Historical and Projected Changes in the Southern Hemisphere Surface Westerlies

The Southern Hemisphere (SH) surface westerlies fundamentally control regional patterns of air temperature, storm tracks, and precipitation while also regulating ocean circulation, heat transport and carbon uptake. Wind‐forced ocean perturbation experiments commonly apply idealized poleward wind shifts ranging between 0.5 and 10 degrees of latitude and wind intensification factors of between 10% and 300%. In addition, changes in winds are often prescribed ad hoc as a zonally uniform anomaly that neglects important regional and seasonal differences. Here we quantify historical and projected SH westerly wind changes based on examination of CMIP5, CMIP6, and reanalysis data. We find a significant reduction in the location bias of the CMIP6 ensemble and an associated reduction in the projected poleward shift compared to CMIP5. Under a high emission scenario, we find a projected end of 21st Century ensemble mean wind increase of ∼10% and a poleward shift of ∼0.8° latitude, although there are important seasonal and regional variations.

in the SH midlatitudes (Grise & Polvani, 2014;Sigmond et al., 2011;Son et al, 2008Son et al, , 2010. While ozone concentrations are expected to recover in the future, the westerly winds are projected to continue to shift poleward and intensify based on high emission climate model experiments. Under these conditions, the effect of greenhouse gases is expected to dominate the opposing influence of ozone recovery in the latter half of the 21st Century (Thompson et al., 2011). Hence, understanding the impact of changing westerly winds on the ocean circulation remains an ongoing focus of research.
Several studies using ocean and coupled climate models ranging from coarse to eddy permitting resolutions have been conducted in the past to understand the influence of projected 21st Century poleward intensification of the surface westerlies on the Southern Ocean and Antarctica (e.g., Bronselear et al., 2020;Delworth & Zeng, 2008;Frankcombe et al., 2013;Spence et al., 2014). Although there have been a few studies that have used regionally and seasonally varying surface forcing (e.g., Bronselear et al., 2020), the majority of studies apply an idealized zonally symmetric intensification and/or poleward shift in the westerly winds in the SH extratropics (generally between 40°S and 60°S). These prescribed changes cause significant impacts on various features of the SH, including the distribution of projected sea level rise (Frankcombe et al., 2013), subsurface warming and circulation changes around the Antarctic continental margin (Spence et al., 2014). However, the applied wind changes tend to be idealized and ad hoc, with no common protocol for applying these wind perturbations to ocean models, including the chosen magnitude of the wind shift and its intensification.
To examine the effect of future changes in surface westerlies, 30-40 previous studies that we are aware of (and possibly many others that we are unaware of) have applied a broad range of poleward shifts and intensifications, with the poleward shift ranging between 0.5° and 10° latitude and wind intensification factors ranging from 10 to over 300% (e.g., Delworth & Zeng, 2008;Downes et al., 2017;Frankcombe et al., 2013;Hogg et al., 2017;Spence et al., 2014;Waugh et al., 2019). Given the wide range of perturbations that have been applied in past studies, some guidance regarding a reasonable estimate of the past and projected changes in the location and strength of the westerly winds in the SH is needed to better facilitate model intercomparison.
In this study, we analyze the historical and projected intensification and poleward shift in the SH surface westerlies across an ensemble of models from the Coupled Model Intercomparison Project 5 & 6 (CMIP5 and CMIP6) along with reanalysis products. We also examine the seasonality and regional variations in these wind stress changes. These details are important for correctly simulating certain aspects of change in the ocean and in Antarctic sea ice. We also examine whether reanalysis products can be used to provide a reliable estimate of the forced anthropogenic change in SH surface westerlies over the last few decades.

Data and Methods
Surface monthly averaged zonal winds (at 10m elevation) from the CMIP5 (Taylor et al., 2012) and CMIP6  archives as well as reanalysis products are used to examine the latitude and strength of the SH surface westerlies. Ocean model simulations employ surface winds to calculate both the surface wind stress and air-sea turbulent heat fluxes; both are primary boundary conditions for ocean models. Surface winds also determine sea-ice advection and wind-driven mixed layer deepening and are therefore central to ocean-sea-ice model forcing fields.
