Over the past decades surface warming in the southern subtropical Indian Ocean (IO) has been greater than that in other oceans. The warming penetrates to a depth of 800 m, in contrast to the off-equatorial surface warming which co-exists with subsurface cooling. We examine the dynamics for this rich structure. Results from the 20th century experiments of the Intergovernmental Panel on Climate Change (IPCC) confirm that the southern subtropical IO surface-to-800 m warming is greater than that in the Pacific and Atlantic Oceans. Outputs from two targeted ensemble sets of coupled model experiments, one with and one without increasing anthropogenic aerosols, show that increasing aerosols strengthen the global Conveyor, and generate a greater poleward shift and intensification of the Agulhas outflow and its retroflection; the process increases the warming rate in the subtropics, and takes heat out of the off-equatorial region generating a cooling.
 Historical data [Rayner et al., 1996] show that since 1950 the warming trend of sea surface temperature (SST) is not uniform across the Southern Hemisphere (SH) subtropical latitudes, but is the largest in the IO (Figure 1a). Such a pattern was not simulated by early climate models, which, under increasing CO2, produced a fast warming rate in the Northern Hemisphere (NH), and a much slower warming rate in the SH subtropical oceans [Cai and Whetton, 2001]. The deficiency was attributed to some long-standing problems including a weakly stratified, overly convective modeled Southern Ocean [Hirst and Cai, 1994; Toggweiler et al., 1989]. Implementation of an eddy-induced transport parameterization [Gent and McWilliams, 1990; Duffy et al., 1995] onto the Cox  isopycnal scheme improved the stratification, however the model problems persisted. The inclusion of increasing aerosol forcing reduces the large warming rate in the NH [Mitchell et al., 1995]. Here we show that aerosol forcing also improves the structure of warming trends in the SH subtropical ocean, particularly the southern IO sector.
 Using a new compilation of historical temperature profiles, the Indian Ocean Thermal Archive (IOTA), Alory et al.  show that the subtropical zonal-mean IO surface warming penetrates to an 800 m depth (Figure 1b), and that the IPCC models running 20th century experiments capture this surface to deep ocean warming trend. The off-equatorial IO surface warming in IOTA is accompanied by a subsurface cooling, which displays two centers, one 10°S at a 100 m depth, and the other 17°S at 400 m. The cooling at 100 m might be due to a greater discharge associated with the stronger El Niños in the post-1980 period, compared with the pre-1980 period [Shi et al., 2007].
 There are, however, several unresolved issues. Firstly, the subtropical gyres are driven by the wind stress curl. The observed curl trend results from decreasing midlatitude westerlies and strengthening high-latitude westerlies as a consequence of a poleward shift of the westerly jet incorporated into the upward trend in the Southern Annular Mode (SAM) [Saenko et al., 2005; Cai, 2006]. The SAM trend over the past decades is related to Antarctic ozone depletion [Thompson and Solomon, 2002; Gillett and Thompson, 2003; Cai and Cowan, 2007]. Yet Alory et al.  found that the IPCC models that include an ozone depletion forcing do not show a greater subtropical IO warming than those without. Thus differences in other forcings must be considered. Secondly, the off-equatorial cooling 17°S at 400 m (Figure 1b) has not been fully explained. Finally, although observations show a greater subtropical surface IO warming, the warming rate at depth relative to that in other oceans is not clear. If surface warming is an indication of deep warming, then Figure 1a means that the subtropical IO deep warming is greater than that in other oceans. This seems to be borne out by the IPCC models. This will be examined.
 Although the majority of the IPCC experiments include an aerosol forcing, there is no output from the same model that singles out the impact of aerosol forcing. Here, we analyze outputs from two newly available ensemble sets, one with aerosols fixed at the pre-industrial level and another that incorporates increasing aerosols through an interactive aerosol scheme [Rotstayn et al., 2007].
