1. Introduction
The Antarctic coasts and shelves undergo snow melting during the summer months of December and January. There is virtually no melting at other times of the year except in the Antarctic Peninsula. Surface melting is not known to significantly influence the mass balance of the Antarctic ice sheet and shelves, by contrast to Greenland. However, surface air temperatures have been reported to be on the rise at many stations in Antarctica, though mainly in the Bellinghausen–Antarctic Peninsula area (Raper et al. 1984; King 1994; Jones 1995; Jacobs and Comiso 1997; Skvarca et al. 1998). Elsewhere, cooling has been reported (Comiso 2000; Doran et al. 2002). As temperature changes, the length of the summer melting period may change and thus contribute to characterize summer climate change at Antarctic coasts and shelves.
More than 20 yr of spaceborne microwave radiometer observations of the surface of the earth are now available from the Scanning Multichannel Microwave Radiometer (SMMR) and the Special Sensor Microwave Imager (SSM/I) sensors (Maslanik and Stroeve 2000). Over this period, the polar-orbiting satellites have provided almost full spatial and daily or semidaily coverage of the Antarctic, except for a small cap south of 87°S for the SSM/I sensor and 84°S for SMMR (more information available online at http://nsidc.org/NASA/GUIDE/docs/dataset_documents/smmr_pathfinder_tbs.html#7), a region of no interest for the surface melting studies. Melting, or even moistening of the surface snow grains, can have a significant impact on the microwave emissivity of the surface (Mote et al. 1993; Zwally and Fiegles 1994). As a consequence, the SMMR and SSM/I data provide signals sensitive to changes in the surface energy balance of Antarctica at temperatures close to melting over the last 20 yr. For instance, a systematic increase in the duration of the summer melt season over 1979–91 in the Antarctic Peninsula (Ridley 1993) and a small and barely significant decline over 1978–87 on the Ronne and Ross ice shelves (Zwally and Fiegles 1994, hereinafter ZF94) have been identified. Although longer-term warming trends have been reported, direct surface temperature retrieval from IR satellite data suggest that the Antarctic continent has been cooling (January temperature) over the last 20 yr (Comiso 2000), a trend that is compatible with the sea ice temporal evolution and screen temperature data. Our objective, here, is to detect and quantify temporal and spatial variations of the melting areas over the last 20 yr.
The satellite data and ground observations used to derive and verify the melting time series are presented in section 2. Data processing to extract melting occurences is described in section 3. The results, including tables, maps and plots of interannual variability, and trends for the Antarctic summer period, are presented and discussed in section 4. A general conclusion summarizes the results and expands the perspectives in section 5.
2. The data
a. Remote sensing data
For the period from April 1979 to July 1987, the data of SMMR on the Nimbus-7 satellite have been used. Data from SSM/I on three satellites of the Defense Meteorological Satellite Program (DMSP), F-8 (August 1987–December 1991), F-11 (January 1992–June 1995), and F-13 (July 1995–March 1999), complete an almost continuous 20-yr series. These data are distributed by the National Snow and Ice Data Center (NSIDC, Boulder, Colorado; Maslanik and Stroeve 2000). Different instruments on different platforms must be cross-calibrated. The SSM/I F-8 data provide a baseline against which other data are adjusted using calibration coefficients from Jezek et al. (1991) for SMMR, Abdalati and Steffen (1995) for SSM/I F-11, and Colton and Poe (1999) for SSM/I F-13. In any case, the melt detection algorithm, as described in section 3, is independent of sensor calibration.
The SMMR and SSM/I instruments provide measurements of the energy emitted by the earth's surface at several frequencies and at both vertical (V) and horizontal (H) polarizations [18SMMR or 19.35SSM/I (V, H), 22.2 (V), 37.0 (V, H), and 85.5 GHz (V, H); Maslanik and Stroeve 2000]. Previous work on snow/icemelt detection (ZF94; Abdalati and Steffen 1997) suggests focusing on the 19- and 37-GHz frequencies. The 22-Ghz channel is used only for water vapor and the 85-GHz frequency is too influenced by water vapor and clouds (scattering) in the atmosphere (Mätzler 2000). The energy emitted by the snow and received by the spaceborne instrument comes from different layers of snow. The depth seen by the instrument (penetration depth; Steffen et al. 1993) depends on the frequency of the signal and on the water content. The lower the frequency, the more responsive it is to melt onset. For melting snow, the penetration depth is strongly reduced. For a given frequency, small amounts of liquid water in snow cause a larger increase of brightness temperature at horizontal than at vertical polarization (ZF94). The depolarization effect is due to a change in dielectric properties at the air–snow interface when snow is wet (Steffen et al. 1993; Abdalati and Steffen 1997). Differential polarization emissivity due to melting is further discussed in section 3.