Data spanning 1850 through to 2099 from the first ensemble from each of multiple CMIP5 and CMIP6 models are used to provide equal weight to each climate model. Data from preindustrial control simulations (200-year runs from 27 CMIP5 and 23 CMIP6 models), historical simulations (1850-2005 for CMIP5 and 1850-2014 for CMIP6), and future projections ( -2099( for CMIP5 and 2015( -2099 for CMIP6) are used in this study (Tables S1 and S2). For the future projections, data from both the intermediate emissions scenario (Representative Concentration Pathway (RCP, Meinshausen et al., 2011) 4.5 for CMIP5 and the Shared Socio-economic Pathway (SSP, O'Neill et al., 2016) 245 for CMIP6 and the high emissions scenario (RCP8.5 for CMIP5 and SSP585 for CMIP6) are analyzed. Both SSP585 (SSP245) and RCP8.5 (RCP4.5) scenarios are designed so that radiative forcing increases by 8.5 W/m 2 (4.5 W/m 2 ) by 2100 relative to preindustrial times, although the emission rates of various greenhouse gases are different while achieving the same radiative forcing by 2100 (O'Neill et al., 2016). The differences in high emissions and moderate emissions scenarios arise because of differences in the projected concentrations of greenhouse gases, aerosols and stratospheric ozone.
Reanalysis data sets from 1979 to 2019 for monthly averaged surface zonal winds (at 10m elevation) from the European Centre for Medium Range Weather Forecasts (ECMWF) Re-analysis (ERA5, Hersbach et al., 2020), Modern-Era Retrospective analysis for Research and Applications, Version 2 (MERRA-2, Gelaro et al., 2017), and the Japanese reanalysis (JRA-55, Kobayashi et al., 2015) are also analyzed. Because of sparse measurements over the Southern Ocean before the satellite era, reanalysis data before the year 1979 are not considered as they do not provide a reliable estimate of the westerly wind changes over the SH. Even though satellite measurements of winds only started in the late 1980s, satellite measurements of other physical quantities helped to appreciably improve the quality of the reanalysis products post 1979. Therefore, the reanalysis wind fields from 1979 on are used in this study. We also considered the National Centre for Environmental Prediction-National Centre for Atmospheric Research (NCEP-NCAR) reanalysis (Kalnay et al., 1996), however, in agreement with Marshall (2003), we found that this data set contains spuriously large trends in high latitude Southern Hemisphere winds (Figures S1 and S2) that are inconsistent with station-based observations. Both the NCEP-NCAR and MERRA-2 reanalysis products show stronger trends in comparison to ERA-5 (or JRA-55) trends, primarily in the South Pacific Ocean basin ( Figure S2). Close agreement was found between ERA-5 and JRA-55 (Figures. S1 and S2) for all analyses presented in this study; hence for simplicity we only present results from the ERA5 reanalysis. All data are first mapped to a common 1° × 1° latitude-longitude grid before conducting the analyses shown below.
The maximum jet strength is defined as the maximum surface zonal wind at each longitude in the SH extratropics between 30°S and 70°S (consistent with the definition of Bracegirdle et al., 2013). The position of the westerly jet is then defined as the latitude where the maximum zonal surface wind speed is located at each longitude between 30°S and 70°S.

Historical Era
A poleward shift and an intensification of the SH surface westerlies is found over the last few decades in both models and reanalysis (Figures 1a and 1b). This poleward intensification projects onto a positive trend in the Southern Annular Mode (SAM, Figure S3)-the dominant mode of variability in the circulation of SH extratropics which relates to a seesaw in atmospheric pressure difference between subtropical and polar southern latitudes. Based on single forcing experiments, this change has been attributed primarily to stratospheric ozone depletion, with greenhouse gases playing a secondary role (Thompson et al., 2011).
