1. Introduction

The wintertime storm tracks, the regions where synoptic-scale eddy kinetic energy (EKE) and poleward temperature flux are maxima, have significant zonal asymmetries in both hemispheres. There are many discussions of the Northern Hemisphere (NH) winter storm tracks, centered in the oceanic basins (e.g., Blackmon et al. 1977; Chang et al. 2002). This zonal asymmetry in the NH storm tracks can be explained by a variety of zonally asymmetric terrestrial conditions such as tropical and midlatitude sea surface temperature (SST) distributions, the Himalayas, and the Rockies. In spite of more zonally distributed SSTs and less topography in the Southern Hemisphere (SH; Fig. 1), the winter storm track there still exhibits marked zonal asymmetry. Figure 2a shows the 300-hPa EKE for synoptic, 2–8-day¹ periods derived from the European Centre for Medium-Range Weather Forecasts (ECMWF) Re-Analysis (ERA-15) June–July–August (JJA) daily data for 1979– 93. The zonal-mean synoptic-scale EKE has a maximum near 45°S. In two dimensions, the largest 300-hPa synoptic-scale EKE region (hereafter the major storm track) stretches from the central Atlantic to the eastern Indian Ocean in the latitudinal band 40°–50°S. The smallest EKE region in these latitudes extends from southeast of Australia to the Straits of Magellan (50°S, 70°W), with a minimum near New Zealand (NZ; 40°S, 170°E). Trenberth (1991) previously described such an overall picture based upon ECMWF data.

To clarify the structure, cross marks have been added in Fig. 2a to show the local maxima of synoptic-scale EKE in each longitude. In Australian longitudes, the major storm track seems to split into two minor storm tracks, one extending to the northeast of Australia (25°S, 150°E) and the other toward the Ross Sea (75°S, 170°W). Chang (1999) describes this feature in terms of the waveguides of baroclinic eddies detected by the application of the one-point lag correlation (Lim and Wallace 1991). He finds that a well-defined waveguide splits into two in the eastern Indian Ocean with a primary branch joining up with the subtropical jet and a secondary one spiraling poleward. Farther to the east, the subtropical branch continues along the equatorial edge of the small EKE region and eventually connects with the major storm track after the abrupt break in the region of the Andes (40°S, 70°W). Taking an overview, the shape of the SH storm track is that of a spiral that starts north of NZ, runs across the South Pacific, breaks over Chile, restarts over the Patagonian Plateau, peaks in the Indian Ocean, decays along the edge of Antarctica, and ends at the Ross Sea. Applying the state-of-the-art tracking technique, B. J. Hoskins and K. Hodges (2004, unpublished manuscript, hereafter HH04) find such a spiral in the density of the tracks of lower-tropospheric cyclonic systems.

The theoretical understanding of the zonal asymmetries in the SH winter storm track is far from complete. The large-scale terrestrial forcing that could be responsible can be classified into tropical SSTs, Antarctica, and a variety of midlatitude forcings. Tropical SST is used to characterize the warm pool from the Indian Ocean to the western Pacific and colder water in the eastern Pacific (Fig. 1). It is known that zonal variations in tropical SST are effective through their influence on convection in forming extratropical stationary eddies in both hemispheres (e.g., Horel and Wallace 1981; Karoly 1989). Recently Sinclair et al. (1997) and Solman and Menéndez (2002) have suggested that during the El Niño events baroclinic eddies are reduced in the southwestern Pacific and preferentially occur in the subpolar waveguide. Tropical SST asymmetries may therefore be an important agent for the SH storm track. Concerning Antarctica, it appears possible that the zonal wavenumber-1 forcing associated with its off-polar location could be important. However, Quintanar and Mechoso (1995) stated that its effect on the SH atmosphere is confined near the South Pole and is negligible away from Antarctica.

