Data and methods

Systematic measurements of meteorological parameters have been carried out at Neumayer since 1981 (König-Langlo & Herber Reference König-Langlo and Herber1996). In March 1992 the observatory programme was transferred to the new Neumayer Station, i.e. the Neumayer II (70°39′S, 8°15′W), located 8 km south-east of the former one. Since 2009 the meteorological observations have been performed at Neumayer III station (70°40′S, 8°16′W). All stations are situated at the Ekström Ice Shelf on a homogenous flat surface, very gently sloping upwards to the south (Fig. 1). Observations of air temperature, air pressure, wind speed, wind direction, dew point temperature, clouds, horizontal visibility, and other several synoptic variables are carried out every three hours (0h00, 9h00, 12h00, 15h00, 18h00, 21h00 coordinated universal time) at Neumayer Station (König-Langlo & Herber Reference König-Langlo and Herber1996, König-Langlo & Loose Reference König-Langlo and Loose2007). The synoptic observations from Neumayer were carried out by a station meteorologist. A variety of redundant instruments as well as data visualization and validation (König-Langlo Reference König-Langlo2012) contribute to minimize errors. A detailed description of the instruments used and their accuracy is given by König-Langlo & Herber (Reference König-Langlo and Herber1996).

Fig. 1

Geographical position of the Neumayer Station (ring) and the dominant directions of the surface wind (arrows). Regional view of Neumayer area (upper right panel).

The main quantities used in this study are the time series of three-hourly observations of air temperature (T), wind speed (U), and sea level pressure (SLP). These records are available from Publishing Network for Geoscientific & Environmental Data network PANGAEA (http://www.pangaea.de) (Rimbu et al. Reference Rimbu, Lohmann, König-Langlo, Necula and Ionita2012). The records used in our study cover the period 1982–2011 (30 years). Each record has 87 656 values (8 x 365 x 30 plus 7 x 8 due to leap years). The values of these variables are not available for all observational hours. However, the number of missing values is relatively small. In the T, U and SLP records only 0.3%, 0.26% and 0.27% values are missing. From the original data (three-hourly observations) we have calculated daily mean time series. The daily means are calculated by averaging all the eight observations within a day. If there is at least one missing observation in a day, then the mean of that day is considered missing. To simplify the analysis we have excluded the 29 February observations from leap years. From the total number of days within the 1982–2011 interval, i.e. 10 950, only 1.81% are missing in the daily T record. The daily U and SLP records have only 1.67% and 1.56% missing values respectively.

Statistical properties of three-hourly and daily T, U and SLP time series used in this study are summarized in Table I. These statistical parameters were calculated separately using all available values (all years), polar day (from 10 November–24 January) and for polar night (from 19 May–27 July). The statistical parameters of T, U and SLP time series, calculated using all available measurements (Table I), are very close to those published in previous papers (e.g. König-Langlo & Loose Reference König-Langlo and Loose2007). The range as well as the interquartile interval of T are higher during polar night than during polar day. The interquartile interval of U as well as the range of SLP variations is higher during polar night than during polar day (Table I).

Table I

Descriptive statistics of temperature (T), wind speed (U) and sea level pressure (SLP) records used in this study. Units are in °C, m s^-1 and hPa respectively.

