Journal of Geophysical Research: Oceans

On the response of Southern Hemisphere subpolar gyres to climate change in coupled climate models

Authors

  • Zhaomin Wang

    Corresponding author
    • School of Marine Sciences and Key Laboratory of Meteorological Disaster of Ministry of Education, Nanjing University of Information Science and Technology (Nanjing Institute of Meteorology), Nanjing, China
    Search for more papers by this author

Corresponding author: Z. Wang, School of Marine Sciences and Key Laboratory of Meteorological Disaster of Ministry of Education, Nanjing University of Information Science and Technology (Nanjing Institute of Meteorology), Nanjing 210044, China. (wzm@nuist.edu.cn)

Abstract

[1] We investigate the responses of the Southern Hemisphere subpolar gyres to projected climate changes over the 21st century by Coupled Model Intercomparison Project Phase 3 and 5 models. Under increased greenhouse gas forcing, the Southern Hemisphere westerly winds consistently become intensified, resulting in increased cyclonic wind forcing in the subpolar region in these models. Under such wind forcing changes, it is a robust feature that there are consistent increases in the westward flow close to the coast of Antarctica, with strong implications to the mass balance of the Antarctic ice shelves and ice sheets. However, there are large discrepancies in the responses of the gyre axes and overall gyre strengths. Some models show equatorward expansions of the southern gyre limbs, resulting in consistent and large gyre strength increases, while some other models show poleward contractions of the gyres and generally small and less consistent gyre strength changes. These uncertainties are primarily a result of the uncertain simulations of eddy-driven circulations in the Antarctic Circumpolar Current. The associated buoyancy forcing changes play a secondary role in driving these oceanic responses.

This study reveals that there are large uncertainties in the projections of the subpolar circulation in the current generation of coupled climate models, although CMIP5 models have considerably smaller inter-model spreads in the present-day and projected gyre strengths. To predict the subpolar circulation changes, future improved modelling studies need to particularly reduce the uncertainties in the projections of the westerly jet and to reduce the uncertainties in the eddy-driven circulation responses to wind forcing changes.

1 Introduction

[2] Much attention has been recently focusing on the Southern Ocean (SO) toward an understanding of its important role in global environmental changes. As such, the performances of global coupled climate models that include atmospheric and oceanic general circulation models for simulating the present-day conditions and future changes in the SO is critical, as these models are the best tools available to predict the future evolution of the climate system. Previous studies have revealed that the simulations of the present-day and future SO circulation in coupled climate models have large discrepancies [e.g., Russell et al., 2006; Sen Gupta et al., 2009; Wang et al., 2011], thus reducing our confidence in projections of the global climate system. In the southern polar ocean, which has direct contact with ice shelves and where dense Antarctic bottom water forms, the discrepancies in the simulated subpolar circulation in these models become even larger [Wang and Meredith, 2008]. In order to provide an improved understanding of the pivotal role of the polar ocean in future climate changes and to facilitate the future model developments, we investigate and document the robustness, uncertainty, and improvements in the historical simulations and projections of the Southern Hemisphere subpolar gyres in CMIP3 (Coupled Model Intercomparison Project Phase 3) and CMIP5 models.

[3] The subpolar gyres are important components in global energy and water cycles, particularly, in the complex coupled atmosphere-ocean-ice (sea ice and ice shelf) system. Changes in these gyres have strong impacts on the ice shelves and adjacent ice sheets [e.g., Rignot and Jacobs, 2002] and therefore global climate change and in particular sea level change over the next century. The subpolar gyres are also important sites of water mass transformations in the SO [e.g., Rintoul et al., 2001], and it has been inferred that the changes in the cyclonicity of the Weddell Gyre play a key role in enhancing and restricting the northward export of Antarctic Bottom Water (AABW) that is formed within it [Meredith et al., 2008].

[4] The southern subpolar circulation is complex and unique. In addition to much stronger bottom topographic effects due to weak stratification, and strong interactions with the overlying atmosphere and sea ice, the southern subpolar gyres lie to the south of the Antarctic Circumpolar Current (ACC) where mesoscale eddies are particularly active. It is thus necessary to consider the changes in local momentum and buoyancy forcing associated with atmospheric and sea ice circulation changes, and to consider the changes in eddy-driven circulation in the ACC, when investigating the changes in the subpolar circulation.

[5] Although the quality and quantity of the SO observation have been greatly improved over recent decades, through the massive increases in robotics, automation, miniaturization, communication, and computing power [Summerhayes et al., 2007], the spatial and temporal coverage of the oceanic observation are still highly limited in the subpolar region. Direct current measurement time series are particularly sparse and short [Klatt et al., 2005], imposing great difficulties for us to derive large-scale structure and the temporal variability of the subpolar circulation. Satellite observations provide great benefits for monitoring large spatial and temporal scale structure of the polar climate system, particularly the sea ice extent [Comiso and Nishio, 2008]. However, altimetry data are still not available in sea ice-covered regions, a key variable to derive the spatial structure and temporal variability of the large-scale subpolar circulation. While much more coordinated and extensive observations have been undertaken during the International Polar Year 2007–2008 with new technology, it will still not be possible to derive the large-scale and long-term temporal variability in the subpolar circulation in the near future. The current available data sets of surface forcing also have large discrepancies in the synthesized momentum and buoyancy fluxes, particularly over the SO [Large and Yeager, 2009].

[6] These observational limitations make it impossible to properly evaluate the simulated broad-scale present-day oceanic conditions and their variability in the subpolar region, and the associated atmospheric and sea ice forcing. The uncertainties in the simulated broad-scale present-day and future subpolar circulation can, however, be assessed to some extent through conducting multi-model intercomparisons. Lessons learned from these multi-model intercomparison projects for the current generation of climate models will be useful to the future development and application of coupled climate models. In the future, more and more climate models will employ ocean components with higher resolutions [e.g., Shaffrey et al., 2009]. These models will eventually also include interactive ice shelf and ice sheet components. As such, an accurate simulation of the subpolar ocean circulation is critical.

[7] In this study, we document the projected changes in the spatial structure and strength of the subpolar gyres in the CMIP3 and CMIP5 models; in particularly, we examine the relationships between the gyre changes and changes in external forcing, including the momentum and buoyancy forcing associated with atmospheric and sea ice circulation, and forcing associated with ACC dynamics. In section 2, the models, data, and the methods are briefly described. The detailed changes of the subpolar gyres are presented in section 3, including the discrepancies and consistent features in the projected changes to subpolar gyres among these models. The processes that drive the subpolar circulation changes are analyzed and are discussed in section 4. Concluding remarks are given in section 5.

2 Models, Data, and Methods

[8] We analyze the output from 19 CMIP3 (http://www-pcmdi.llnl.gov/ipcc/model_documentation/ipcc_model_documentation.php) and 14 CMIP5 models [Taylor et al., 2012]. These models are listed in Tables 1 and 2, along with their horizontal and vertical resolutions, and eddy parameterizations. For CMIP3 models, we mainly use the output from runs with sresa1b emission scenario that have generally consistent increases in westerly winds, whereby CO2 is increased from the present-day level to 703 ppm at the end of the 21st century; for CMIP5 models, the output of one future scenario run, Representative Concentration Pathway (RCP) 8.5, is analyzed. RCP8.5 is a high emission scenario, for which the number 8.5 refers to approximate radiative forcing increase (in W/m2) at the end of 21st century. Though RCP4.5 is a medium mitigation scenario and its radiative forcing increase is closer to that in sresa1b scenario for CMIP3 models, the changes in atmospheric forcing are much less consistent for this scenario than for RCP8.5 (not shown). (Note that sresA2 is a higher emission scenario and is hence closer to RCP8.5 than sresa1b, but there are less models that have sresA2 runs; this limited our selection of models for detailed analysis: for example, there is no sresA2 output for model miroc3 _2_hires, which is an only high-resolution model.) Analyzing ocean circulation changes in models with relatively consistent atmospheric forcing changes can help us identify uncertain oceanic processes, which is the main purpose of this paper.