The forcing in midlatitudes contains a number of different aspects; in particular, midlatitude SST, the South African Plateau, and the Andes. First, midlatitude SST forcing would be important if the meridional SST gradient directly controls the baroclinicity above it (Hoskins and Valdes 1990). In SH winter, midlatitude SSTs have a clear zonal wavenumber-1 structure with the largest meridional gradient to the south of Madagascar (40°S, 45°E; Fig. 1). A close relationship between the sharp SST front and the SH storm track is suggested by Nakamura and Shimpo (2004), who show that this relationship holds on both the seasonal and interannual time scales. Inatsu et al. (2002, 2003) also showed this relationship in an aquaplanet experiment with zonal wavenumber-1 SST gradient forcing centered at the storm track axis. Second, effects of the Andes have frequently been investigated. Cyclogenesis (cyclolysis) is prominent down- (up-) stream of the Andes (Sinclair 1995; Gan and Rao 1994; HH04). Another aspect that may be relevant to cyclogenesis is the moisture transport by northerly winds in Brazil caused by the blocking effect of the Andes (James and Anderson 1984). Finally, an effect of the South African Plateau is hinted at by the results of HH04, whose analysis of the genesis of cyclone systems shows cyclogenesis on the west coast of South Africa possibly feeding the development of baroclinic eddies in the Indian Ocean.

The extent to which the various tropical, midlatitude, and polar forcings create zonal asymmetry in the SH winter storm track is the subject investigated here. Following the above discussion, we investigate four aspects: tropical SSTs, midlatitude SSTs, the South African Plateau, and the Andes. Antarctica is not considered separately. In order to measure the effect of one surface condition, atmospheric general circulation model (AGCM) experiments are performed without each of the conditions in turn and comparison is made with the control. In this paper, we concentrate on austral winter, as the SH summer storm track has less zonal asymmetry.

Section 2 gives the experimental setup in this paper. In section 3, the upper-tropospheric EKE storm track in the experiments is discussed. The mechanism by which tropical SSTs are able to influence the storm track is investigated in section 4. Section 5 contains results for the lower-tropospheric storm track. Finally a summary and some discussion are presented in section 6.

2. Experiments

The AGCM used is the Hadley Centre Atmospheric Model, version 3 (HadAM3), with a 2.5° latitude by 3.75° longitude grid and 19 levels in the vertical. (See Pope et al. 2000 for more details.) In all of the experiments, the model was integrated for 9 model yr after the spinup (with the seasonal solar cycle included), and then a daily mean dataset was archived.

The runs performed were the control (CTR) with climatological SSTs and full mountains (Fig. 1), and five experiments named ZTS, ZMS, NMT, NSA, and NAD, to examine, respectively, the effect of removing tropical SST asymmetries, midlatitude SST asymmetries, SH topography except Antarctica, the South African Plateau, and the Andes (Table 1). The effect of each surface condition could be assessed by comparison of the relevant experiment without this aspect with the CTR run. In the ZTS run, the SSTs within 20°S–20°N are replaced by the zonal average of their climatology, beyond 35° they are climatological, and in the latitudinal range 20°– 35° there is a smooth transition. In contrast, the ZMS run has zonally symmetric SST south of 35°S and climatological SSTs north of 20°S, with a smooth transition between these latitudes. The NMT run has no mountains (zero topographic height) between 60° and 15°S, with a smooth change in surface height from 15°S to the equator. In this run, the SSTs are climatological, and the roughness is specified as zero where mountains have been removed. The NSA and NAD runs are the same as the NMT run, except that only the mountains in the South African and South American continents, respectively, are removed.

3. Upper-tropospheric storm-track results

Figures 2c–f, 3, and 4 show the simulated synoptic-scale EKE and stationary eddies at 300 hPa in SH winter. The CTR run (Fig. 2c) has an EKE that is generally about 10% weaker but captures the large-scale and most of the detailed features of the observed upper-tropospheric storm track (Fig. 2a), including the spiral structure from Australia to the Ross Sea (cross marks in Figs. 2a,c), and the largest EKE from the central Atlantic to the western Indian Ocean within the band 40°–50°S. The stationary eddies in the CTR run (Fig. 2d) are also quite similar to those observed (Fig. 2b). South of 40°S, the largest amplitude stationary cyclone is centered near 50°S, 60°E and extends from the Atlantic to the Indian Oceans. Two stationary anticyclones are centered southwest of NZ (50°S, 160°E) and over the Bellingshausen Sea (60°S, 100°W). The anticyclone near NZ is sandwiched between two stationary cyclones in the subtropical and polar regions, corresponding to a strong subtropical jet over Australia and a modest subpolar jet near Antarctica (not shown). These are downstream of the single jet in the Indian Ocean along the northern edge of the large stationary cyclone. Given its realism, the CTR run will be regarded as a suitable basis for comparison with the other experiments.

