Variability and Trends in Antarctic Surface Temperatures from In Situ and Satellite Infrared Measurements (original) (raw)

1. Introduction

Global records have indicated an overall increase of 0.57°C in surface air temperature from 1961 to 1997 with slightly greater increase in the Southern than the Northern Hemisphere (Jones et al. 1999). The records show that the warmest year recorded in this century occurred in 1998, the warmest 10 years in succession occurred in 1989–98, and the warmest decade is expected to be the 1990s. In a greenhouse-warming scenario, increases in global surface temperature are expected to be amplified in the polar regions due to feedback effects associated with the high albedo of the surface in these regions (Budyko 1966; Manabe et al. 1992). Surface air temperatures have indeed been reported to be on the rise in many stations in Antarctica, especially in the Antarctic peninsula, where the trends are as high as 0.5°C decade−1 (Raper et al. 1984; King 1994; Jones 1995; Jacobs and Comiso 1997; Skvarca et al. 1998). A thorough evaluation of such a phenomenon and how it is reflected in the entire Antarctic region is important, since the mean surface temperature on the ice shelves of western Antarctica during the summer is only about −8°C (Ohmura et al. 1996).

It is useful to note that the Antarctic temperature record is based mainly on only a few stations, most of which are located around the periphery of the continent. The surface temperature distributions are also different from station to station and the trends can have opposite signs even in adjacent stations (Jacka and Budd 1991; Weatherly et al. 1991; Smith et al. 1996). The uncertainties related to inadequate spatial sampling have been studied by Karl and Knight (1994) and can be considerable. This paper aims to provide an updated study of the Antarctic trends using station data in combination with a spatially more detailed observation of surface temperatures from space.

Surface temperature maps of the polar regions have been previously derived from satellite infrared data for the period 1979–85, using the Temperature Humidity Infrared Radiometer (THIR) data (Comiso 1983, 1994). Satellite infrared systems provide spatially coherent and continuous coverage during both day and night in cloud-free regions. Passive microwave satellite data have also been used to obtain surface temperature information (e.g., Shuman et al. 1994) but more research is needed to evaluate their accuracy and spatial consistency. In this study, infrared data from the Advanced Very High Resolution Radiometer (AVHRR) for the period 1982–98 were processed and combined with those of the THIR dataset for an extended time period coverage. Since clouds affect surface temperature and the retrieved data are those for cloud-free conditions only, the bias in the spatial distribution of the retrieved monthly temperatures will also be evaluated.

There have been two 20-yr time periods during the last century when the rate of increase in temperature was especially large, namely, 1925–44 and 1978–97 (Jones et al. 1999). It is fortuitous that the available satellite dataset covers basically the second period since it provides the opportunity to examine details associated with the high warming rate. Except for 1979 and 1992, when infrared data for the entire year were processed, the derived temperature data that are currently available are only for a winter (July) and a summer (January) month during the 20-yr time series. However, when combined with station data, the joint datasets provide a formidable tool and a means to gain useful insights into the specific warming phenomenon observed. The satellite data also provide a means to understand many processes related to the Antarctic climatology. For example, surface temperature affects the growth and decay of sea ice, the heat and salinity fluxes between the ice-covered ocean and the atmosphere, and the ice extent. Also, it affects the areal coverage of continental surface melt, evaporation and sublimation, accumulation rate, the stability of the ice sheet, and the variation in the circulation patterns in the atmosphere and the boundary layer.

2. Surface observations from Antarctic stations

The Antarctic station data can be classified into two types: manned station data and unmanned station data. Manned stations usually provide more accurate data than unmanned stations because in the former, the attention needed for the proper maintenance of instruments in polar environments is afforded and the sensors are calibrated on a regular basis. The unmanned stations, usually called the automatic weather stations, have been used to supplement the small number of manned stations and to obtain a better general overview of the continent (Stearns and Wendler 1988). Some of the datasets from unmanned stations were indeed found to be less consistent and have larger gaps in the data record than those of manned stations due to occasional malfunction of the sensors or the latter being buried by snow. Considering the large extent of the ice sheet and its largely varying elevation, the limited number of ice stations (some of which are clustered together in the same general areas) do not provide a good spatial coverage of the entire continent. In situ surface temperature information over sea ice around the continent is even more limited since most of the information is provided by temporary and moving platforms such as buoys and ships.

The lengths of surface air temperature records around Antarctica varied from one station to another because of differences in installation and operation times. There are stations with relatively long and reliable measurements such as the Faraday, South Pole, Halley, Vostok, and McMurdo stations. Other manned stations that were established a much later dates are the Neumayer, Mirny, Great Wall, Mizuko, and Belgrano II stations. Also, many automated stations were installed in the 1980s when technology allowed for self-contained systems in which measurements are obtained almost in real time through satellite links.