Statistical properties of data can also be summarized in histograms (Fig. 2). The histogram of all year T data (Fig. 2a) is skewed to the left. The relatively abrupt end of the right tail (Fig. 2a) is related to the melting barrier of the snow and ice in the region. In response to a favourable temperature increase, melting can take place. The associated modifications in the snow-ice surface energy budget lead to a limitation of temperature increase. The histogram of yearly T also shows two relatively small local maxima which are related to the T values corresponding to cold and warm seasons in the region (Fig. 2b & c). The T distribution for polar day (Fig. 2b) is unimodal and asymmetric. Also in this case the abrupt decrease in the right tail of the distribution function is related to the snow-melting barrier effect. The distribution of T for polar night (Fig. 2c) presents less asymmetry than the corresponding distribution for polar day (Fig. 2b). Because all temperatures are negative, the decrease in the frequency in the right tail of the histogram is less abrupt than in the case of the polar day histogram. The histogram of polar night temperatures (Fig. 2c) suggests also a possible multimodality in the distribution function. The synoptic disturbances, which are associated with relatively warm air advection from the east toward Neumayer (König-Langlo & Loose Reference König-Langlo and Loose2007), are responsible for the dominant maximum in the polar night temperature histogram (Fig. 2c). The katabatic winds, blowing from the south bringing very cold air to Neumayer (König-Langlo & Loose Reference König-Langlo and Loose2007), are responsible for the relatively small peak of T distribution near -33°C during polar night (Fig. 2c). The SLP histograms all show bell-shaped distributions (Fig. 2d–f) suggesting a Gaussian probability distribution function of this variable. The histograms of U (Fig. 2g–i) shows a typical Weibull distribution (e.g. Justus et al. Reference Justus, Hargraves, Mikail and Graber1978). The histograms of the wind direction (Fig. 2j–l) show one prominent peak at about 90° that is related to the typical synoptic disturbances in the region (König-Langlo & Loose Reference König-Langlo and Loose2007). These histograms also show another two less pronounced peaks at about 170° and 240° which are associated with the katabatic and supergeostrophic winds, respectively, at Neumayer Station (König-Langlo & Loose Reference König-Langlo and Loose2007).

Fig. 2

Histogram of three-hourly measurements during the period 1982–2011 of a. temperature for all years, b. temperature for all polar days, c. temperature for all polar nights, d. sea level pressure for all years, e. sea level pressure for all polar days, f. sea level pressure for all polar nights, g. wind speed for all years, h. wind speed for all polar days, i. wind speed for all polar nights, j. wind direction for all years, k. wind direction for all polar days, l. wind direction for all nights.

To establish the atmospheric circulation patterns related to T, U and SLP variability at Neumayer, we have used the daily 500 hPa geopotential height (Z500) extracted from the National Centers for Environmental Predictions (NCEP) and the National Center for Atmospheric Research (NCAR) (NCEP/NCAR) reanalysis database (http://www.esrl.noaa.gov/psd/data/reanalysis/reanalysis.shtml, accessed January 2012) for the period 1982–2011. The NCEP/NCAR reanalysis dataset (hereafter NCEP1) is based on a global data assimilation system which uses input data from the original NCEP operational global T62 spectral model and various observations from multiple sources. The spatial resolution of the Z500 field used in our study is 2.5° longitude x 2.5° latitude. A detailed description of the NCEP/NCAR global reanalysis data can be found in Kalnay et al. (Reference Kalnay, Kanamitsu and Kistler1996) and Kistler et al. (Reference Kistler, Kalnay and Collins2001). An improved version of the NCEP/NCAR global reanalysis data is the NCEP-Department of Energy (DOE) Reanalysis 2 dataset (hereafter NCEP2). The NCEP2 fixed errors and updated parametrizations of physical processes in NCEP1 (Kanamitsu et al. Reference Kanamitsu, Ebisuzaki, Woollen, Yang, Hnilo, Fiorino and Potter2002). However, the results based on NCEP1 data were qualitatively the same as those based on NCEP2 so that only the results based on NCEP1 data will be discussed in this paper.

The time series of the MJO phase, as described by Wheeler & Hendon (Reference Wheeler and Hendon2004), for the period 1982–2011, was downloaded from the Centre for Australian Weather and Climate Research (CAWCR) webpage (http://cawcr/.gov.au/staff/mwheeler/maproom/RMM/RMM1RMM2.74toRealtime.txt, accessed January 2012). To better assess and confirm the origin of intraseasonal oscillations from Neumayer we compare the frequency of MJO phases for the periods characterized by high and low values of the time coefficients of the dominant pattern of T, U and SLP.