The largest change in the asymmetric structure is found in the ZTS run with zonally uniform tropical SST (Fig. 2e), and this will be discussed first. The major storm track has a broader latitudinal scale and starts farther eastward. The reduction in the EKE near NZ is now weaker, and the minima are now much farther east. Taking a broad view, the storm track in this experiment has the appearance more of a loop rather than a spiral (cross marks in Fig. 2e). The changes in stationary eddies are consistent with the fact that they are largely responsible for the storm-track changes. The amplitudes of zonal wavenumbers 1 and 2 are much smaller, and the positions of the stationary cyclones and anticyclones have changed drastically (Fig. 2f). The largest cyclone in the south Indian Ocean associated with the broader major storm track has disappeared. The strong anticyclones near NZ and over the Bellingshausen Sea have weakened and merged, consistent with the associated weaker minimum in EKE. As a result, the subpolar jet intrudes into the central Pacific, and there is little diffluence of the jet in the eastern Indian Ocean.

In contrast, in the runs without aspects of the midlatitude forcing, the stationary eddies are generally quite similar to those in the CTR run (Figs. 2d and 3; the NSA and NAD runs are not shown). The extended cyclone in the Atlantic and Indian Oceans and the NZ and Bellingshausen Sea anticyclones are still present in all these experiments. As shown in Fig. 4, the spiral storm track structure is also maintained (cross marks in Fig. 4). However, the storm track is significantly modified. First, considering the case with zero zonal asymmetry in midlatitude SST (the ZMS run), the EKE in the Indian Ocean decreases by 10%–20%, and that off Chile increases by 10%–20% (Fig. 4a). The EKE minimum region near NZ and the other features are retained in this run. Second, in the absence of all the SH mountains except those in Antarctica (the NMT run), the zonal asymmetry in EKE is much reduced, but the maximum and minimum of EKE are not shifted (Fig. 4b). The spiral storm-track structure is also somewhat reduced (cross marks in Fig. 4b). That this is probably a combined effect of the South African Plateau and the Andes can be seen from the other two experiments. Removing only the South African Plateau (the NSA run), the major storm track starts in the central Atlantic as in the CTR run, but it reaches a peak south of South Africa and is weaker in the Indian Ocean (Fig. 4c). This is consistent with there being less cyclogenesis in the absence of the South African Plateau. The main impact of removing the Andes instead (the NAD run) is that the start of the major storm track in the Atlantic is much weaker (Fig. 4d). However, it still attains a clear peak in the central Indian Ocean, as in the CTR run. This is consistent with there being less cyclogenesis downstream of South America in the absence of its topography.

In summary, only the removal of zonal asymmetries in tropical SST drastically changes the stationary eddies. These changes appear to give associated storm track changes, in particular removing the EKE minimum region near NZ and broadening the EKE maximum in the Indian Ocean. In contrast, midlatitude forcing has only a modest effect on the stationary eddies, but it is still important in the buildup of the peak of the major storm track in the central Indian Ocean.

4. The tropical SST asymmetric forcing mechanism

The focus in this section is on how tropical SST asymmetric forcing affects the stationary eddies and thus the storm track in the SH. Tropical SST zonal asymmetries give asymmetries in convective heating and the associated vertical circulations. The divergent flow due to the convection can act as the source of Rossby waves that propagate into the extratropics. To investigate this, we will compare the ZTS and CTR runs using two diagnostics: the wave activity flux (Takaya and Nakamura 2001) and the Rossby wave source (RWS) ideas (Hoskins and Sardeshmukh 1988). Definitions and meanings of these diagnostics are given in the appendix, along with a discussion of their significance.

A variety of diagnostics for the CTR run are shown in Fig. 5. The stationary eddy 200-hPa streamfunction is given in Fig. 5a. Comparing with Fig. 2b, the cyclone is again found in the southern Indian Ocean and the anticyclone south of NZ. Now, however, the strong westerlies in the Indian Ocean are seen to be associated more with the strong anticyclone in the tropical region and the weak westerlies in the NZ region also with a cyclone to the northwest of it. The wave activity flux vectors in Fig. 5a indicate that wave activity propagates from the eastern equatorial Indian Ocean and the anticyclone there to the higher latitude cyclone, eastward to the anticyclone south of NZ, and then equatorward to the cyclone region. The enhanced westerlies in the Indian Ocean and the reduced westerlies in the NZ region are then seen as being associated with a Rossby wave train that originates in the tropical Indian Ocean and follows an almost great circle path poleward, eastward, and then equatorward.

