Variability of the atmospheric circulation in the southern hemisphere during the last glacial maximum



[1] A high-resolution atmospheric climate model has been used to investigate short-term variability of the atmospheric circulation in the Southern Hemisphere during the Last Glacial Maximum (LGM). The model was forced at the lower boundary by sea surface temperatures (SSTs) and sea ice fields from a lower-resolution coupled model simulation of the LGM. Although the semiannual oscillation (SAO) in the control simulation showed a different annual cycle than in the National Centers for Environmental Prediction (NCEP) reanalysis, the LGM SAO was similar to that of today, albeit slightly weaker, with the onset of the second peak delayed by a month. Nearly one third of the interannual variability of monthly mean 500-hPa height anomalies was accounted for by the Southern Annular Mode. Empirical orthogonal function analysis showed that the next two most important modes looked similar to those associated with El Niño-Southern Oscillation (ENSO). However, correlations of the associated time series with those of the coupled model's tropical sea surface SSTs were much smaller than for the observed present-day ENSO correlations.

1. Introduction

[2] The general circulation of the atmosphere is driven by a range of different forcings operating on a range of timescales from seasonal to millions of years. During the last hundred years, Earth's global average surface temperature has increased around 0.6°C largely because of the increase of greenhouse gases in the atmosphere [e.g., Intergovernmental Panel on Climate Change (IPCC), 2001]. Modeling the trend in global warming into the future has led, among other conclusions, to predictions that the occurrence of droughts and heavy rainfall in some regions will increase [e.g., IPCC, 2001].

[3] In order to increase the reliability of such predictions, as well as to improve our understanding of the impact of forcings on the climate system, models have been used often to simulate past climates. If those simulations show consistent correlations between forcings and climate patterns, it will give greater confidence in the use of those models to simulate future climates. The advantage of simulating a climate of the past is that the output can be compared against geological proxy data. The relationships between forcings and climate patterns are therefore constrained, and the impact of the different forcings on the atmosphere can be ranked.

[4] A past climate that is often simulated is that of the Last Glacial Maximum (LGM), which was about 20,000 years ago. It has been suggested that the main forcings of the colder climate of the LGM were a weaker or reduced thermohaline circulation, the large continental ice sheets, particularly in the Northern Hemisphere, and the general reduction of CO2 levels. There have been many studies on aspects of the circulation in the Northern Hemisphere [e.g., Dong and Valdes, 1998; Hewitt et al., 2001; Kageyama et al., 1999; Kitoh et al., 2001; Toracinta et al., 2004], but relatively only a few on the circulation of the Southern Hemisphere during the LGM [Drost, 2006; Drost et al., 2007a; Wyrwoll et al., 2000]. This is partly due to the fact that most interest has been on changes in the Northern Hemisphere circulation and that the Northern Hemisphere has good geological proxy data coverage. It has been more difficult to compare output from climate models against proxy data in the Southern Hemisphere since the Southern Hemisphere is only sparsely covered by available proxy data.

[5] Although it is possible to simulate the mean state of past climates through climate modeling, it is not possible to evaluate relatively high-frequency variability of past climates [for instance, Southern Annular Mode (SAM) and seasonal variations]. The time resolution of geological proxy data is too coarse to be able to discern climatic changes on short-term timescales. Also, the impact of climate variations like El Niño-Southern Oscillation (ENSO) could possibly mask the true representation of geological proxy data. ENSO alters the climatic background on which high-frequency variability takes place, thereby largely reducing the relevancy of the proxy data as a true climate indicator for short-term fluctuations. For instance, proxy data indications of a temporarily warm regime cannot uniquely be ascribed to a change in the background flow (caused by ENSO) or to high-frequency variability. However, although variability might not be validated by proxy data, climate modeling does present a unique opportunity to investigate the possible variability that might have occurred during past climates. Accounting for the variability inherent to the model, it can ascribe simulated variability to well-known patterns as found in the current climate, or could even suggest different causes for some climate fluctuations.

