Quantifying uncertainty in future Southern Hemisphere circulation trends


Corresponding author: P. A. G. Watson, Atmospheric, Oceanic and Planetary Physics, Clarendon Laboratory, University of Oxford, Parks Road, Oxford OX1 3PU, UK. (watson@atm.ox.ac.uk)


[1] The Antarctic polar night jet has intensified during spring in recent decades due to stratospheric ozone depletion and rising greenhouse gas (GHG) concentrations and this has had substantial effects on the region's climate. GHG concentrations will rise over the 21st century whereas stratospheric ozone is expected to recover and there is uncertainty in future southern hemisphere (SH) circulation trends. We examine sensitivity to the physics parameterisation of the 21st century SH circulation projection of a coupled atmosphere-ocean General Circulation Model and the sensitivity of the contribution from stratospheric ozone recovery. Different model parameterizations give a greater range of future trends in the position of the tropospheric jet than has been found in previous multi-model comparisons. Ozone recovery causes a weakening and northward shift of the DJF tropospheric jet. Varying the physics parameterization affects the zonal wind response to ozone recovery of the SON stratosphere by ∼10% and that of the DJF troposphere by ∼25%. The projected future SAM index changes with and without ozone recovery and the SAM index response to ozone recovery alone are found to be strongly positively correlated with projected 21st century global warming.

1. Introduction

[2] This paper presents results from a perturbed physics ensemble (PPE) of a coupled atmosphere-ocean General Circulation Model (AOGCM) in the climateprediction.net (CPDN) experiment to project how the southern hemisphere (SH) circulation will evolve in the twenty-first century, isolating the response to the recovery of ozone concentrations. The effect of perturbing the physics parameters in the model is examined in order to measure and better understand their effect on the circulation and quantify the uncertainty due to uncertainty in the parameters.

[3] The Southern Annular Mode (SAM) is a large-scale pattern in the SH troposphere that dominates the mid–high latitude circulation on weekly and monthly timescales. High values of the SAM index are associated with a stronger and more poleward eastward tropospheric jet around Antarctica, with a weakening of the wind on the equatorward side of the jet. Several studies have found that since the 1970s there has been an increase in indices of this pattern during the southern summer (December–January) [Thompson and Solomon, 2002; Thompson and Wallace, 2000; Marshall, 2003]. Increasing stratospheric winds during the southern spring and summer have also been observed and these have been found to couple with summer tropospheric winds and the SAM [Thompson and Solomon, 2002].

[4] Changes in the southern winds have been found to explain most of the recent cooling of the Antarctic interior and half of the warming of the Antarctic peninsula that has occurred since the 1970s [Thompson and Solomon, 2002]. The SAM index has also been linked to temperature and precipitation anomalies in the southernmost countries in the SH [Gillett et al., 2006; Thompson et al., 2011].

[5] Concentrations of stratospheric ozone at high southern latitudes in the southern spring have decreased since the mid-1970s by about half [Solomon, 1999]. The trend towards a later break-up of the polar vortex has been linked to this decline [Zhou et al., 2000] and spring and summer stratospheric winds and the SAM index for a given year have been linked to variability in stratospheric ozone concentrations [Thompson and Solomon, 2002].

[6] Previous studies that have modelled the circulation over the twentieth century have variously concluded that GHG forcings alone can explain up to half of the observed trends and that a combination of GHG and ozone forcings are able to replicate the trends well [e.g., Gillett and Thompson, 2003; Shindell and Schmidt, 2004; Marshall et al., 2004; Arblaster and Meehl, 2006; Cai and Cowan, 2007; Son et al., 2009]. Studies that have modelled the effect of the ozone forcing alone have found that this can explain much of the trends in the spring and summer months [Gillett and Thompson, 2003; Arblaster and Meehl, 2006]. It has been found that natural forcings alone have not contributed significantly to the trends [Arblaster and Meehl, 2006] and that the trends are well outside the range of natural variability [Marshall et al., 2004].

