Data

The data used in this study are monthly surface observations from 13 Antarctic stations (see Fig. 1) obtained from the Antarctic READER (Reference Antarctic Data for Environmental Research) Project (Turner et al. Reference Turner, Colwell, Marshall, Lachlan-Cope, Carleton, Jones, Lagun, Reid and Iagovkina2004) (http://www.antarctic.ac.uk/met/READER/data.html, accessed August 2010). This study uses two global datasets: the National Centers for Environmental Prediction-National Center for Atmospheric Research (NCEP-NCAR) re-analysis NCEP-NCAR global re-analysis and the European Centre for Medium-Range Weather Forecasts (ECMWF) 40 Year Re-analysis ERA-40 re-analysis. The NCEP-NCAR global re-analysis is based on the NCEP operational Eta model of 10 January 1995 with a reduced horizontal resolution of T42 (209 km) and 28 vertical levels. Its temporal coverage is four times per day for NCEP I from 1 January 1948 to the present. Detailed descriptions of the NCEP-NCAR global re-analysis are given by Kalnay et al. (Reference Kalnay, Kanamitsu, Kistler, Collins, Deaven, Gandin, Iredell, Saha, White, Woollen, Zhu, Chelliah, Ebisuzaki, Higgins, Janowiak, Mo, Ropelewski, Leetmaa, Reynolds and Jenne1996) and Kistler et al. (Reference Kistler, Kalnay, Collins, Saha, White, Woollen, Chelliah, Ebisuzaki, Kanamitsu, Kousky, van Den Dool, Jenne and Fiorino2001). The ERA-40 re-analysis contains the data from September 1957–August 2002 with a horizontal resolution of T159 (125 km) and 60 vertical levels (23 standard pressure levels). More descriptions about ERA-40 can be found in the work of Uppala et al. (Reference Uppala, Kållberg and Simmons2005). The two re-analysis datasets are on a 2.5°×2.5° latitude-longitude grid. The primary atmospheric variables include mean sea level pressure and 2 m air temperature for the period from 1958–2001. Although Bromwich & Fogt (Reference Bromwich and Fogt2004) found a low level of skill in ERA-40 data before 1978, ERA-40 re-analysis data are also compared and validated (Uppala et al. Reference Uppala, Kållberg and Simmons2005). Justino et al. (Reference Justino, Setzer, Bracegirdle, Mendges, Grimm, Dechiche and Schaefer2010) found the seasonal cycle of the temperature 2 m above ground level over Antarctica may be biased in ERA-40. Yu et al. (Reference Yu, Zhang, Zhou, Zhong, Lenschow, Hsu, Wu and Sun2010) also compared two re-analysis from 1979–2001 with observational data and found at the interannual timescale, atmospheric pressures at different height levels in ERA-40 are in better agreement with the observed pressure than that in the NCEP-NCAR re-analysis. The ERA-40 re-analysis also outperforms NCEP-NCAR re-analysis in atmospheric temperature except the surface layer atmosphere where the biases are somewhat larger. Phillpot (Reference Phillpot1991) and Schwerdtfeger (Reference Schwerdtfeger1984) found that for stations located at high elevation, the reduction to sea level is ambiguous. Because of high elevation of Amundsen–Scott (2800 m) and Vostok (3488 m), surface pressure at these two stations is used to calculated trends and correlation coefficients, but at the other 11 stations sea level pressure is used.

Fig. 1 Location map of the Antarctic stations used in this study and the Antarctic topography.

In addition, the Southern Oscillation Index (SOI) was derived from http://www.cgd.ucar.edu/cas/catalog/climind/soiAnnual.html, accessed August 2009.

Analysis method

The AAO, PSA1 and PSA2 indices are defined as the standardized leading three time series of EOF analysis of ERA-40 monthly sea level pressure poleward of 20°S. Total linear trends are obtained as the slope of a straight line by least-square fit at each grid point for specific periods. The trends that are linearly congruent with the monthly AAO, PSA1 and PSA2 indices can be obtained following the method of Thompson & Wallace (Reference Thompson and Wallace2000b): a) regressing monthly values of the time series of the grid point onto the AAO index, b) multiplying the resulting regression coefficients by the linear trend in the AAO index, c) the trends that are linearly congruent with the monthly PSA indices are estimated by means of similar method to the AAO index. The residual trend is defined as the trend attained by subtracting the trends in the AAO and PSA1 (PSA2) indices from total trend.

