An analysis of the simulated Southern Hemisphere, mid-to-high-latitude climate of the Last Glacial Maximum

Abstract

Climates of the past, such as the Last Glacial Maximum (LGM), give a useful insight into the response of atmosphere and ocean processes to various forcings that were different to the present day. The use of General Circulation Models aids in understanding past climates in combination with the available proxy data. In this study, a selection of models from the Paleoclimate Modelling Intercomparison Project was used to assess the changes in Southern Hemisphere atmospheric variability on seasonal to interannual time-scales. The evidence from two of the models suggests a seasonal deepening of the circumpolar trough (CPT) during the LGM relative to present day, with enhanced tropospheric westerlies in all seasons. The third model suggested there was a weakening of the CPT in all seasons except austral spring, where a strengthening was simulated. Finally, the work considers how the Southern Annular Mode may have been different in the LGM for the three models under consideration.

Keywords:

• Southern Hemisphere
• Last Glacial Maximum
• semiannual oscillation
• southern annular mode
• general circulation model
• PMIP2
• circumpolar trough

Introduction

The development of long-term proxy data for past climates has enabled a more thorough understanding of previous climate regimes that existed throughout the Earth's history. The work in this study focuses on the climate of the Last Glacial Maximum (LGM) approximately 21,000 years before present (yr BP) when extensive ice sheets occurred throughout the Northern and Southern Hemisphere mid-to-high-latitudes (see Jansen et al. 2007 for a review). The presence of these ice sheets and the associated changes in surface temperature had a profound effect on the global climate, with surface air temperatures several degrees colder than at present.

The circulation of the Southern Hemisphere (SH) has been shown to be particularly sensitive to temperature changes at high latitudes associated with the LGM and also at the termination of the LGM. The studies by Govin et al. (2009) and Anderson et al. (2009) indicated that the SH westerlies migrated north during the LGM, which reduced the transport of relatively warm deep waters into the surface ocean around Antarctica, allowing ice shelves and sea ice to extend into the Southern Ocean. Likewise, at the termination of the LGM, Toggweiler (2009) suggested that Antarctic warming caused a poleward shift in the SH westerlies, which increased the upwelling of relatively warmer waters and promoted a feedback to reduce the northward extent of Antarctic ice shelves and sea ice. The apparent sensitivity of the circulation in SH mid-to-high-latitudes is likely to be an important factor controlling the way the system may respond to future climate forcing.

The aim of this study is to analyse the changes in the SH mid-latitude circulation in several General Circulation Model (GCM) integrations for the LGM on both seasonal and interannual timescales. Section 2 (Materials and methods) describes the models used and Section 3 (Results and discussion) looks at how each model represents the semiannual oscillation, which is a seasonal mode of variability in the SH. The Results and discussion section also looks at how the SH westerlies vary in each of those models and links those changes to the semiannual oscillation (SAO) and also to the hemispheric meridional temperature gradient. The final part of the Results and discussion identifies the leading mode of SH seasonal to interannual variability and how it may have differed at the LGM while linking the results to the previous two sections. The conclusions are given in the final section.

Materials and methods

PMIP2 data and models

To analyse the differences in Southern Hemisphere (SH) climate between the LGM and the present day we make use of the Paleoclimate Modelling Intercomparison Project phase 2 (PMIP2) data archive (see Joussaume & Taylor 2000; for more detail on the experiment along with http://pmip2.lsce.ipsl.fr/). The PMIP2 repository contains data for several atmospheric variables at various time averages and follows a similar set-up to the Multi-Model Dataset (MMD), held at the Program for Climate Model Diagnostics and Intercomparison (PCMDI) and employed by the Intergovernmental Panel on Climate Change (IPCC; see Randall et al. 2007).

To choose suitable models, we reviewed an initial analysis of all the available PMIP2 models by Braconnot et al. (2007); see their Table 3, which identified that some models had drifts in surface air temperature (at 2 m) greater than or equal to 0.05 K/century. Braconnot et al. (2007) indicated that these models may not have had a fully closed energy budget; these models were excluded from the analysis.

