and Dixon, 2006; Cai et al., 2006]. Of interest in the ALL ensemble is the sharp decline in the THC index around 1990. This may be a delayed response caused by the decline in sulfate emissions from non-Asian sources in the late 1970s, as this feature is not seen in the AS or AXA ensembles. The THC evolution in the Mk3.6 is similar to that shown by the Mk3 R21 with the ALL ensemble displaying a strengthening trend from 1960 onward (Figure S3a, supporting information).
 The northward heat transport across the equatorial Atlantic for each of the ensembles (Figure 2b) confirms the minor role of Asian aerosols. The addition of Asian aerosols (comparing AXA to AS) does not lead to an increase in northward heat transport, which shows a weak long-term decrease in both ensembles, consistent with a small reduction in the NADWF. The further addition of non-Asian aerosols (comparing ALL to AS) leads to enhanced northward heat transport, offsetting the suppression due to greenhouse gases. The northward heat transport also increases in the Mk3.6 ALL ensemble, while century-long declines are seen in the AS and AXA ensembles (Figure S3b). The physical interpretation of the aerosol effect, according to Delworth and Dixon , is that aerosols reduce the heat flux into the mid- to high-latitude oceans, or enhance the heat loss out of the deep water formation region, leading to a cooling of the North Atlantic, destabilizing the water column and enhancing the THC and northward heat transport. The increased heat transport to the NADWF region leads to an oceanic heat content increase and an enhancement of heat loss to the atmosphere [Cai et al., 2006]. In agreement, a modeling study by Ming and Ramaswamy  with aerosol forcing (although without isolated emission realizations) simulated a northward transport of moist static energy in the atmosphere across the entire SH to 20°N.
4 Changes in Southern Tropical Ocean Temperatures
 One conclusion reached by Cai et al.  was that the southern subtropical warming of the Indian Ocean, observed at the end of the 20th century [Alory et al., 2007; Cai et al., 2007], can be explained by a poleward shift in the subtropical gyre, in part contributed by NH aerosols. The shift led to a region of warming centered in the midlatitude oceans (35–50°S) and subsurface cooling in the subtropics. Both AXA and AS produce widespread heating of the SH midlatitude subsurface below 800 m (Figures 3a and 3c). A large contributor to the hemispheric wide midlatitude warming is the Indian Ocean (Figures 3b and 3d); however, a pattern of subtropical subsurface cooling, as seen in observations in the late 20th century [Alory et al., 2007], is not simulated. The emergence of a cooling signal amid the general warming trend in the subsurface tropical waters of the SH is only simulated when non-Asian aerosols are included (Figures 3e and 3f); this feature is also seen in the Mk3.6 (Figure S4, supporting information). Thus, Asian aerosols alone cannot generate sufficient cooling in the NH to promote cross-equatorial heat transport, thereby not promoting the loss of heat from the subtropical Indian Ocean as part of the intensification of the global Conveyor. However, this also raises the question as to whether these processes will reverse as a result of declining emissions of global anthropogenic aerosols.
5 Evidence of a 1980s Reversal Due to a Decline in Non-Asian Aerosols?
 The removal of aerosols, predominantly reflecting species such as sulfates from the climate system, is predicted to have significant impacts on regional and global warming [Stern, 2006]. For example, rises in surface temperature have been noted across Europe [Philipona et al., 2009] and China [Streets et al., 2008] since the 1980s, with decreasing aerosols thought to play a role. Using a GCM, Mickley et al.  showed that the removal of aerosols across the United States during the mid 21st century results in an annual mean surface temperature increase of 0.4–0.6 K across the eastern states. For the ocean, Chang et al.  suggested that a reversal of trends commenced in the cross-hemispheric Atlantic meridional SST gradient during the 1980s, due to a decline in sulfate aerosols. Similar multidecadal changes have been seen in North Atlantic SSTs, with anthropogenic aerosols being implicated as the dominant driver [Booth et al., 2012].
 Here using the Mk3 R21, we examine if similar patterns in the subsurface ocean, emerging since the 1980s when global sulfate emissions peaked and started to decline, are present [see Rotstayn et al., 2007, Figure 1]. Figure 4 shows the evolution of the temperature zonally averaged over the subsurface cooling region in the subtropical southern global ocean (region of cooling in Figure 3e). This region covers 10–25°S, 100–400 m incorporating the majority of the observed cooling region [Cai et al., 2007]. A weak cooling is observed in two observational estimates: Ishii09 [Ishii and Kimoto, 2009] and Levitus09 [Levitus et al., 2012] (Figure 4, right columns); however, the observed temperature evolution exhibits substantial interannual variability. As shown in Figure 3, both AXA and AS in the Mk3 R21 do not simulate a long-term cooling in the southern subtropical global ocean over the 20th century, whereas the ALL ensemble shows a weak cooling. To test the robustness, we analyze the CSIRO Mk3.6, as well as the IPSL CM5A-LR, that also has both ALL and AXA experiments (six and four runs, respectively). All three models show a similar evolution in their AXA and AS ensembles (IPSL CM5A-LR did not run AS simulations) with a century-long subsurface warming (Figures 4a–4c and 4, middle column), punctuated by small interannual declines. The late 20th century declines around the early 1980s and 1990s may be a result of stratospheric aerosol injections after volcanic eruptions. The three models show a similar evolution in their ALL ensembles up to the mid-20th century, followed by a multi-decade long cooling or a hiatus period (Figure 4, right column). The warming commences again after the mid-1980s and appears to accelerate in the IPSL CM5A-LR after 1990, highlighting a lag between the aerosol forcing and the oceanic response. The simulated warming follows the observed trajectory, which continue to the end of the 2000s. A common response for the three models is a slowdown in the subsurface cooling trend during the 1980s, followed by a warming, which is suggested to be in part due to the decline in non-Asian aerosols (although this cannot be tested in the IPSL CM5A-LR due to the lack of an AS experiment).
 Since the mid-20th century, emissions of Asian aerosols have been steadily climbing, despite global sulfate emissions peaking in the 1980s. Previously, it has been shown that the global-scale response of the oceans to anthropogenic aerosols included a pan-oceanic heat redistribution along the pathway of the global Conveyor. A key finding of this study is that Asian aerosols contribute very little to such processes. This may be because the forcing is too localized and not widespread on the continental scale (compared to aerosol emissions from Europe and North America). As such, Asian aerosols do not mitigate the greenhouse gas-induced weakening of the MOC, or generate a subsurface cooling in the southern subtropical oceans. Thus, any future declines in Asian aerosols may not have the same impact on large-scale ocean changes as declines for industrialized nations outside Asia have had on such trends. Furthermore, this study suggests that observed recent warming in the southern subtropical global ocean subsurface at the start of the 21st century may be in part a result of the reduction in global aerosol emissions, manifested through changes in the THC and subsurface temperatures; the latter of which is a robust feature in three climate models forced with global aerosols. These results highlight the need for further studies to increase our understanding of the potential impact of declining aerosols on oceanic circulation, which have historically mitigated the influence of increasing greenhouse gases.
 This work is supported by the Australian Climate Change Science Programme. We thank Ariaan Purich and Evan Weller, and the two anonymous reviewers for their constructive suggestions, which greatly benefited the manuscript. We particularly thank Leon Rotstayn and Martin Dix for conducting the experiments using the CSIRO Mk3 R21 model.
 The Editor thanks Nick Dunstone and an anonymous reviewer for their assistance in evaluating this paper.