Seasonal correlation between the PSA and surface air temperature are derived from the period 1958–2001 with 42 degrees of freedom. Significance of the correlations is tested by applying Student's t-test. Normally the four seasons are defined: summer (December–February), autumn (March–May), winter (June–August), and spring (September–November). To ease the comparison of the different contributions from three different indices at significant levels, autumn (April–May) and spring (October–November) are defined.

Relationship to PSA

The seasonal relationship between the AAO index and surface air temperature has been discussed previously by Marshall (Reference Marshall2007) using similar observational data. Therefore, we will focus only on the relationship between PSAs (PSA1 and PSA2) indices and surface air temperature.

The correlation between the PSA and Antarctic surface air temperatures are done for each of the four seasons during the 44 yr (1958–2001) period. In general, the strength and location of the low and high pressure centres associated with the PSA1 and PSA2 indices vary with seasons (Lau et al. Reference Lau, Sheu and Kang1994), and the strength and location of corresponding meridional wind also change, which together lead to significant seasonal variations of surface air temperature. The Antarctic Peninsula, Bellingshausen Sea, Amundsen Sea, Ross Sea and Weddell Sea are mainly areas influenced by the PSA during the entire year. There are, however, seasonal differences in the influence area coverage and significance of the relationship between the PSA and temperature.

In summer there are significant positive correlations (at a significance level of > 98%) between the PSA1 and temperature at the stations over the Ross Sea (Scott Base) and the Weddell Sea (Halley). At Amundsen–Scott, Davis and Faraday stations the positive correlations are significant at a significance level of > 90%. At Mawson, Esperanza, Orcadas, Syowa, and Vostok stations the correlation is positive but not statistically significant. At the other three stations (Casey, Dumont d'Urville, and Mirny) the negative relationship is weak and not significant. During the same season the PSA2 has a remarkable effect on the surface temperature in the northern Peninsula (Orcadas, Esperanza and Faraday) with high positive correlation coefficients, particularly at Orcadas where the relationship is significant at > 99% level. On the contrary the correlations between PSA2 index and temperature at other Antarctic stations are not significant.

The spatial pattern of the PSA-temperature correlations is similar in autumn to that in summer, and only the differences will be discussed here. The correlation between the PSA1 index and surface temperature strengthens in the northern Antarctic Peninsula with an increase in the correlation coefficient from 0.29–0.51 (at > 99.9% level) at Faraday, from 0.18–0.32 (at > 95% level) at Orcadas, and from 0.02–0.23 at Esperanza. On the contrary, the influence over the eastern Weddell Sea weakens with a decrease in correlation from 0.38 at > 98% level to 0.24 at < 90% level at Halley. The relationship between the PSA1 index and temperature at Scott Base remains unchangeable. The correlation between PSA2 index and temperature varies with station locations. The correlation at Scott Base increases from -0.01 to -0.32 at > 95% level, but a decrease occurs in the northern Antarctic Peninsula, especially for Esperanza and Faraday where the relation is less significant.

In winter the correlations between the PSA1 index and temperature at stations in the northern Peninsula are the most significant at > 99.9% level, but the correlation is much lower at other stations. Similarly, for PSA2, there is only one station (Dumont d'Urville) where the correlation between PSA2 index and surface temperature is significant at > 99% level and two other stations in East Antarctica and one station in the northern Peninsula where the correlation is found to be significant at > 90% significance level.

The spatial coverage of the significant correlation between PSA1 index and temperature is the largest in spring with seven stations at > 90% significance level. Only in spring are temperatures at Davis, Vostok, and Mirny significantly correlated with PSA1 index. The correlation between PSA1 index and Faraday temperatures is strongest in spring with a magnitude of 0.69 and a significance level > 99.9%. The physical mechanisms for the strong relationship involve the poleward advection of warm air across the northern Peninsula to the east coast where Faraday is located (not shown). At Orcadas, north of Faraday, the correlation coefficient is lower, 0.32 with > 95% level, but at Esperanza near Faraday, the correlation is not significant. These differences among the three stations in northern Peninsula may be related to the local circulation induced by topography. The correlations between PSA1 index and temperature at all 13 stations are positive. The strongest correlation between PSA1 index and Scott Base temperature also appears in spring, with correlation coefficient 0.48 at > 99% level. The region influenced by the PSA2 in spring is the Weddell Sea (Halley, Esperanza and Orcadas), particular at Orcadas where the strongest correlation among the four seasons is found (0.56 at > 99.9% level).