Typical long-term fluctuations in Antarctic surface air temperatures are provided by the Faraday and Vostok stations as shown in Fig. 1a. The plots show contrasting seasonal cycles between the two stations, with an average amplitude of about 12°C at Faraday, located at the western side of the Antarctic peninsula, and about 37°C at Vostok, which is located at the Antarctic plateau. Also, the summer and winter surface temperatures are much lower at Vostok than at Faraday. Furthermore, the peaks and dips are apparently not correlated since at Vostok, anomalously cold winters occurred in 1977 and 1983 while warm winters occurred in 1963, 1981, and 1987. On the other hand, at Faraday, anomalously cold winters occurred in 1959 and 1987 while warm winters occurred in 1962, 1971, 1985, and 1989. Similar mismatches are also apparent during summer periods. It is thus not surprising that when linear regression analysis was applied to these two datasets, with the seasonal cycles removed (i.e., by subtracting the multiyear monthly averages from each monthly average), the trends in temperature from 1958 to 1998 are both positive but differ by more than an order of magnitude (e.g., 0.066° and 0.004°C yr−1, respectively). Such discrepancies are indications that Antarctic trends can be very variable.

To obtain a better idea about spatial variability of trends, data from other Antarctic stations were examined. Such trends have been reported previously by various investigators (e.g., Raper et al. 1984; Sanson 1989;Jacka and Budd 1991; Weatherly et al. 1991; Jones 1995). This study updates previous reports and puts emphasis on station data with record lengths of about 45 yr. It is also required that the data gap for each year is less than 2 months. For the latter, the gaps were filled in by polynomial interpolation. The location of the stations and the trends after the seasonal cycle is removed are shown in Figs 2a and 2b, for approximately 45- and 20-yr record length, respectively. The ones with positive slopes are in coded rectangular symbols while those with negative trends are in similarly coded elliptical symbols. The numerical values of the trends and corresponding standard deviations (σ) are given in Table 1. Actual record lengths for each station are given in Table 1 as rl45 and rl20 and averages 41 and 18 yr, respectively. The trends are shown to vary considerably from station to station, and at times adjacent stations even have opposite signs.

The average of the 45-yr trends from the 21 stations is 0.012°C yr−1 with 17 having positive trends and only 4 having negative trends. On the other hand, the average of the 20-yr trends is −0.008°C yr−1 with 9 stations having positive trends and 12 stations having negative trends. Yearly averages were also utilized in the trend analysis and the results are given in parentheses in Table 1. The trend results from the monthly and yearly datasets provide very similar values, especially when data gaps are minimal. Except for Signy Island, the trend values varied considerably from the 20-yr record to the 45-yr record. Most of the previous studies provided positive trends but usually for the longer record. It is thus quite surprising that despite apparent increase in global temperatures during the last 20 years (e.g., Jones et al. 1999), the Antarctic region in general shows slight cooling during the period. Such cooling could partly explain the slight positive trend in sea ice extent observed during the 1979–96 period by Cavalieri et al. (1997).

This leads to a key question about trend analysis: How long must a historical record be before trend results become viable? The answer likely depends on parameter, measurement accuracy, and interanannual variability. To test the sensitivity of trends in surface temperature to the length of the record, data from Antarctic stations with long and reliable records were analyzed. Data from each station are evaluated to determine how the trend would vary as a function of the length of record, starting with 2, 3, and up to about 42 yr. Using Faraday and Vostok station data, such analysis showed that the trend fluctuates substantially up to a record length of about 10 yr, more moderately between 10 and 20 yr of data, and becomes nearly stable after 20 yr (Fig. 3a). This test was repeated using 1979–98 data and the results show basically the same pattern (Fig. 3b). Similar plots using data from six other stations (as listed) indicate that this dependence of trend on record length is generally true (Figs. 3c and 3d). The mean standard deviations of the trends during the first, second, and third decades for all eight stations are 0.250°, 0.019°, and 0.009°C yr−1, respectively. Thus, trend results are apparently meaningless for record lengths of less than 10 yr while the trend appears to stabilize in the second decade and is even more stable after two decades. Beyond two decades, the changes in trends are slow but can vary significantly from one decade to another as shown in Table 1.

3. Infrared data, cloud-masking, and retrieval techniques

The National Oceanic and Atmospheric Administration AVHRR aboard satellites is a cross track scanner operating at the following wavelengths: 0.58–0.68 _μ_m (channel 1), 0.73–1.1 _μ_m (channel 2), 3.5–3.9 _μ_m (channel 3), 10.3–11.3 _μ_m (channel 4), and 11.5–12.5 _μ_m (channel 5). Channels 1 and 2 are in the visible wavelengths and used for albedo estimates and cloud masking, channels 4 and 5 are thermal infrared channels used for surface temperature estimates and cloud masking, while channel 3 is an intermediate channel used mainly for cloud masking. The resolution of the basic data, which is sometimes referred to as local area coverage (LAC) or high resolution picture transmission, is about 1 km2 at nadir, while at other scan angles the resolution gets degraded. Because of more complete global coverage than LAC, we use global area coverage (GAC) data in this study. GAC data have an effective resolution of 5 km by 3 km and are constructed as follows: Along the 2048-pixel scan line, the radiances from four successive spots are averaged while the next spot is skipped. Also, only every third scan line is recorded while skipping two scan lines. We also use data from the 11.5-_μ_m channel of THIR that has a resolution of 6.7 km × 6.7 km. The THIR data, however, have been mapped into the polar stereographic format (used for SSMI ice data) with a grid size of 25 km × 25 km, while the GAC data have been mapped into the same format but at a grid size of 6.25 km × 6.25 km. When the two sets of data are combined for interannual analysis, the resolution of the GAC is degraded to that of the THIR.