Preprocessed data distributed by NSIDC are available on a rectangular grid on a polar stereographic projection covering most of Antarctica. Each pixel is a 25 km × 25 km square. All observations made within a day and within a given grid square are accumulated to provide a mean daily sample.
b. In situ data
There is no available direct observation of surface melting in Antarctica. Melting can occur only if the surface temperature is at 0°C, but very few meteorological stations in Antarctica, and apparently none in coastal Antarctica, report surface (skin) temperature. Temperature closest to the surface is generally reported at 2 or 3 m. If the wind is not strong enough to efficiently mix the air in the lowest few meters, the temperature difference between the surface and 2–3 m in the atmosphere can be significant. In particular, in case of surface inversion, the temperature recorded at 2–3 m may be positive with no melting at the surface. Alternately, melt may occur with negative air temperature when the radiation balance is positive. Using atmospheric temperature to detect surface melting thus bears some uncertainty that must be kept in mind, even if surface air mixing seems to be effective most of the time on the coasts.
Stations have been selected to, as much as possible, verify satellite-derived melting in widely different regions. When several stations are available in the same area (e.g., on the Ross ice shelf), those farthest from rocks and having the most homogeneous environment on scales of a satellite grid square have been preferred. Data from both staffed and automatic weather stations [(MWS) and (AWS)] are available. Data from AWS have been downloaded from ftp://ice.ssec.wisc.edu and MWS data have been retrieved from the European Centre for Medium-Range Weather Forecasts (ECMWF) observation archive. The frequency of reports varies with stations. To compare with satellite data, we only select the maximum recorded daily temperature, which is less prone to the inversion effect and corresponds to the time when melting occurs. Table 1 list the geographic locations of all stations used.
3. Processing of the microwave data
a. Data selection process
Electric power supply limitations of Nimbus-7 have only allowed operation of SMMR every other day (until July 1987). In order to take into account the missing days, ZF94 twice counted each occurrence of melting. For convenience, we simply linearly interpolate brightness temperatures over missing days. Statistically, there should be no significant difference between the two methods. In addition, several gaps are found in the series (usually due to instrument failures). Most gaps last only 1 or 2 days. For gaps of duration less than 3 days, the missing brightness temperatures are linearly interpolated. For gaps equal or longer than 3 days, linear interpolation may induce significant errors. Years with gaps longer than a month in a row, for which summer is badly undersampled, are simply ignored. As a consequence, the 1981/82 and 1987/88 summers are not taken into account in the present study. For shorter gaps, in order to avoid spurious interannual trends, the missing days of a particular year are removed from all the other years in the 20-yr series. For instance, in 1988, the period from 24 to 28 December is unavailable everywhere at the coast of Antarctica, except in the region of the Ross ice shelf. The data for 24–28 December are thus dropped for all years, everywhere except in the Ross region. This reduction of the time series, repeated for all gaps, results in December being particularly affected (only 14 days in most regions). Sampling reduction results in underestimating the frequency of summer melting events. Sensitivity experiments are described in section 4 to both tentatively correct this underestimation and test the robustness of the interannual variability and trends estimated from the reduced series.
All grid squares are sampled at least once a day, but the density of satellite tracks at the surface increases with latitude. For melting events lasting significantly less than a day, for example, transient tenuous moistening of surface snow grains near midday, the probability of missing the event thus increases when latitude decreases. Also, measurement failures inducing false reports [e.g., meltinglike peaks of brightness temperature (Tb) in full winter] occur randomly. Such events last generally one day only and can be avoided if melting events recorded over less than two successive days are ignored.
For 19 GHz, horizontal polarization, the emissivity of bare soils (approximately 0.9) or ocean (0.6) is different from the emissivity of dry snow or ice (0.8) (Mätzler 1987). Melt signals in a grid square potentially containing significant rock or ocean surfaces are thus not reliable. A land–sea mask provided by NSIDC (Maslanik and Stroeve 2000) is used to avoid ocean pixels. Because most of the significant rock outcrops are associated with boulders and mountains peaking higher than 1500 m of altitude, where melting is unlikely, a topographic map is used to eliminate surfaces higher than 1500 m.
b. The algorithm
The 19-GHz horizontal polarization channel is chosen to detect the melt onsets. Melt induces large increases of brightness temperature. An annually and regionally varying threshold is thus calculated and all values of brightness temperature above the annual mean plus this threshold are associated with melting. This threshold is proportional to the standard deviation of the signal, thus taking into account the spatial variability of its amplitude.