In summary, stations in the Ross Sea and the Weddell Sea, which are subject to a strong influence of Rossby wave train (Fig. 2b & c), exhibit high correlation between PSA and surface temperature, with the highest correlation found in the northern Antarctic Peninsula where northerly winds prevail.

Trend analysis

Figure 3 and Table II display the estimated summertime trends in surface temperature from 1958–2001 that include total trend, trend congruent with AAO, trend congruent with PSA2 and residual trend. Total trends in surface temperature are generally positive except at Amundsen–Scott, Mawson, Mirny, and Casey where a cooling trend exceeds 0.5°C. The strongest warming trend occurs in the northern Peninsula with a slope of 1.67°C 44 yr^-1 at Esperanza. The significant warming trend in the northern Antarctic Peninsula was also found by Turner et al. (Reference Turner, Colwell, Marshall, Lachlan-Cope, Carleton, Jones, Lagun, Reid and Iagovkina2005). 44 yr (1958–2001) summertime (December–February) linear trends in surface air temperature using ERA-40 re-analysis data are also computed and the results are shown in Fig. 4a. A positive trend occurs over the Antarctic continent with maximum values as high as 3°C 44 yr^-1. The gridded ERA-40 re-analysis data show variation of trend with topography, which is not captured by data from the surface stations that are located mostly over coastal area. For NCEP re-analysis there is a negative linear December–February trend in surface air temperature in East Antarctica with the minimum value of -6°C 44 yr^-1, which is opposite to the result of ERA-40 re-analysis and a positive trend occurs in East Antarctica in Fig. 5a.

Fig. 3 The estimated trends in Antarctic surface temperature for 1958–2001 in summer (December–February) for observational data. a. Total trend, b. trend congruent with AAO, c. trend congruent with PSA2, and d. residual trend.

Table II The estimated change in Antarctic surface temperature for 1958–2001 caused by total trends, trends in the AAO, PSA1 and PSA2 and residual trends. Units: °C.

Fig. 4 44 yr (1958–2001) linear December–February trends in surface air temperature for ERA-40 re-analysis data. a. Total trends, b. & c. the components of the trends that are linearly congruent with the monthly AAO, PSA2, and d. residual trends. Units are °C for surface air temperature.

Fig. 5 44 yr (1958–2001) linear December–February trends in surface air temperature for NCEP re-analysis data. a. Total trends, b. & c. the components of the trends that are linearly congruent with the monthly AAO, PSA2, and d. residual trends. Units are °C for surface air temperature.

Antarctic Oscillation contributes to warming over the northern Antarctic Peninsula (Esperanza, Faraday, and Orcadas) related to increasing westerly, but caused significant cooling at the rest of the ten surface station locations (Fig. 3b) (Thompson & Wallace Reference Thompson and Wallace2000a, Van den Broeke & Van Lipzig 2004, Justino & Peltier Reference Justino and Peltier2008). Similar result was also found by Marshall (Reference Marshall2007). Van den Broeke & Van Lipzig (2004) explained that during strong vortex (positive AAO) East Antarctic cooling has two components: suppressed meridional air exchange, and additional cooling near the surface in areas where a weakening of near-surface winds has intensified the surface temperature inversion. The warming in the Antarctic Peninsula may result from stronger northerly wind. Orr et al. (Reference Orr, Cresswell, Marshall, Hunt, Sommeria, Wang and Light2004, Reference Orr, Marshall, Hunt, Sommeria, Wang, van Lipzig, Cresswell and King2008) found the warming over the Antarctic Peninsula involving northerly wind and adiabatical warming induced by strong down-slope wind. The regresses of zonal and meridional wind on summer AAO show positive AAO index corresponds to increasing north-westerly wind over the northern Antarctic Peninsula (not shown). The largest AAO related warming (0.67°C) is at Esperanza and the largest cooling (-0.72°C) occurs at Mirny. The proportion of total temperature change in autumn that is attributed to AAO varies significantly across Antarctica. At Faraday the AAO related warming is one tenth of what is actually observed. At Amundsen–Scott AAO explains majority of the total cooling trend. At some stations the temperature changes congruent with AAO is opposite to the total trend.