[6] The specific aim of this study is to investigate the variability in the Southern Hemisphere midlatitudes in a simulation of the climate during the LGM. The main modes of variability in the Southern Hemisphere are the semiannual oscillation (SAO) and the interannual variability. Interannual variability of the Southern Hemisphere is mainly explained by the Southern Annular Mode (SAM, or Antarctic Oscillation/High-Latitude Mode) and the patterns associated with ENSO. Section 2 will briefly outline the setup of the model used. Section 3 will discuss the semiannual oscillation, and section 4 will investigate the interannual variability. Section 5 presents the summary.

2. Method

[7] The model used was the high-resolution atmosphere-only HadAM3H configuration of the Hadley Centre's Unified Model (the UM [Cullen, 1993]). HadAM3H has been derived from the atmospheric component (HadAM3) of the coupled model HadCM3 [Gordon et al., 2000; Pope et al., 2000], which has been the Hadley Centre's “standard” climate model configuration for a number of years. The horizontal resolution of HadAM3H is 1.24° latitude × 1.88° longitude (HadAM3 was 2.5° × 3.75°) and has 19 layers in the vertical which are based on a hybrid vertical coordinate system [Simmons and Burridge, 1981]. The model is based on a grid point scheme and has an integration time step of 15 min. The specifications and performance of HadAM3H have been discussed by Pope and Stratton [2002] and Hudson and Jones [2002], and the reader is referred to those papers for details of the model.

[8] In this study, a control simulation was set up under preindustrial conditions [Drost, 2006; Drost et al., 2007a]. The initial conditions, sea surface temperatures (SSTs), and sea ice fields came from a HadCM3 simulation which was run under the same preindustrial settings [Hewitt, 2000]. That particular simulation was run for 500 years, after which it was determined that the model had run to an equilibrium state. In order to speed up the spin-up process toward this equilibrium state, Haney forcing was used. The control simulation ran parallel to the coupled model's simulation for the last 30 years. Although the control simulation stabilized within months, only the last 21 years were used for analysis so as to stay well away from any possible spin-up effects.

[9] HadCM3 was also configured to simulate the LGM [Hewitt, 2000]. The coupled model's simulation was set up as much as possible according to the specifications of the Paleoclimate Modeling Intercomparison Project (PMIP [World Climate Research Programme-111, 2000]). HadAM3H's LGM simulation was initialized from the end of the 500-year HadCM3 LGM simulation, which was also sped up using Haney forcing. SSTs and sea ice were also derived from the coupled model's LGM simulation. Soil and vegetation cover, greenhouse gases, continental ice sheet cover, and orography were updated and interpolated onto the higher-resolution grid [Drost, 2006; Drost et al., 2007a]. HadAM3H LGM simulation ran for 30 model years, but only the last 21 years were used for analysis.

[10] The setup of the simulation and the discussion on the aspects of the mean zonal circulation in the midlatitudes in the Southern Hemisphere during the LGM were described by Drost et al. [2007a]. Their main findings were that the polar trough had shifted slightly equatorward and that the amplitude of the wave number 3 pattern of the general circulation was slightly larger. That study mainly compared the simulated climate of the LGM against the output from a control simulation. In this study, the simulated variability during the LGM is compared both against the National Centers for Environmental Prediction (NCEP) climatology of 1971–2000 [Kalnay et al., 1996] and against the variability in the control simulation. Comparing output from the control simulation with the reanalysis data will indicate possible biases in the model. Once these biases are known, it is possible to compare the simulated variability of the climate of the LGM with the variability in today's climate. The control simulation will be referred to as CONTROL, and the Last Glacial Maximum simulation will be referred to as LGM.

3. Semiannual Oscillation

[11] The SAO is the main seasonal variation of the zonal flow in the middle-high latitudes of the Southern Hemisphere. It is characterized by a twice yearly variation of both the strength and the position of the polar trough of low pressure surrounding Antarctica. From March to June, and from September to December, the circumpolar zone of low pressure expands equatorward and weakens, where as from June to September, and from December to March, it contracts and intensifies [Meehl, 1991; Mullan and McAvaney, 1995; van Loon, 1967]. As the result of the different heat capacities between the midlatitude (50°S) and high latitude (65°S) (ocean versus polar continent), warming and cooling have different lag times. Because of these different lag times, the temperature contrast between the two zonal bands has two maxima: one during the austral autumn (March/April) and one during spring (September/October). Differences between the cooling and heating rates at these times give rise to a larger maximum during autumn. The variations in temperature gradient modulate the strength of the westerlies and cyclonic activity with both experiencing maxima following the SAO. The strongest westerly winds occur during March and September south of about 50°S and during June and December north of about 50°S.