Some techniques for retrieving surface temperatures over snow and ice from AVHRR and THIR have been reported (Comiso 1983; Key and Haefliger 1993; Comiso 1994; Steffen et al. 1993; Massom and Comiso 1994;Key et al. 1997). The accuracy in the retrieval of surface skin temperatures from satellite infrared data depends strongly on the success in cloud masking. The ability to discriminate cloud-free areas from cloud-covered areas over open ocean is usually facilitated by the large contrast in the albedo or emissivity of clouds and that of water. Clouds over sea ice and the ice sheets in the polar regions are more difficult to mask because of the presence of cold snow cover that substantially reduces the contrast in the albedo and emissivity of clouds and the surface (Yamanouchi et al. 1987; Allison et al. 1993).

With AVHRR data, cloud masking is usually done by a thresholding technique that utilizes one of the visible channels or the difference between channels 3 and 4 and/or between channels 4 and 5. Because of instrumental noise, a noise-reduction procedure (Simpson and Yhann 1994) is applied to the channel 3 data before they are utilized. Another technique, called daily differencing method and previously used for THIR data (Comiso 1994), takes the difference of daily orbital data and uses a threshold, based on the change in observed radiances due to the movement of clouds. The method found to be most effective in the polar environment is a combination of all these techniques with emphasis on the use of the daily differencing method using channel 4 data and the thresholding technique using the difference between channels 3 and 4 data. In the open ocean, an additional masking is applied with the aid of near-coincident ice concentration maps from passive microwave data. Areas in the open ocean with retrieved temperatures less than −4°C were taken out and assumed to be cloud contaminated. For consistency, the THIR dataset reported in Comiso (1994) was enhanced using this additional open ocean requirement that provided a better definition of the ice edge. For optimum effectiveness, slightly different masking procedures were applied separately to open ocean, sea ice, and ice sheet surfaces, each of which required different thresholds. Also, ascending (day) data were analyzed separately from descending (night) data. Masking for night data is more difficult than for day data because during darkness the visible channels cannot be used and the difference between channels 3 and 4 does not provide good cloud discrimination.

There are other reasons for processing night and day data separately. For example, it allows for flexibility in studying diurnal changes in the emissivity of cloud-free surfaces associated with surface melting during the day and refreezing of the same surface at night. Ancillary data on surface melt would be useful for this purpose. Secondly, the impact of some polar features (i.e., polynya or melt areas) can be better quantified with night data separated from day data. Furthermore, the correlation of temperature with elevation loses its meaning when the surface starts to melt during daytime but may not be the case during nighttime.

Typical cloud patterns in the Antarctic region, as observed in the visible during the summer and in the infrared during the winter, are shown in Figs. 4a and 4c, respectively. The areas in these images identified as cloud-covered areas are shown in white in Figs. 4b and 4d, respectively. The technique is reasonably effective in masking much of the cloud cover identified by inspection in channel 1 during the summer. The effect of temperature inversion can be considerable, as pointed out by Phillpot and Zillman (1970), especially in winter and at the high elevations of Antarctica. This is apparent in the 15 July 1992 AVHRR image (Fig. 4c) in which cloud tops are shown to be warmer than the surface. A special differentiation technique for masking out such clouds was developed and when applied to the data in Fig. 4c, the result is shown in Fig. 4d. The cloud-masking technique is generally effective but some residuals remain. The effects of such residuals were further examined by analyzing the differences between the monthly average and daily maps from the same month. The averages of the differences are about 2°C in the summer and 4°C in the winter over ice and 1°C in the summer and 2°C in the winter over ocean. Since these values include actual deviations between monthly and daily surface values, the actual cloud residuals must be even less than these averages.

To correct for atmospheric attenuation, the split-window algorithm (McClain 1981; Bernstein 1982; Walton et al. 1990) was implemented for the open ocean, while a slightly modified version was used for the sea ice and continental ice sheet regions. The latter was developed through the use of regression analysis that made use of Antarctic radiosonde data. The results from this procedure have been compared and show good consistency with those derived using the Antarctic parameters provided by Key et al. (1997). The AVHRR radiances are converted to surface (skin) temperatures using the same emissivities for ice and water used by Comiso (1994). Also, ice-covered and open-water surfaces are separated using ice concentration maps derived using the Bootstrap algorithm (Comiso 1995) over the sea ice region.