2. Data and Model and Experiments
 The atmospheric model is a low-resolution, flux adjusted (spectral R21) version of the CSIRO atmospheric general circulation model with 18 hybrid vertical levels and a horizontal resolution of approximately 5.6° in longitude and 3.2° in latitude. The oceanic component, based on the Cox-Bryan code [Cox, 1987], has the same horizontal resolution but 21 levels in the vertical. The coupled model has a comprehensive aerosol scheme in the simulations from 1871 to 2000, with and without the effects of increasing anthropogenic aerosols. We focus on the period since 1951. An eight-run ensemble with all of these forcings (ALL ensemble) and a further eight runs with all forcings except those related to anthropogenic aerosols (AXA ensemble) were performed. In ALL, the aerosol species treated interactively are sulfate, particulate organic matter, black carbon, mineral dust and sea salt, and their historical emission inventories are used. Other forcings included are those due to changes in long-lived greenhouse gases, ozone, volcanic aerosol and solar variations. The AXA ensemble only differs from that of the ALL ensemble in that the anthropogenic aerosol emissions are fixed at their 1870 levels. A multi-century control climate is used to assess the significance of the difference between ALL and AXA, as described by Cai et al. .
3. Model Evidence of Greater Subtropical Indian Ocean Warming
 The ensemble mean of the IPCC models used in the Alory et al.  study reproduces a stronger surface warming in the subtropical IO than that in other oceans (Figure 1c) and the latitude-depth structure (Figure 1d) with warming penetrating to 800 m. The subsurface cooling 10°S at 100 m does not show up because most models do not produce a greater ENSO discharge after 1980. The contrast between the IO and the Pacific-Atlantic sector is also seen in depth-averaged warming (Figure 2). The ensemble IO warming (black curve) is greater than that in the IOTA (blue line), and the Pacific and Atlantic Oceans (orange curve), however the feature of warming maximum located in the 40°S–50°S latitude band is well simulated. At the latitude of largest warming, 42°S, the standard deviation of the spread is 0.31°C century−1, far smaller than the ensemble trend at this latitude.
 These contrasts between the IO and the Pacific-Atlantic sector are produced in both ALL (Figures 3a–3c) and AXA (Figures 3d–3f), but important differences exist between the two sets. Firstly in ALL, except in the subtropical IO, where the warming is more pronounced, warming trends are generally smaller. This is because the increasing aerosols cool both hemispheric oceans through a strengthening cross-hemispheric transport from the SH oceans to the NH oceans [Cai et al., 2006], particularly to the North Atlantic, mitigating an increasing CO2-induced slowdown of the North Atlantic Deep Water Formation (NADWF). Secondly, the contrast between the subtropical IO deep warming and that in other oceans is sharper in ALL. Finally, the off-equatorial subsurface cooling 17°S at 400 m is only produced in ALL; as will be discussed later, in ALL, the stronger cross-hemispheric heat transport to the NH oceans is derived mostly along the Conveyor pathway of the SH oceans, including the subtropical IO.
4. Dynamics for a Stronger Subtropical IO Warming
 The fundamental forcing of the southern subtropical IO warming is a poleward shift of the southern subtropical gyre. There are two reasons for a stronger warming in the IO than in the other oceans. Firstly, the climatological wind stress curl over the subtropical IO is far greater than that over the other oceans [Cai, 2006, Figure 2 (lower panels)]. This is determined, in turn, by a sharper meridional SST gradient in the subtropical IO, because north of the subtropical latitudes there is significant heat input by the Indonesian Throughflow (ITF). The stronger meridional SST gradient supports stronger meridional gradients of zonal wind, and hence stronger wind stress curls, and the wind-driven subtropical gyre, than in the other oceans. Secondly, the subtropical IO gyre is a part of the global Conveyor pathway, carrying an additional 10 Sv of buoyancy-driven flow of Gordon  in the Agulhas outflow, some of which eventually compensate the NADWF. As a result, the climatological IO subtropical gyre is far larger than in other oceans [Gordon, 1985]. Thus for an equal extent of a poleward shift, a greater warming is generated in the IO than in other oceans. This explains the greater warming in the subtropical IO than in the subtropical Pacific and Atlantic in AXA (see Figures 3e and 3f). However, in AXA the contrast between the IO warming rate and that in other oceans is weak (Figures 3a and 3d).