The annual mean 19-GHz H brightness temperature is calculated for each year from 1 April to 31 March in each grid square for the 20-yr series. By calculating an annual mean, we take into account interannual variations of the signal. In some regions like on the Larsen ice shelf, melting lasts as much as four months (November–February). In such cases, means are badly biased toward values typical of melting and do not reflect average values of unaffected surfaces. Strong melt signals are thus simply filtered out by eliminating values more than 30 K above the mean (the 30-K threshold is derived from the ZF94 study as discussed below). The process is recursive: the first time, the mean is calculated using all the values (without any filter); then strong melt signals are filtered out and the calculation is repeated twice. Then, for each year and each grid square, the standard deviation σ is computed using the same filter. The threshold T above which a brightness temperature value is considered to be a melt signal is the mean (M) plus N standard deviations (σ): T = M + Nσ.
Although M, σ, and, thus, T vary with year and grid square, the value of the constant N does not. Meteorological observations are used to select and adjust N, so that, as much as possible, all and only real melting events are detected. The N is chosen so that most melt events correspond to daily maximum temperatures above −5°C as reported by weather stations at test sites listed in Table 1. For N = 3, the percentage of temperatures above −5°C for all the available stations is greater or equal to 98%. This percentage remains high (85%) for temperatures above 0°C. Figure 1 illustrates how selective the N = 3 threshold is at four stations only, but all the available stations show similar results.
4. Spatial and temporal variability of surface snowmelting
a. Processing of the melting signal
Removing the randomly missing data periods of one particular year from all other years (the reduced series in section 3) avoids biasing the interannual variability and trends of the melt series. In the following, the results will be referred to as those of the “reduced” calculations. The reduced calculations do not apply to the full summer. A number of other calculations were made to ensure that the unbiased trends obtained from the reduced series are robust. For instance, in order to avoid the problem of the many missing days of December, calculations restricted to the second part of summer (January–March) were done, which broadly confirm the calculations obtained with the reduced series. Also, complete series were constructed by filling gaps of a particular year with the brightness temperatures averaged over the corresponding days of the other years. The results, hereinafter referred to as those from “filled-in” calculations, will be presented (Tables 4–6) and discussed along with, and as a sensitivity test with respect to, those of the reduced calculations. In fact, when a cumulative result is presented (e.g., the mean total number of melting days in summer, Fig. 2), the filled-in option, rather than the necessarily underestimated reduced option, is preferred. When variability and trends are presented (e.g., Figs. 5,6), the reduced calculations are preferred.
Because the interannual variability of melting is spatially coherent over large regions (Figs. 3–4), we select seven areas over which synthetic calculations are performed (Fig. 2). These areas coincide with those selected by ZF94, so that our results can extend theirs. Those areas are designed as follows: peninsula, Filchner-Ronne, Dronning Maud Land (DML), Amery, Wilkes Land, Ross and Marie Byrd Land (MBL).
Figure 2 shows the mean annual number of melting days over the full 18-yr period (20 yr minus the summers 1981/82 and 1987/88; see section 3). Maps of spatial extent and duration anomalies of surface melting, are given in Figs. 3 and 4 for four different areas and for each summer period between 1979/80 and 1998/99 (with two missing summers; see section 3a). The regional dependence of temperature, and thus of surface melting can be illustrated by the very different duration of the melting period of the peninsula region (50 days on average) versus the rest of the coasts and shelves (from 5 to 20 days; Table 6).
Three different synthetic parameters of interest are discussed:
The cumulative melting surface (CMS; day km2) is the annual sum of the pixel days where melting occurs, multiplied by the pixel area (25 × 25 km2).
The maximum melting surface (MMS; km2) is the surface over which melting is detected at least once during a summer.
The mean melt duration (MMD; day) yields information about the duration of the melting period and is calculated by dividing the CMS by the MMS of the summer.
b. Melt variability and trends
Depending on the zone, the information given by the melt indices is not the same. For instance, the CMS of the peninsula is mainly dependent on the duration of melting because the melting surface variability is low. On the other hand, the large ice shelves undergo very short melting periods, but their melting surface is highly variable.
ZF94 report on Antarctic melt over the period 1978–87. We use a more sensitive and accurate detection algorithm (see section 3) and various alternate data processing to show that the means, variability, and trends (maxima and minima, order of magnitude, and sign of the trends; Tables 4–6) are at least qualitatively robust. We thus extend ZF94 results not only in time, but also in significance.