Other models, which did not show significant drift in the equilibrium climate states, were available for the LGM in the PMIP2 data archive. One model was IPSL-CM4-V1-MR, which was used in the analysis by Rojas et al. (2009) but was found by Ackerley and Renwick (2010) to be deficient in representing SH mid-to-high-latitude seasonal climate variability (in particular the SAO) and so was not considered in this study. The other models listed in Braconnot et al. (2007) were not used as they were deficient in the diagnostics needed to undertake the analysis in this study (such as air temperature at 500 hPa).

The models that did fit the above requirements (little drift, capable of representing SH mid-to-high-latitude climate and data availability) were CCSM3, FGOALS-g1.0 (from now FGOALS) and HadCM3M2. The CCSM3 and HadCM3M2 models lie within the spread of the Coupled Model Intercomparison Project 3 (CMIP3) ensemble for various climate variables analysed by Randall et al. (2007), which further justifies their use in this study. However, Masson and Knutti (2011) indicate that the version of FGOALS used by CMIP3 does not represent surface air temperature well relative to reanalysis data. The work by Randall et al. (2007), however, suggests that the biases in surface air temperature in FGOALS are mainly within the Northern Hemisphere and the representation of the Southern Hemisphere climate is within the spread of the CMIP3 ensemble, which (along with the assessment by Ackerley and Renwick 2010) justifies the use of FGOALS in this study. Details on the configuration of each model can be found in Table 1. The data downloaded for each of these models were 100 years of monthly means for the Pre-Industrial control run (from now PI) and the 21,000 yr BP LGM run (from now LGM). All references to the seasons are given as DJF (December, January and February), MAM (March, April and May), JJA (June, July and August) and SON (September, October and November), corresponding to summer, autumn, winter and spring.

Table 1  Configurations of the models used in this study.

Model name (PMIP2 database)	Model name in the text	Atmosphere resolution (levels)	Ocean resolution (levels)	Model reference
CCSM3	CCSM3	T42 (26)	1°×1° (40)	Otto-Bliesner et al. (2006)
FGOALS-g1.0	FGOALS	2.8°×2.8° (26)	1°×1° (33)	Yu et al. (2002, 2004)
HadCM3M2	HadCM3M2	3.75°×2.5° (19)	1.25°×1.25° (20)	Gordon et al. (2000)

Results and discussion

The Semiannual Oscillation (SAO)

(a) SAO climatology

The SAO is the intra-annual variation in the strength and spatial extent of the SH circumpolar trough (CPT). In the current climate, the SAO is characterized by a contraction and strengthening of the CPT from June to September/October and December to March, with an expansion and weakening from September to December and March to June.

An index for the SAO was developed by van Loon (1967) as the monthly, zonal mean difference in 500 hPa air temperature between 50°S and 65°S (from now DT₅₀₀). DT₅₀₀ for NCEP reanalysis data and each of the models used in this study can be seen in Fig. 1A. The biannual nature of the SAO can be seen in the NCEP data with the strongest (weakest) temperature gradients occurring in March and October (June and December) when the CPT is in the contracted and intensified (expanded and weakened) state. The SAO also exhibits variability on annual to decadal time scales (van Loon et al. 1993). For example, Meehl et al. (1998) and Simmonds and Jones (1998) showed that the strength of the September peak reduced after 1979, leading to enhanced cyclonic activity in November and December with a reduction in August and September. Despite the weaker DT₅₀₀ in spring than autumn, Walland and Simmonds (1999) suggest cyclonic activity is generally stronger in the spring and attribute this to reduced static stability in SH mid-latitudes.

Figure 1  The SAO index. A, For NCEP reanalysis data (black solid line) and each of the PMIP2 models' PI (solid lines) runs. B, For the PMIP2 models' PI (solid lines) and LGM runs (dashed lines). The colour key for the models is CCSM3, blue; FGOALS, green; HadCM3M2, red.