Using monthly data for both January and July from 1984 to 1998, the resulting monthly surface temperatures from AVHRR are shown to be consistent with corresponding monthly in situ air temperatures from Antarctic stations (Fig. 5a), the correlation coefficient being 0.98 while the standard deviation is 3.2°C. In this comparison, only in situ data that matches AVHRR cloud-free data are used in the monthly average. The true monthly average in situ data are compared with AVHRR data in Fig. 5b and regression analysis yielded a correlation coefficient of 0.96 and a standard deviation of 3.7°C. A plot similar to Fig. 5a but using the seasonal data in 1992 is shown in Fig. 5c and analysis yielded a correlation coefficient of 0.97 with a standard deviation of 3.1°C. If only the January data in Fig. 5a are used in the analysis, the correlation coefficient is 0.96 and the standard deviation is 2.5°C, while the corresponding values for the July data only are 0.96 and 3.8°C. It should be noted, however, that the observations from the Antarctic stations are not without errors and the sampling areas for satellite and station measurements are different. Considering that the seasonal and interannual fluctuations of sea ice and ice sheet temperatures are much larger than those of the ocean, where less than 1°C rms errors have been quoted (e.g., Bernstein 1982), these rms errors for ice are reasonably good. A similar comparative study done using THIR data by Comiso (1994) yielded approximately the same error. A comparison of THIR and AVHRR data during periods of overlap (1982–85) is shown in Fig. 5d. The scatterplot indicates good agreement of THIR and AVHRR data in almost all regions including the open ocean with correlation coefficient of 0.995 and standard deviation of 1.52°C. For a smooth transition from THIR to AVHRR data in the time series, the average of the two datasets were taken during the overlap period.

Comparisons of six station data with satellite temperature data over a seasonal cycle in 1992 are shown in Fig. 6 with location and elevations as indicated in the plots. The plots show that the seasonality as observed from the two datasets are generally similar. In some station locations, the AVHRR temperatures (dotted lines) are higher than the station values while in other stations, the opposite is true. The differences may be associated with differences in measurement area (station vs satellite), time resolution (satellite passes are not exactly at the same time as station measurement), and spatial variations in surface emissivity. Generally, however, the differences are within the 3°C standard deviation indicated earlier.

The monthly averages derived from the infrared data are not true monthly averages since they are just averages of surface values during cloud-free conditions. The magnitude of the associated error has been studied by taking the differences between the true monthly averages, using station data, and the monthly averages of cloud-free data (as identified by the AVHRR cloud mask) from the same stations. The results show that the cloud-free only monthly average is colder than the true monthly average by about 0.3°C with a standard deviation of about 0.6°C during summer and 0.5°C with a standard deviation of 1.5°C during the winter.

4. Spatial and seasonal variability

The color-coded monthly maps presented in Fig. 7 show the month-to-month variations in surface temperature and illustrate how the spatial distributions of temperatures develop over a seasonal cycle in the Antarctic region in 1992. In each of these maps, the averages of night and day data were used. As has been noted previously (Comiso 1994), the seasonal fluctuation of surface temperature over the Antarctic ice sheet is substantially larger than that of sea ice. The spatial patterns of the ice sheet temperature are also well defined, persistent, and strikingly similar to those of surface elevation maps (Drewry 1983) of the continent. While there is an apparent latitudinal dependence of temperature, the effect of elevation is shown to be more dominant since the lowest temperatures in the maps are located not at the South Pole but at the higher elevations in the Antarctic plateau.

Figure 8a shows plots of monthly averages during day and night, of ocean, sea ice, and ice sheet surfaces while Fig. 8b shows differences between day and night values. The difference between night and day temperatures is largest during summer and is gradually reduced to almost zero during winter, as expected. The ice sheet temperature ranges from −46° to −17°C compared to from −16° to −2°C over sea ice and from −1° to 1°C over open ocean south of 55°S latitude (see Fig. 7). The difference in surface temperature between day and night ranges from about 5°C in summer to about 0°C in winter. The differences are also about the same for four months in summer and for five months in winter. The coldest average temperature occurred in September while the warmest temperature occurred in January, indicating that the cooling period takes a much longer time than the warming period. Similar asymmetric behavior in the growth and decay characteristics of the Antarctic sea ice cover has been observed (e.g., Zwally et al. 1983; Gloersen et al. 1992).

To gain additional insight into the seasonal variability, the plots in Fig. 9a compare the averages over sea ice and the ice sheets separately with corresponding minimum values in the region using day (ascending) data. Substantial drop in the surface temperature minimum occurred from January to March in both sea ice and the ice sheet but the minimum was relatively constant from March to June. A further drop in temperature occurred after June with the coldest temperature being in August for both sea ice and ice sheet. A slight phase difference of about half a month is also apparent between the surface temperature distribution of sea ice and that of the ice sheet. During nighttime (Fig. 9b), the coldest temperature over sea ice also occurred in August, but the coldest over the ice sheet was in September. This retardation in the occurrence of the coldest time period is well known and due to the longwave outgoing radiation (Stearns et al. 1993; King and Turner 1997). The minimum temperature over the ice sheet also shows a drastic drop from February to March and a drastic increase from October to November. Analysis of the locations of the temperature minimum indicates that in general, the coldest area in the ice sheet is confined between 76° and 82°S latitude in the Antarctic plateau. The actual location of the coldest spot, however, covers the longitudinal range from 40° to 122°E with the most frequent location being between 43° and 78°E in winter. In the sea ice-covered regions, the low temperatures usually occur near the drainage basins, as identified by Parish and Bromwich (1987), and the ice shelves.