CMIP5 and older generation climate models are known to have a large equatorward bias ( Figure 1a) in the zonal mean location of the SH surface westerlies  possibly due to biases in the shortwave cloud forcing in the models as compared to reanalysis (Ceppi et al., 2012). Biases in the shortwave cloud forcing can induce surface temperature anomalies in the midlatitudes which affect the meridional temperature gradient, which in turn affects the mean latitude of the westerlies. Negative biases in shortwave cloud forcing lead to equatorward biases in the latitude of the westerlies. There is a notable reduction in the equatorward bias (compared to ERA5) in the zonal mean location of the maximum SH surface westerlies (see also Bracegirdle, Holmes, et al., 2020) reducing from 1.3° in CMIP5 models down to 0.3° in the CMIP6 multimodel mean, averaged over 1979 to 2005. While the bias has been reduced, two-thirds of models still have a zonal maximum situated further north than the reanalysis estimate ( Figure 1a). In contrast, the CMIP5 multimodel mean (MMM) has an almost identical mean strength for the SH surface westerlies as compared to ERA5, while the CMIP6 MMM is 4% too strong (see Figure 1b). When limiting this intergenerational CMIP comparison to include just the subset of models that are common to both CMIP5 and CMIP6 (i.e., 12 models; see Tables S1 and S2), we again find a significant reduction in the equatorward bias (reduced bias of ∼0.7° latitude in CMIP6 MMM; Figure S4a). In contrast, we do not find any significant intergenerational difference in the strength of SH surface westerlies between CMIP5 and CMIP6 ( Figure S4b).
Studies examining the ocean response to historical changes in surface winds usually rely on atmospheric reanalyses for their forcing fields. However, changes over the relatively short reanalysis period may be strongly influenced by internal climate variability and may be a poor representation of the anthropogenic forced change. To test if the recent (1979-2019) trends in the zonal mean location and strength in the ERA5 reanalysis lie outside the range of internal climate variability, a Monte-Carlo analysis was carried out by calculating trends over large numbers of random contiguous 41-year periods from the 200-year preindustrial control simulations of 50 CMIP models (27 CMIP5 and 23 CMIP6; Figure S5). This test assumes that the model variability is representative of the observed internal climate variability. The trend in the location of the SH westerlies calculated from the ERA5 reanalysis lies well within the distribution of trends associated with internal variability. However, the trend in the strength of the westerlies is unlikely to be explained by internal variability alone (P < 0.1). Given the model differences in the representation of internal variability we repeat the analysis using individual CMIP5 and CMIP6 models. Similar results are obtained in more than 90% of the models for both the position and strength of the surface westerlies ( Figures S6-S9). A seasonal analysis further finds that trends in both position and strength and for both model generations are significant in summer (DJF, Figures S10 and S11). This is consistent with the strong influence of stratospheric ozone depletion in the summertime, as discussed by   (1901-1925 average) and future (2075-2099) for both CMIP5 and CMIP6 models and the thick line in the middle of the bars represent the multimodel mean. Swart et al., (2015) for CMIP5 models. In all other seasons and for both metrics, the reanalysis trends cannot be robustly distinguished from internal variability, although the larger autumn westerly strength trends are much closer to being significant compared to winter and spring trends ( Figure S11b). This is consistent with recent pacemaker model simulations by Schneider et al. (2015) and D. Yang et al. (2020), who found that a substantial component of recent multidecadal westerly wind variability could be accounted for in model experiments forced by observed tropical SST variations, independent of anthropogenic forcing.
Most previous ocean model studies that have examined the effects of SH wind changes have done so by prescribing zonally symmetric changes in wind latitude and strength (e.g., Delworth & Zeng, 2008;Downes et al., 2017;Frankcombe et al., 2013;Hogg et al., 2017;Spence et al., 2014;Waugh et al., 2019). Zonal differences in the changes in SH westerlies have only been examined in a few studies (e.g., Bracegirdle et al., 2013;Waugh et al., 2020). The climatological zonal mean location of the surface westerlies is more poleward in the Pacific and western Indian Ocean compared to the Atlantic and eastern Indian basins (Figure 2a). This is also a consistent feature in the climate models. In the ERA5 reanalysis, there is an 8° meridional difference in the most poleward (∼56°S) and equatorward locations ( (b) and (c) represent multimodel mean trends from CMIP5 and CMIP6 respectively and shading represents the interquartile range. White circles represent the regions where trends are significant based on a two-tailed student t-test after considering the autocorrelation in the data sets.