The SAO is a strongly coupled ocean-atmosphere process and therefore provides a good test for GCMs. The SAO index for each of the models' PI runs is given in Fig. 1A. HadCM3M2 (red solid line) displays a peak in the DT₅₀₀ in March, which is stronger than that in the NCEP data. The DT₅₀₀ is generally too strong in December to July and too weak between August to November for HadCM3M2. CCSM3 (blue line) agrees well with the NCEP data from October to March for the DT₅₀₀ but is much higher than the NCEP data between April and September. FGOALS (green line), like CCSM3, also shows higher values of DT₅₀₀ than the NCEP data between April and September, which persist through to January (with the exception of October). From the results in Fig. 1A we infer that the CPT is generally too strong in FGOALS and CCSM3 throughout the year and for HadCM3M2 during December to July but HadCM3M2 tends to have a weaker than observed CPT during SON. All models show an autumn peak and exhibit approximately the correct magnitude of annual mean DT₅₀₀ but none really fully captures the semiannual nature in terms of the secondary peak in spring.

(b) SAO at the LGM

The SAO index (DT₅₀₀) for each of the models given in Table 1 can be seen in Fig. 1B for the PI (solid lines) and LGM (dashed lines) runs. The first point to note is that both FGOALS and CCSM3 show an increase in DT₅₀₀ in all seasons whereas the response in HadCM3M2 is more complex, with a weakening from November to July and a strengthening from July to November. The shape and magnitude of the LGM SAO in HadCM3M2 is more comparable with the NCEP data in Fig. 1A than the PI run (this was also noted in Drost et al. 2007). As the FGOALS and CCSM3 PI runs have a generally stronger CPT throughout the year than in the NCEP data, the increase in DT₅₀₀ may in part represent an enhancement of the systematic bias already noted above. However, there are differences in the responses between FGOALS and CCSM3, with the latter model showing a stronger DT₅₀₀ increase than the former in September to November.

The suggestion from Fig. 1B is that CCSM3 has a more active CPT during SON than during MAM (relative to PI), with a general increase in activity throughout the year, especially in June to November. The results for FGOALS (see Fig. 1B) suggest the CPT is stronger throughout the year also, particularly in March to November. Finally, from Fig. 1B, we would expect a weakening of the CPT for HadCM3M2 in December to May, with little overall change in June to August and a slight strengthening in September to November.

Seasonal and annual zonal mean winds

As a first step in looking at the strength of the CPT, the analysis in this section focuses on the strength of the zonal mean westerly component of the wind in the SH. Studies by Anderson et al. (2009), Govin et al. (2009), Hesse and McTanish (1999) and Shulmeister et al. (2004), and references therein, suggest that the SH westerlies may have been different during the LGM. Similarly, the differences in the SAO between the LGM and PI simulations highlighted in the previous section suggest that the modelled westerlies may also be different for the LGM relative to the PI. Therefore the overall mean change in the SH westerlies for the LGM relative to the PI will be investigated here and compared to available proxy-data evidence.

Also, previous modelling studies have largely focused on the annual mean, summer and winter climate only (for example, Justino & Peltier 2006; Braconnot et al. 2007; Rojas et al. 2009). However, work by Ackerley and Renwick (2010) has shown that changes in the SAO index, forced by changes in the model boundary conditions, may also be important for SH climate in the ‘transitional’ seasons (autumn and spring). Therefore this section will also investigate the seasonal changes in the SH westerly winds for the LGM relative to the PI simulations in each model to identify any sub-annual variations in the westerly wind strength.

(a) Annual mean

In both CCSM3 and FGOALS there was an increase in the SH westerlies throughout the troposphere, particularly between 40^°S and 60^°S, which can be seen in Fig. 2A and 2B, respectively. The strengthening of the annual mean westerlies agrees well with the increase in DT₅₀₀ throughout the year for LGM relative to PI in Fig. 1B for both CCSM3 and FGOALS. The increase in the westerlies is stronger in FGOALS than in CCSM3 even though the latter has a larger relative change in DT₅₀₀ throughout the year from PI to LGM (see Table 2). However, the value of DT₅₀₀ only spans a narrow latitude band and does not account for the temperature gradient between the pole and the sub-tropics. The values of the mean difference in air temperature at 500 hPa for the LGM relative to PI over wider latitude spans can be seen in Table 2 for CCSM3 and FGOALS. While the gradient change at high latitudes is slightly stronger in CCSM3 than in FGOALS, between the sub-tropics and the pole the change in DT₅₀₀ is much stronger in FGOALS than CCSM3, which would cause the stronger increase in the westerlies in FGOALS relative to CCSM3.