[12] The SAO is not constant and varies each year. Meehl [1991] found that an altered seasonal cycle of SSTs and ocean heat storage at 50°S can result in a change in the SAO signal. Meehl et al. [1998] showed that the SAO was stronger before and weaker after 1979. They ascribed this to an anomalous change in the temperature gradient between 50°S and 65°S, which was the result of a warming trend. This resulted in a “flattening of the seasonal cycle of baroclinicity.” Stammerjohn et al. [2003] also found that the SAO responds to the meridional temperature gradient as modulated by, among other forcings, global warming. Consequently, since the meridional temperature gradient was steeper during the LGM than it is today [Drost, 2006], the question arises as to whether the SAO was different during the colder climate of the LGM.

[13] Figure 1 shows the SAO derived from the NCEP climatology and the control and LGM simulation, computed from the difference in monthly mean 500-hPa temperatures between 50°S and 65°S. The SAO of CONTROL is analyzed first in order to investigate how well the model can reproduce the observed signal. The NCEP data show the normal SAO pattern, with peaks during late summer/early autumn and during spring. CONTROL has simulated the monthly varying meridional temperature gradient during the first half of the year quite well, both in intensity as well as in time. However, CONTROL fails to model the peak in the SAO during the austral spring. Although the settings of CONTROL were more representative of the preindustrial period, it is not likely that the meridional temperature gradient was much different to present during that time. The fact that CONTROL does not reconstruct the peak in the SAO signal during the austral spring is because HadCM3 fails to produce accurately the annual cycle of SSTs at 40°S–60°S [Gordon et al., 2000]. Figure 2 presents the monthly mean 500-hPa temperature profiles at 50°S and 65°S and gives a clearer picture of why the SAO in CONTROL is different. During the austral summer months, the amplitude of the SAO in CONTROL is slightly larger than the amplitude of the SAO in NCEP. This is primarily the result of CONTROL having lower temperatures at 65°S. During autumn, the temperature at 65°S in CONTROL remains cooler than in NCEP, and the temperature difference between the two zonal bands in CONTROL remains larger than in NCEP. If other conditions are equal, the implication is that CONTROL has slightly stronger winds and higher baroclinicity than NCEP during March-April-May (MAM) at high latitudes. The main difference in SAO between NCEP and CONTROL, however, is during the austral winter and spring. While the amplitude of the SAO in NCEP increases after June, this does not happen in CONTROL until September. The result is that the SAO signals are out of phase during June-July-August (JJA) and September-October-November (SON). The anomalous pattern of the SAO in CONTROL is related to the warming at 50°S which lags the warming in the NCEP climatology. This has been documented by Gordon et al. [2000] who showed that the SSTs in HadCM3 at 40°S–60°S were too low and that warming of HadCM3 SSTs lags the warming of SSTs in NCEP. They also found that the peak SST errors in the Southern Ocean are in regions of strong SST gradients and therefore could possibly be associated with the incorrect positioning of the Antarctic Circumpolar Current (ACC).

Figure 1.

Semiannual oscillation for NCEP, CONTROL, and LGM computed from the monthly mean 500-hPa temperatures at 50°S and 65°S.

Figure 2.

Monthly mean 500-hPa temperature profiles at 50°S and 65°S for NCEP, CONTROL, and LGM. The shading enables clear identification of the differences between NCEP and CONTROL at 50°S and 65°S. Note, in particular, the lag time between NCEP and CONTROL and between NCEP and LGM at 50°S around July-September.