We next examine recent regional trends in the ERA5 reanalysis to examine whether they can be accounted for by intrinsic variability, or whether they can provide a reliable estimate of the forced signal. To do this, we compute regional trends in the location and strength of surface westerlies in the ERA5 reanalysis, as well as in CMIP5 and CMIP6 models, for the modern period (1979-2019). Major regional differences between ERA5 and ensemble mean modeled trends in the meridional location of the westerlies can be seen (Figure 2b). Regional differences in trends in the meridional location of westerlies from either model generations are not consistent with the ERA5 trends. Indeed, even though the MMM averages over a large component of the internal variability inherent in individual models, we still find no consistency in the regional pattern of trends between the CMIP5 and CMIP6 MMM (Figure 2b). For example, in the east Pacific ERA5 shows a strong positive trend, in contrast to the CMIP5 MMM which shows a negative trend and CMIP6 MMM which has almost no trend (Figure 2b). We conclude that over the relatively short reanalysis period (i.e., 41 years from 1979 to 2019), the regional differences in trends in both the latitude and the strength of westerlies are likely dominated by natural interannual to decadal climate variability (Figures 2b and 2c, Figure S12).
For the models, we extend the above analysis to cover the full 20th Century, to see if robust regional patterns in the trends emerge. Using the longer period for both the CMIP6 and CMIP5 models, similar regional patterns in MMM trends in the position of westerlies are found, with significant poleward trends identified everywhere except in the western Pacific, (Figure 3b), with spatial correlation coefficient of 0.7 (P < 0.05) between CMIP5 and CMIP6 MMM trends. Similar regional patterns are also found in trends in the strength of the westerlies (spatial correlation coefficient of 0.8 (P < 0.05) between CMIP5 and CMIP6 MMM trends) with strong trends found in the eastern Indian and western Atlantic Oceans basins (Figure 3c).
Changes in the zonal mean position and strength of the westerlies also show consistent seasonal differences over the historical time period (1900-1999, Figure S13). While a poleward shift is found in all four seasons in both CMIP5 and CMIP6 MMM ( Figure S13a), the strongest trends are found during summer and weakest trends during winter ( Figure S13a). Similar seasonality is also found in the wind strength trends, with stronger trends in summer compared to winter ( Figure S13b).

Future Projections
Future changes in the SH surface westerlies are expected to be affected by the competing effects of increasing greenhouse gases (GHGs) and stratospheric ozone recovery (Thompson et al., 2011). While both GHGs and ozone have acted in concert in the past, as ozone recovers it is expected that the two effects will oppose each other in the future (e.g., Eyring et al., 2010;Goyal et al., 2019;Newman et al., 2006). After ozone recovery stabilizes, it is expected that changes in the westerlies will be largely determined by changes in GHGs.  Table S3). As with the historical period, there are also major differences in these trends by season (Figures 3d and 3e). In particular, a poleward shift is found in all seasons with the largest shift projected during autumn and summer (compared to only in summer during the historical era), and a weaker shift projected for winter and spring (Figure 3d and Figure S13a). Strengthening of the westerlies is also projected in all seasons with the weakest trends in summer, in contrast to the historical era, when summertime trends were the strongest (Figure 3e and Figure S13b). As discussed earlier, the projected changes in the SH westerlies are expected to be affected by the competing effects of increasing GHGs and stratospheric ozone recovery (Bracegirdle, Krinner, et al., 2020b). While the effect of GHGs acts in all seasons, stratospheric ozone primarily affects the SH during summer because of the breakdown of the stratospheric polar vortex during spring (Arblaster & Meehl, 2006). Weaker summertime trends in the 21st Century are therefore expected because of the opposing contributions of GHGs and stratospheric ozone forcing in that season (Thompson et al., 2011). This suggests that the role of GHGs becomes much more important in the future under the high emission scenario, particularly given the expected recovery of stratospheric ozone. Consistent results are found for projected changes in both the latitude and the strength of westerlies in CMIP5 models, although trends are stronger in the CMIP5 MMM (Figures 3d and 3e). It is interesting to note that the projected strengthening of westerlies in the high emission scenarios of both CMIP5 and CMIP6 models during the 21st Century occurs throughout the year, but is strongest in winter and spring, whereas the projected shift in westerlies is considerably larger in summer and autumn compared to winter and spring (Figures 3d and 3e). This is counter to the expectation that the changes in the latitude and strength of westerlies operates in tandem , suggesting that different factors might be affecting the projected seasonal trends in both the poleward shift and the strengthening of westerlies in the SH.