Empirical Orthogonal Functions (EOF) analysis (Von Storch & Zwiers Reference Von Storch and Zwiers1999) was used to identify the dominant patterns of T, U and SLP variability. The EOF method reduces a large number of variables to a few independent modes retaining much of the variance of the original data. Empirical Orthogonal Functions also filter out the noise from the data (Von Storch & Zwiers Reference Von Storch and Zwiers1999). We use correlation maps (Von Storch & Zwiers Reference Von Storch and Zwiers1999) to derive atmospheric circulation patterns associated to the dominant modes of intraseasonal variability of T, U and SLP measured at Neumayer. Singular spectrum analysis (SSA) (Ghil et al. Reference Ghil, Allen and Dettinger2002) was used to extract quasi-periodic signals from the time series. Singular spectrum analysis is a method for decomposition of a time series into a sum of a small number of independent and interpretable components such as slowly varying trends, oscillatory components and structureless noise (Ghil et al. Reference Ghil, Allen and Dettinger2002). The multi-taper method (MTM) was used for spectral estimation, which proved to be a useful tool for exploration of the spectral properties of signals that contain both broadband and oscillatory components (Ghil et al. Reference Ghil, Allen and Dettinger2002). Non-stationarity in the oscillatory signals is investigated using wavelet analysis (Torrence & Compo Reference Torrence and Compo1998). Wavelet analysis is a spectral technique for analysis of localized variations of power within a time series. By decomposing a time series into time-frequency space, one is able to determine both the dominant modes of variability and how those modes vary in time (Torrence & Compo Reference Torrence and Compo1998).

Mean annual cycle of temperature, wind speed and sea level pressure

The mean annual cycle of all variables considered in our study is defined as the average of all available measurements for each observational hour within a year over the period 1982–2011. The resolution of mean annual cycle is therefore three hours.

Temperature recorded at Neumayer presents strong variations within a year (Fig. 3a). The black curve in Fig. 3a represents the mean annual cycle of temperature while the red curve stands for its low-pass filtered (timescales longer than 150 days) component, based on a Fourier filter applied to the mean annual cycle. This low-pass filtered component contains both annual and semi-annual oscillations. The mean temperature decreases continuously from January to July-August and increases relatively quickly from September to December. Similar variations were reported in previous studies for Neumayer (e.g. König-Langlo & Loose Reference König-Langlo and Loose2007) and Faraday (Van den Broeke Reference Van den Broeke1998) weather stations.

Fig. 3

a. Time series of the mean annual cycle of temperature with a three-hour resolution (original) and its low frequency part (low-pass). Low frequency part refers to timescales longer than 150 days. b. High frequency part of the mean annual cycle defined as the difference between mean annual cycle and its low frequency part represented in a. through black and respectively red curves.

At Neumayer (70°39′S, 8°15′W) the sun stays permanently above the horizon (polar day) from 19 November–24 January and permanently below the horizon (polar night) from 19 May–27 July. The polar day and polar night periods are associated with two distinct temperature variability regimes (Fig. 3b). High frequency variations are visible during polar day while relatively low frequency variations dominate during the polar night (Fig. 3b). A spectral analysis (not shown) reveals that polar day temperature variations are dominated by the diurnal cycle. This cycle is almost absent during the polar night. The relatively low frequency variation of temperature during polar night (Fig. 3b) reflects an increasing role of the atmospheric circulation and relatively low influence of radiative processes on temperature variability at Neumayer Station.

From mid-February to mid-March the wind speed shows an abrupt increase (Fig. 4a). The wind shifts back to lower values at the end of November and beginning of December. The low frequency component (timescale longer than 150 days) contains both annual and semi-annual oscillations. Within one year, the sea level pressure is relatively high at the end of December, January and June (Fig. 4b). Similar variations in the annual cycle of pressure were recorded at different Antarctic coastal stations close to the Neumayer Station (Van den Broeke Reference Van den Broeke1998).