[14] Some studies have indicated that the strength of the SAO might be related to the amplitude of the wave number 3 pattern of the general circulation [e.g., Trenberth, 1986; Trenberth and Mo, 1985]. It was noticed that there was a weakening in the SAO in the late 1970s, which was associated with an intensification of the wave number 3 pattern of the extratropical circulation, a deepening and equatorward movement of the polar trough, and an increase of the strength of the westerlies [e.g., Burnett and McNicoll, 2000; Chen and Yen, 1997; Hurrell and van Loon, 1994; Meehl et al., 1998; van Loon et al., 1993]. Similar features were described by Drost [2006] and Drost et al. [2007a] in their analysis of the performance of the control simulation. They found that the wave number 3 pattern was enhanced during JJA and the modeled westerly wind was stronger than in NCEP, particularly during SON. These two features correspond well to the smaller amplitude of the simulated SAO during the second half of the year. It was therefore concluded that the SAO in CONTROL was a true reflection of the seasonal variation in the simulated climate patterns between 50°S and 65°S, and that the weak amplitude of the SAO during the second half of the year was the result of the anomalous modeled SSTs at high latitudes.

[15] The SAO during the LGM is more similar to the SAO of the NCEP climatology than to the SAO of CONTROL (Figure 1). Unlike CONTROL, the 50°S and 65°S LGM temperature profiles are synchronized with each other during JJA and SON, resulting in a clear peak during SON in the SAO (Figure 2). Given the model's experimental design, it is not possible to conclusively state that the similarity with the SAO of the NCEP climatology was the result of more realistic SSTs in the LGM simulation, or whether the SAO was actually stronger in the spring during glacial times. Similar to CONTROL, however, the mean zonal temperature at 50°S in the LGM simulation lags the temperature rise during the latter part in JJA by about a month compared to NCEP (Figure 2). As indicated earlier, Gordon et al. [2000] have shown that the ocean part of HadCM3 has its main errors between 40°S and 60°S, and the possible anomalous lag time could be more a reflection of an inconsistency in the model rather than a feature of the LGM. The fact that both simulations show a similar lag time is not investigated any further in this study.

[16] A particularly important point that comes out of the SAO analysis is that there could be an opposite climatic response during SON when comparing midlatitude circulation features of the LGM with CONTROL and with the NCEP climatology. It was noted by Drost [2006] and Drost et al. [2007a] that the largest anomalies in CONTROL occurred during SON. Since the amplitude of the SAO is indicative of the level of baroclinicity, the cyclonic activity in the LGM was enhanced and closer to the pole than in CONTROL during SON. These conclusions are consistent with the increased meridional temperature gradient and meridional heat transport during that time of the year [Drost, 2006; Drost et al., 2007a]. However, a note of caution needs to be placed here. Although the meridional temperature gradient determines the seasonal evolution of the SAO in baroclinicity, it is the static stability that determines the amplitudes of its maxima [Walland and Simmonds, 1997, 1999]. Following Walland and Simmonds [1999], baroclinic response during the LGM would be different than in CONTROL. As a result of the stronger cooling at the surface than at height, the colder climate of the LGM has a higher static stability than the warmer climate of the control simulation [Drost, 2006]. The reduced lapse rate at the LGM would result in a weaker baroclinic response to an increase in temperature gradient. For that reason, the difference in baroclinicity between the two simulations during the first half of the year would be slightly larger than what the difference between their SAO implies, with the cyclonic activity during the LGM being reduced even more. By comparing the SAO of the LGM with that of the NCEP climatology and not accounting for the higher static stability during the LGM, the SAO indicates that baroclinicity during the LGM was stronger during December-January-February (DJF) and weaker during JJA and SON than in the present day. The onset of the second peak in the SAO of the LGM lags a month behind than in the NCEP climatology. However, the main difference between the two signals is in their amplitudes. The smaller amplitude of the SAO of the LGM is consistent with an enhanced wave number 3 circulation and an equatorward shift of the subpolar trough during the LGM, which has been documented by Drost [2006] and Drost et al. [2007a].

4. Interannual Variability

[17] The Southern Hemisphere circulation exhibits a range of climatic variations after the mean seasonal cycle is filtered out. Kidson [1999] gives an overview of what signals can be detected within these variations, their timescales, and their proportion of the total variance. Kidson [1988a] looked at the time and spatial variation of the Southern Hemisphere 500-hPa zonal wind and found that there are several signals in the variations. These variations can be understood by examining the 500-hPa geopotential height fields and are well described through empirical orthogonal function (EOF) analysis [e.g., Ghil and Mo, 1991; Karoly, 1990; Kidson, 1988b, 1991].