The shape of the seasonal distribution changes from V shape to U shape from warm to cold regions and therefore depends on location, elevation, and proximity to the ocean (King 1994). On a large scale, the seasonal temperature distribution of satellite-observed temperatures in the entire Antarctic Peninsula also show a generally V-shape distribution compared to other regions, because of a much briefer winter period. Much of the west Antarctic region, including the peninsula region, are more vulnerable to warming than in other regions because the annual mean temperatures are close to melting temperatures.

5. Interannual variability in surface temperatures

a. Winter and summer anomalies

The time sequence of July images from 1979 to 98 (Fig. 10) show larger year-to-year variability in both sea ice and ice sheet surface temperatures than the January images (Fig. 11). In the sea ice–covered areas of the Weddell Sea, some years (e.g., 1979, 1980, 1982, 1986, 1987, 1994, and 1995) tend to be colder (more green areas) than other years (e.g., 1981, 1984, 1989, 1996, and 1998). Year-to-year variability in the ice sheet temperatures is even more dramatic. At the Antarctic plateau, the coldest temperatures are represented by different shades of gray and are shown to have different shapes and extents from one year to another. In 1985, the surface temperatures at the Ross Ice Shelf and the Antarctic plateau are shown to be much colder than average. This phenomenon coincided with unusually cold temperatures recorded at the McMurdo and Vostok stations during the same period. Other unusually cold periods, especially in east Antarctica, occurred in 1979, 1985, 1986, and 1987, while unusually warm periods occurred in 1980, 1981, 1982, 1988, and 1995.

The July 1995 image stands out as showing the highest average temperature at the Antarctic plateau during the last decade although globally 1997 and 1998 were warmer years (Jones et al. 1999). Unusually high temperatures during this period are also observed at Mawson, Davis, Mirny, and Vostok stations (i.e., −9°, −8°, −9°, and −57°C, compared to mean values of −18°, −17°, −17°, and −67°C, respectively). It should be noted that 1980 and 1981 were also relatively very warm years in the Antarctic plateau.

The corresponding January images (Fig. 11) show large interannual variability in the spatial distribution of sea ice but relatively uniform temperatures within the pack. Changes in spatial patterns from one year to the next are associated with interannual variations in time of occurrences and location of melt and breakup. During summer, the sea ice surface temperatures are very close to those of open water and sometimes it is difficult to identify the location of the ice edge. However, subfreezing temperatures (e.g., oranges) appear to be still dominant in many parts of the inner region. In the continental ice sheet, the temperatures show much larger interannual variability. For example, surface temperatures in the east Antarctic region are significantly higher in 1983, 1984, 1986, 1988, 1990, and 1991 than other years. Also, the coldest temperature in the last two decades occurred in January 1995, which preceded one of the warmest in winter (July 1995) during the same period.

To gain additional insight into the large interannual variability in July, anomaly maps were generated from differences between each month and the 20-month average for the same month (Fig. 12). In each of these maps, the scope of warming and cooling in various regions is better represented. Also, unexpected changes over sea ice and the continent are better identified for each year. In addition, the warm events in 1980, 1981, and 1995 in east Antarctica become very apparent. Alternately, relatively cold events in 1979, 1983, 1985, 1986, and 1987 stand out in the same region. It should also be noted that three unusually warm years (1980, 1981, and 1982) were followed 2 yr later by 3 unusually cold yr (1985, 1986, and 1987). Since 1994, 1995, and 1996 appear to be unusually warm years, it will be interesting to see what the temperature will be like in 1999 through 2001.

SST has been shown to be negatively correlated with sea ice extent by White and Peterson (1996). The negative correlation of surface air temperatures in the Antarctic peninsula with sea ice extent in the Amundsen and Bellingshausen Seas has also been observed by Jacobs and Comiso (1997). In the anomaly maps, the outer limit of the white area is the farthest north that the ice cover has reached during the 20-yr period (as inferred from passive microwave data). It is apparent that in areas where substantial retreat in sea ice cover occurred (i.e., those with large white areas), the adjacent sea ice area is also anomalously warm (red or purple). Conversely, relatively cold sea ice regions (blue or green) are usually located in areas where the ice extent is most expansive. This phenomenon is coherent with anomaly maps of ice concentrations (Fig. 13) derived from passive microwave data for the same period. In particular, the reds and purples in the images, which represent areas of reduced ice concentration, are located in abnormally warm areas in the temperature anomaly maps (Fig. 12). The locations of retreats in sea ice are also qualitatively consistent with the statistical analysis of the yearly frequency of ice in each pixel (Parkinson 1998). However, the link between ice extent and temperature is not always clear-cut, as shown in the 1987 image, especially in the eastern Weddell Sea (Fig. 12). While this appears to be more of an exception, the lack of negative correlation has also been observed by Raper et al. (1984) who attributes it to environmental factors, like wind and ocean currents.