Fig. 4

a. Time series of the mean annual cycle of wind speed with a three-hours resolution (original) and its low frequency part (low-pass). Low frequency part refers to timescales longer than 150 days. b. As in a. but for sea level pressure.

Daily to intraseasonal oscillations of temperature, wind speed and sea level pressure and related atmospheric circulation patterns

The spectral structure of the variability at timescales from three hours to 150 days is shown by a wavelet analysis of the three-hourly resolution time series of T (Fig. 5a), U (Fig. 5b) and SLP (Fig. 5c) recorded during the year 2010. We chose this particular year because the time series show temporal structures typical of the other years of the analysed period. Prior to the wavelet analysis the low frequency variations, i.e. timescales longer than 150 days, were removed using the Fourier filter. The wavelet spectrum of T shows enhanced variability in several spectral bands (Fig. 5a). A persistent but non-stationary oscillation with a period of one day is clearly visible in the wavelet spectrum (Fig. 5a). It is relatively strong during the Antarctic summer and almost absent during Antarctic winter. A diurnal cycle is not evident in the wavelet spectrum of U (Fig. 5b) and SLP (Fig. 5c). Enhanced variability at ∼4 to ∼8 days is indicated by wavelet spectra of all variables, most clearly in U (Fig. 5). In the intraseasonal band the temperature record shows enhanced variability at ∼ 40 days and ∼ 80 days (Fig. 5a) while dominant periods for U are ∼ 16 days and ∼ 60 days (Fig. 5b). Sea level pressure record shows a broad peak corresponding to a period of between ∼ 25 and ∼ 35 days (Fig. 5c). The intraseasonal variability is particularly strong during Antarctic winter for all variables (Fig. 5).

Fig. 5

a. The continuous wavelet spectrum of the high frequency component (timescales less than 150 days) of the 2010 year temperature (left) and the corresponding global power spectrum (right). The thick black contour is the 80% significance level against red noise. The cone of influence where edge effects might be relevant is indicated as thick black curve. Colours show power (or variance). The dashed line in the right panel represents the 80% significance level against the red noise. More details of the method are found in Torrence & Compo (Reference Torrence and Compo1998). b. & c. as in a. but for 2010 year wind speed and sea level pressure respectively.

The correlation maps of T, U and SLP time series with Z500 for the period 1982–2011 shows different patterns (Fig. 6). Temperature (Fig. 6a) and U (Fig. 6b) variations are related with a wave-train pattern that extends from the South Pacific to the South Atlantic and Indian oceans. The highest correlations are recorded near Neumayer Station. During the positive phase of this pattern, enhanced advection of relatively warm air from the north-east induces relatively high temperatures at the station. This is consistent with the positive correlation (r = +0.54) between U and T records. A reverse situation, i.e. low U and negative T anomalies are recorded at Neumayer during the negative phase of this pattern. The correlation map of SLP with Z500 (Fig. 6c) shows an annular structure similar to SAM/AAO. There is a relatively weak projection of T or U and SLP related atmospheric circulation patterns (Fig. 6c).

Fig. 6

Correlation map of a. temperature (T), b. wind speed (U), and c. sea level pressure (SLP) with 500 hPa geopotential height. Low frequency components (timescales longer than 150 days) were removed from the data prior to the correlation.

Patterns of temperature, wind and sea level pressure variability and associated atmospheric circulation anomalies

The variables analysed here are obviously not independent of each other. The T is significantly positively correlated with U (r = +0.54) and negatively correlated with SLP (r = -0.20). Wind speed is negatively correlated with SLP (r = -0.37). To identify the common patterns of variability we perform an EOF analysis of the normalized daily time series of these variables. The time variations with timescales longer than 150 days were filtered out from these time series.