[7] Studies that have modelled the SH circulation in the twenty-first century have found that GHG forcings contribute to a trend of increasing SH zonal mean zonal wind (ZMU) and that a recovery in stratospheric ozone concentrations will counteract this trend in summer. Some studies find that overall there is a weak positive trend in the summer SAM index [Shindell and Schmidt, 2004; Arblaster and Meehl, 2006; Arblaster et al., 2011]. However, projections from models participating in the Chemistry-Climate Model Validation project phase 2 (CCMVal2) and in the Intergovernmental Panel on Climate Change Fourth Assessment Report (AR4) indicate both positive and negative future trends [Son et al., 2010], indicating there is uncertainty arising from the model structure.

[8] A major source of uncertainty in model output is that due to the parameterizations used in the model to represent physics that is not resolved in the model grid (e.g., cloud processes), on top of that due to choices made for the structure of the model (e.g., the grid resolution) and internal variability due to the choice of initial conditions. One way of addressing this uncertainty is to use perturbed physics ensembles (PPEs), in which the same base model is run many times with different parameter values and initial conditions [e.g., Stainforth et al., 2005]. Previous PPE experiments have found that multi-model comparisons do not sample all the uncertainty in global mean temperature projections associated with the physics parameterisation [e.g.,Rowlands et al., 2012], and the same may be true for projections of future circulation trends. Thus here we quantify uncertainty in projections of future circulation and in the circulation response to ozone recovery alone associated with the physics parameterisation.

2. Methodology

[9] The ensembles of simulations used the HadCM3L climate model and were run from the year 2000 to 2080. HadCM3L is an AOGCM developed by the UK Met Office which has a three-dimensional dynamic atmosphere and ocean with a horizontal resolution of 2.5° in latitude and 3.75° in longitude. The atmosphere and ocean have 19 and 20 levels in the vertical respectively and the model top is at 5 hPa. The model solves a quasi-hydrostatic version of the primitive equations and parameterized processes include land-surface processes, boundary layer physics, gravity wave drag, large-scale cloud and precipitation, radiation and convection [Frame et al., 2009].

[10] Three different stratospheric ozone scenarios are considered. One is a time-evolving projection of ozone concentrations for the twenty-first century, produced using an atmospheric chemistry model assuming that emissions of ozone-depleting substances decrease in accordance with the Montreal Protocol and that climate changes in accordance with a previous HadCM3 experiment [Johns et al., 2003]. This scenario is used in projections of 21st century ZMU (section 3.1). Two scenarios have constant annual cycles of ozone for the years 2000 and 2075, used in the ozone recovery experiment (section 3.2). The 2000 and 2075 cycles were created by taking an eleven-year mean concentration for each month from the time-evolving case, from December 1994 to November 2005 inclusive for the 2000 case and from December 2069 to November 2080 for the 2075 case.Figure 1a shows the SON mean column ozone south of 60°S for the scenarios. Figures 1b and 1c show the annual cycle of column ozone as a function of latitude for the constant 2000 and 2075 scenarios. The largest differences in ozone levels occur polewards of 60°S at pressures around 100 hPa. The SON season shows the largest difference in ozone levels between 2000 and 2075, and previous studies have found strong DJF tropospheric signals due to ozone depletion/recovery [e.g., Arblaster and Meehl, 2006; Son et al., 2008], so results for these seasons are the focus of this paper.

Figure 1.

(a) A timeseries of the SON area-weighted mean column ozone south of 60°S in Dobson units (DU) for the ozone scenarios considered in this study. (b and c) The column ozone as a function of season and latitude for the constant 2000 and 2075 annual cycles, showing a large deficit in the 2000 case south of 65°S in the southern spring.

[11] The results from eight separate physics parameterizations were analyzed. The parameter sets were manually selected from those that give realistic hindcasts of global mean surface temperature and which do not require large top-of-atmosphere flux adjustment. The 2000s mean SON and DJF ZMU has a good resemblance to that in the ECMWF reanalysis (ERA-40) [Uppala et al., 2005] in all the parameterizations (not shown). There is not a consensus on how to use more detailed information on model climate to decide if a model projection is plausible [Knutti, 2010], so we did not place further restrictions on the parameterizations we could choose from. The range of 2000–2080 global mean surface temperature changes across these paramaterizations is 2.6–4.3 K, which is similar to the range considered likely by Rowlands et al. [2012] under the SRES A1B GHG scenario based on observational constraints. Since we show in section 3.3 that our projected circulation trends are very well correlated with the projected global warming, this also implies that the range of our projected circulation trends is reasonable. The parameter values are provided as auxiliary material. For each parameterization and ozone scenario an ensemble of ∼100 simulations was run in which each was started with a different atmospheric and oceanic state in order to sample sensitivity of the results to the initial conditions. The use of large ensembles reduces initial condition uncertainty so that effects solely due to the parameterizations can be observed.