Table 1. Nineteen CMIP3 Models Used in This Studya
No.Model NameCountryResolution (x;y;z)Eddy Parameterization
  1. aThe relevant references for GM90, Gent95, Griffies98, Griffies05, Treguier97, Visbeck97, and Wright97 are Gent and McWilliams [1990], Gent et al. [1995], Griffies [1998], Griffies et al. [2005], Visbeck et al. [1997], and Wright [1997], respectively.
1cccma _cgcm3 _1Canada1.875°; 1.875°; 29 levelsGM90
2cccma _cgcm3 _1_t63Canada1.4°; 0.94°; 29 levelsGM90
3cnrm _cm3 _0France2°; 1.5°–0.5°; 31 levelsGM90
4csiro _mk3 _0Australia1.87°; 0.84°; 31 levelsGM90, Griffies98
5csiro _mk3 _5Australia1.87°; 0.84°; 31 levelsGM90, Griffies98
6gfdl _cm2 _0USA1°; 1°–0.33°; 50 levelsGriffies98, Griffies05
7gfdl _cm2 _1USA1°; 1°–0.33°; 50 levelsGriffies98, Griffies05
8giss _aomUSA4°; 3°; 16 levelsnone
9giss _model_e_rUSA5°; 4°; 13 levelsVisbeck97, Griffies98
10ingv _echam4Italy2°; 2°–1°; 33 levelsTreguier97
11ipsl _cm4France2°; 2°–1°; 31 levelsTreguier97
12miroc3 _2_hiresJapan0.28°; 0.19°; 47 levelsGent95
13miroc3 _2_medresJapan1.4°; 1.4°–0.56°;43 levelsGent95
14miub _echo_gGermany/Korea2.8125°; 2.79°–0.5°; 20 levelsnone
15mpi _echam5Germany1.5°; 1.5°; 40 levelsGent95, Griffies98
16mri _cgcm2 _3_2aJapan2.5°; 2°–0.5°; 23 levelsGM90
17ncar _ccsm3 _0USA1.125°; 0.53°–0.27°; 40 levelsGM90, Griffies98
18ncar _pcm1USA1.125°; 0.53°–0.27°; 40 levelsGM90, Griffies98
19ukmo _hadcm3UK1.25°; 1.25°; 20 levelsGent95,Visbeck97,Wright97
Table 2. Fourteen CMIP5 Models Used in This Studya
No.Model NameCountryResolution (x;y;z)Eddy Parameterization
  1. aHorizontal resolutions are at 50°S. See the caption of Table 1 for the relevant references of the eddy parameterization schemes, while Eden08 and Eden09 refer to Eden and Greatbatch [2008] and Eden et al. [2009], respectively.
1bcc-csm1-1China1.000°; 1.000°; 40 levelsGriffies98, Griffies05
2CanESM2Canada1.406°; 0.930°; 40 levelsGM95
3GFDL-ESM2GUSA1.000°; 1.000°; 50 levelsGriffies98, Griffies05
4GFDL-ESM2MUSA1.000°; 1.000°; 50 levelsGriffies98, Griffies05
5GISS-E2-RUSA1.250°; 1.000°; 32 levelsGM90
6HadGEM2-CCUK1.000°; 1.000°; 40 levelsVisbeck97, Griffies98
7HadGEM2-ESUK1.000°; 1.000°; 40 levelsVisbeck97, Griffies98
8IPSL-CM5A-LRFrance2.000°; 1.267°; 31 levelsTreguier97
9IPSL-CM5A-MRFrance2.000°; 1.267°; 31 levelsTreguier97
10IPSL-CM5B-LRFrance2.000°; 1.267°; 31 levelsTreguier97
11MIROC-ESMJapan1.406°; 0.930°; 44 levelsGWMM95
12MIROC-ESM-CHEMJapan1.406°; 0.930°; 44 levelsGWMM95
13MRI-CGCM3Japan1.000°; 0.500°; 51 levelsVisbeck97
14NorESM1-MNorway1.125°; 0.534°; 70 levelsEden08,Eden09

[9] A gyre strength is defined in the following way. Along a longitude λ, we first calculate the meridional integral of vertically integrated zonal transport from the Antarctic coast, ϕs, to the gyre axis (defined as the northern boundary of westward transport region), ϕn, i.e., inline image , where u is the zonal velocity and a is the radius of earth. Then we find the minimum value (largest westward flow) of these integrals in each gyre domain, and this minimum (and negative) value is defined as the gyre strength. So, the change in a gyre strength could be caused by changes in the zonal velocity or by meridional shifts of the gyre axis or by both.

[10] Similarly, the ACC transport is calculated as a vertical integral of zonal velocity over all depths and from land to land, across the Drake Passage.

[11] The linear trends are derived from the slopes of best linear fits (least square) to the time series of the annual means of relevant variables over the 21st century. A t test is performed for the significance of the trends.

3 Changes in the Subpolar Gyres

3.1 Mean Strengths of the ACC and the Subpolar Gyres

[12] The CMIP3 ensemble mean ( ± 1 std) ACC transport is 155 ± 79 Sv (1 Sv = 106 m 3/s) (Table 3), slightly larger than the observationally based estimate of 134 ± 11 Sv [Cunningham et al., 2003]. However, the inter-model spread is large, ranging from 49 Sv in model 11 to 325 Sv in model 18. The ensemble mean strengths of the three gyres are comparable with the best available observed results [Wang and Meredith, 2008]. However, as reflected by large inter-model standard deviations, some models have too strong or too weak gyres, with the weakest Weddell Gyre being simulated in model 9 and the strongest one in model 18; in some models, the Ross Gyre (models 1, 2, 13, and 16) and the Australian-Antarctic Gyre (models 4, 10, and 14) are too weak.

Table 3. Averaged ACC Transports (Sv) Through Drake Passage and the Gyre Strengths (Sv) of the Weddell Gyre (WG), Ross Gyre (RG), and Australia-Antarctic Gyre (AG) over the 21st Century in Their sresa1b Runs of 19 CMIP3 Modelsa
No.Model NameACCWGRGAG
  1. aStandard deviations (Sv) calculated from annual mean time series are also given. The last row shows the ensemble means and inter-model standard deviations.
1cccma _cgcm3 _197 ± 1 − 15 ± 1 − 3 ± 0 − 6 ± 0
2cccma _cgcm3 _1_t63116 ± 1 − 20 ± 0 − 5 ± 0 − 5 ± 0
3cnrm _cm3 _054 ± 1 − 43 ± 2 − 10 ± 1 − 27 ± 1
4csiro _mk3 _0313 ± 3 − 28 ± 4 − 39 ± 3 − 3 ± 2
5csiro _mk3 _5152 ± 1 − 38 ± 2 − 16 ± 1 − 13 ± 1
6gfdl _cm2 _0106 ± 2 − 52 ± 2 − 13 ± 1 − 18 ± 2
7gfdl _cm2 _1131 ± 3 − 43 ± 6 − 12 ± 1 − 15 ± 1
8giss _aom145 ± 6 − 58 ± 6 − 23 ± 3 − 36 ± 6
9giss _model_e_r268 ± 2 − 10 ± 3 − 12 ± 2 − 15 ± 1
10ingv _echam4112 ± 1 − 27 ± 1 − 15 ± 3 − 4 ± 1
11ipsl _cm449 ± 1 − 62 ± 2 − 8 ± 1 − 38 ± 2
12miroc3 _2_hires137 ± 2 − 101 ± 10 − 42 ± 7 − 51 ± 9
13miroc3 _2_medres183 ± 3 − 45 ± 3 − 3 ± 0 − 7 ± 1
14miub _echo_g83 ± 1 − 21 ± 2 − 12 ± 1 − 1 ± 0
15mpi _echam5158 ± 3 − 87 ± 4 − 50 ± 2 − 32 ± 5
16mri _cgcm2 _3_2a92 ± 2 − 20 ± 1 − 4 ± 0 − 17 ± 1
17ncar _ccsm3 _0193 ± 2 − 43 ± 2 − 25 ± 1 − 18 ± 2
18ncar _pcm1325 ± 9 − 134 ± 9 − 80 ± 5 − 115 ± 21
19ukmo _hadcm3226 ± 2 − 31 ± 2 − 59 ± 10 − 13 ± 3
 Ensemble Mean155 ± 79 − 46 ± 31 − 23 ± 21 − 23 ± 26

[13] For CMIP5 models (Table 4), the ensemble means ( ± 1 std) of the ACC transports ( 145 ± 38 Sv) and gyre strengths (Weddell Gyre: − 44 ± 22 Sv; Ross Gyre: − 24 ± 15 Sv; Antarctic-Australia Gyre: − 19 ± 10 Sv) are close to those for CMIP3 models, but the inter-model spreads are considerably smaller. The much smaller inter-model spreads indicate that there are less extreme simulated transports in these CMIP5 models. However, some models still simulated too weak gyres (e.g., models 8, 9, and 10 for the Weddell Gyre, and models 2 and 10 for the Australian-Antarctic Gyre) and too strong gyres (e.g., model 5 for the Ross Gyre).