 The forcing of increasing aerosols improves the IO trend pattern. Firstly, in the off-equatorial IO, the cooling 17°S at 400 m depth is only present in ALL. As discussed by Cai et al.  and Delworth and Dixon , aerosol-induced cooling leads to a more stable NADWF. The upper branch of the Conveyor strengthens in ALL with an increasing ITF and strengthening flows toward the North Atlantic (Figure 4a), as opposed to a weaker global Conveyor and ITF in AXA (Figure 4b). In ALL, the heat for supporting the stronger NADWF is derived along the Conveyor pathway of the SH oceans, including the off-equatorial IO, leading to the subsurface cooling 17°S at 400 m depth (Figure 4a, contour, and Figure 3b). Figure 4a suggests that such a subsurface cooling takes place in the off-equatorial Pacific and Atlantic Oceans as well. By contrast, in AXA there is an accumulation of heat in the off-equatorial region (Figure 4b). Secondly, increasing aerosols sharpen the contrast between the SH subtropical IO warming and that in the other oceans (Figure 4a, contour). This is because the poleward shift and intensification of the subtropical IO gyre are far greater in ALL than in AXA, as further illustrated in Figures 5a and 5c: in ALL the streamfunction trend has a maximum of 4.9 Sv at 45°S, compared with a value less than 1 Sv at 42°S in AXA.
 That the origin of the large warming in ALL lies in the ocean is supported by a heat loss trend in the midlatitude IO where the warming is largest (figure not shown). The poleward intensification of the subtropical IO gyre is so powerful and generates such a large warming that it has to lose heat to the atmosphere. It turns out that a positive air-sea feedback ensues: the stronger Agulhas outflow and retroflection due to the stronger NADWF initially shift the location of the maximum SST gradient and maximum atmospheric baroclinicity polewards, reducing the storm activity and westerlies to the north, and taking less heat out of the ocean there; to the south storms and westerlies increase, cooling the ocean [Inatsu and Hoskins, 2004]. This leads to a greater meridional SST gradient, which generates stronger westerlies polewards and increasing wind stress curl, leading to stronger wind and curl trends in ALL (Figures 5b and 5d). The stronger curl trend in turn supports the strong poleward shift and intensification of the subtropical IO gyre.
 Over the past decades, the surface warming trend in the southern subtropical IO has been greater than that in other southern subtropical oceans. The surface warming penetrates to an 800 m depth, in contrast to the off-equatorial region, where surface warming is accompanied by subsurface cooling. We show that the structure is influenced by an anthropogenic aerosol forcing; with forcing of increasing aerosols, the IO trend pattern is better simulated. This is because increasing aerosols mitigate the global Conveyor from an increasing CO2-induced slow-down, generating a greater cross-hemisphere heat transport. The heat is derived along the pathway of the Conveyor. In association a stronger Agulhas outflow and retroflection are generated. The stronger outflow leads to a stronger warming rate in the subtropical latitudes and takes heat out of the off-equatorial IO, generating the off-equatorial subsurface cooling. We note that the cooling effect of increasing aerosols in our model may be over-estimated as it leads to a warming rate in the SH oceans that is generally too small. However, it improves the spatial structure of warming trends in the southern IO, particularly the generation of off-equatorial subsurface cooling, concentrated warming in the subtropical IO, and its contrast with the subtropical Pacific and Atlantic Oceans.
 This work is supported by the Australian Greenhouse Office. We thank members of the CSIRO Climate Model and Applications Team for developing the model. We thank G. Alory for reviewing the paper before submission, and two anonymous reviewers for their helpful comments.