[18] The leading EOF signals of interannual variation in the Southern Hemisphere mean sea level pressure from NCEP are presented in Figure 3. The EOFs are determined over the period 1971–2000 from monthly data from which the monthly mean was removed. These three EOFs show the characteristic Southern Annular Mode (SAM, or the “Antarctic Oscillation” or the “High-Latitude Mode,” EOF1), the South-Pacific Dipole (EOF2), and the “wave train” (EOF3). These patterns explain 24.7%, 11.3%, and 9.5% of the interannual variability, respectively. Correlation of the time series of the latter two patterns with tropical SSTs indicated that these patterns are ENSO related (see Figure 4 [e.g., Karoly et al., 1998; Mo, 2000; Renwick, 1998]). The calculated EOF patterns are robust to sampling variability. EOFs were calculated separately for the first and second half of the time series of NCEP and CONTROL data (not shown). Spatial correlations of the equivalent EOF patterns of the NCEP time series were above 0.9, and those of CONTROL were slightly less. The lower correlation values for the latter are ascribed to the shorter time series of the CONTROL data.

Figure 3.

Three leading EOFs of Southern Hemisphere variability in NCEP. The variability explained by each mode is placed at the bottom right of each picture. Positive values are shaded and the zero-line is marked as a thick line. Contour interval is 0.1. Contours are nondimensional.

Figure 4.

Correlation of the time series of EOF2 (top) and EOF3 (bottom) with SST (NCEP). Positive values are shaded and the zero-line is marked as a thick line. Contour interval is 0.1.

[19] EOF analysis in CONTROL followed the same method as to that for NCEP except only 21 years have been used. Interannual variability in CONTROL seems to be explained by the same patterns as in the NCEP climatology since the first three EOFs are similar, although there are some differences (Figure 5). The first EOF accounts for nearly 30% of the variance, indicating that the SAM in CONTROL is more prominent than in the NCEP climatology. The maximum over the Indian Ocean has moved eastward and is much closer to the maximum over the western Pacific. The eastward shift is a direct consequence of the enhanced wave number 3 pattern in the Southern Hemispheric circulation during the LGM [Drost et al., 2007a]. Simulated mean sea level pressure south of Australia in CONTROL was lower than in the reanalysis during virtually the whole year [Drost, 2006; Drost et al., 2007b, Figure 4]. Both the Indian Ocean and western Pacific maxima have become stronger. The trough in the East Pacific is now farther to the west. The SAM in CONTROL is associated with stronger westerly winds than in NCEP, especially over the western Pacific.

Figure 5.

Three leading EOFs of Southern Hemisphere variability in CONTROL. The variability explained by each mode is placed at the bottom right of each picture. Positive values are shaded and the zero-line is marked as a thick line. Contour interval is 0.1.

[20] Although it was shown that HadCM3 was capable of simulating ENSO [Collins et al., 2000], Drost [2006] had shown that HadCM3 in this control simulation failed to properly reconstruct the extent and strength of the interannual variability in the Pacific SSTs. The main difference between the two simulations was that the current control simulation underwent Haney forcing in order to speed up the spin-up process to provide balanced initial conditions for the CONTROL simulation [Hewitt, 2000]. Such a surface heat flux restoring term could have affected tropical SST variability [e.g., Cai and Chu, 1996]. Figure 6 clearly shows that the standard deviation of tropical SSTs is much lower in CONTROL (middle picture) than in NCEP (top picture) and has a maximum slightly more toward the central Pacific. It is therefore expected that analysis of the ENSO variability in NCEP and CONTROL will show some differences between the two climatologies. However, the second and third EOF patterns in CONTROL are quite similar to the EOF patterns of the NCEP climatology. The wave train pattern now accounts for a similar amount of the variability as the Pacific dipole pattern (both about 10%). The wave number 3 pattern in the second EOF in CONTROL (the “wave train”) is slightly weaker than in the NCEP climatology and shows larger maxima over the Pacific region. The wave train seems to start more from the south, rather than the east, of Australia.

Figure 6.

Annual standard deviation of SST in NCEP (top), CONTROL (middle), and LGM (bottom; note the enlarged landmass). Land and regions that at any time were covered in sea ice are plotted in black. Contour interval is 0.1°C. Light-shaded regions are 0.4–0.8°C and dark-shaded regions are 0.8–1.2°C.