To obtain a view of the tropical asymmetric circulation, Fig. 5b gives the divergence and divergent flow vectors at 200 hPa. Generally, there is divergence above the deep convection that is predominantly in the NH, with a divergent flow toward the southern winter hemisphere. However there is much enhanced divergence above the northern summer monsoon regions of Africa, Southern Asia, and the Americas. In particular, the southward divergent flow vectors in the southern tropical Indian Ocean are prominent. To investigate how these could indicate the origin of the Rossby wave train, Fig. 5c gives the divergent flow vectors and absolute vorticity contours for this region. It is clear that there is a large divergent flux of negative absolute vorticity out of the region. As discussed in the appendix, this is the integral of the RWS over the region and, in the SH, constitutes a source of anticyclonic vorticity, consistent with the anticyclone there seen in Fig. 5a. It is also seen in the total streamfunction shown in Fig. 5c. In the frictionless, steady balance described in the appendix, the rotational flow must flux negative vorticity into the region on average. However, given the larger magnitude of the rotational flow and its large cancellation between vorticity fluxes into and out of the region that is apparent from the full streamfunction contours in Fig. 5c, the net flux would require numerical calculation.

These ideas discussed with reference to a Rossby wave train forced by summer monsoonal asymmetries north of the equator in the Indian Ocean region are supported by the diagnostics of the ZTS run given in Fig. 6. The stationary eddy streamfunction (Fig. 6a) is much weaker and has a totally different structure than in the CTR run. The wave activity flux (Fig. 6a) in the eastern Indian Ocean sector has reversed direction, suggesting high-latitude forcing of the tropical pattern. The upper tropospheric divergence field (Fig. 6b) indicates much-reduced asymmetries in tropical convection and the associated divergent wind predominantly reflects the zonally averaged Hadley cell structure. The small remaining zonal asymmetries in convection are probably produced by the land–sea contrast in the NH. The difference in the magnitude of the divergent winds in the Indian Ocean sector, particularly away from the equator, is clear (Fig. 6c), and the integrated RWS as given by the divergent flux of absolute vorticity out of the region is much smaller.

From the diagnostics of these two runs presented here, it is apparent that the asymmetries in tropical heating generate a Rossby wave pattern that leads to the dominant zonal asymmetries in the SH upper-tropospheric storm track through its production of elongated strong westerlies in the Indian Ocean and weak westerlies near NZ.

5. Lower-tropospheric storm track

Indications of the origin of the zonally asymmetric structure of the lower-tropospheric storm track, as deduced from the various experiments, will now be discussed. The basic field used for this discussion is the 850-hPa meridional flux of temperature by synoptic time-scale (2–8-day period) eddies (hereafter denoted VT850). For ERA-15 and the model control, CTR, VT850 fields are given in Figs. 7a,b. The latter also shows the regions of large meridional SST gradient in shading. The major storm track is again seen to start in the Atlantic, reach its maximum in the western Indian Ocean, and decrease in the eastern Indian Ocean. Eastward of this, latitudinal maxima spiral in toward Antarctica, with a local maximum occurring there near the date line. In the upper troposphere, there is also a secondary storm track in the eastern Pacific near 35°–40°S. These features are again quite well captured by the model, CTR, though with maximum amplitudes reduced some 15%. The exception is the eastern Pacific secondary storm track, which is not represented well enough for deductions to be made from the experiments.

Zonal tropical SSTs (ZTS shown in Fig. 7c) do not produce such a marked change here as they did in the upper-tropospheric storm track (Figs. 2c,e). The largest change is that the major storm track maximum is broader meridionally. As was true in the upper troposphere, this is consistent with the lack in this case of the upper anticyclone to the north and the cyclone to the south, which give the strong upper westerlies in between in CTR (Figs. 2d,f). There is an indication of a slightly less abrupt end to the storm track south of Australia, but the behavior in the NZ sector is generally similar to that in CTR. The upper-tropospheric changes associated with the absence of the propagating Rossby wave also have a signature in the lower-tropospheric storm track, but it is not as strong as in the upper troposphere.