Figure 2  The annual (A–C), DJF (D–F), MAM (G–I), JJA (J–L) and SON (M–O) mean change (LGM minus preindustrial) in westerly wind component for CCSM3, FGOALS and HadCM3M2. Figures are labelled with the models and seasons. Solid lines indicate increases in the westerly wind strength and dashed lines indicate decreases (LGM minus preindustrial).

Table 2  The difference (LGM–PI) in DT₅₀₀, annual mean temperature gradient over different latitude spans for CCSM3 and FGOALS.

Latitude span for DT500	Change in DT500 (LGM–PI) in CCSM3 (K)	Change in DT500 (LGM–PI) in FGOALS (K)
50–65°S	1.14	1.03
40–65°S	1.13	1.12
30–65°S	0.72	1.40
20–65°S	0.73	1.81

The increase in westerly wind strength in CCSM3 and FGOALS agrees with the synthesis study by Shulmeister et al. (2004) and references therein, which suggested that there is evidence (albeit circumstantial or possibly influenced by external factors beyond wind speed) ‘for enhanced westerly flow at the LGM’. However, both of the models indicate only modest (approximately 1–2 m s⁻¹) increases in the surface wind speed with the dominant increases occurring in mid- to upper-levels of the troposphere.

HadCM3M2 (Fig. 2C) has a general weakening of the westerlies throughout the troposphere between 40°S and 60°S with slight increases to the north of 40°S. This result is consistent with the study by Hesse and McTanish (1999), who suggest that, from the analysis of the size distribution of aeolian dust deposits in the Tasman Sea, there is no evidence for enhanced westerly winds at the LGM. The weakening in the modelled westerlies in HadCM3M2 indicates that the reduction in DT₅₀₀ from November to July overcomes the small increase in DT₅₀₀ from August to October. HadCM3M2 is the only model of these three to show a general weakening of the SH westerlies during the LGM, whereas CCSM3 was the only model to show an increase in Rojas et al. (2009), suggesting that the spread in responses is large for the LGM relative to PI in the PMIP2 models.

(b) DJF

For DJF, both CCSM3 and FGOALS (Fig. 2D and 2E, respectively) show a strengthening of the westerlies. However, the centre of the increase is located further north in FGOALS than CCSM3 (centred approximately on 45°S and 55°S, respectively) and the increase is stronger in FGOALS than CCSM3. However, despite the general increases in the westerlies in FGOALS, the strongest increases are restricted to between approximately 40°S and 50°S, which is not the case in CCSM3. Both CCSM3 and FGOALS have their weakest increases in the zonal mean westerly wind component during DJF, which agrees with the small change in DT₅₀₀ in Fig. 1B.

Again, HadCM3M2 has a persistent weakening of the SH westerlies throughout a narrow band between 45°S and 65°S with a strengthening of the winds from 30°S to 45°S. The reduction in the westerlies’ strength agrees well with the reduction in DT₅₀₀ in Fig. 1B during DJF and is a stronger reduction than in any other season (which also agrees with Fig. 1B). The reduction of the SH high-latitude westerlies and increases to the north are indicative of either a northward shift of the CPT or a shift to the negative phase of the Southern Annular Mode (SAM; e.g. Thompson & Wallace 2000). The SAM will be discussed more below.