1) Interannual variability
Interannual variability for CMS in ZF94 (their Figs. 6 and 7) and in our calculations (our Figs. 5 and 6) are very similar over the common time period covered. The first two summers in common (1979/80 and 1980/81) behave similarly in all seven zones, showing a clear melt decrease. Then, as pointed out by ZF94, a 2-yr oscillation is found for the summers 1982/83, 1984/85, and 1986/87. This oscillation is seen in all zones and in the whole of Antarctica, except for Filchner and Wilkes. This is also in agreement with ZF94.
The next decade undergoes the strongest summer melting for all the zones. It takes place in 1992/93 in the peninsula and Marie Byrd Land zones and is not seen anywhere else (it even coincides with a very weak melting period elsewhere except for the Wilkes zone). On the contrary, in the DML, Ross, Filchner, Amery, and Wilkes zones, it takes place one year earlier, in 1991/92.
Principal components analysis (PCA) and calculation of empirical orthogonal functions (EOF) are common data processing methods to extract modes of variability and spatial patterns of coherency from spatially distributed time series. Unfortunately, such methods are not appropriate for melt time series. Because melting is a threshold process, it is not normally distributed, and, for instance, in some pixels the time series contain zeros for all but a few years. For the same reason, complex EOF analysis (e.g., Yuan and Martinson 2000) to detect evolutionary patterns of correlation, for example, traveling waves, is not applicable. As a first approach to objective analysis of the interannual variability in the melt indices, we simply use spectral analysis on the bulk indices. Common periodic variability in the different zones should sign in with similar periods. Also, a circumpolar traveling wave, for example, melt variability related to the Antarctic circumpolar wave (Gloersen and White 2001), should sign in with a phase evolving broadly linearly with time around Antarctica. Because the series are short, we use the multitaper method (MTM; Thomson 1982), which is sensitive and provides estimates of the statistical significance of detected frequencies.
The results are displayed in Table 2. Periods of 2–3 yr are mainly found, consistent with visual inspection of the series. The mean Antarctic MMD index has a 2.5-yr period. Longer periods (4–5 yr) are also found. However, beyond this, no clear spatially coherent picture emerges from Table 2. In particular, although the phases vary from region to region, they are not consistent with a circumpolar traveling wave.
Dynamical connections between the Tropics and the high southern latitudes at the El Niño–Southern Oscillation (ENSO) pace have been found in the Pacific sector (e.g., Trenberth and Caron 2000). Correlation of Antarctic climate variability with the Southern Oscillation index (SOI) have also been suggested (e.g., Yuan and Martinson 2000; Bromwich et al. 2000). Negative correlations between melt indices and the summer-mean SOI are found above the 99% significance level in only two regions (Amery and Ross) and for the Antarctic as a whole. However, because we looked for correlations in eight regions and for three indices, we significantly increase the likelihood of finding merely random correlations and thus decrease the statistical significance of the correlations. In particular, the correlations are not robust with respect to the various melt indices (i.e., in most cases, a significant correlation is found for one index only). In addition, over 1980–99, significant trends are seen in the melt indices (section 4c) and in the SOI, possibly affecting the significance of correlations in terms of year-to-year variability. However, the correlations are weakly modified, and in fact, slightly improved, when 20-yr trends are removed. Our melt results thus appear to hint at a possible implication of the SOI in Antarctic climate variability.
In fact, the striking 2-yr oscillation seen around 1985 in several regions and melt indices is not a characteristic of the SOI, but rather bears some resemblance to the dominant (annular) mode of oscillation of the Antarctic atmospheric pressure, often simply referred to as the Antarctic Oscillation and quantified by the Antarctic Oscillation index (AOI; Gong and Wang 1999). Visual comparison of the melt series with the AOI suggests other similarities in the 1980–99 period, and this is confirmed by correlations significant at the 99% level in all regions (Table 3). The correlations are stronger and more convincing (e.g., more robust with respect to melt index) than for the SOI. Interestingly, correlations between a 4-month running mean of the AOI and surface melt indices were blindly searched for with various phase leads and lags, but significant correlations showed only for phases near zero, that is with the AOI averaged over the four summer months in common with melt calculations. (October–January). Therefore, there appears to be a relation between summer melt and the distribution of air mass between the mid- and high latitudes in the Southern Hemisphere in summer. A lower correlation in the peninsula than elsewhere possibly reflects that the Antarctic Oscillation has less amplitude at lower latitude. The correlations are systematically negative, and in some cases very high, indicating that higher pressure at the Antarctic periphery coincides with less melting. The coastal weather patterns (temperature, surface energy balance, and thus, e.g., cloudiness) that are associated with pressure higher than normal and that can affect summer melting thus need to be investigated. As for the SOI, correlations after removing long-term trends were calculated, this time with consequences in some regions. On average over Antarctica, though, much of the AOI–melt correlation is reflected on a year-to-year basis.