Alternating warm and cold as well as high and low ice concentration anomaly patterns around the continent are apparent from the winter data in the sea ice region (Figs. 12 and 13). By inspection, a year-to-year circumpolar motion of the anomalies is suggested but is sometimes not so apparent because of large interannual fluctuations in surface temperatures and changes in the size and shape of the anomaly patterns. The overall effect is coherent with that postulated as the effect of the Antarctic circumpolar wave (ACW) as described by White and Peterson (1996). However, the patterns indicate that the Antarctic sea ice cover is predominantly mode-3 wave (e.g., 1980, 1981, 1987, and 1989) as opposed to mode-2 wave (e.g., 1992) proposed by White and Peterson (1996). The patterns also clearly indicate that the ACW affects not just the ice edges but also the inner regions of the pack. In some cases, the identification of the wave mode is difficult because of the general warming patterns (e.g., 1990s) that obscure the signal. A mode-2 wave would be more consistent with the observed period of the El Niño–Southern Oscillation, which has been postulated as having links with the Antarctic anomalies (Smith and Stearns 1993; Peterson and White 1998).

Anomaly maps in the summer (Fig. 14) show somewhat different patterns than those in winter. Relatively cold summers in the continent (e.g., 1980, 1981, 1982) are sometimes followed by relatively warm winters. Conversely, relatively warm summers (e.g., 1986, 1987), are followed by relatively cold winters. This is consistent with the modulated ice extent time series distribution in which low extents in the summer are followed by high extents in the subsequent winter (Zwally et al. 1983; Comiso and Gordon 1998) and vice versa. During the last two decades, 1988 appears to be the only year when warming in summer and winter occurred simultaneously, while 1995 appears to be an exceptionally cold summer.

b. Trend analysis

To quantify yearly variability, average temperatures in January and July were calculated separately for open ocean (>55°S), sea ice, and ice sheet regions (Fig. 15). The year-to-year changes over the ice-free ocean are difficult to evaluate since the expected fluctuation of about 1°C approaches the accuracy of the data. It is, however, encouraging to find year-to-year consistency in the open ocean values since this suggests effectiveness of the retrieval technique, especially that for the cloud masking. Interannual temperature fluctuations over sea ice and the ice sheet are also small during the summer but departures from average values are more significant. In winter, the fluctuations over the open ocean are again small but those of sea ice and ice sheets are considerably higher. The largest interannual fluctuations occur in the Antarctic Plateau where the range in temperatures is as high as 14°C.

Regression analysis of the January 1979–January 1998 dataset yielded a trend of 0.054 ± 0.012°C yr−1 for open ocean, 0.196 ± 0.032°C yr−1 for sea ice, and −0.050 ± 0.043°C yr−1 for the ice sheet. The corresponding values for the July data are 0.035 ± 0.011°C yr−1 for open ocean, 0.078 ± 0.034°C yr−1 for sea ice, and −0.034 ± 0.130°C yr−1 for the ice sheet. The trend analysis of the January and July data combined, with seasonality subtracted by taking the difference of each month and the average for the same month over the 20-yr dataset yielded 0.044 ± 0.008°C yr−1 for open ocean, 0.137 ± 0.024°C yr−1 for sea ice, and −0.042 ± 0.067°C yr−1 for the ice sheet.

The slight cooling detected for the ice sheet is relatively higher but of the same sign and within errors when compared with averages from station data (−0.008) for the same period. The satellite values have been compared directly with some station data (e.g., Fig. 16) and the results indicate general agreement. Again, the cooling is compatible with the trend in the sea ice extent as observed by Cavalieri et al. (1997). It should be noted, however, that the standard deviation of the trend is quite large and the 95% confidence level for the trend ranges from −0.177 to 0.094. The trends in temperature over sea ice are positive and quite high, however, especially in the summer, but these trends have larger errors due to persistent cloud cover and to physical changes (e.g., ice concentration and thin ice fraction) in the sea ice cover.

An entire year of infrared data was processed in 1992, as well as in 1979, to better understand the seasonal characteristics of surface temperature in the region. Annual averages generated from these 2 yr (Figs. 17a and 17b) are shown to be generally in good agreement, especially within the continent. It is encouraging that distinct features in the continent observed in 1979 also show up in the 1992 image and that the temperature distributions within the ice pack are similar. The difference map (Fig. 17c), however, indicates significant differences both in the continental region as well as the sea ice–covered region. Interpretation of the changes is difficult because of large interannual fluctuations (see Figs. 10 and 11) but overall, the general warming in some areas and cooling in other areas is consistent with that observed in station data. The results of a trend analysis on a pixel by pixel basis (Fig. 17d), using data in Figs. 10 and 11, also shows patterns very similar to those shown in Fig. 17c, especially in the continent.