(c) MAM

As with DJF, both CCSM3 (Fig. 2G) and FGOALS (Fig. 2H) show a strengthening of the SH westerlies whereas HadCM3M2 (Fig. 2I) has a weakening between 45°S and 65°S with a strengthening of the winds from 30°S to 45°S. However, there have been some changes to the structure and strength of the SH westerlies compared to DJF. For MAM, the westerlies intensified slightly in CCSM3, agreeing with the relatively small increase in DT₅₀₀ for LGM relative to PI (compared to June–September; see Fig. 1B).

In FGOALS, the increase in the westerlies is stronger in MAM than in DJF, which agrees with the increases in DT₅₀₀ from PI to LGM in Fig. 1B. The strongest changes in wind speeds exhibit a ‘tilt’ from north to south (going upwards from the surface), which looks similar to the annual mean pattern given in Fig. 2B. The increases in the westerlies extend throughout the SH from the equator to approximately 80°S, showing a general increase in the winds during the LGM.

HadCM3M2 has a very similar pattern for the change in zonal mean wind speed in MAM compared to DJF (Fig. 2F and 2I, respectively), which again agrees with the weakened SAO from December to May (and beyond). However, the winds have a weaker easterly (westerly) component between 45°S and 65°S (30°S–45°S) in MAM compared to DJF.

(d) JJA

There are large increases in the zonal mean tropospheric wind in CCSM3 for the LGM relative to PI (Fig. 2J), agreeing with the large increase in DT₅₀₀ during this season (see Fig. 1B). The same can also be said about FGOALS, where the westerlies have increased further compared to MAM, with a strong increase in the upper level westerlies at approximately 30°S and 60°S (see Fig. 2K). The patterns are very different in HadCM3M2 for JJA (Fig. 2L) than in either DJF or MAM (Fig. 2F and 2I, respectively), which indicates that JJA is the ‘transition season’ between the weakened period of the SAO from November to July and the strengthened period from July to November, for the LGM relative to PI. The changes in wind speed are almost zero throughout the SH troposphere, with a small region of increased westerly winds at mid- to upper-levels south of 60°S.

(e) SON

The SON season for CCSM3 is very similar to JJA and is the season with the strongest increase in the SH westerlies for LGM relative to PI (see Fig. 2M), again agreeing with the largest increases in DT₅₀₀ (Fig. 1B) occurring during the spring. The difference between the LGM and PI winds decrease from SON into DJF, as can be seen by comparing Fig. 2M and 2D.

The differences in the westerlies between the LGM and PI simulations reduced in FGOALS during SON relative to JJA (Fig. 2H and 2K, respectively), which agrees with the reduction in the difference in DT₅₀₀ between the LGM and PI during spring before reducing further in summer. HadCM3M2 (Fig. 2O) now has a small increase in the SH westerlies throughout the troposphere, which is centred on 65°S with a reduction to the north, which is almost a reversal of the December–May pattern and agrees with the increases in DT₅₀₀ during the LGM compared to PI (see Fig. 1B).

The Southern Annular Mode (SAM)

The leading mode of atmospheric variability (at monthly to seasonal time scales), derived from Empirical Orthogonal Function (EOF) analysis, in the extra-tropical Southern Hemisphere is characterized by a ‘meridional seesaw’ of mass between high and mid-latitudes with a strong zonally symmetric structure known as the Southern Annular Mode (SAM; see Thompson & Wallace 2000 for a review of Northern and Southern Hemisphere annular modes) or the high-latitude mode (HLM) given in Karoly (1990). In this study, the EOFs were calculated from the monthly anomalies of the 500 hPa geopotential height field relative to the overall monthly mean (for each month individually) field averaged over all years of the model simulation. The EOFs from the monthly means and anomalies were calculated separately for the PI and LGM simulations. The leading mode of variability from the EOF analysis undertaken on the generated SH 500 hPa geopotential height anomaly field will be referred to as the SAM. The 500 hPa geopotential height field has been used in earlier work undertaken by Rogers and van Loon (1982) and Kidson (1988) and was chosen as it lies above the extensive ice sheets and sea ice over Antarctica, which may influence the model-derived sea level pressure (which has been used in other studies such as Drost et al. 2007 and Ackerley & Renwick 2010) and 500 hPa heights have been used effectively in more recent studies of SH variability such as Renwick (2002) and Justino and Peltier (2006) using data from the NCEP-NCAR reanalyses (Kalnay et al. 1996)