[12] All the simulations used the same solar and volcanic forcings and the SRES A1B GHG and sulphate scenario. The volcanic forcing is a single 80-year sample from estimated forcing since 1400 and includes eruption events, and is described in detail byRowlands et al. [2012]. Thus differences between the simulations are due solely to the physics parameterization, the initial conditions and the ozone scenario.

[13] Perturbing the parameters of HadCM3L will sample a wide variety of plausible tropospheric behaviours. HadCM3L does not have a fully resolved stratosphere and the possible stratospheric behaviours will probably not be fully explored. Models without a well resolved stratosphere have previously been used to show a tropospheric response to ozone depletion/recovery in agreement with observations and results from models with a well resolved stratosphere [e.g., Shindell and Schmidt, 2004; Arblaster and Meehl, 2006; Cai and Cowan, 2006; Arblaster et al., 2011]. A model with a well-resolved stratosphere and coupled chemistry may have additional uncertainty on top of what is found here.

3. Results and Discussion

[14] For brevity this section presents plots from just one parameterization, for which the ZMU trends are intermediate amongst those of all the parameterizations. The main differences between the parameterizations in the SH circulation 21st century projections and response to ozone recovery are in the SON-mean polar night jet (PNJ) strength and the DJF-mean tropospheric winds between ∼30–70S, as detailed in the following sections.

3.1. Projections of 21st Century ZMU

[15] Figure 2ashows in color the 2020s–2070s trends in the SON ZMU in the SH as a function of latitude and pressure averaged over the ensemble members run with the time-evolving ozone scenario, with the ensemble and temporal mean ZMU contours for the 2020s plotted on top to show the position of the jet. The trend is calculated from the 2020s as a large volcanic eruption in the simulations disturbs the trends in the 2010s, so that the mean ZMU in the SH increases linearly only from the 2020s onwards. This shows an asymmetric dipole pattern in the ZMU trend, with a strong intensification of the eastward winds in the stratosphere south of the position of maximum ZMU of up to 0.16 ms−1 yr−1 (which may be contributed to by both changes in the PNJ maximum ZMU and in the timing of its breakdown) and a slight decrease in intensity just to the south of up to 0.07 ms−1 yr−1. The pattern of the trends is qualitatively similar for all the parameterizations and the ranges on these values are 0.12–0.31 ms−1 yr−1 and 0.05–0.09 ms−1 yr−1 respectively.

Figure 2.

Projections of 21st century zonal mean zonal wind (ZMU): (a) the ensemble mean SON ZMU trends for the 2020s to the 2070s (colors) and the ensemble and temporal mean SON ZMU for the 2020s in ms−1 (contours) for one parameterization. (b) The same for DJF ZMU below 100 hPa. The ZMU response in the ozone recovery experiment: (c) the ensemble and temporal (2020s–2070s) mean differences between the SON ZMU in the constant 2075 and constant 2000 ozone scenarios for one parameterization. (d) The same for the DJF ZMU below 100 hPa. Trends in Figures 2a and 2b and differences in Figures 2c and 2d are shown where they are greater than two standard errors from zero.

[16] Figure 2b shows the same for the DJF troposphere. Below the 200 hPa level, there is a slight increase in intensity to the north of about 50°S of ∼0.01 ms−1 yr−1and a decrease of similar magnitude to the south, indicating a northward shift of the tropospheric jet. The pattern resembles the CCMVal2 multi-model mean trend in structure and magnitude [Son et al. 2010], showing HadCM3L is able to produce a realistic tropospheric response similar to that of stratosphere-resolving models. The pattern is variable for the different parameterizations, with just two out of the eight showing a dipole structure in the wind trends like that inFigure 2b, three showing a dipole structure of the opposite sign and the other three showing an intensification of the ZMU across the width of the jet. Here we define the SAM index as the difference between the mean 500 hPa ZMU at 50–70S and 30–50S, which are the latitude ranges encompassing the flanks of the jet where the dipolar ZMU anomalies associated with the SAM are centred [Thompson and Wallace, 2000]. (The DJF-mean SAM index thus defined in ERA-40 between 1979–2002 has a correlation of 0.9 with that produced by the NOAA Climate Prediction Centre, which is based on a principal component time series. Our data does not have the temporal resolution required to construct an index based on principal component analysis.) Trends in this index range between −0.015 and 0.030 ms−1 yr−1 across the parameterizations, with errors ∼0.0003 ms−1 yr−1, with six of the parameterizations giving an increase in the SAM index. The trend in the position of the 850 hPa ZMU maximum ranges between −0.027°/yr and 0.008°/yr. This is a considerably larger range than reported by Son et al. [2010, Figure 5d] in the CCMVal2 models (−0.008°/yr to 0.012°/yr) and the AR4 models (−0.014°/yr to 0.012°/yr), and only part of these ranges is due to differences in model parameterizations.