Table 4. Averaged ACC Transports (Sv) Through Drake Passage and the Gyre Strengths (Sv) of the Weddell Gyre (WG), Ross Gyre (RG), and Australia-Antarctic Gyre (AG) over the 21st Century in Their rcp8.5 Runs of 14 CMIP5 Modelsa
No.Model NameACCWGRGAG
  1. aStandard deviations (Sv) calculated from annual mean time series are also given. The last row shows the ensemble means and inter-model standard deviations.
1bcc-csm1-1152 ± 2 − 39 ± 3 − 23 ± 1 − 15 ± 1
2CanESM2157 ± 1 − 34 ± 2 − 9 ± 0 − 4 ± 0
3GFDL-ESM2G112 ± 2 − 41 ± 3 − 23 ± 2 − 32 ± 2
4GFDL-ESM2M127 ± 2 − 50 ± 2 − 12 ± 1 − 20 ± 2
5GISS-E2-R242 ± 2 − 31 ± 3 − 67 ± 4 − 10 ± 2
6HadGEM2-CC167 ± 3 − 77 ± 5 − 33 ± 3 − 24 ± 4
7HadGEM2-ES164 ± 2 − 78 ± 4 − 33 ± 3 − 21 ± 3
8IPSL-CM5A-LR111 ± 8 − 12 ± 3 − 13 ± 3 − 14 ± 4
9IPSL-CM5A-MR134 ± 6 − 10 ± 3 − 11 ± 1 − 17 ± 5
10IPSL-CM5B-LR89 ± 5 − 18 ± 2 − 23 ± 2 − 8 ± 2
11MIROC-ESM168 ± 3 − 51 ± 2 − 13 ± 1 − 16 ± 2
12MIROC-ESM-CHEM171 ± 2 − 50 ± 1 − 14 ± 1 − 16 ± 2
13MRI-CGCM3108 ± 1 − 57 ± 5 − 15 ± 3 − 29 ± 2
14NorESM1-M133 ± 2 − 67 ± 4 − 41 ± 1 − 43 ± 3
 Ensemble Mean145 ± 38 − 44 ± 22 − 24 ± 15 − 19 ± 10

3.2 Changes in the ACC Transport and the Subpolar Gyre Strengths

[14] It has been found that despite consistent intensifications and poleward shifts of westerly winds, some CMIP3 models show increased ACC transports and some show decreased ACC transports in their sresa1b runs [see Wang et al., 2011]. To have a comparison with the results from CMIP5 models and to facilitate the examinations on the linkages between the ACC transport and the subpolar gyres, the linear trends in the ACC transport in CMIP3 and CMIP5 models are listed in Tables 5 and 6, respectively, along with the linear trends in the gyre sizes (represented by westward flow areas in the subpolar gyres) and strengths.

Table 5. Linear Trends T (Sv/century for Transport, and 10 10m2 ∕ dec for the Westward Flow Areas) and Their Significance Levels S (%) in ACC Transport Through Drake Passage, Westward Flow Area (WF) of the Subpolar gyres, the Weddell Gyre Strength (WG), Ross Gyre Strength (RG), and Australia-Antarctic Gyre Strength (AG) in Their sresa1b Runs of 19 CMIP3 Modelsa
No.Model NameACC T(S)WF T(S)WG T(S)RG T(S)AG T(S)
  1. aThe adjusted trends (see text) are also given after the semicolon in each column.
1cccma _cgcm3 _14(0); 220(0); 28 − 4(0); − 4 − 1(0); − 10(4); 0
2cccma _cgcm3 _1_t6312(0); 12 − 13(0); − 130(11); 00(4); 00(72); 0
3cnrm _cm3 _0 − 12(0); 850(0); 4 − 10(0); − 4 − 3(0); − 1 − 9(0); − 4
4csiro _mk3 _0 − 8(23); − 2413(15); 13 − 2(53); − 2 − 14(0); − 14 − 1(42); − 1
5csiro _mk3 _510(0); 10 − 17(0); − 175(0); 5 − 4(0); − 44(0); 4
6gfdl _cm2 _015(0); 8 − 19(0); − 191(33); 10(44); 02(0); 2
7gfdl _cm2 _1 − 8(3); − 822(0); 22 − 2(73); − 2 − 3(0); − 3 − 6(0); − 6
8giss _aom − 88(0); − 8869(0); 69 − 18(0); 71(47); 1 − 21(0); 1
9giss _model_e_r − 3(2); 421(0); 21 − 8(0); − 84(1); 40(75); 0
10ingv _echam4 − 12(0); − 1210(0); 14 − 3(0); − 3 − 7(7); − 7 − 7(0); − 5
11ipsl _cm47(0); 717(0); 28 − 3(0); − 5 − 7(0); − 8 − 6(0); − 6
12miroc3 _2_hires12(0); 13 − 16(0); − 230(87); 0 − 8(8); − 8 − 11(1); − 11
13miroc3 _2_medres1(43); 118(2); 18 − 4(27); − 4 − 6(0); − 60(74); 0
14miub _echo_g − 2(0); − 211(0); 115(19); 53(5); 30(9); 0
15mpi _echam57(0); 71(81); 10(75); 05(0); 5 − 5(0); − 5
16mri _cgcm2 _3_2a − 1(31); − 3 − 10(0); − 100(72); 00(91); 01(3); 1
17ncar _ccsm3 _0 − 19(0); − 1917(0); 17 − 9(0); − 9 − 4(0); − 4 − 7(0); − 7
18ncar _pcm1 − 20(0); − 2015(0); 1519(0); 19 − 19(0); − 19 − 19(4); − 19
19ukmo _hadcm36(0); 61(34); 1 − 3(1); − 3 − 2(76); 1512(0); 12
Table 6. Similar to Table 5, but for Their rcp8.5 Runs of 14 CMIP5 Models
No.Model NameACC T(S)WF T(S)WG T(S)RG T(S)AG T(S)
1bcc-csm1-12(36); 25(11); 5 − 9(30); − 9 − 4(1); − 4 − 3(0); − 3
2CanESM24(0); − 10(79); 11 − 4(0); − 4 − 1(0); − 1 − 2(0); − 3
3GFDL-ESM2G8(0); 80(96); 0 − 11(0); − 11 − 18(0); − 18 − 2(2); 0
4GFDL-ESM2M1(32); 110(0); 10 − 4(2); − 4 − 6(0); − 6 − 2(4); − 2
5GISS-E2-R − 10(0); − 100(97); 00(93); 014(27); 143(9); 3
6HadGEM2-CC − 27(0); − 2714(21); 14 − 24(0); − 24 − 21(0); − 21 − 8(2); − 8
7HadGEM2-ES − 17(0); − 17 − 7(35); − 7 − 16(0); − 16 − 15(0); − 150(63); 0
8IPSL-CM5A-LR28(0); 2842(0); 420(95); 0 − 10(0); − 10 − 14(3); − 14
9IPSL-CM5A-MR39(0); 395(27); 5 − 8(0); − 81(0); 1 − 14(16); − 14
10IPSL-CM5B-LR − 6(0); − 1554(0); 54 − 15(0); − 22 − 19(0); − 19 − 4(8); − 4
11MIROC-ESM − 1(70); 638(0); 29 − 6(0); − 6 − 11(0); − 12 − 4(0); − 4
12MIROC-ESM-CHEM − 1(22); 232(0); 32 − 6(0); − 6 − 11(0); − 11 − 4(0); − 4
13MRI-CGCM3 − 17(0); − 1542(0); 30 − 16(0); − 16 − 16(0); − 16 − 2(0); − 2
14NorESM1-M15(0); 15 − 16(0); − 23(10); 3 − 8(0); − 82(15); 2

[15] Note that the adjusted trends are also listed in Tables 5 and 6, which are obtained in the following way. We first calculated the trends and their significance levels over the whole periods of their pre-industrial runs. We consider that there are spurious drifts only in those models with statistically significant trends in their pre-industrial runs. These statistically significant trends are then deducted from the corresponding trends in their future runs to get the adjusted trends. As indicated by the adjusted trends in future runs, Table 5 shows that CMIP3 models 1, 2, 3, 5, 6, 9, 11, 12, 15, and 19 have significant increases, while models 4, 7, 8, 10, 14, 16, 17, and 18 have significant decreases in the ACC transport.