To assess how well the models represented the leading mode of SH variability in the 500 hPa geopotential height field the output from each of the models' PI phases was compared to NCEP reanalysis data taken between 1979 and 2008. The leading mode of variability in the NCEP 500 hPa data can be seen in Fig. 3 and displays a generally zonal structure with one notable meridional extension into the Southeast Pacific and two other lesser extensions toward South Africa and Southeast Australia (wavenumber 3 pattern). The polarity of the SAM dictates the strength of the SH westerlies in the CPT. In the positive (negative) phase of the SAM there is generally a negative (positive) pressure anomaly over the pole and positive (negative) pressure anomalies at mid-latitudes leading to a strengthening (weakening) of the westerlies in the CPT.

Figure 3  The leading EOF from the 500hPa geopotential height field for the NCEP reanalysis data from 1979 to 2008. Solid lines indicate positive values and dashed lines indicate negative values.

The leading mode of SH variability in the 500 hPa geopotential height field can be seen in Fig. 4A–4C for each of the PMIP2 PI runs. The variance explained by the SAM along with the spatial correlations between each PMIP2 model and the NCEP reanalysis are given in Table 3. In each model the SAM explains a larger proportion of the monthly variability in 500 hPa geopotential height than in the NCEP reanalysis data. This may be caused by either the SAM being more dominant in each of the models or the influence of other modes of variability (such as those influenced by the El Niño Southern Oscillation, ENSO) have been reduced. The work by Ackerley and Renwick (2010) analysed the variance explained by the three leading modes of SH sea level pressure data in the preindustrial control simulations of five PMIP2 models. The results of Ackerley and Renwick (2010) suggest that the results of the EOF analysis given in Table 3 are likely to be a combination of a poor representation of ENSO (also see Zheng et al., 2008, for a discussion of ENSO in the PMIP2 models) and a higher dominance of the SAM relative to the NCEP data. The CCSM3 SAM pattern has a high spatial correlation with that from the NCEP data, as does HadCM3M2. FGOALS, on the other hand, has a relatively weak correlation with NCEP. The reason for the results in Table 3 is apparent in Fig. 4A–4C. CCSM3 and HadCM3M2 have a much more zonally symmetric structure to the 500 hPa geopotential height field whereas FGOALS has a strong wavenumber 3-4 pattern, which is less zonally symmetric. Zheng et al. (2008) indicate that the ENSO cycle in the FGOALS model has a much larger amplitude relative to the other PMIP2 models they analysed, which may be responsible for forcing the strong wavenumber 3-4 pattern in Fig. 4. Also, Ackerley and Renwick (2010) show that the time series of the first and second EOFs of sea level pressure are highly correlated with sea surface temperatures in the tropical Pacific, further suggesting that the representation of ENSO in FGOALS is likely to be forcing the wave pattern in Fig. 4 and subsequently reducing the correlation coefficients. CCSM3, however, lies within ‘±20 of the observation’ for the amplitude of ENSO according to Zheng et al. (2008) and displays a more zonally symmetric structure in the 500 hPa geopotential height field (HadCM3M2 was not analysed by Zheng et al. 2008).

Figure 4  The leading EOF in the 500 hPa geopotential height field for each of the models used in this study. A–C, The PI control phase. D–F, The LGM phase. The figures are labelled to indicate which model each plot is taken from. Solid lines indicate positive values and dashed lines indicate negative values.

Table 3  The variance explained by the leading EOF in each model and the spatial correlation between the leading EOF in the NCEP reanalysis and each of the PMIP2 models used in this study.