In contrast to the high emission scenario, no significant trends are found in the moderate emissions scenario in both CMIP6 (SSP245) and CMIP5 (RCP45) MMM for both the latitude (except during autumn in CMIP5) and strength (except during autumn and spring in CMIP5) of the surface westerlies (Figures 3d and  3e). In these cases, greenhouse forcing stabilizes at a much lower level and stratospheric ozone forcing can largely compensate the increase in greenhouse gases.
Projected 21st Century trends from CMIP6 models in the latitude of the maximum westerlies also show large regional differences, with the strongest poleward trends over the Atlantic and east Pacific Oceans, and somewhat weaker poleward trends in the Indian Ocean (Figure 3b). Both CMIP5 and CMIP6 show similar regional patterns in the MMM trends in the meridional location of the westerlies (with a spatial correlation, R = 0.83). However, CMIP6 MMM trends in the meridional location are weaker as compared to CMIP5 MMM trends (Figure 3b). The weaker poleward shift in CMIP6 MMM as compared to CMIP5 MMM is consistent with the reduction in the equatorward bias in the meridional location of westerlies in CMIP6 MMM as compared to CMIP5 MMM, as models with a larger equatorward bias also tend to show a larger projected poleward shift Kidston & Gerber, 2010). Significant projected trends in the strength of westerlies under the SSP585 scenario of CMIP6 are evident at all longitudes, with stronger trends centered south of Australia and within the Drake Passage (Figure 3c). Again, consistent regional patterns are found between both the model generations (R = 0.9, Figure 3c). However, the projected 21st Century trends are again stronger in the CMIP5 MMM as compared to CMIP6 MMM in all regions except for the Atlantic (Figure 3c).

Summary and Discussion
In the past, a wide range of wind shifts and accelerations have been used to force ocean models in order to examine the response of the Southern Ocean and the Antarctic margin to past and projected changes in SH westerlies. Understanding future changes has also been hampered by the fact that CMIP5 models showed a significant equatorward bias in the location of the SH westerlies. Previous work has shown that projected wind changes are sensitive to the model's mean state. In particular, models with larger equatorward biases tend to show larger projected poleward wind shifts . As such, an anomalous wind shift based on a climate model projection (or from an ensemble of models) will retain a signature of the model's mean state bias (e.g., Duran et al., 2020).
In this study, we found a significant reduction in the equatorward bias in the location of SH westerlies in CMIP6 models as compared to CMIP5 models, with the location of maximum surface westerlies in closer agreement with the position of maximum surface westerlies in the ERA5 reanalysis. Given the sensitivity of model projections to mean state biases, CMIP6 models thus likely offer a more credible estimate of past and future changes in SH westerlies for forcing ocean model simulations. A recent initiative called the Flux-Anomaly-Forced Model Intercomparison Project (FAFMIP, Gregory et al., 2016) have suggested common protocols based on CMIP5 to look at the projected changes in CMIP6 models. Our results suggest that there might be an important benefit in using projections from CMIP6 models instead of CMIP5. We also found that the reanalysis time period (41 years from 1979 to 2019) is too short to provide an estimate GOYAL ET AL.  of the forced trends in the SH westerlies, as the trends over this multidecadal period appear to be strongly influenced by internal climate variability (see also Schneider et al., 2015;D. Yang et al., 2020). Moreover, it is likely that any anthropogenic forced component of regional or seasonal differences in the reanalysis trends is dominated by internal variability. MMM regional and seasonal trend patterns in both the latitude and strength of the maximum winds only become consistent between CMIP5 and CMIP6 when considering longer (e.g., centennial) time-scale trends.