[16] In their RCP8.5 runs of CMIP5 models, the strengths of the westerly jets also consistently increase in all models, albeit with different magnitudes [Bracegirdle et al., 2013], resulting in consistently increased cyclonic wind forcing in the subpolar region, similar to the wind stress curl changes in this region in CMIP3 models (see Figure 1). As shown by the adjusted trends in Table 6, the changes in the ACC transport are still diversified. Six models (models 3, 8, 9, 11, 12, and 14) have significant increases in the ACC transport, and the other six models (models 2, 5, 6, 7, 10, and 13) have significant decreases in the ACC transport.

Figure 1.

Gyre strength trends (Sv/century) against wind stress curl trends (10 − 9/m3/dec) in CMIP3 sresa1b runs and in CMIP5 rcp8.5 runs. The black lines are the best linear fits. Here the adjusted gyre strength trends are used, which are obtained by subtracting trends with significance level being less than 5% in Tables A1 and A2 from their corresponding trends in Tables 5 and 6. The wind stress curl trends in each panel are averaged over the corresponding gyre domains that are defined as the regions with negative barotropic stream function values in the Weddell-Enderby basin (60°W– 75°E), Ross basin (160°E–90 °W) and Antarctic-Australia basin (75 °E–135 °E), respectively, as illustrated in Figure 5. Note that using non-adjusted trends does not qualitatively change the results as indicated by the best linear fits.

[17] The numbers of models that have significantly intensified and weakened gyres are given in Table 7 for both CMIP3 and CMIP5 models. There are much more models that have intensified gyres than that have weakened gyres, particularly for CMIP5 models. These results suggest that the subpolar gyres generally become intensified in response to increased local cyclonic wind forcing (see Figure 1) in a warmer climate that has intensified and poleward shifted westerly winds. It is interesting to note that there are more consistent gyre changes in CMIP5 models than in CMIP3 models, likely due to that the increase in RCP8.5 radiative forcing is much larger than that in sresa1b.

Table 7. Numbers of Models (Left in Each Column) That Have Significant Negative Trends (Intensified Gyres) in Gyre Strength and (Right in Each Column) That Have Significant Positive Trends (Weakened Gyres) in Their sresa1b Runs of CMIP3 Models and in Their RCP8.5 Runs of CMIP5 Modelsa
 WG − T ∼ + TRG − T ∼ + TAG − T ∼ + T
  1. aThe adjusted trends in Tables 5 and 6 were used, which were obtained by subtracting those statistically significant trends in their pre-industrial control runs in Tables A1 and A2 from the trends in future runs.
CMIP37 ∼ 310 ∼ 29 ∼ 5
CMIP510 ∼ 012 ∼ 18 ∼ 1

[18] However, as shown in Figure 1, the changes in the gyre strengths (negative trends mean strengthening and positive ones mean weakening) show great diversity, despite that the wind forcing becomes more cyclonic in the subpolar region in almost all models. Under increased cyclonic wind forcing, some models even show reductions in the gyre strength, particularly for CMIP3 models. The best linear fits in Figures 1b– 1d further indicate that among these models, the gyre strengths do not increase with the increased cyclonic wind forcing. These facts strongly suggest that other processes also play important roles in controlling the gyre strength changes. Note that these results remain unchanged if non-adjusted trends are used in Figure 1.

[19] We now examine if the uncertain gyre strength changes are related to the uncertain changes in the ACC transport in these models. We first calculate the ensemble means of changes in gyre size and gyre strengths in those models with increased ACC transports and in those models with reduced ACC transports (Table 8). Note that the trends before adjustments are used in this analysis, because here we want to show how the ACC transport changes are dynamically related to gyre changes. The averaged gyre size becomes smaller in those CMIP3 models with increased ACC transports, while it becomes larger in those models with reduced ACC transports. For CMIP5 models, the averaged size becomes larger in both cases, but the change is much smaller when averaged over models with increased ACC transports than over models with decreased ACC transports. The best linear fits in the scatter plots (Figures 2a and 2b) also indicate that the subpolar gyre areas increase with reduced ACC transports, particularly for CMIP3 models, which was originally reported in Wang et al. [2011]. This relationship was confirmed in Meijers et al. [2012] through analyzing the results from CMIP5 models only.

Table 8. Ensemble Means for Those Models with Significantly Increased ACC Transports (Left in Each Column) and for Those with Significantly Reduced ACC Transports (Right in Each Column) of Trends (T) (Sv/century for Transport, and 1010 m2/dec for the Westward Flow Areas) in Westward Flow Area (WF) of the Subpolar Gyres, the Weddell Gyre Strength (WG), Ross Gyre Strength (RG), and Australia-Antarctic Gyre Strength (AG) in Tables 5 and 6a
 WF TWG TRG TAG T
  1. aNote that the trends before adjustments are used, in order to show the dynamical relationship between the ACC transport and the gyre strengths.
CMIP3 − 4.8 ∼ 27.5 − 1.5 ∼ − 4.8 − 1.5 ∼ − 5.1 − 0.8 ∼ − 12.0
CMIP513.4 ∼ 48.3 − 8.5 ∼ − 18.2 − 7.5 ∼ − 18.1 − 6.6 ∼ − 5.4
Figure 2.

Gyre westward flow area trends (1010m2/dec) against ACC transport trends (Sv/century) (a) in CMIP3 sresa1b runs and (b) in CMIP5 rcp8.5 runs.

[20] Table 8 shows that there are much larger increases in the gyre intensity in those models with reduced ACC transports than in those models with increased ACC transports, except for the Australian-Antarctic Gyre in CMIP5 models. The scatterplots (Figure 3) show that generally, the subpolar gyre strength increases when the ACC transport decreases, except for the Antarctic-Australia gyre in CMIP5 models. The similar changes in the Australian-Antarctic Gyre strength in both cases in Table 8 and the reversed relationship between the gyre strength changes and the ACC transport changes in Figure 3f are apparently caused by the large increases in this gyre strength in models 8 and 9 that have increased ACC transports. These two models have been found to have the largest equatorward biases in the westerly jet position [Bracegirdle et al., 2013] and to have generally weak subpolar gyres (Table 4). Removing these two outliers to re-plot Figure 3 (not shown) leads to a consistent feature among these panels, i.e., the gyre strength increases when the ACC transport decreases in all the panels. Therefore, there is an indication that the subpolar gyres become more intensified in those models with reduced ACC transports than in those models with increased ACC transports. These results remain qualitatively unchanged in a similar figure using adjusted trends in gyre strengths and ACC transport (not shown).

Figure 3.

Gyre strength trends (Sv/century) against ACC transport trends (Sv/century) in CMIP3 sresa1b runs (top row) and in CMIP5 rcp8.5 runs (bottom row) for the Weddell Gyre (left column), Ross Gyre (middle column), and Antarctic-Australia Gyre (right column).

[21] Below, in order to understand what drive these gyre strength changes, we analyze detailed changes in the subpolar gyres and link the gyre changes to external momentum and buoyancy forcing, and processes associated with the ACC dynamics.

3.3 Detailed Changes in the Subpolar Gyres Over the 21st Century

[22] In the following, we particularly present results from three CMIP3 models, csiro _mk3 _5 (model 5), gfdl _cm2 _1 (model 7), miroc3 _2_hires (model 12), and three CMIP5 models, GFDL-ESM2G (model 3), MRI-CGCM3 (model 13), NorESM1-M (model 14), to illustrate the detailed gyre changes and how these changes are related to ACC transport changes. (We also analyzed results from other models with relatively large ACC transport changes and found that the results drawn from these six selected models are generally robust.) We select these six models because (i) they have reasonable simulations of the ACC transport and the gyre strengths and (ii) the changes in the ACC transport and the gyre strength are clearly different from the model drifts in the corresponding control runs. Note that in model miroc3 _2_hires, the Weddell Gyre strength around the Prime Meridian is 101 ± 10 Sv (see Table 3), much larger than the observed value of 56 ± 8 Sv [Klatt et al., 2005], but this model is still included since this high resolution model is the only model that comes close to being eddy permitting in the ACC region.