The first EOF (loadings are +0.59, +0.65, -0.47) describes 59% of the variance and captures the out-of-phase variations of T and U with SLP. This is consistent with the correlations between these variables reported above. The correlation of the associated time coefficients (PC1) with Z500 (Fig. 7) shows a wave-train pattern extending from the South Pacific to the Indian Ocean. The highest correlation (r = +0.6 significant at 95% level) is recorded near Neumayer Station. The correlations are relatively low in the South Pacific Ocean. When this pattern is in a positive phase, enhanced advection of relatively warm air from north-east toward the station leads to positive temperature anomalies at the station. At the same time, SLP near the station is relatively low (Fig. 7). This pattern projects well on the T (Fig. 6a) and U (Fig. 6b) associated patterns, consistent with relatively high loadings of these variables in the leading EOF.

Fig. 7

Correlation map of the PC1 of temperature (T), wind speed (U) and sea level pressure (SLP) with 500 hPa geopotential height for lag zero. Low frequency components (timescales longer than 150 days) were removed from the data prior to the correlation.

A lag-correlation analysis reveals that the wave-train pattern associated to PC1 (Fig. 7) propagates eastward. The correlation map of PC1 with Z500 anomalies when Z500 leads with six days (Fig. 8a) shows a wave-train pattern with relatively high correlations extending to the South Pacific sector relative to the lag-zero pattern (Fig. 7). Correlations in this area become non-significant for lag +6 days (Fig. 8b). Instead correlations increase in the Atlantic sector as lag-time increases from -6 days to 6 days (Fig. 8). This suggests a propagation of the signal from the South Pacific to the Atlantic sector.

Fig. 8

Correlation map of the PC1 of temperature (T), wind speed (U) and sea level pressure (SLP) with 500 hPa geopotential height for a. lag = -6 days, and b. lag = 6 days. Low frequency variations (timescales longer than 150 days) were removed from the data prior to the correlation.

The eastward propagation of the atmospheric circulation pattern associated with PC1 (Fig. 8) suggests a possible connection of intraseasonal climate variability in the Pacific sector with intraseasonal variability recorded at Neumayer. One important source of intraseasonal variability in the Pacific sector, which is the origin of the wave-train pattern associated with PC1 (Fig. 8), is the tropical region (e.g. Mo & Higgins Reference Mo and Higgins1998). As the tropical variability presents well defined oscillations in the intraseasonal band (e.g. Mo & Higgins Reference Mo and Higgins1998) we look for a direct connection between PC1 and the MJO. For this purpose we calculate the frequency of the MJO phases (Wheeler & Hendon Reference Wheeler and Hendon2004) for the days when PC1 was higher than +2 standard deviation and compared with the frequency of MJO phases during the days when PC1 was lower than -2 standard deviation (Fig. 9). The results remain qualitatively the same for reasonable changes of these threshold values. As shown in Fig. 9, the frequency of 1 to 4 MJO phases is higher for periods characterized by high values of PC1 relative to the periods characterized by low values of PC1. A reversed situation is observed for phases 5 to 8 (Fig. 9). The highest difference (significant at 95% level) is recorded for phase 8 of the MJO.

Fig. 9

Frequency of Madden-Julian Oscillation phases for periods characterized by high values of PC1 (black bars) and low values of PC1 (red bars).

Because the focus of this study is on the variability of Neumayer data at intraseasonal timescales we performed a wavelet analysis of a five-day mean PC1 values (Fig. 10). The wavelet spectrum shows that the intraseasonal oscillations are strongly non-stationary. Most of the significant oscillations occur at ∼ 40 days and ∼ 80 days, respectively. This is confirmed also by the global spectrum (Fig. 10, right panel) which shows significant (95% level) peaks at these timescales. To better assess and confirm the periodic signals shown by the wavelet spectrum (Fig. 10) we performed a Singular Spectrum Analysis of five-day mean PC1 values. The eigenspectrum of PC1 (not shown) presents two pairs of eigenvalues which correspond to two quasi-periodic signals (e.g. Ghil et al. Reference Ghil, Allen and Dettinger2002). The power spectrum of the reconstructed signal from these two SSA components shows a prominent peak at ∼ 40 days and ∼ 80 days (not shown), consistent with the wavelet spectrum (Fig. 10).