Positive trends caused by PSA2 lie at coastal stations from 60°E westwards to 120°W except for Davis and Dumont d'Urville. Negative trends are found at other six stations in East Antarctica (Fig. 3c). At most stations PSA2 related temperature changes (with maximum change of 0.4°C) are less than those associated with AAO. The trends in surface air temperature associated with PSA2 correspond to the wind direction of atmospheric circulation induced by PSA2 (Fig. 3b). In other words, the region with southerly (northerly) prevailing wind has a cooling (warming) trend.

The residual trends in summer are positive at all stations, with the largest positive trend of 0.94°C 44 yr^-1 at Vostok (Fig. 3d). In contrast, the warming in East Antarctica determined from the ERA-40 re-analysis data (Fig. 4d) are > 2.5°C 44 yr^-1. There is a cooling trend in surface temperature over the Ross Sea, which is absent from the Scott Base (Fig. 3d). In summer significant differences in the surface temperature trends between ERA-40 and the surface stations occur in East Antarctica, where the residual trend reaches a maximum value. However, for the NCEP re-analysis the trends related to AAO and PSA2 explain a large part of total trends especially in East Antarctica, the magnitude of residual trends is smaller than that of ERA-40 (Fig. 5b–d).

In autumn (April–May, Fig. 6a) positive trends in surface temperature are at Orcadas, Esperanza, Faraday and Scott Base, particularly at Esperanza where the warming trend is 4.09°C 44 yr^-1. The warming trend over the northern Antarctic Peninsula is consistent with the result of Steig et al. (Reference Steig, Schneider, Rutherford, Mann, Comiso and Shindell2009). The other nine stations show negative trends, with the most cooling of -3.39°C at Halley. The trends determined using ERA-40 re-analysis data are positive everywhere in Antarctica and the rate of warming at Esperanza nearly doubles that from the station data (Fig. 7a). For NCEP re-analysis the spatial pattern of the trend in Antarctica are similar to that of ERA-40 except for the Weddell Sea (Fig. 8a). It indicates that ERA-40 and NCEP re-analysis data are not accurate in describing the total trend in autumn surface air temperature.

Fig. 6 The estimated trends in Antarctic surface temperature for 1958–2001 in autumn (April–May) for observational data. a. Total trend, b. trend congruent with AAO, c. trend congruent with PSA1, and d. residual trend.

Fig. 7 44 yr (1958–2001) linear April–May trends in surface air temperature for ERA-40 re-analysis data. a. Total trends, b. &c. the components of the trends that are linearly congruent with the monthly AAO, PSA1, and d. residual trends. Units are °C for surface air temperature.

Fig. 8 44 yr (1958–2001) linear April–May trends in surface air temperature for NCEP re-analysis data. a. Total trends, b. &c. the components of the trends that are linearly congruent with the monthly AAO, PSA1, and d. residual trends. Units are °C for surface air temperature.

The influence of AAO on surface temperature is stronger in autumn than in summer although the spatial pattern is similar between the two seasons (Figs 3b & 6b). The largest AAO related warming (2.4°C 44 yr^-1) is found at Esperanza while the strongest cooling (-1.53°C 44 yr^-1) occurs at Mawson. PSA1 contributes to warming at most stations except for Casey, Dumont d'Urville, Syowa, and Vostok, and the warming rate is < 1°C. The trends associated with AAO and PSA1 determined using ERA-40 and NCEP re-analyses are consistent with those derived from surface station data. Strong zonal westerlies induced by the strengthening of AAO and anomalous northerly winds associated with high PSA1 contribute to warming over the Antarctic Peninsula and West Antarctica (Fig. 6b & c). Similar to summer season, the trend attributable to AAO is more than to PSA2.

The station residual trends in autumn (Fig. 6d) are not always positive as seen in the ERA-40 residual trends (Fig. 7d). The similar case also occurs in the NCEP residual trends (Fig. 8d). The residual trends are negative at five stations and the largest cooling rate of -3°C 44 yr^-1 is found at Halley which is nearly equal to the total trend. This indicates that forces other than AAO and PSA1 are responsible for the surface temperature changes at Halley and other stations.