[23] There are also some other reasons to select these six models. In these six models, two CMIP3 models, csiro _mk3 _5 and miroc3 _2_hires, and two CMIP5 models, NorESM1-M and GFDL-ESM2G, have increased ACC transports, but there are gyre strength decreases in the former models (csiro _mk3 _5 and NorESM1-M) and increases in the latter models (miroc3 _2_hires and GFDL-ESM2G). CMIP3 model gfdl _cm2 _1 and CMIP5 model MRI-CGCM3 have reduced ACC transports and increased gyre strengths. As such, the selection of these six models provides a representation sample of possible changes in the ACC transport and gyre strength from models listed in Tables 1 and 2. This can thus help us understand the processes that drive the gyre changes and can facilitate the analysis of the robust and uncertain aspects of simulated gyre responses to climate changes in these models. In particular, by choosing these models with different ACC transport changes, we are able to examine how the ACC transport changes affect the gyre strength changes.

[24] In the following, the results of these six models are presented in such a way that the three CMIP3 models are at upper three panels and the three CMIP5 models are at lower three panels. In order to facilitate the examinations on the links between the gyre changes and the changes in the ACC transport, the two models in the left column have increased ACC transport and mainly weakened gyres; in the middle column, model miroc3 _2_hires has increased ACC transport but no significant changes in the overall strengths of the Weddell Gyre and Ross Gyre, and model GFDL-ESM2G has increased ACC transport and also increased gyre strengths; the two models in the right column have decreased ACC transports and generally intensified gyres.

[25] Figure 4 shows the linear trends in vertically integrated zonal transport over the 21st century, to the south of 50°S. In the left and middle column, there are broader and larger positive trends in the ACC region than in the right panels, consistent with ACC transport changes in these models. To the south of the gyre axes, except for Figure 4c that has some positive trends mixed with negative trends, there are negative trends in these models, but these negative trends in the last three panels (Figures 4d– 4f) are much larger, consistent with their overall gyre strength increases. On the gyre axes, there are generally negative trends in the last three panels, and positive trends at various locations in the first three panels.

Figure 4.

Trends in vertically integrated zonal transport over the 21st century. The black lines mark the axes of the subpolar gyres (defined as the locations of minimum barotropic stream function that is obtained by integrating the zonal transport from the coast of Antarctica to the north) averaged over the first two decades of their 21st century runs (2001–2020 for CMIP3 models and 2006–2025 for CMIP5 models).

[26] The changes in the locations of the gyre axes are shown in Figure 5, along with the averaged barotropic stream functions over the first two decades of the 21st century. Though the barotropic stream functions show diversity in the gyre size and intensity, these models are able to simulate the three subpolar gyres, namely, the conventionally defined Weddell Gyre and Ross Gyre, and the recently defined deep Antarctic-Australian gyre [Donohue et al. 1999; Bindoff et al. 2000; McCartney and Donohue, 2007; Wang and Meredith, 2008]. Note that the double gyre structure for the Weddell Gyre, which is observed by Orsi et al. [1993] and simulated by Beckmann et al. [1999], does not appear in these barotropic stream functions averaged over the first two decades of the 21st century.

Figure 5.

The averaged barotropic stream functions over the first two decades of the 21st century runs and gyre axes derived from these barotropic stream functions (green) and from those averaged over 2081–2100 (yellow) for the subpolar gyres. The north-south oriented ticks mark the subpolar gyre centers for the first two decades of the 21st century (green) and for 2081–2100 (yellow).

[27] At various places, there are no changes in the locations of the gyre axes, reflected by the yellow lines overlapping the green ones. Generally, the shifts of the subpolar gyre axes are small, but the gyre center location changes in zonal direction in Figure 5d and 5f are relatively large.

[28] As indicated by the changes in the ACC transport and in the westward flow area in Figure 2, in the first three panels (Figure 5a– 5c) with increased ACC transports, the subpolar gyres contract poleward (decreased total westward flow area), while in the last two panels (Figures 5e and 5f) with reduced ACC transports, the gyres expand equatorward (increased total westward flow area). Interestingly, in Figure 5d, though there are relatively large poleward and equatorward shifts of the gyre axes at a number of places, the total westward flow area changes little (see Table 6).

[29] Though the gyre structure changes appear generally to be small, the overall gyre strength changes are relatively large and quite different in these models. As shown in Figure 3, the gyre strength changes are relatively small in those models with increased ACC transports and large in the models with reduced ACC transports. The reasons for these contrasting features will be further analyzed below.

[30] The meridional-depth cross sections across the Weddell Gyre (averaged over a 45°longitude sector to the east of the Weddell Gyre centers, marked by yellow ticks in Figure 5 where the westward transport changes are larger than at the locations of green ticks that are obtained from the averaged barotropic stream function over the last two decades of the 21st century) of the linear trends in zonal current and its means over the first two decades of the 21st century are shown in Figure 6. In all the panels of Figure 6, there is a robust feature that the westward currents become intensified close to the coast of Antarctic. But the differences between models are large, with current intensifications being confined to the very upper layer in Figures 6a– 6c, and broader and deeper intensifications in Figures 6d– 6f. In the left column, the positive trends around the thick black solid lines push the zero zonal currents toward the Antarctica, as illustrated by the dashed lines. In contrast, in the last four panels, the shifts of the zero zonal currents are much smaller with either poleward shifts in Figures 6c and 6f or equatorward shifts in Figures 6d and 6e. The large poleward shifts in the location of zero zonal current in Figures 6a and 6b are associated with the increased ACC transports. The results derived from Figures 5 and 6 through analyzing the six selected models are generally supported by the results from analyzing other models with significant changes in the ACC transport (not shown).

Figure 6.

Meridional-depth cross sections of linear trends over the 21st century in zonal current averaged over a 45°longitude sector to the east of the Weddell Gyre centers (yellow ticks in Figure 5) derived from averaged barotropic stream function over 2081–2100. The black contour lines (dashed: negative values) with a contour interval of 0.005 m/s show the zonal velocity averaged over the first two decades of the 21st century (2001–2020 for CMIP3 models and 2006–2025 for CMIP5 models). The thick solid line is the line of zero zonal velocity for the first two decades of the 21st century, and the thick dashed line for 2081–2100.

[31] Averaged potential density changes for the 21st century over the upper 1000 m and over the same corresponding longitude sectors as in Figure 6 are used to illustrate the corresponding density structure changes related to the zonal current changes (Figure 7). We note that the zonal current is determined by the zonal momentum equation, and in the zonal momentum equations of models with coarse resolutions, different eddy viscosity parameterizations are often used for the numerical stability reason, so potential density changes obtained in this way cannot exactly reflect the zonal current changes in Figure 6, in particular when the zonal current changes are small. Nevertheless, there are some notable features in Figure 7. Consistent with the increases in westward flow in the upper layer close to the coast of the Antarctica, the meridional gradients of averaged potential density become larger in these six models. At lower latitudes, the meridional gradient changes are much larger in Figures 7a and 7b, consistent with increased ACC transports in these two models, as baroclinic ACC transport is determined by meridional density gradient (see section 4.2) [Olbers et al., 2004].

Figure 7.

Meridional profiles of potential density averaged over the upper 1000 m and over a 45°longitude sector to the east of the Weddell Gyre centers (yellow ticks in Figure 5) derived from averaged barotropic stream function over 2081–2100. The solid lines are for the averaged potential density anomalies over the first two decades of the 21st century. The dashed lines are obtained by adding linear trends in kg/m 3/century over the 21st century to the averaged potential density anomalies over the first two decades of the 21st century.

[32] The above changes in zonal current, along with changes in zonal transport shown in Figure 4 and gyre axis changes shown in Figure 5, can be used to explain the range of changes in gyre strengths shown in Tables 5 and 6 and in Figures 1 and 3 and to explain the contrasting changes shown in Tables 7 and 8. The shifts of the gyre centers can affect the overall transports in two ways, i.e., through changing the distances between the coast of the Antarctic (or the western boundary) and the gyre centers and through impacting the ocean currents. There are consistently increased westward flows to the south of the gyre axes. However, there are generally poleward shifts of the gyre axes in the models that have increased ACC transports, and the increases in westward flow are small. In contrast, for the models that have reduced ACC transports, there are generally relatively large increases in westward flow and small changes in gyre axes. These changes lead to generally small and less consistent changes in the gyre strengths (as defined in section 2) for the models with increased ACC transports, while there are more consistent and relatively large increases in the gyre strengths for the models with reduced ACC transports.