[21] Although the second and third EOFs have similar spatial patterns to those based on NCEP data, their temporal variability exhibits some differences with the NCEP climatology. The correlation patterns of the time series of the second and third EOFs from CONTROL with the computed SSTs only weakly indicate a connection with ENSO variability in the tropical Pacific (Figure 7). Both correlation patterns show a maximum in the (eastern) Pacific, but their amplitude and extent are anomalous. As indicated earlier, these anomalies are related to the inadequacy of the control simulation in reconstructing the tropical Pacific SSTs. Although the second and third EOFs are not related to the classical ENSO patterns, they are referred to as ENSO signals throughout this paper because of their similarity to the NCEP-based ENSO-related EOFs. Using the criterion of North et al. [1982], the second and third eigenvalues overlap slightly, but are well separated from those associated with the higher orders of variability.

Figure 7.

Correlation of the time series of EOF2 (top) and EOF3 (bottom) with SST (CONTROL). Positive values are shaded, and the zero-line is marked as a thick line. Contour interval is 0.1.

[22] Further analysis of the time series of the EOFs shows more differences. Figure 8 shows the distribution of the amplitudes of the first three EOFs in nondimensional units. The SAM (EOF1), which is commonly characterized by a negative and positive phase, shows up quite distinctively in the NCEP climatology. It indicates that the SAM in NCEP is most often in either one of its phases and less often in the “transition” zone between the phases (bimodal distribution). The control simulation, however, does not show the two peaks and is much flatter across the whole distribution (unimodal distribution). The SAM in CONTROL has a bias toward the positive phase but shows a few more months in a strong negative phase. The amplitude distribution belonging to the “dipole” pattern (EOF2 NCEP versus EOF3 CONTROL, the middle graph in Figure 8) shows a very symmetrical, Gaussian-shaped distribution for NCEP with a small bias toward a negative distribution. CONTROL shows a similar distribution but with a shift more toward the positive values. The wave train pattern (EOF3 NCEP versus EOF2 CONTROL, the bottom graph in Figure 8) sees a distinctive positive peak and a weak negative peak in the NCEP climatology. The positive phase is associated with El Niño and the negative with La Niña. It is evident that the 1971–2000 climatology experienced more El Niños than La Niñas. In general, the distribution in time of the positive and negative phase of EOF2 in CONTROL is quite evenly spread. It does show a small positive peak (El Niño) but fails to model the negative peak (La Niña), although there is a small bias toward negative values. From the time series analysis, it can be concluded that although the three leading EOFs of the interannual variability as modeled by CONTROL account for a similar percentage of the total variability as they do in the NCEP climatology, the occurrence and strength of their phases differ noticeably.

Figure 8.

Distribution of the EOF coefficients for the first three EOFs for NCEP, CONTROL, and LGM. x axis is nondimensional amplitude, and y axis is number of months. NCEP is over the period of 1971–2000 and CONTROL and LGM are over 21 years, hence the difference in the total numbers. Note that NCEP EOF2 is compared with EOF3 in CONTROL and LGM and NCEP EOF3 with EOF2 in CONTROL and LGM.

[23] Although the interannual variability in CONTROL does not completely match the present-day climatology, it is interesting to use the model to investigate the variability of the Southern Hemisphere circulation under glacial forcings. Figure 9 presents the first three EOFs of the interannual variability of the Southern Hemisphere circulation during the LGM. These patterns are determined in an identical way and over the same number of model years as to those of CONTROL. The LGM EOF patterns are to a large degree similar to the NCEP EOF patterns. The circulation associated with the SAM is again slightly stronger over the southwest Pacific and over the Indian Ocean than in the NCEP climatology. The spatial extent of the maximum over the southwest Pacific is reduced and slightly enlarged over the Indian Ocean compared to CONTROL. The SAM during the LGM accounts for the interannual variation slightly more than that it did in CONTROL (31.2% versus 29.3%). Analysis of the time series shows that the SAM during the LGM does not show the two distinct peaks associated with the vacillation of the negative and positive phase. The distribution is slightly more toward the positive during the LGM than in CONTROL and, as in CONTROL, has its maximum on the positive side (Figure 8). This would imply that the SAM during the LGM can be attributed to slightly stronger westerly winds at high latitudes than in CONTROL. However, this might also be the effect of a slightly reduced westerly wind in the mean zonal circulation at those latitudes during the LGM.