In spring (Fig. 9a, Table II) positive trends are found at most stations except for four coastal stations in East Antarctica (Syowa, Mawson, Mirny and Casey). The strongest warming of 2.58°C 44 yr^-1 occurs at Scott Base. The ERA-40 re-analysis data (Fig. 10a) indicate a strong warming trend distributed across the Antarctic continent with the maximum value of > 6°C 44 yr^-1 in the Weddell Sea. The large differences between the ERA-40 trends and the station trends suggest that the surface air temperature in ERA-40 re-analysis data is unreliable in the spring season. Although an increasing trend exists in most of the Antarctic continent for the NCEP re-analysis, the magnitude is smaller than that of ERA-40 (Fig. 11a). The increase in surface temperature caused by AAO recedes northwards with AAO related warming occurring at only two stations. But the warming caused by PSA1 covers almost the entire Antarctic continent except for the South Pole. Like for summer and autumn, ERA-40 and NECP re-analyses show a high skill in describing the influence of AAO and PSA1 on springtime temperature. The springtime residual trends exhibit a positive spatial pattern except for four stations (Orcadas, Esperanza, Halley, and Syowa; Fig. 9d). The increasing trend of 2.08°C 44 yr^-1 at Scott Base needs other explanation. Like for summer and autumn, the ERA-40 residual trends in spring (Fig. 10d) shows a strong warming over the entire Antarctic. The NCEP residual trends in spring also perform an increasing trend, but the centre of positive values lie over the Ross Sea (Fig. 11d).

Fig. 9 The estimated trends in Antarctic surface temperature for 1958–2001 in spring (October–November) for observational data. a. Total trend, b. trend congruent with AAO, c. trend congruent with PSA1, and d. residual trend.

Fig. 10 44 yr (1958–2001) linear October–November trends in surface air temperature for ERA-40 re-analysis data. a. Total trends, b. & c. the components of the trends that are linearly congruent with the monthly AAO, PSA1, and d. residual trends. Units are °C for surface air temperature.

Fig. 11 44 yr (1958–2001) linear October–November trends in surface air temperature for NCEP re-analysis data. a. Total trends, b. & c. the components of the trends that are linearly congruent with the monthly AAO, PSA1, and d. residual trends. Units are °C for surface air temperature.

The relative magnitudes of the trends are compared by averaging over the 13 stations for each of the three seasons and the results are shown in Table II. The change in surface air temperature is the largest in autumn (April–May), even surpassing the sum of the changes in summer (December–February) and spring (October–November), while the smallest is in spring. The same applies to the trends related to AAO. Besides the large trend in AAO, the more southward position of circumpolar trough in autumn is also the reason for the largest change connected with AAO. The change associated with PSA (PSA1 or PSA2) is most evident in spring and least in summer. Although the residual trend is the largest in autumn, the ratio of the temperature change explained by AAO and PSA to the total change is also the largest. In all three seasons, AAO appears to exert larger effect on surface temperature than that by PSA. The strong negative trend (-3.01°C 44 yr^-1) found at Halley does not appear to be connected to AAO and PSA. Other factors, e.g. katabatic wind may be responsible for the temperature trend at Halley in autumn and at other stations in different seasons (e.g. Vostok in summer and Scott Base in spring). Hence from Table II we can consider that under most circumstances during three seasons the AAO exert a larger effect on surface temperature than that of the PSA.

The comparison with ERA-40 re-analysis shows that the magnitude of temperature change in ERA-40 is larger than that of the station observations, especially in the total trends and the residual trends. The apparent strong ERA-40 warming trends in all three seasons in the entire Antarctic continent are not supported by the station data used in this study. Similar conclusions are also drawn by Johanson & Fu (Reference Johanson and Fu2007). The summertime total and residual trends for NCEP re-analysis in East Antarctica are opposite to those of ERA-40 re-analysis. In October–November and April–May, the two re-analysis datasets show great similarity. However, the trends associated with the AAO and PSA determined based on ERA-40 and NCEP re-analyses data are mostly consistent with those determined from the station data.

Relationship to PSA

In summer a positive correlation between sea level pressure and PSA1 at > 90% significance level is only found at Dumont d'Urville. In autumn positive correlations exist at Faraday and Esperanza at > 99% significance level, and over South Pole at > 90% level. In winter the correlation in the northern Peninsula weakens, and negative correlations occur at Scott Base and Syowa at a level > 95%. In spring the sea level pressure is positively correlated only with PSA1 at Amundsen–Scott. The regions under the influence of PSA1 vary with season. Compared to the temperature correlation patterns, the correlation for sea level pressure is more scattered.