4 External Processes Responsible for Changes in the Subpolar Gyres

4.1 Wind Stress Forcing

[33] The simplest form of barotropic (vertically integrated from ocean bottom − H to surface η) vorticity equation [see 11.13.6 of Gill, 1982]

display math(1)

where β is the meridional gradient of Coriolis parameter, inline image, ρ is density, and inline image is wind stress (or wind stress minus bottom stress) and inline image is defined as wind stress curl.

[34] This is the so-called Sverdrup balance, which describes that a gyre strength is controlled by wind stress curl. This balance can be used to describe the large-scale interior flow of the subtropical gyres satisfactorily. Using a regional coupled ocean-ice model with fixed buoyancy forcing and northern boundary conditions, Beckmann et al. [1999] found that the seasonal variability of the Weddell Gyre strength is well correlated with the area-averaged wind stress curl over the Weddell Gyre domain.

[35] Figure 8 shows the trends in wind stress curl over the 21st century. Generally, there are positive trends at lower latitudes and negative trends at higher latitudes. Around the gyre axes and to the south of them, the wind forcing becomes more cyclonic. Negative trends can also be seen to the north of the gyre axes in parts of Figure 8b and in parts of the Weddell and Ross Gyres in Figure 8f, or even to the far north of the gyre axes in Figures 8c– 8e. Despite these differences, the cyclonic wind forcing becomes consistently increased in the subpolar region in these six models (in fact in almost all models as shown in Figure 1). As already shown in Figure 1, however, the overall gyre strength changes are not simply related to the changes in wind stress curl averaged over gyre domains in the future runs of CMIP3 and CMIP5 models. For example, models csiro _mk3 _5 and NorESM1-M have large negative trends in wind stress curl (more cyclonic wind forcing) at higher latitudes, but the strengths of some subpolar gyres become reduced (see Tables 5 and 6 and Figure 1).

Figure 8.

Linear trends in wind stress curl over the 21st century. The black lines mark the gyre axes.

[36] To have a further thorough quantitative examination, trends in wind stress curl are averaged in several ways, namely, over a box domain (60°W–75°E, 80°S–55°S), over the Weddell Gyre domains, over the southern limbs of the Weddell Gyres, and over the gyre axes of the Weddell Gyre (from gyre centers to 75°E) (Table 9). Clearly, none of these averaging methods leads to a consistent relationship between the gyre strength changes and changes in wind stress curl among these models that would support a conclusion that more cyclonic wind forcing leads to increases in overall gyre strengths. These results indicate that processes other than local wind stress curl forcing are also important factors in driving changes in the subpolar gyres in these CMIP3 and CMIP5 models.

Table 9. Averaged Trends (10 − 9N/m3 dec − 1) in Wind Stress Curl over a Box Domain (from the Coast of Antarctica to 55°S and from 60°W to 75°E; Second Column), over the Weddell Gyre Domain (Region with Negative Barotropic Stream Function in the Weddell-Enderby Basin; Third Column), over the Southern Half of the Weddell Gyre (Region with westward transport in the Weddell-Enderby basin; fourth column), and over the eastern part of the gyre axis (from the gyre Centers, Marked by the Green Ticks, to the Eastern Boundaries of the Weddell Gyres, as Shown in Figure 5; fifth column)
Model nameBoxGyreSouthernGyre Axis
csiro _mk3 _55.0 − 2.4 − 14.85.8
miroc3 _2_hires − 5.2 − 4.3 − 4.75.1
gfdl _cm2 _1 − 3.6 − 9.9 − 5.9 − 29.7
GFDL-ESM2G − 2.1 − 11.1 − 7.9 − 14.8
MRI-CGCM3 − 5.1 − 21.6 − 44.4154.0
NorESM1-M − 2.3 − 8.9 − 12.2 − 29.8

4.2 The Role of Eddies in the ACC

[37] In the subpolar region, Ekman pumping resulting from cyclonic winds brings denser deep water upward, and strong eddy processes transport lighter water toward this region from lower latitudes across the ACC in the upper layer [Bryden, 1979]. Thus, surface wind forcing tends to overturn the isopycnals, and the eddy-induced buoyancy transport tends to flatten them. Previous studies [e.g., Marshall and Radko, 2003] have shown that these are two primary and competing processes in setting the isopycnal structure in the SO. The domed isopycnal structures in the subpolar region are thus determined by the large-scale wind forcing and the interior buoyancy transport induced by mesoscale eddies. In the models used in this study, the effects of mesoscale eddies are either parameterized or marginally resolved (eddy-permitting in miroc3 _2_hires).

[38] According to the 2-D (meridional-depth) residual-mean circulation theory, the mesoscale eddy process yields inline image [e.g., Andrews et al., 1987; Gent and McWilliams, 1990], where Ψ *  is the overturning stream function driven by eddies, v ′  and ρ ′  are perturbations from their means of meridional velocity and potential density, and ρz is the vertical gradient of potential density. This eddy driven circulation largely balances the wind-driven overturning circulation, inline image (ρo is reference density), in the SO.

[39] In coupled climate models used in this study, due to low resolutions in the ocean component, Ψ *  is generally parameterized as Ψ *  = Ks ρ by adopting the closure inline image [e.g., see Marshall and Radko, 2003], where the mixing coefficient K is either a constant (in most models that employ Gent-McWilliam type eddy parameterization) [Gent and McWilliams, 1990] or a function of sρ [e.g., see Visbeck et al., 1997], and inline image is the slope of isopycnal surface. Thus, in these models, the strengths of eddy-induced buoyancy transports depend on the mixing coefficients and isopycnal slopes (for various eddy parameterizations used in the models, see Tables 1 and 2).

[40] According to previous studies [e.g., Marshall and Radko, 2003], in CMIP3 and CMIP5 models, under increased wind forcing, more cyclonic wind stress curl (see Figures 1 and 8) tends to increase isopycnal slopes, which can consequently cause stronger eddy-induced poleward buoyancy transports in the upper layer. The wind-driven and eddy-driven overturning circulations are two competing processes in changing the locations of the isopycnal domes, with the eddy processes tending to decrease the density just to the north of the domes and hence tending to push the domes poleward, and the Ekman pumping processes tending to increase the density there and hence tending to move the domes equatorward. The net effects on the isopycnal domes depend on the relative strengths of these two competing processes, which are further determined by the local wind forcing and local stratifications at lower and higher latitudes on both sides of the ACC, and the parameterized (or partially resolved in model miroc3 _2_hires) poleward eddy-induced buoyancy transport in these models.

[41] Using thermal wind balance inline image and inline image, where u is geostrophic zonal current and K is a constant mixing coefficient, we have inline image, which offers a dynamic relationship between geostrophic baroclinic transport and parameterized eddy-induced poleward buoyancy transport. (More complicated, but consistent, relationship will be obtained if K is a function of isopycnal slope.) This relationship illustrates that the eddy-induced poleward buoyancy transports are generally stronger in those models with increased ACC transports (hence larger increases in baroclinic transport; for example, see Figures 6a and 6b) than in those models with decreased ACC transports (for example, see Figures 6e and 6f). Therefore, in some models with increased ACC transports and hence increased eddy-induced poleward buoyancy transports, the subpolar gyre axes are pushed poleward, which does not favor the increases in the gyre strengths. (Note that there are also some increases in baroclinic transport in the ACC region in some models with reduced ACC transport, as shown in Figures 6e and 6f, but these increases are too small to lead to increases in total ACC transport due to the narrowing of the ACC region [Wang et al. 2011])

[42] There are certainly some regional differences in gyre size changes for some models with relatively large ACC transport changes (e.g., see Figure 5d and for some other models, not shown). These regional differences cannot be explained by only using the ACC transport. We thus need to do detailed budget analysis in the density equation. This particularly needs three-dimensional eddy-induced buoyancy transport data, which have to be provided in future model inter-comparison projects, or in a more controlled study using a single model.

4.3 Surface Buoyancy Forcing

[43] For the North Atlantic subpolar gyres, previous studies have analyzed the buoyancy forcing to understand the gyre changes [e.g., Bryan et al., 1995 and Häkkinen and Rhines, 2004], through the link between the isopycnal structure and the geostrophic current. In the southern high latitudes, the density change is dominated by salinity change, as the sea water temperature is close to freezing point [see Figure 4 in Wang and Meredith, 2008]. We thus concentrate on the analysis of changes in freshwater forcing to examine the role of buoyancy forcing in driving the changes in the subpolar gyres in these models. Since there are no water flux data available for model miroc3 _2_hires, the result from model gfdl _cm2 _0 is used instead. Note that in this model, the ACC transport increases and the gyre strength changes are small and generally not statistically significant (Table 5).