[17] These differences show that the projected SH circulation trend is sensitive to the parameterization, and that uncertainty associated with the physics parameterization has not been fully explored in previous multi-model comparisons.

3.2. Ozone Recovery Experiment

[18] The differences in ZMU trends between ensembles run under the time-evolving and constant 2000 ozone scenarios were not highly statistically significant, so to see the effect of varying the parameterization on the response to ozone recovery alone, the 2020s–2070s mean differences between simulations run with the constant 2075 and 2000 ozone scenarios were used. The correlation coefficient of the ZMU differences in consecutive decades is approximately 0.1 or less at each latitude and pressure, so each decade is taken to be independent for significance tests. The constant 2075 and 2000 scenarios do not show significantly different ZMU or zonal mean temperature trends (not shown), indicating that the influence of ozone recovery does not change substantially as climate changes. Thus the time-averaged difference can be interpreted as the mean of the steady-state difference between simulations with ozone concentrations at 2075 levels and at 2000 levels.

[19] Figure 2c shows the 2020s–2070s mean difference in the SON ZMU between the constant 2075 and the constant 2000 ozone scenarios. The largest difference in ZMU occurs at about (60°S, 30 hPa) where the winds for the constant 2075 ozone case are about 5.9 ms−1 less than those for the constant 2000 ozone case, or about 10% of the maximum ZMU, with errors ∼0.01–0.1 ms−1. The differences are qualitatively similar for all the parameterizations and the maximum ZMU difference ranges between 5.1–5.9 ms−1. The maximum difference from the centre of this range is therefore ∼10% of the central value, which gives a measure of the uncertainty of the response to ozone recovery.

[20] Figure 2d shows the DJF tropospheric 2020s–2070s mean ZMU differences between the constant 2075 and the constant 2000 ozone scenarios. These differences are negative near 60°S and a smaller positive difference is centred at about 40–45°S throughout the depth of the troposphere, indicating a weaker and more northerly jet and a more negative SAM index. The difference in the SAM index is −2.1 ms−1, with error 0.02 ms−1.

[21] The pattern of DJF ZMU differences is qualitatively similar for all the parameterizations. The range on the SAM index difference is −1.3 ms−1 to −2.2 ms−1. The difference from the centre of this range therefore reaches up to 25%. This quantifies the uncertainty of the response of the troposphere to ozone recovery associated with tropospheric physics. The correlation between the SAM index difference and the maximum stratospheric SON ZMU difference across the parameterizations is 0.77, indicating that ∼60% of the variance in the tropospheric behaviour between parameterizations is associated with variations in the stratospheric behaviour.

3.3. Relationship Between the Circulation Response and Projected Global Warming

[22] Arblaster et al. [2011] showed that the SAM response to increasing GHGs is positively correlated with the climate sensitivity across different CMIP3 models. Here we confirm this result and extend this work by showing the DJF SAM index responses to ozone recovery and projections of the 21st century SAM index are also strongly correlated with the projected 21st century global mean surface temperature change across the parameterizations.