PSA2 mainly changes the sea level pressure in the northern Peninsula with the strongest influence occurring in winter when three stations show negative correlation at 99% level. The main reason for these negative correlations is that the centre of the larger negative anomaly is situated in the northern Peninsula. In autumn the sea level pressure at three stations (Syowa, Mirny and Casey) in East Antarctica is correlated with PSA2 at > 90% level, while in winter positive correlations occur at Scott Base and Vostok.

Trend analysis

The estimated changes in Antarctic sea level pressure for 1958–2001 associated with total trends, trends in AAO, PSA1 and PSA2, and residual trends during three seasons are shown in Figs 12–17 and Table III. Those trends in sea level pressure for NCEP are offered as supplemental material (http://dx.doi.org/10.1017/S095410201100054X) and are not shown in this paper.

Fig. 12 The estimated trends in Antarctic sea level pressure for 1958–2001 in summer (December–February) for observational data. a. Total trend, b. trend congruent with AAO, c. trend congruent with PSA2, and d. residual trend. For Amundsen–Scott and Vostok stations, station surface pressure is used.

Fig. 13 44 yr (1958–2001) linear December–February trends in sea level pressure for ERA-40 re-analysis data. a. Total trends, b. &c. the components of the trends that are linearly congruent with the monthly AAO, PSA2, and d. residual trends. Units are Pascal for sea level pressure.

Fig. 14 The estimated trends in Antarctic sea level pressure for 1958–2001 in autumn (April–May) for observational data. a. Total trend, b. trend congruent with AAO, c. trend congruent with PSA1, and d. residual trend. For Amundsen–Scott and Vostok stations, station surface pressure is used.

Fig. 15 44 yr (1958–2001) linear April–May trends in sea level pressure for ERA-40 re-analysis data. a. Total trends, b. & c. the components of the trends that are linearly congruent with the monthly AAO, PSA1, and d. residual trends. Units are Pascal for sea level pressure.

Fig. 16 The estimated trends in Antarctic sea level pressure for 1958–2001 in spring (October–November) for observational data. a. Total trend, b. trend congruent with AAO, c. trend congruent with PSA1, and d. residual trend. For Amundsen–Scott and Vostok stations, station surface pressure is used.

Fig. 17 44 yr (1958–2001) linear October–November trends in sea level pressure for ERA-40 re-analysis data. a. Total trends, b. &c. the components of the trends that are linearly congruent with the monthly AAO, PSA1, and d. residual trends. Units are Pascal for sea level pressure.

Table III The estimated changes in Antarctic sea level pressure for 1958–2001 caused by total trends, trends in the AAO, PSA1 and PSA2 and residual trends. Units: Pascal.

*For Amundsen–Scott and Vostok stations station surface pressure is used.

In summer (Fig. 12a) all stations in the Antarctic continent experience a decrease in sea level pressure, with the most remarkable fall of -8 hPa happening at Mirny. Only in South Orkney Islands (Orcadas station) does a positive trend of 1.44 hPa 44 yr^-1 exist. The ERA-40 sea level pressure trends in summer (Fig. 13a) also are negative 60°S, but the pressure trend at the location near Mirny station is larger than the observed. The NCEP trends in sea level pressure in East Antarctica continent interior are positive and opposite to those of ERA-40. The similar case also occurs in trends in surface temperature. As a result of positive AAO index, all stations show a negative trend in sea level pressure (Fig. 12b). The decrease in pressure associated with positive PSA2 index occurs at all but four stations (Orcadas, Amundsen–Scott, Scott Base and Dumont d'Urville; Fig. 12c). The ERA-40 spatial patterns (Fig. 13b & c) are in agreement with observed data. Although the NCEP spatial patterns are similar to those of ERA-40 except for a small part of the Antarctic continent interior, the values of trends for NCEP is less than those of ERA-40 (not shown). The residual trend is an increasing trend except for three stations (Vostok, Casey and Mirny). For two re-analyses data a positive trend appears in the Antarctic continent, but the values in NCEP is larger than those in ERA-40 especially in East Antarctica. As shown in Table III, the changing values associated with PSA2 and the residual trend are further less than these caused by the AAO and the total trend. Hence the higher AAO index plays a crucial role in surface pressure variance (Thompson & Solomon Reference Thompson and Solomon2002).