Fig. 10

The continuous wavelet spectrum of the PC1 of temperature (T), wind speed (U) and sea level pressure (SLP) (left) and the corresponding global power spectrum (right). The thick black contour is the 95% significance level against red noise. The cone of influence where edge effects might be relevant is indicated as light shadings. Colours show power (or variance). The dashed line in the right panel represents the 95% significance level against red noise. More details of the method are found in Torrence & Compo (Reference Torrence and Compo1998).

The second mode of T, U, and SLP intraseasonal variability captures in phase variations of these variables (loadings are +0.52, +0.13, +0.83). It describes 25% of the intraseasonal variance of these variables. The correlation of the associated time coefficients (PC2) with Z500 shows an annular structure (Fig. 11) that projects well on the negative phase of the SAM/AAO. It is similar to the circulation pattern associated with SLP variations (Fig. 6c). This was expected because the SLP loading dominates the EOF2. This pattern is associated with weak positive temperature anomalies at Neumayer Station. Consistent with this result, a recent study (Schneider et al. Reference Schneider, Okumura and Deser2012) shows that SAM/AAO is negatively correlated with the temperature at Neumayer Station.

Fig. 11

Correlation map of PC2 of temperature (T), wind speed (U) and sea level pressure (SLP) with 500 hPa geopotential height. Low frequency components (timescales lower than 150 days) were removed from the data prior to the correlation.

The wavelet spectrum of the five-day mean of PC2 values (Fig. 12) also shows significant oscillations in the intraseasonal frequency band at ∼ 35 days, ∼ 60 days and ∼ 120 days. An SSA of PC2 (not shown) also reveals the existence of periodic signals at ∼ 35 days, ∼ 60 days and ∼ 120 days in PC2.

Fig. 12

As in Fig. 10 but for PC2.

Discussions and conclusions

We investigated the daily to intraseasonal variability of T, U and SLP recorded at Neumayer Station, located in the north-eastern Weddell Sea, Antarctica (Fig. 1). The records cover a period of 30 years (1982–2011) with three-hour resolution. The mean annual cycle of T presents specific spectral characteristics during the polar day (from November–January) relative to the polar night (from May–July). Polar day variations of T are dominated by the diurnal cycle while intraseasonal oscillations dominate the T variations during polar night. The diurnal cycle does not appear clearly in the three-hourly records of U and SLP wavelet spectrum during either polar day or polar night. However, both U and SLP records show strong intraseasonal oscillations. During the polar night the temperature contrast between Antarctica and the surrounding oceans increases which allows warm air masses from the north to penetrate more deeply into Antarctica (Van den Broeke Reference Van den Broeke1998). At longer timescales the mean annual cycle of these variables are dominated by semi-annual and annual oscillations (Fig. 4). Van den Broeke (Reference Van den Broeke1998) showed that, for the annual cycle of surface pressure recorded at several stations along the east Antarctic coast, the semi-annual oscillation accounts for 36–38% variance. A significant coupling between the semi-annual cycle of surface pressure and air temperature was detected and related with changes in meridional atmospheric circulation (Van den Broeke Reference Van den Broeke1998).