Based on these findings, we can provide a set of recommendations for forcing ocean model simulations with past and projected changes in SH surface winds: (1) Recent observed wind trends over the Southern Ocean likely include a substantial component of internal decadal variability, and thus should not be assumed to be indicative of forced changes. Even in summer when the trends are significant, the magnitude of change will be strongly affected by intrinsic variability.
(2) CMIP6 models should be used instead of CMIP5 models for guiding the forcing used in ocean model simulations, for both past and future changes in the SH westerlies, given the much reduced mean state biases. (3) Seasonal variations in trends in both the location and the strength of the westerlies should be considered for simulations where seasonal changes are important (e.g., for studies examining seasonal changes in mode water formation, or Antarctic sea ice variability). (4) As ocean circulation is sensitive to the position of the wind maximum/wind stress curl, prescribed wind forcing should also include regional variations in surface wind trends. This is particularly relevant for projections where regional differences in trends can be as large as 150% for the location and 90% for the strength of the westerlies (Figures 3b and 3c).
While we have focused on ensemble average hindcasts and projections for CMIP5 and CMIP6 simulations, using the multimodel mean to construct zonal-mean wind forcing anomalies presents some problems. For example, only prescribing a zonal wind anomaly is not dynamically consistent if no changes are made to the meridional winds. In addition, the application of a zonal wind perturbation to daily reanalysis fields will distort the geometry of storms. Tapering regions by applying wind anomalies over a particular latitude band in the SH extratropics can also create spurious wind stress curl anomalies (e.g., Maher et al., 2018). One option to minimize these limitations is to use output from individual models as boundary forcing (e.g., Naughten et al., 2018), something commonly done for atmospheric downscaling (e.g., Evans et al., 2014). This is a more viable option now that CMIP6 models have less equatorward bias in the SH westerlies as compared to CMIP5. Using multiple models would also provide a means to estimate uncertainty in the projections.
Under a high emission scenario, a poleward intensification of the SH surface westerlies is projected to continue in the future despite the projected recovery of stratospheric ozone, because greenhouse gas forcing dominates the future trends across all seasons. We have provided quantitative information on the past and projected future changes in zonal mean position and strength of the surface westerlies over both annual and seasonal time scales (Table S3). This can be used to guide the forcing of idealized ocean model simulations with zonally averaged past and future changes in the SH westerlies.

Acknowledgments
Authors would like to thank Darryn Waugh from the Johns Hopkins University and four anonymous reviewers for their comments which helped improve the quality of the manuscript. This study was supported by the Australian Research Council (grants CE170100023, FL150100035). Rishav Goyal is supported by the Scientia PhD scholarship from the University of New South Wales. Matthew H. England is also supported by the Earth Science and Climate Change Hub of the Australian Government's National Environmental Science Program (NESP) and the Center for Southern Hemisphere Oceans Research (CSHOR), a joint research center between QNLM, CSIRO, UNSW and UTAS. Analysis were conducted on the National Computational Infrastructure (NCI) facility based in Canberra, Australia. The authors acknowledge the World Climate Research Programme's Working Group on Coupled Modeling, which is responsible for CMIP, and the authors thank the climate modeling groups for producing and making their model output available. For CMIP, the U.S. Department of Energy's Program for Climate Model Diagnosis and Intercomparison provides coordinating support and led development of software infrastructure in partnership with the Global Organization for Earth System Science Portals. NCEP-NCAR reanalysis data was provided by the NOAA/OAR/ ESRL PSD, Boulder, Colorado. ERA5 data were obtained from the ECMWF Climate Data Store (CDS), JRA-55 data were provided by the Japan Meteorological Agency (JMA) and MERRA-2 data were provided by NASA Goddard Earth Sciences (GES) Data and Information Services Center (DISC).