Figure 9.

Three leading EOFs of Southern Hemisphere variability in LGM. The variability explained by each mode is placed at the bottom right of each picture. Positive values are shaded and the zero-line is marked as a thick line. Contour interval is 0.1.

[24] The interannual variability in tropical SSTs in LGM indicates a more classical ENSO pattern than in CONTROL (Figure 6, bottom graph). Whether this is a realistic simulation of tropical SST variability is hard to verify, since there are hardly any accurate climatic records of short-term variability that could indicate that ENSO was a common feature of the LGM climate. However, the possible occurrence of ENSO during the LGM has been indicated by studies based on proxy data [e.g., Koutavas et al., 2002] and by modeling studies [e.g., An et al., 2004]. Although it is very important to first investigate whether ENSO was part of the climate of the LGM before studying any possible variance associated with this phenomenon, we have temporarily sidestepped that issue and have assumed that the LGM did experience some kind of climatic variability that could be associated with today's ENSO. As was mentioned earlier, this discussion did not set out to solve high-frequency climate fluctuations during the LGM, but merely suggests possible characteristics of short-term climate patterns during the LGM.

[25] As in CONTROL, correlating the LGM time series of EOF2 and EOF3 with the SSTs does not show the classical ENSO patterns either, although the largest correlations are found in the tropical Pacific (Figure 10). The latter is expected since the LGM experienced larger interannual variability in the tropical Pacific than CONTROL. The correlation pattern for the third EOF in particular is different since it seems to indicate that the variability is linked to SST variability in the western and central Pacific rather than in the eastern Pacific. Whether such a pattern shows up in other fields in the equatorial tropics and what their impact is on the climate was not investigated.

Figure 10.

Correlation of the time series of EOF2 (top) and EOF3 (bottom) with SST (LGM). Positive values are shaded and the zero-line is marked as a thick line. Contour interval is 0.1.

[26] The second EOF shows the characteristic “wave train” pattern. The start of this pattern is now more comparable to the NCEP EOF3 pattern since it starts off from the east coast of Australia rather than from the south of Australia as in CONTROL. The difference in the origin of the wave train patterns could be related to the differences in tropical SST variability between the simulations. Strong tropical SST variability, as in NCEP and LGM (Figure 6), may result in the wave train originating east of Australia since it propagates from the tropics, whereas the wave train may come from the south of Australia in CONTROL because tropical SST variability might be too weak to force much of a wave train from the tropics. The gradient in the LGM wave train pattern in the southwest Pacific, and especially over the New Zealand region, is reduced. The wave train has a stronger continuation and can easily be identified going into the South Atlantic all the way to southern Africa. Analysis of the time series indicates symmetry in the occurrence of an “El Niño” and a “La Niña” event, with peaks in the distribution at both 1 and −1 (Figure 8). Whether that is an artifact of the model or is inherent in the LGM simulation has not been investigated further.

[27] The dipole pattern of the third EOF shows a remarkable change. The characteristics of the dipole, a negative and a positive anomaly over the low and high latitudes of the Pacific, respectively, are still there. But the flanks of the low over Antarctica have moved equatorward and are now shouldering the positive anomalies over the high latitudes. This results in the third EOF also showing a “wave train” pattern, albeit about 90° out of phase with EOF2, suggestive of the pair of EOFs depicting wave propagation across the south Pacific. Using the criterion of North et al. [1982], the second and third eigenvalues overlap slightly, implying some mixing between these patterns. Linear combinations of EOFs 2 and 3 could be configured to more closely resemble those from the NCEP data. The two different phases of the wave train pattern result in different regional climatologies. With the interannual variability associated with ENSO during the LGM being of a similar magnitude as in CONTROL, it would seem that ENSO should have been felt in a country as New Zealand during the LGM. Although it is difficult to obtain an interannual time resolution in sedimentary samples, Pepper et al. [2004] found the characteristic ENSO period in the spectral analysis of laminae thicknesses of samples in New Zealand. Both EOFs are well separated from the higher-order EOFs.