[44] The trend patterns are quite different in these models (Figure 9). Generally, there are large negative trends (reduced freshwater input into ocean or increased freshwater loss) at lower latitudes and large positive trends at higher latitudes, except for models GFDL-ESM2G and gfdl _cm2 _1. (Note that the positive change in freshwater flux tends to make the surface water less dense, and the negative one tends to make the surface water denser.) Around the gyre axes of, for example, the Weddell Gyre, there are large positive trends in Figures 9a, 9b, 9d, and 9e, particularly in Figures 9a and 9b, while there are large negative trends in Figures 9c and 9f. To the south of the gyre axes and closer to the coast of Antarctica, there are also large discrepancies in the simulated trends: there are generally broad or large positive trends in Figures 9a, 9b, 9c, and 9f, and some large negative trends in Figures 9d and 9e. These patterns of large freshwater flux changes are mainly determined by changes in freshwater transport associated with sea ice circulation changes in a warmer climate. Since freshwater flux consists of precipitation minus evaporation and net sea ice freezing/melting, this can be demonstrated by examining the linear trends in precipitation minus evaporation. In the subpolar region, there are generally smaller positive trends (not shown), induced by increased atmospheric moisture transport associated with the westerly wind strengthening. The locations of the large negative trends in Figure 9 are determined by the location of the simulated sea ice margins in each model, which reflect the large discrepancies in the simulated sea ice extent. gfdl _cm2 _1 has significantly lower sea ice extent than the observed for the satellite record [Parkinson et al. 2006], so the large negative trends appear at very high latitudes (Figure 9e).

Figure 9.

Linear trends in freshwater flux into ocean over the 21st century. Negative (positive) trends indicate reduced (increased) freshwater water input into ocean or increased (reduced) freshwater loss from ocean. The black lines mark the gyre axes.

[45] The pattern of increased freshwater input around the gyre axes and reduced freshwater input at the southern gyre rims tends to decrease the meridional density gradient and hence to weaken the gyres for models csiro _mk3 _5, NorESM1-M, and gfdl _cm2 _1 in Figures 9a, 9b, and 9e, while the pattern in Figures 9c and 9f tends to strengthen the gyres, particularly for the Weddell Gyre and the Australia-Antarctic gyre. However, the gyre strengths in these models do not change in such a way as implied by these changes in buoyancy forcing, suggesting that buoyancy forcing plays a secondary role in driving the changes in the strengths of the subpolar gyres.

[46] Note that these changes in surface buoyancy forcing have more direct effects on the surface layer, but the stratification of water column are strongly affected by the buoyancy transport by the ocean circulation. The above analysis has suggested that the changes in isopycnal structure in the subpolar region can be strongly affected by Ekman pumping and eddy-induced buoyancy transports, with the effects of surface buoyancy forcing being less important.

4.4 Discussions

[47] A more complete form of the barotropic vorticity equation (vertically integrated from ocean bottom -H to surface η) can be written as

display math(2)

where inline image is the vertically integrated friction term, inline image is the vertically integrated nonlinear term, and pb is the bottom pressure [Hughes, 2000]; Hughes and de Cuevas, 2001]. ∇ pb × ∇ H is also called bottom pressure torque term. In the subpolar region, the weak density stratification leads to an equivalent-barotropic current structure [Killworth, 1992] and large bottom velocities [e.g., Klatt et al., 2005 for the observed case]. This leads to non-negligible interactions between the bottom current and bottom topography.

[48] Using results from a high resolution ocean model, it has been found in Hughes [2000] and Hughes and de Cuevas [2001] that in a zonal integral, the vertically integrated friction term and the vertically integrated nonlinear term are small, indicating that the ocean is basically inviscid.

[49] The ocean components of these CMIP3 and CMIP5 models, however, have coarse resolution, except for model miroc3 _2_hires. Large viscosity values are used in the momentum equations in these models for reasons of model stability, and to impose a strong vorticity source within the boundary layers that have weak lateral shears, hence making the balance between wind stress and bottom form stress in a zonal integral invalid. Large viscosity may also distort the momentum changes through large momentum diffusion, leading to unrealistic gyre strength changes. It is thus important to use high resolution models that use smaller viscosity values to simulate the subpolar gyres. This is another reason why model miroc3 _2_hires that has an eddy-permitting ocean component has been included in this analysis. In contrast to the results from several coarse-resolution models that have increased ACC transport, model miroc3 _2_hires has increased ACC transport and contracted subpolar gyres, but no reductions in the overall gyre strengths. This might be due to the lower viscosity used in this high-resolution model. However, results from more high resolution models are needed to confirm the result from this particular model, because this model has subpolar gyres that are too strong and can hence have biased gyre changes in response to climate change.

[50] In equation (2), wind stress curl still clearly remains as an external driving force. The increased cyclonic wind forcing in the subpolar region is responsible for the consistent increases in the westward flow, and is responsible for the generally increased gyre strengths, particularly in CMIP5 models, despite the opposing effects caused by eddy-induced buoyancy transports. We also suggest that the large variations in wind stress curl in zonal direction appear to be an external force to drive the relatively large shifts of the Weddell Gyre centers (see Figures 5d, 5f, 8d, and 8f).

[51] Our analysis has revealed that eddy-driven circulation in the ACC is also important to the gyre changes, as reflected by the analyzed relationship between the gyre changes and the changes in the ACC transport. In future studies, to have a more quantitative analysis of gyre changes, the eddy-induced buoyancy transports have to be estimated to quantify the changes in isopycnal structure in the subpolar region. Future model inter-comparison projects should provide necessary data for such diagnostics, because most CMIP3 and CMIP5 models are using non-constant mixing coefficients in their eddy parameterization schemes that are calculated as functions of stratification at every time step.

[52] Though the main purpose of this paper is to examine the external processes that drive the gyre changes, internal processes in equation (2) need to be analyzed to offer more quantitative understanding of the gyre strength changes. Those terms in equation (2) can be integrated along each gyre axis from the gyre center to the eastern boundary to quantify the contributions of these terms to the changes in the integrated southward flow. In equation (2), the bottom pressure torque term, being generated by interactions between the bottom topography and currents originating from wind and surface buoyancy forcing, can be calculated by using bottom pressure and topography data. Bottom pressure and bottom topography data have been provided for some CMIP5 models, but more model implementation details are still needed to have an accurate calculation of this term, such as how the bottom pressure data are filtered or smoothed or interpolated, and what kind of schemes are used to treat the bottom topography. Also, the available monthly mean data may not been fine enough to estimate the nonlinear terms. These analyses should be done in future studies when modelling centers can provide relevant data and their model details.

5 Concluding Remarks

[53] Westerly winds consistently increase in the future runs of CMIP3 and CMIP5 models, albeit with different magnitudes. This leads to consistent increases in the cyclonic wind forcing in the subpolar region and an intensification of the westward flow close to the coast of Antarctica. This is generally consistent with increased overall gyre strength, particularly in the CMIP5 models.

[54] However, there are large discrepancies in the simulated intensifications of the westward flow and in the changes to gyre axes, with the discrepancies being larger in CMIP3 models than in CMIP5 models. The intensifications of the westward flow are strong, deep, and broad in those models with decreased ACC transports, in contrast to the changes in the models with increased ACC transport. In the models with decreased ACC transports, the gyre axes generally move equatorward or have little shifts, while there are relatively large poleward shifts of the gyre axes in those models with increased ACC transports. The equatorward or small shifts of the gyres, along with the increases in the westward flow, lead to consistent, significant, and relatively large increases in gyre strengths in those models with decreased ACC transports; in contrast, the poleward contractions of the gyres in those models with increased ACC transport, though accompanied with some increases in the westward flow close to the coast of Antarctica, lead to relatively small and less consistent changes in the gyre strengths. Given that the wind stress changes are relatively consistent across models, the differences between model projections for the gyres are mainly due to different processes taking place within the ocean component of the models.