[23] Figure 3ashows the 2020s–2070s DJF SAM index trend due to non-ozone forcings in simulations run with the constant 2000 ozone scenario, which is principally the response to rising GHGs, plotted against the 2000s–2070s mean temperature change for each parameterization. The correlation of 0.84 is a bit stronger than that found byArblaster et al. [2011] in their comparison of models forced by rising GHGs. Figure 3b shows the same for the mean 2020s–2070s differences in the SAM index response in the ozone recovery experiment (section 3.2), which exhibit a strong positive correlation of 0.90. Thus the SAM response to ozone recovery is less negative and weaker for parameterizations with larger 21st century temperature changes. Figure 3c shows the same for the SAM index trend in projections of the 21st century (section 3.1) and again these show a strong positive correlation of 0.87, so the projected SAM trend is stronger for parameterizations with greater global warming. The correlations are similar if the SAM index is defined according to the ZMU at mid-tropospheric levels other than 500 hPa. Thus parameterizations that project greater global warming also project a more positive SAM response due to rising GHGs, a less negative SAM response due to ozone recovery, and consistent with these results a more positive trend overall.

Figure 3.

(a) Plot of the 2020s–2070s SAM index trend in runs with constant 2000 ozone for each parameterization against the ensemble, decadal and global mean surface temperature change between the 2000s and the 2070s. (b) Plot of the SAM index 2020s–2070s difference in the ozone recovery experiment against the global surface temperature change. Note the SAM index axis values are negative. (c) Plot of the SAM index 2020s–2070s trend in projections of the 21st century with time-evolving ozone against the global surface temperature change. A strong positive correlation is found in each case.

[24] The reason for these strong correlations is not clear. The SAM index response has correlations above 0.9 with the DJF SH tropical (equatorwards of 20°S) and mid-latitude (40–60°S) tropospheric zonal mean temperature responses and with the ZMU response around 70°S and 40 hPa in the ozone recovery experiment and in the 21st century projections with and without ozone recovery. The correlations of these responses with the projected global warming are also above 0.9. Whether changes in these regions cause a SAM response or whether these changes are themselves just part of changes in the SAM is not clear from these experiments. The projected global warming has a correlation of 0.77 with the SON-mean PNJ strength, defined as the maximum ZMU at 30 hPa in the 2000s. This has a correlation of 0.84 with the SAM response in the ozone recovery experiment and correlations near 0.5 with SAM trends in 21st century projections with and without ozone recovery. Therefore parameterizations with greater global warming tend to have a stronger SON-mean PNJ, which may respond differently to ozone and GHG forcing and give a different tropospheric response. These correlations are less strong than between the SAM index responses and the global warming, however, so other factors may be more important. There is no clear relationship between the SAM index response and any individual model parameter.

4. Discussion and Conclusions

[25] We have discussed the sensitivity to changing model physics parameters of 21st century projections of spring and summer SH circulation and the role of the recovery of stratospheric ozone concentrations in determining these changes, using results from PPEs in the CPDN project. The parameterizations disagree about whether the DJF tropospheric ZMU will intensify and whether there will be a northwards or southwards shift of the tropospheric jet, with greater spread in the trend in the jet position than found in the AR4 and CCMVal2 multi-model ensembles bySon et al. [2010], showing that these multi-model ensembles do not fully sample uncertainty associated with the physics parameterization.

[26] We have shown the circulation response to recovery of stratospheric ozone is sensitive to the model parameterization. The uncertainty of the maximum SON ZMU difference resulting from recovery of stratospheric ozone with regard to varying the model parameterization is ∼10%. The response to ozone recovery of the DJF troposphere is a northwards shift and a reduction in strength of the tropospheric jet. The uncertainty of the magnitude of this pattern is ∼25%. This helps to constrain the error on previous modelling results due to the choice of model parameterization.

[27] Furthermore, we have found that 21st century projections of DJF ZMU and its response to ozone recovery and non-ozone forcings separately are closely linked to the projected 21st century global mean surface temperature change. Parameterizations with greater global warming exhibit a smaller, less negative SAM response to ozone recovery and a greater, more positive change over the 21st century both with and without ozone recovery, with correlations around 0.9. This implies that the differences in model physics that give rise to different amounts of global warming and to different circulation trends are closely related.


[28] Peter Watson is supported by a Natural Environment Research Council studentship. David Karoly was supported by the Australian Research Council through the Discovery Projects funding scheme (project FF0668679). Myles Allen was supported by the Natural Environment Research Council HYDRA project (NE/I00680X/1). David Lee was supported by the UK Department for Transport through a contract with Manchester Metropolitan University. We wish to thank the anonymous reviewers for their comments on an earlier draft of this paper.

[29] The Editor thanks the two anonymous reviewers for their assistance in evaluating this paper.