6. Discussion and conclusions

Among the key results of this study are the following:(a) satellite infrared data provide spatially detailed maps of surface temperature in the Antarctic region with an accuracy of 3°C; (b) a predominance of positive trends in surface air temperature is observed in station data when a 45-yr record length is used, while a slightly negative trend is observed from both station and satellite data when the 20-yr record length is used; (c) satellite data reveal anomalously high temperatures in large areas of the Antarctic plateau (e.g., 1980, 1981, and 1995) and anomalously low temperatures in the ice shelves (e.g., Ross Ice Shelf in 1985); and (d) alternating warm and cold anomalies in the sea ice region around the continent are observed and shown to be correlated with the limits in the extent (and concentration) of the sea ice cover and the expected effect of the Antarctic circumpolar wave (ACW).

The surface temperature maps derived from infrared data from 1979 through 1998 are shown to be spatially and temporally coherent with estimated rms error of about 3°C. The actual accuracy of the satellite data is likely better since the station temperatures may have some errors associated with occasional measurement problems in such an extremely adverse environment. The accuracy of the satellite data may be further improved with better cloud-masking techniques and more spatially extensive coverage of radiosonde data to better correct for atmospheric effects. However, these temperature data already provide substantial improvement to what is currently available for process studies and are useful for comparison with those derived by global circulation models.

Using station data with long record lengths, a sensitivity study indicates that the trend tends to stabilize after a record length of two decades is available. The current historical satellite dataset is thus long enough to make the trend analysis meaningful. The 20-yr trend from the satellite data over the ice sheet was estimated to be −0.042 ± 0.067°C yr−1 and is consistent with −0.008 ± 0.025°C yr−1 derived from station data for the same period. Analysis of the 45-yr trend from station data, however, yielded a warming trend of about 0.0124 ± 0.0076°C yr−1, which is consistent with previous studies (e.g., Raper et al. 1984; Sanson 1989; Jacka and Budd 1991; Weatherly et al. 1991; Jones 1995).

The slight cooling of the entire ice sheet observed in both in situ and satellite records during the last 20 yr is intriguing since during the same time period a general warming is being observed globally (Jones et al. 1999). A general warming in the pheriphery of the Antarctic Peninsula accompanied by a declining sea ice cover in the adjacent seas have also been reported (King 1994; Jacobs and Comiso 1997). However, the slight cooling detected in the entire Antarctic region is compatible with a slightly positive trend in the sea ice extent that has been observed from passive microwave data (Cavalieri et al. 1997). Considering that contradictory results in the Antarctic ice extent trend have been reported (e.g., De la Mare 1997; Bjorgo et al. 1997) a longer time series would be most desirable to gain more insights into these trend results.

The trend study is also complicated by the presence of alternating warm and cold anomalies in the sea ice region around the continent, especially during winter months. The data suggest an eastward-propagating temperature regimes consistent with effects of the ACW reported previously by White and Peterson (1996). However, in this study, both satellite temperature and ice concentration anomaly data show a predominance of mode-3 wave instead of the mode-2 wave reported earlier. They also clearly indicate that the ACW propagates into the ice pack and is not just an ice edge phenomenon.

The satellite data provide the only means to evaluate large-scale patterns of warming and cooling because of the paucity of Antarctic stations. The extent of the warm event in 1995 and the cold event in 1985 would be difficult to assess otherwise. Although the Vostok and McMurdo stations provide confirming results, they are only point measurements, whereas the satellite data provide the spatially detailed patterns of anomalously warm and cold areas in the Antarctic plateau where yearly shifts in the general locations of the coldest spots are evident. Anomalies are also shown to he specially large at the Ronne and Ross ice shelves.

This study shows that a combined satellite infrared and in situ data provide a powerful means for assessing large-scale spatial, seasonal, and interannual variability of surface temperatures over the ice sheet, sea ice, and the adjacent oceans. Although an improvement in the cloud-masking technique would help, the uncertainties in the retrieved temperatures are relatively small, especially when compared to the large fluctuations in surface temperatures observed over the ice sheet and sea ice regions. Future work will include the processing of the entire AVHRR dataset for a continuous temperature record from 1979 to the present.

Acknowledgments

I am deeply grateful to Larry Stocks of Caelum, Inc., for excellent programming support in the processing and analysis of AVHRR data, to Will Manning of Raytheon STX for help in the analysis of Antarctic station data, and to Stan Jacobs of Lamont Earth Observatory for detailed comments and suggestions. The detailed reviews and excellent suggestions by Dr. K. Steffen and an anonymous reviewer are also greatly appreciated. Furthermore, I am grateful to Drs. R. H. Thomas, H. J. Zwally, R. Bindschadler, and P. Wadhams for useful comments and encouragements. The AVHRR GAC data used in this study were put together by Dr. C. Tucker and the Pathfinder project at the Goddard Space Flight Center. The Antarctic station monthly data have been quality checked and put together by Dr. T. H. Jacka of the University of Tasmania, P. D. Jones of the University of East Anglia, and by L. M. Keller, G. A. Weidner, C. R. Stearns, and M. T. Whittaker of the University of Wisconsin. The research was supported by the Cryosphere Program and the Earth Science Enterprise/EOS project at NASA.