5. Summary and Discussion

[28] We have analyzed the atmospheric variability in the Southern Hemisphere under LGM conditions. Output was compared against both the NCEP climatology and a control simulation. Because several anomalies in the simulation were model-related, this study cannot make absolute statements about the Southern Hemisphere interannual variability that might have existed during the LGM. A better understanding and an improved version of the general circulation model could possibly lead to a better simulation of LGM variability. Also, only one simulation has been analyzed. The significance of the analysis would be enhanced if an ensemble study of similar climate simulations were to be carried out. Since the authors are not aware of any progress toward such a study, such an investigation is left for the future. Hence the analysis done in this study should not be regarded as a precise description of the Southern Hemisphere's interannual variability during the LGM. However, the facts and ideas coming out of this study serve as a supplement to our knowledge of the climate of the LGM gained by analyzing geological proxy data. The temporal resolution of proxy data is too coarse to obtain information of variability on short timescales, so that climate modeling studies are the only means to investigate short-term climatic patterns during past climates. The main seasonal variation of the zonal mean flow is the SAO. Although the control simulation did not reconstruct the SAO particularly well during the second half of the year, the SAO of the LGM simulation looked quite similar to that of the present day. This would mean that the polar trough, together with its associated strong winds, shifted meridionally as it does today: During summer and winter, the polar trough is located nearer to the pole and shifts equatorward during autumn and spring. The amplitudes of the SAO during the LGM are slightly smaller than in NCEP. A weaker SAO is consistent with an enhanced wave number 3 circulation and an equatorward shift of the subpolar trough during the LGM, which have been documented by Drost [2006] and Drost et al. [2007a]. The onset of the peak during spring is delayed by 1 month. EOF analysis has shown that the modeled interannual variability in the Southern Hemisphere under glacial forcings can still be explained predominantly by the SAM and ENSO patterns. Of these two patterns, the SAM is most robust. The SAM explained a slightly larger amount of variance than CONTROL (31.2% versus 29.3%), but was more dominant than in the NCEP climatology (only 24.7%). It implies that the SAM during the LGM can be attributed to slightly stronger westerly winds at high latitudes than during the present day.

[29] As a result of the model's incapability in correctly reconstructing SSTs patterns in the tropics, ENSO was simulated only very weakly at best. This study did not set out to investigate the occurrence of ENSO during the LGM, however, but mainly the variability associated with a possible ENSO during the LGM. If ENSO had occurred, how would it have affected the climate of the Southern Hemisphere? EOF analysis of the variability in the Southern Hemisphere during the LGM showed similar ENSO-related patterns as in the NCEP climatology and explained 9.6% and 8.6% of the variance. The patterns resemble the characteristic “wave train” pattern and a weaker Pacific dipole pattern. The latter also showed a “wave train” pattern but 90° out of phase with the former.

[30] In order to improve our reconstructions of short-term fluctuations of past climates, it is essential to be able to accurately simulate high-frequency climate fluctuations in today's climate. Once the underlying mechanisms of climate variability are understood and are properly represented in a model, one might feel confident about theoretical reconstructions of past climate high-frequency variability. This study was not able to show that the model was capable of simulating ENSO patterns in today's climate correctly. Nevertheless, it showed that variability in an LGM simulation was relatively similar to that of today's climate with the SAO and SAM being the strongest components of the zonal mean circulation variability in the Southern Hemisphere.


[31] The lead author would like to thank Victoria University of Wellington for awarding him a PhD scholarship. Further funding for this study came from FRST contract C01X0202, “Adaptation to Climate Variability & Change,” and Marsden Grant Contract UOC301, “Orbital or Thermal Causes of Glaciation in New Zealand.” The lead author is very grateful to Chris Hewitt of the UK Met Office for the kind provision of his coupled model output and the Hadley Centre for using their Unified Model. Many thanks go out to the people at NIWA, Wellington, for providing a great research environment. Special thanks go out to the IT staff, Graham Rickard, and Phil Andrews for helping with many computer issues and the UM model. Finally, the authors would like to thank the two anonymous reviewers for their excellent review of the paper. Their comments have helped the authors in clarifying some of the issues raised.