2) Trends
For each index, in each zone, we determine a 20-yr trend by calculating a linear regression (slope and standard deviation of the slope) across the 18 available years. We estimate the significance of the results from a t test (Tables 4–6). Trends with a confidence level below 85% are not considered to be significantly different from 0. We draw the attention of the reader to the fact that the trends are affected by missing data (especially in December). However, trends computed from filled-in rather than reduced series (section 4) are qualitatively similar.
From a general point of view, the MMS, the MMD, and, thus, the CMS have all decreased over Antarctica. For the Antarctic CMS, the decrease amounts to approximately −1.8% ± 1% yr−1 (Fig. 5). This confirms the work of ZF94 who found −2.4% yr−1 for the first 9 yr of the period. The MMS trend of the whole Antarctica is also negative but below the significance threshold. The MMD shows no trend at all, suggesting that the melting-period duration over the whole of Antarctica is stable (approximately 20 day yr−1, from the filled-in calculations) whereas the melting surfaces are shrinking.
Only the peninsula zone shows positive trends: the CMS trend is positive but not significant; MMD increases by 1.2% ± 0.7% yr−1 (Table 6), which means that the length of the melting period increases. At the same time, the MMS decreases, and this combination explains why the CMS shows no significant positive trend. On the contrary, the indices of the other zones decrease strongly (−3% to −7%). All trends in MBL are statistically insignificant.
c. Relation between melt indices and meteorological data
Meteorological data (Table 1) have been used to calibrate the melt threshold of the algorithm used in this study (section 3). The consistency of the variability and trends in the melt indices with the meteorological data is now tested. For that purpose, we calculate the percentage of warm events per summer (November–February) for a selection of meteorological stations in each zone, except MBL, for which no appropriate data were found. A warm event is detected when air temperature is above the mean December–January temperature of the station. The alternation of warm and cold years as detected in the microwave and the meteorological data are in general agreement (Figs. 5 and 6). Thus, 2- and 2.5-yr oscillations are often confirmed (peninsula, DML, Amery), and the strong melting periods of 1991/92 (Filchner, DML, and Amery) and 1992/93 (peninsula and MBL) are clearly seen in the meteorological data.
Summer mean (December–January) temperature for four zones [peninsula, DML, Amery, and Wilkes; data from J. Jacka, Antarctic Comparative Research Centre (CRC); more information available online at http://www.antcrc.utas.edu.au/∼jacka/temperature.html] are also compared with the annual filled-in CMS over the 18-yr period. The results of linear fits are given in Fig. 7 {CMS = Slope × (〈T°〉Dec–Jan − Tthreshold), where CMS is in 106 day km2, Slope is in 106 day km2 °C−1, and 〈T°〉Dec–Jan [summer mean (December–January) temperature] and Tthreshold are in °C}, in which corresponding plots are also displayed. The DML area has the most important mean slope. On average over the four zones, melting increases by 9.3 × 106 day km2 (sum of the four slopes) when mean summer temperature increases by 1°C. The Tthreshold reflects the average summer temperature below which no significant melting occurs. The peninsula, DML, and Wilkes areas have mean Tthreshold close to −4.0°C, versus only −1.3°C for Amery. ZF94 make a similar analysis over the period of 1978–87. The two studies confirm a good correlation between the temporal evolution of the regional CMS indices and the local air temperature.
The climate significance of our results is further reinforced if compared with the annual trends reported by Comiso (2000) from meteorological data over the last 20 yr. A positive trend is found at Rothera Point (0.090° ± 0.027°C yr−1). This may be related to a significant positive MMD trend for the peninsula region, although the MMS slowly decreases. Trends reported by Comiso (2000) as significantly different from 0 in the DML (Syowa station), Wilkes (Casey station), and Filchner-Ronne (Halley Bay station) zones are negative, ranging from −0.040° ± 0.021° to −0.079° ± 0.025°C yr−1.