REFERENCES

Fig. 1.

Fig. 1.

Fig. 1.

(a) Surface air temperature records in Faraday and Vostok stations, (b) surface air temperatures at Vostok with interannual monthly averages subtracted from each month, and (c) surface air temperatures at Faraday station with interannual monthly averages subtracted from each month. Trend lines derived from linear regression are drawn in (b) and (c).

Citation: Journal of Climate 13, 10; 10.1175/1520-0442(2000)013<1674:VATIAS>2.0.CO;2

Fig. 2.

Fig. 2.

Fig. 2.

Location map of the Antarctic stations with coded symbols to indicate trend values (rectangular for positive trends and elliptical for negative trends) in each of the stations. The trends for each station are derived using linear regression with the interannual monthly averages subtracted from each month.

Citation: Journal of Climate 13, 10; 10.1175/1520-0442(2000)013<1674:VATIAS>2.0.CO;2

Fig. 3.

Fig. 3.

Fig. 3.

Trends in temperatures as a function of record length (2 yr and greater) in (a) Faraday (bold line) and Vostok (dashed line) stations, 1955–98; (b) Faraday and Vostok, 1979–98; (c) six other stations, 1955–98; (d) six other stations, 1979–98.

Citation: Journal of Climate 13, 10; 10.1175/1520-0442(2000)013<1674:VATIAS>2.0.CO;2

Fig. 4.

Fig. 4.

Fig. 4.

AVHRR GAC data from a single orbital pass over the Antarctic showing (a) channel 1 (albedo) on 10 Jan 1992; (b) channel 4 (thermal infrared) on 10 Jan 1992 with cloud masking; (c) channel 4 on 15 Jul 1992; and (d) channel 4 on 15 Jul 1992 with cloud masking. The color scale on the left is for the 10 Jan data while the color scale on the right is for the 15 Jul data.

Citation: Journal of Climate 13, 10; 10.1175/1520-0442(2000)013<1674:VATIAS>2.0.CO;2

Fig. 5.

Fig. 5.

Fig. 5.

Comparison of in situ surface ice temperatures with those derived from AVHRR data using (a) matched monthly Jan and Jul data from 1984 to 1998; (b) unmatched data using true monthly in situ data for Jan and Jul from 1984 to 1998; and (c) matched monthly data from Jan to Dec 1998. (d) Comparison of monthly AVHRR data with monthly THIR data during overlap period.

Citation: Journal of Climate 13, 10; 10.1175/1520-0442(2000)013<1674:VATIAS>2.0.CO;2

Fig. 8.

Fig. 8.

Fig. 8.

(a) Plots of average surface temperatures over open ocean, sea ice, and the ice sheet for night and day data from Jan to Dec 1992 as derived from AVHRR observations. (b) Plots of the differences in surface temperatures between night and day over ocean, sea ice, and the ice sheet from Jan to Dec 1992.

Citation: Journal of Climate 13, 10; 10.1175/1520-0442(2000)013<1674:VATIAS>2.0.CO;2

Fig. 9.

Fig. 9.

Fig. 9.

(a) Plots of average and coldest surface temperatures over sea ice and the ice sheet in the Antarctic for each month in 1992 during daytime and (b) plots of average and coldest surface temperatures over sea ice and the ice sheet in the Antarctic for each month in 1992 during nighttime.

Citation: Journal of Climate 13, 10; 10.1175/1520-0442(2000)013<1674:VATIAS>2.0.CO;2

Fig. 15.

Fig. 15.

Fig. 15.

(a) Interannual variability of surface temperatures from Jan 1979 to Jan 1998 over open ocean, sea ice, and ice sheet, (b) interannual variability of surface temperatures from Jul 1979 to Jul 1998 over open ocean, sea ice, and ice sheet. Trend lines derived using linear regression are shown in (a) and (b).

Citation: Journal of Climate 13, 10; 10.1175/1520-0442(2000)013<1674:VATIAS>2.0.CO;2

Fig. 17.

Fig. 17.

Fig. 17.

(a) Color-coded mean annual temperatures derived from monthly averages from Jan to Dec 1992, (b) mean annual temperatures derived from monthly averages from Jan to Dec 1979, (c) the difference between the mean temperatures in 1992 and 1979, and (d) trend results for each pixel using the Jan and Jul monthly average data from 1979 to 1998.

Citation: Journal of Climate 13, 10; 10.1175/1520-0442(2000)013<1674:VATIAS>2.0.CO;2

Table 1.

Temperature trends and standard deviations in Antarctic stations.

Table 1.

Table 1.