The total trend in autumn (Fig. 14a) is very similar to that in summer except for a strengthening of positive trend in the northern Peninsula. But the autumn trends in ERA-40 and NCEP re-analyses (Fig. 15a) are all negative in Antarctica including the Antarctic Peninsula. All stations manifest a decreased trend in sea level pressure (Fig. 14b). The pressure trend congruent with the PSA1 displays a positive trend except at five stations in East Antarctica. The residual trends at nine stations are positive trends with maximum value of 4.17 hPa, and the other four stations are negative with minimum value of -6.13 hPa. However, in Fig. 15d the trend in Antarctic continent shows positive except for the East Antarctic coastal zone. In contrast to ERA-40 re-analysis the area of positive trends in NCEP re-analysis is smaller (not shown).

In spring (Fig. 16a) ten stations display a negative trend with the strongest fall of surface pressure at Mirny. The opposite occurs at Antarctic Peninsula stations (Esperanza and Faraday) and island station (Orcadas). The total trends in ERA-40 and NCEP re-analyses is also negative in most of Antarctica (Fig. 17a). It seems that the total trends in ERA-40 and NCEP re-analysis data agrees with that of observations. Like the other two seasons the changes in sea level pressure corresponding AAO is negative, but the magnitude is somewhat less in spring. Except for Orcadas and South Pole, stations with positive trends are distributed in East Antarctica. Similar to the residual trend in autumn, only four stations in East Antarctica exhibit negative trends. There are some disagreements in trend between two re-analyses and observations for the residual trend.

In Antarctica sea level pressure has decreased in 44 yr, particularly at Mirny station where the falling trend is most pronounced in all three seasons. A large part of the pressure decrease is unlikely to be connected to either AAO or PSA. Like temperature, the magnitude of the pressure change averaged over the 13 stations is found to be the largest in autumn and the smallest in spring. The impact of AAO on sea level pressure varies with seasons with the least impact in spring. Among the three seasons AAO plays a more important role in explaining the summertime surface pressure change pattern. In most cases the influence from AAO exceeds that from PSA. The sea level pressure in ERA-40 and NCEP re-analyses appears to be in general more reliable than surface temperature.

Although AAO, PSA1 and PSA2 are main elements in EOF analysis of monthly mean sea level pressure, ENSO also influence strongly the Antarctic surface temperature and pressure leading to the Antarctic dipole (Yuan Reference Yuan2004). So the effect of the SOI on Antarctic surface temperature and pressure should be taken into consideration. Table I shows 44 yr (1958–2001) monthly linear trends of SOI. No monthly trend exceeds the 90% threshold. The trends for each of the four seasons in a year are also calculated. The trends in autumn (March–May), winter (June–August), spring (September–November) and summer (December–February) are -0.91, -0.51, -0.17 and -0.12, respectively. Only the trend in autumn is at 90% significance level.

Correlation analyses between SOI and surface temperature and pressure are performed and significant correlations are found only in springtime at several stations (not shown). The contributions of SOI to the Antarctic surface temperature and pressure in summer (December–February), autumn (April–May) and spring (October–November) are estimated and the results are shown in Fig. 18 and Table IV. The largest SOI related change occurs in autumn which is related to the largest trend of -0.86°C in autumn as compared to the trend of -0.12°C and -0.10°C in summer and spring. Although ENSO affects surface temperature and pressure over Antarctica differently in different seasons, the overall effect (with the largest value of < 0.5°C and 1 hPa in autumn), is smaller than that of AAO or PSA. The smaller contribution of ENSO to the Antarctic temperature and pressure trends may be a result of smaller change of SOI in the 44 years compared to the changes of AAO and PSA during the same periods.

Fig. 18 The trends in Antarctic surface temperature (a., c. & e.), and sea level pressure (b., d. & f.), congruent with the SOI for 1958–2001. a. & b. In summer (December–February), c. & d. in autumn (April–May), and e. & f. in spring (October–November). For Amundsen–Scott and Vostok stations, station surface pressure is used.

Table IV The estimated changes in Antarctic surface temperature and sea level pressure for 1958–2001 caused by the SOI. Units are °C for surface air temperature and Pascal for sea level pressure.