Our EOF analysis isolated two patterns of daily to intraseasonal variability of T, U and SLP. The first pattern captures the out-of-phase variations of T and U with SLP. The time coefficients associated with this pattern (PC1) show enhanced variability at ∼ 40 days and ∼ 80 days. Fluctuations at timescales of ∼ 40 days were identified in other Antarctic climate variables. Kushner & Lee (Reference Kushner and Lee2007) isolated a regional atmospheric circulation pattern which propagates eastward within the SAM/AAO pattern. They identified two main periodicities, one of them being ∼ 40 days. The existence of such an oscillation was confirmed by Pohl et al. (Reference Pohl, Fauchereau, Reason and Rouault2010). Some authors related this oscillation to the tropical MJO which presents enhanced variability at these timescales (e.g. Madden & Julian Reference Madden and Julian1994). However, it is not clear if the ∼ 40 day oscillation can be related to the MJO signature in Antarctic climate or if it is an intrinsic mode of SAM/AAO variability. Our lag-correlation analysis suggests that the ∼ 40 day periodicity in the Neumayer variables is related to South Pacific intraseasonal variability which is strongly influenced by the MJO (e.g. Donald et al. Reference Donald, Meinke, Power, Maia, Wheeler, White, Stone and Ribbe2006). Furthermore, a composite analysis reveals that the frequency of phase 8 of MJO is significantly higher during periods characterized by low values of PC1 relative to periods characterized by high values of PC1. This is another indication that the atmospheric circulation wave-train associated with PC1 and the corresponding periodicities are related to tropical MJO.

The atmospheric circulation pattern associated with PC1 of T, U and SLP from Neumayer contains elements of the Pacific South American (PSA) patterns (e.g. Mo & Higgins Reference Mo and Higgins1998). It was shown that one of the PSA patterns is the response to the El Niño Southern Oscillation (Karoly Reference Karoly1989) which suggests a tropical origin of eastward propagating wave-train patterns that influence Southern Hemisphere variability. Mo & Higgins (Reference Mo and Higgins1998) showed that the two PSA modes represent the intraseasonal oscillation in the Southern Hemisphere with periods of roughly 40 days and the evolution of the PSA modes shows a coherent eastward propagation. The pattern associated with temperature, wind and sea level pressure variability, as captured by the PC1, shows also an eastward propagation (Fig. 8). Strong connection of tropical processes with Antarctic climate was identified (Carvalho et al. Reference Carvalho, Jones and Ambrizzi2005, Fogt & Bromwich Reference Fogt and Bromwich2006, Fogt et al. Reference Fogt, Bromwich and Hines2010) which is an additional argument for a tropical origin of the atmospheric circulation pattern associated to the dominant mode of T, U, and SLP intraseasonal variability recorded at Neumayer Station.

PC2 of T, U and SLP shows enhanced variability in the intraseasonal band with peaks at ∼ 35, ∼ 60 and ∼ 120 days. The atmospheric circulation pattern associated with this mode (Fig. 11) projects strongly on the SAM/AAO. Empirical mode decomposition applied to the daily SAM/AAO index reveals the existence of strong periodic signals at 35 days, 63 days and 117 days (Pohl et al. Reference Pohl, Fauchereau, Reason and Rouault2010), which can be related to the corresponding peaks in our PC2 spectrum. Although these peaks are in the range of tropical MJO variability, they cannot be attributed to the MJO signature on SAM/AAO variability. A recent study (Pohl et al. Reference Pohl, Fauchereau, Reason and Rouault2010) showed that the interaction between MJO and SAM/AAO is very weak. The MJO is associated with regional-scale cyclonic or anticyclonic anomalies that tend to propagate eastward and which cannot produce a generalized mass transfer between the polar region and mid-latitudes as is the case for AAO. Therefore, these peaks in the PC2 spectrum are probably related to internal dynamic processes of the polar atmosphere including those related to the SAM/AAO.

The results presented in this paper suggest that tropical processes and higher latitude processes are related to distinct anomaly patterns of the meteorological variables measured at Neumayer, and that these patterns have specific temporal oscillatory components. In a forthcoming paper we analyse the variability of climatic variables recorded at Neumayer from semi-annual to decadal timescales. The results presented in this paper could lead to a significant improvement in the interpretation of the ice core variability and its relationship with atmospheric circulation in the Southern Hemisphere.