Making the midlatitude SSTs zonal, ZMS makes the storm track much more zonally uniform (Fig. 7d). The elongated maximum in the Indian Ocean is lost and the maximum amplitude is considerably reduced. This is consistent with the fact that the control has a meridional SST gradient of more than 12 K/10° latitude south of Madagascar (45°E), which is about 3 times larger than that to the east of NZ. As in the upper troposphere, west of Chile the storm track is slightly enhanced in this case. Despite the lack of the region of maximum SST gradient that might have been thought to be associated with the storm track spiraling in toward Antarctica, the ZMS run still seems to show this feature and also the date line maximum.

Removing the Andes results in little qualitative change in the storm track VT850 (NAD; Fig. 7f). However, as in the upper troposphere, the intensity is somewhat reduced downstream of the Andes, in the Atlantic through to the eastern Indian Ocean maximum. This is consistent with the cyclogenesis regions found by HH04 in the lee of the Andes. Also, as was the case in the upper troposphere, removing the South African topography, the NSA run (Fig. 7e) gives similar reductions in magnitudes downstream of it and a maximum value now occurring some 30° farther east than in CTR. Also in support of this, the CTR 850-hPa temperature field given in Fig. 8 shows much-enhanced baroclinicity on the western side of South Africa as well as near the Andes (the region below the topography should be ignored).

In summary, the most important ingredient in the asymmetry of the major lower-tropospheric storm track is the midlatitude SST distribution. This is in agreement with the previous studies of Inatsu et al. (2002, 2003) and Nakamura and Shimpo (2004). However, tropical asymmetries influence the structure of the Indian Ocean maximum in the major storm track and its abrupt end. The topography of South America and South Africa increases the intensity of the storm track downstream of each of them. The date line maximum near Antarctica is present in all of the experiments, which suggests it may be associated with Antarctic topography.

6. Discussion and conclusions

From the experiments diagnosed in this paper, it has been deduced that Indian Ocean Rossby wave forcing associated with the zonal asymmetry in tropical SSTs creates the SH stationary eddies that are the primary control of both the major minimum and maximum of upper-tropospheric synoptic-scale EKE. Midlatitude forcings give small changes to the stationary eddies but are important for the storm tracks. In particular, midlatitude SST asymmetries are very important for the lower-tropospheric storm-track distribution as measured by the meridional heat flux. The Andes contribute to their downstream storm track intensity through cyclogenesis in their lee. More surprisingly, cyclogenesis on the western side of the South African Plateau is also important in this regard.

Referring to Fig. 9, the arguments given above have been in terms of the passive response of the storm track to stationary eddies forced by tropical-SST-related convection (the left and top solid arrows) or directly to terrestrial forcings (the two right solid arrows in Fig. 9). However, there are some important feedback issues (the dashed arrows in Fig. 9) that should be recognized.

First, the storm-track activity in general has a feedback onto time-mean zonal wind (arrow A in Fig. 9). The thermal effect, as indicated by VT850, is to weaken the lower-tropospheric mean baroclinicity, particularly the largest values imposed by the SST distribution. Following Hoskins et al. (1983), the mechanical forcing by the storm-track eddies can be diagnosed using E = (υ′² − u′², − u′υ′), where the prime denotes synoptic-scale variability. The divergence (convergence) of E indicates rotational forcing consistent with westerly (easterly) wind acceleration by synoptic-scale eddies. Figure 10 shows the zonal wavenumber-1 and -2 components of the barotropic zonal wind and of the synoptic-scale E and its divergence. It is evident that the eddy activity tends to accentuate the asymmetries in the basic flow. As usual, the baroclinic damping and barotropic forcing of the mean flow give a mixed negative and positive feedback on the coupled asymmetric mean flow and storm track.

The mean low-level zonal wind tendency due to synoptic-scale eddies (Fig. 10) may be expected to affect the ocean circulation in the fully coupled atmosphere– ocean system (arrow B in Fig. 9). This implies the feedback loop suggested in some previous studies (arrows A, B, and C in Fig. 9; e.g., Hoskins and Valdes 1990; Watanabe and Kimoto 2000). Enhanced zonal wind drives enhanced ocean gyres, and at the western boundary the stronger confluence of the ocean currents creates larger SST gradients, which in turn creates stronger synoptic-scale eddies. In the SH, the surface zonal wind leads to the large SST gradient off Argentina where the Malvinas and Brazil Currents are confluent. This can be viewed as being the origin of the larger SST gradient in the Atlantic Ocean that has been shown to be important particularly for the lower-tropospheric storm track.