Figure 7 also displays how the Antarctic CMS relates to the Microwave Sounding Unit (MSU; Christy et al. 2000) global Antarctic tropospheric temperature (December–February mean anomalies). MSU temperatures are less subject to cloud clearing, aerosol, and contamination than are IR measurements and are representative of the energy available for melting at the surface on a more global scale than meteorological stations. The correlation is high (r = 0.82), highly significant, and positive, confirming the results reported above.
5. Discussion and conclusions
The monitoring and depiction of climate variability and change improve as our knowledge of the spatial and temporal structure and of the nature of climate change expands. Here, we have contributed to characterizing climate variability and change in Antarctica by building a history of summer melt events over the last 20 yr of the twentieth century. This is performed by processing microwave brightness temperature data from spaceborne instruments. Satellite data allow large spatial and temporal coverage; however, the very nature of melt in Antarctica limits this coverage to the summer coastal, shelf, and peninsula regions. We show that the interannual variability of melt events has spatial coherence over large regions and has partial coherence over the whole Antarctic ice sheet. Also, 2- to 5-yr oscillations are identified in the series, which are reminiscent of those reported by Comiso (2000), Cavalieri et al. (1997), and White and Peterson (1996) for sea ice extent. Strong melt events occur during the period 1991–93. The signatures of the Antarctic Oscillation and also possibly, but less clearly, the Southern oscillation are found in this variability. On the other hand, there is no clear sign of a circumpolar traveling wave, for example, the Antarctic circumpolar wave.
A decreasing trend is found in four out of seven main regions of Antarctica. The melting duration at sites where melting occurs ranges from about 50 days in the peninsula (with maxima up to 100), to less than 10 days in the Filchner-Ronne and the Ross ice shelves. On average over the ice sheet, this number is about 20 days and does not significantly change over the 20-yr period. On the other hand, the surface affected by melt decreases over time, from about 1.6 × 106 km2 at the beginning to about 1.0 × 106 km2 at the end of the period (i.e., from 11.4% to 7.2% of the total continental surface). A decrease in summer melt characteristics (duration or extent) is particularly significant in the Dronning Maud Land, Wilkes Land, Amery, and Ross regions. No region shows any significant increase, except possibly the peninsula for the mean melt duration. The interannual variability shown by melt characteristics appears consistent with the interannual variability of the MSU tropospheric temperatures and the number of warm temperature events at a few meteorological stations where relevant data are available. Because melt is the result of a positive surface energy balance when surface temperature is at 0°C, it is not a direct and unique function of atmospheric temperature as provided by meteorological stations. Melt indices thus provide an additional characterization of climate.
Our results extend those of ZF94, and the general trend of CMS they found over the 1979–87 period (−2.4% yr−1) is essentially confirmed over the last 20 yr (−1.8% ± 1% yr−1). Moreover, the correlations between regional CMS indices and mean summer temperatures (December–January) at meteorological stations remain significant and in qualitative agreement with ZF94. Surface melting occurs when the December–January average temperature is above a threshold between −4.0° and −1.3°C. For the four areas as a whole (peninsula, DML, Amery, and Wilkes Land), surface melting increases by about 9.3 × 106 day km2 per degree Celsius of summer temperature increase.
A recent negative continental trend for the January surface temperature is reported by Comiso (2000). Our results are consistent with a mean January cooling on the continent, which in turn, appears compatible with a recent slightly positive trend in the summer sea ice extent (Cavalieri et al. 1997). Our study thus contributes to characterizing a previously identified cooling trend in Antarctica over the last 20 yr of the twentieth century. Reported longer-term trends suggest a mean warming over the second half of the century (e.g., Comiso 2000). Global temperatures are also seen to increase, and this is mostly associated with increased greenhouse gas concentration (Houghton et al. 2001). Some climate models suggest that climate change may be slower around Antarctica than in other regions of the globe because of the efficient uptake of additional surface heat by oceanic convection (e.g., Manabe et al. 1992). In addition, the coastal climate of Antarctica may fluctuate in response to naturally changing patterns of the atmospheric circulation, blurring the signature of anthropogenic warming. Altogether, the climate interpretation of our evidence and other evidence of Antarctic climate trends in the last 20 yr or more is not obvious. Climate models and meteorological analyses, which give access to synthetic data and (for climate models) climate processes, are natural complements to the field of remotely sensed data in such interpretational work.
Acknowledgments
This work is supported by the French Programme National d'Etude de la Dynamique du Climat, project “Anthropique.” Three anonymous reviewers contributed to improve the quality of the paper.