Model	Variance explained in PI	Spatial correlation
NCEP reanalysis	20.6	-
CCSM3	26.3	0.87
FGOALS	21.3	0.68
HadCM3M2	21.8	0.92

The SAM for the LGM in each of the PMIP2 models can be seen in Fig. 4D–4F and shows very little difference in structure for CCSM3 and HadCM3M2 compared to PI. The LGM pattern in FGOALS, however, (Fig. 4E) is much more zonally symmetric than in PI and resembles the patterns in CCSM3 and HadCM3M2. This may be caused by a reduction in the amplitude of ENSO for the LGM simulation (see Zheng et al. 2008) and therefore the forcing mechanism for the wave pattern seen in the preindustrial simulation in Fig. 4b.

Conclusions

The work undertaken here has highlighted some of the possible differences in SH extra-tropical climate variability during the LGM in comparison to PI conditions. The models were chosen to provide a different viewpoint on the analyses by Rojas et al. (2009) and Drost et al. (2007) by looking at the seasonal and inter-annual variability in the CPT in a selection of PMIP2 models. The main results in this study are:

•	The representation of the SAO in each of the models varies considerably, with a tendency towards too high an index during March–September for CCSM3 and FGOALS and a poor representation of the October peak in HadCM3M2.
•	The strength of the SAO index increases throughout the year in both CCSM3 and FGOALS for the LGM in comparison to PI.
•	The SAO index in HadCM3M2 weakens from November to July and strengthens from July to November.
•	The response in the zonal, annual mean westerly winds agrees with the analysis of the SAO with CCSM3 and FGOALS showing a strengthening and for HadCM3M2 a weakening.
•	The seasonal changes in the SAO agree with the seasonal changes in the tropospheric westerlies in each of the models. However, there is a large degree of variation between the seasonal responses (of the westerlies) in each model.
•	The changes in the SAO for the LGM relative to the preindustrial simulations are therefore a good measure of the changes in seasonal variability of the westerly winds in any individual model.
•	The SAM is the leading mode of variability in each of the model PI runs; however, the variance explained was larger in each of the models than for the NCEP data (particularly CCSM3).
•	The spatial pattern of the SAM changed very little for the LGM relative to PI in CCSM3 and HadCM3M2 but became more zonally symmetric in FGOALS.

Overall, the models give varying responses to the LGM forcing, with two out of three models suggesting a strengthening of the CPT and one suggesting there was an overall weakening. The results from Rojas et al. (2009) stated that there was a weakening of the westerlies during JJA for three out of the four models used (which included HadCM3M2 and CCSM3) with small or little change during summer. The work here suggests that FGOALS (not included in Rojas et al. 2009) also displays stronger westerlies in JJA (and throughout the year), which would imply that of the five models investigated between the work presented here and in Rojas et al. (2009), three simulated a weakening and two simulated a strengthening of the westerlies.

Due to the spread in these models, we cannot draw any solid conclusions about the strength of the CPT during the LGM relative to today. This work, in conjunction with Rojas et al. (2009), has highlighted that there are still large uncertainties in the representation of specific modes of Southern Hemisphere climate and climate variability in GCM simulations of the LGM. Agreement between the output from any individual model and paleo-proxy data should be treated with caution due to the large differences between the climates simulated by the models. By identifying these differences between the modelled climates, the work presented in this study can provide useful information to the SH paleo-proxy and modelling communities as to how GCMs represent the climate of the SH during the LGM.

Acknowledgements

This work was funded by the New Zealand Foundation for Research Science and Technology (FRST) contract UOAX0714. We acknowledge the international modelling groups for providing their data for analysis, and the Laboratoire des Sciences du Climat et de l'Environnement (LSCE) for collecting and archiving the model data. The PMIP2/MOTIF Data Archive is supported by CEA, CNRS, the EU project MOTIF (EVK2-CT-2002-00153) and the Programme National d'Etude de la Dynamique du Climate (PNEDC). The analyses were performed using version 04-02-2009 of the database. More information is available on http://pmip2.lcse.ipsl.fr/ and http://motif.lsce.ipsl.fr/. Thanks also to James Renwick for extremely useful input when reviewing a draft version of this paper and to the two anonymous reviewers for helping to improve the manuscript significantly.