[55] The isopycnal structures in the subpolar region are primarily determined by large-scale wind forcing and meso-scale eddies [e.g., Marshall and Radko, 2003]. These are two competing processes in determining the locations of the isopycnal domes, i.e, the locations of the gyre axes. Our analysis has shown that the ACC transport responses to relatively consistent wind forcing changes show great diversity, reflecting large uncertainties in the simulated eddy-driven circulation. These uncertainties lead to inconsistent changes in the gyre axes, and consequently lead to inconsistent changes in the overall gyre strengths.

[56] During the past several decades, the westerly jet intensified on its poleward side [Thompson and Solomon, 2002; Marshall, 2003]. An associated response of the subpolar gyres to this wind forcing change has, however, not been observed. The model results show that the increased cyclonic wind forcing can intensify the westward flow in the subpolar region (particularly in the upper layer and close to the coast of the Antarctic). It is, however, not clear how the gyre axes changed. A recent analysis [Böning et al. 2008] suggested that the isopycnal structure in the SO has changed very little. If this was also the case for the locations of isopycnal domes in the subpolar region, the gyre axes might remain unchanged and thus the overall gyre strengths might have been intensified.

[57] The intensification of the westward flows in the southern limbs of the gyres have strong implications for the mass balance of ice shelves and the stability of the Antarctic ice sheets. Such intensifications could re-direct the coastal currents that can penetrate below ice shelves [Hellmer et al. 2012]. Also, more cyclonic wind forcing could bring more heat upward from deeper ocean. Such changes may have led to the enhanced bottom melting of polar ice shelves that has been observed during the past decades [Rignot and Jacobs, 2002]. This could be a partial explanation for some of the observed freshening in the subpolar regions (e.g., the Ross Sea) [see Jacobs et al., 2002], for which melting of glacial ice has been suggested as a cause.

[58] Our analysis has revealed that there are uncertainties in the projections of the subpolar circulation in the current generation of climate models, particularly in the projections of the gyre axes. The changes in the gyre axes are controlled by a number of key processes that are represented quite differently among models. They are part small-scale, such as the oceanic eddy processes, and part large-scale, such as the atmospheric momentum and buoyancy forcing.

[59] In the current coarse resolution climate models, eddy processes can only be parameterized, which is a crude approximation and varies among models. This considerably adds to the uncertainty about the projection of the future evolution of the ACC and the subpolar gyres. It is thus very important to push the development of global coupled climate models that include an eddy-resolving ocean.

[60] Under increased greenhouse gas forcing and forcing associated with ozone changes, these models generally project intensifications of the jet on its poleward side, but with different magnitudes and spatial patterns. During the first half of the 21st century, Chemistry-Climate Model Validation (CCMVal) models, which have improved resolving of key stratospheric processes and interactive atmospheric chemistry processes, project the weakening of the westerlies under increased greenhouse gas forcing in the summer season. This is likely caused by the simulated recovery of the ozone hole and better representations of the stratospheric processes [Eyring et al., 2007; Son et al., 2008]. These uncertainties associated with atmospheric processes have to be substantially reduced as well before we can make a reliable projection of the subpolar ocean circulation.

Appendix A: Model Drifts in Pre-industrial Runs

[61] Due to short spin-up times in some models, there are quite large internal drifts in their control runs that can continue into the subsequently forced runs. In order to ensure that the changes in the projection runs are not caused by these internal drifts, we diagnosed the model internal drifts in each control run (Tables A1 and A2). Note that there are no pre-industrial data available for some models, namely, models 15 and 18 for CMIP3 models, and models 1, 4, and 8 for CMIP5 models. These changes in the model control runs are compared with those listed in Tables 5 and 6 to determine the extents to which the projected changes are affected by the model spurious unforced trends, and are used to adjust the forced trends in the future runs. They are particularly useful for us to select models that have significant gyre changes caused by external forcing rather than by the internal drifts.

Table A1. Linear Trends T (Sv/century for Transport, and 1010m2/dec for the Westward Flow Areas) and Their Significance Levels S (%)in ACC Transport Through Drake Passage, Westward Flow Area of the Subpolar Gyres (WF), Weddell Gyre Strength (WG), Ross Gyre Strength (RG), and Australia-Antarctic Gyre Strength (AG) over the Whole Period of Their picntrl (pre-industrial) Runs of 17 CMIP3 Modelsa
No.Model nameACC T(S)WF T(S)WG T(S)RG T(S)AG T(S)
  1. aNote that data of models 15 and 18 are not available.
1cccma _cgcm3 _12(0) − 8(0)0(95)0(84)0(37)
2cccma _cgcm3 _1_t630(12)0(89)0(51)0(92)0(66)
3cnrm _cm3 _0 − 20(0)46(0) − 6(0) − 2(0) − 5(0)
4csiro _mk3 _016(0) − 5(36)5(10)11(12)1(40)
5csiro _mk3 _50(32) − 5(16)1(31)0(40)0(39)
6gfdl _cm2 _07(0) − 18(6) − 10(30)0(70) − 1(12)
7gfdl _cm2 _1 − 2(29)4(30) − 3(82)0(77) − 1(16)
8giss _aom − 11(53)0(99) − 25(0) − 4(23) − 22(0)
9giss _model_e_r − 7(0)4(21)0(89) − 2(81)1(62)
10ingv _echam4 − 5(15) − 4(3)2(30)0(41) − 2(1)
11ipsl _cm40(21) − 11(0)2(0)1(2)0(10)
12miroc3 _2_hires − 1(2)7(3)0(79)0(96) − 1(37)
13miroc3 _2_medres0(87)6(65) − 2(63)0(0) − 1(13)
14miub _echo_g − 1(29)0(94) − 3(12)0(90)0(96)
16mri _cgcm2 _3_2a2(0)1(53) − 1(8)0(99)0(67)
17ncar _ccsm3 _00(80)2(56)0(64)0(98)0(77)
19ukmo _hadcm313(11)11(7) − 3(8) − 17(0) − 4(38)
Table A2. Linear Trends T (Sv/century for Transport, and 1010m2/dec for the Westward Flow Areas) and Their Significance Levels S (%) in ACC Transport Through Drake Passage, Westward Flow Area of the Subpolar Gyres (WF), Weddell Gyre Strength (WG), Ross Gyre Strength (RG), and Australia-Antarctic Gyre Strength (AG) over the Whole Period of Their piControl (Pre-Industrial) Runs of 11 CMIP5 Modelsa
No.Model nameACC T(S)WF T(S)WG T(S)RG T(S)AG T(S)
  1. aNote that data of models 1, 4, and 8 are not available.
2CanESM25(0) − 11(2)1(16)0(68)1(0)
3GFDL-ESM2G − 1(78)14(38)1(56) − 3(14) − 2(2)
5GISS-E2-R0(60)0(88)1(47) − 3(32)4(22)
6HadGEM2-CC − 3(13)7(39) − 3(37) − 3(21) − 2(37)
7HadGEM2-ES0(89)4(62) − 1(55) − 1(52) − 6(6)
9IPSL-CM5A-MR − 5(21)11(45)0(87)1(44)0(93)
10IPSL-CM5B-LR9(0) − 7(31)7(0)0(45)1(74)
11MIROC-ESM − 7(0)9(0) − 2(6)1(0)0(29)
12MIROC-ESM-CHEM − 3(0)3(10)1(11)0(38)0(53)
13MRI-CGCM3 − 2(0)12(3) − 4(17)0(2)1(24)
14NorESM1-M1(31) − 14(1)3(21)2(8)0(95)

Acknowledgments

[62] We acknowledge the modeling groups, the Program for Climate Model Diagnosis and Intercomparison (PCMDI) and the WCRP's Working Group on Coupled Modelling (WGCM) for their roles in making available the WCRP CMIP3 multi-model data set, and the British Atmospheric Data Center for their great work on archiving CMIP5 data. We thank Mike Meredith, Andrew Meijer, and Jan Zika for helpful discussions and are grateful to Meredith's comments on an earlier draft. The constructive comments on the manuscript from three anonymous reviewers are acknowledged. Part of this work was done at the British Antarctic Survey, and monthly mean data were processed at the server of the British Atmospheric Data Center. This work is supported by Chinese National Key Basic Research Program (2010CB950301), by China National Natural Science Foundation (NSFC) Project (41276200), by the Special Program for China Meteorology Trade (grant GYHY201306020), and by a project funded by the Priority Academic Program Development of Jiangsu Higher Education Institutions (PAPD)