REFERENCES
Abdalati, W., and K. Steffen, 1995: Passive microwave-derived snow melt regions on the Greenland ice sheet. Geophys. Res. Lett., 22 , 787–790.
Abdalati, W., and K. Steffen, 1997: Snow melt on the Greenland ice sheet as derived from passive microwave satellite data. J. Climate, 10 , 165–175.
Bromwich, D. H., A. N. Rogers, P. Kallberg, R. I. Cullather, J. W. C. White, and K. Kreutz, 2000: ECMWF analyses and reanalyses depiction of ENSO in Antarctic precipitation. J. Climate, 13 , 1406–1420.
Cavalieri, D., P. Gloersen, C. Parkinson, J. Comiso, and H. Zwally, 1997: Observed hemispheric asymmetry in global sea ice changes. Science, 278 , 1104–1106.
Christy, J., R. Spencer, and W. Braswell, 2000: MSU tropospheric temperatures: Data set construction and radiosonde comparisons. J. Atmos. Oceanic Technol., 17 , 1153–1170.
Colton, M. C., and G. A. Poe, 1999: Intersensor calibration of DMSP SSM/I's: F-8 to F-14, 1987–1997. IEEE Trans. Geosci. Remote Sens., 37 , 418–439.
Comiso, J. C., 2000: Variability and trends in Antarctic surface temperatures from in situ and satellite infrared measurements. J. Climate, 13 , 1674–1696.
Doran, P. T., and Coauthors. 2002: Antarctic climate cooling and terrestrial ecosystem response. Nature, 415 , 517–519.
Gloersen, P., and W. B. White, 2001: Reestablishing the circumpolar wave in sea ice around Antartica from one winter to the next. J. Geophys. Res., 106 , 4391–4395.
Gong, D., and S. Wang, 1999: Definition of Antarctic oscillation index. Geophys. Res. Lett., 26 , 459–462.
Houghton, J. T., Y. Ding, D. J. Griggs, M. Noguer, P. J. van der Linden, and D. Xiaosu, Eds.,. 2001: Climate Change 2001: The Scientific Basis: Contribution of Working Group I to the Third Assessment Report of the Intergovernmental Panel on Climate Change. Cambridge University Press, 896 pp.
Jacobs, S. S., and J. C. Comiso, 1997: A climate anomaly in the Amundsen and Bellingshausen Seas. J. Climate, 10 , 697–711.
Jezek, K. C., C. Merry, D. Cavalieri, S. Grace, J. Bedner, D. Wilson, and D. Lampkin, 1991: Comparison between SMMR and SSM/I passive microwave data collected over the Antarctic ice sheet. Byrd Polar Research Center Tech. Rep. 91-03, The Ohio State University, Colombus, OH, 62 pp.
Jones, P. D., 1995: Recent variations in mean temperature and the diurnal temperature range in the Antarctic. Geophys. Res. Lett., 22 , 1345–1348.
King, J. C., 1994: Recent climate variability in the vicinity of the Antarctic Peninsula. Int. J. Climatol., 14 , 357–369.
Manabe, S., M. Spelman, and R. Stouffer, 1992: Transient responses of a coupled ocean–atmosphere model to gradual changes of atmospheric CO2. Part II: Seasonal response. J. Climate, 5 , 105–126.
Maslanik, J., and J. Stroeve, 2000: 1990–March 2000 DMSP SSM/I Daily Polar Gridded Brightness Temperatures. National Snow and Ice Data Center, Boulder, CO, CD-ROM.
Mätzler, C., 1987: Applications of the interaction of microwaves with the natural snow cover. Remote Sens. Rev., 2 , 259–387.
Mätzler, C., 2000: A simple snowpack/cloud refletance and transmittance model from microwave to ultraviolet: The ice-lamella pack. J. Glaciol., 46 , 20–24.
Mote, T. L., M. R. Anderson, K. C. Kuivinen, and C. M. Rowe, 1993: Passive microwave-derived spatial and temporal variations of summer melt on the Greenland ice sheet. Ann. Glaciol., 17 , 233–238.
Raper, S. C. B., T. M. L. Wigley, P. R. Mayes, P. D. Jones, and M. J. Salinger, 1984: Variations in surface air temperatures. Part III: The Antarctic, 1957–82. Mon. Wea. Rev., 112 , 1341–1353.
Ridley, J., 1993: Surface melting on Antarctic Peninsula ice shelves detected by passive microwave sensors. Geophys. Res. Lett., 20 , 2639–2642.
Skvarca, P., W. Rack, H. Rott, and T. Ibarzábal y Donángelo, 1998: Evidence of recent climatic warming on the eastern Antarctica Peninsula. Ann. Glaciol., 27 , 628–632.
Steffen, K., W. Abdalati, and J. Stroeve, 1993: Climate sensitivity studies of the Greenland ice sheet using satellite AVHRR, SMMR, SSM/I, and in situ data. Meteor. Atmos. Phys., 51 , 239–258.
Thomson, D. J., 1982: Spectrum estimation and harmonic analysis. Proc. IEEE, 70 , 1055–1096.
Trenberth, K. E., and J. M. Caron, 2000: The Southern Oscillation revisited: Sea level pressure, surface temperatures, and precipitation. J. Climate, 13 , 4358–4365.
White, W., and R. Peterson, 1996: An Antarctic circumpolar wave in surface pressure, wind, temperature and sea-ice extent. Nature, 380 , 699–702.
Yuan, X., and D. G. Martinson, 2000: Antarctic sea ice extent variability and its global connectivity. J. Climate, 13 , 1697–1717.
Zwally, H. J., and S. Fiegles, 1994: Extent and duration of Antarctic surface melting. J. Glaciol., 40 , 463–476.
Number of melting days vs in situ max daily temperatures for four representative stations and for the chosen threshold (3σ)
Citation: Journal of Climate 16, 7; 10.1175/1520-0442(2003)016<1047:VATOTS>2.0.CO;2
Map of the mean duration of surface melting (days) for the 18 yr of data available, plus the location of the seven zones. The color version of this figure was available online at http//lgge.obs.ujf-grenoble.fr/~christo/teledet/torinesi.htm at the time of writing
Citation: Journal of Climate 16, 7; 10.1175/1520-0442(2003)016<1047:VATOTS>2.0.CO;2
Surface melting anomaly (days per 25 × 25 km2 pixel) during the summer periods between 1979/80 and 1988/89 for four different zones (see Fig. 2), calculated by the filled-in calculation. The summers of 1981/82 and 1987/88 are not displayed owing to too many missing data (see Web site listed in Fig. 2 for color version)
Citation: Journal of Climate 16, 7; 10.1175/1520-0442(2003)016<1047:VATOTS>2.0.CO;2
As in Fig. 3 but during the summer periods between 1989/90 and 1998/99 (see Web site listed in Fig. 2 for color version)
Citation: Journal of Climate 16, 7; 10.1175/1520-0442(2003)016<1047:VATOTS>2.0.CO;2
CMS (106 day km2), MMS (km2), and MMD (day) for three zones plus the whole continent; reduced calculation. Linear regressions and relative slopes (% yr−1) values are plotted when statistically significant (see Tables 4, 5, and 6). Warm events (%) are calculated with the max daily temperatures of Rothera Point for the peninsula area, Syowa and Molodeznaja for DML, and Lettau and Ferrel for the Ross area. The summer of 1979/80 is labeled 1980 and so on
Citation: Journal of Climate 16, 7; 10.1175/1520-0442(2003)016<1047:VATOTS>2.0.CO;2
As in Fig. 5 but for four different zones. Warm events (%) are calculated with the max daily temperatures of Belgrano for the Filchner area; Mawson, Zhongshan, and Davis for Amery; and Casey Air Strip for the Wilkes area
Citation: Journal of Climate 16, 7; 10.1175/1520-0442(2003)016<1047:VATOTS>2.0.CO;2
Correlation between the CMS (106 day km2) and the Dec–Jan temperature average (T; °C) of 11 stations and four zones. Argentine Island station (65°15′S, 64°16′W) was not used earlier because no daily data were available. (bottom right) Antarctic CMS vs Dec–Feb tropospheric temperature anomalies (70°–80°S). The percentage of significance (S; 99+ when S is very close to 100%) is also given
Citation: Journal of Climate 16, 7; 10.1175/1520-0442(2003)016<1047:VATOTS>2.0.CO;2
List of Antarctic sites at which meteorological data are compared with satellite data
Periods (yr) and phases (°) calculated by MTM on complete melting index series, above the 90% significance level
Significant correlations (above the 99% confidence level) between the summer-mean AOI (Oct–Jan) and the surface melting index
CMS index calculations. Trend, σ: std dev of the trend, RT: relative trend [%-trend × 100/(mean over the 20 yr)], S: significance of the trend (%) derived from the t test, and mean value over the 20 yr for the reduced calculation. Mean value over 20 yr, RT (%), and S (%) for the filled-in calculation. Trends are given when S > 85%; 99+ when S is very close to 100%
As in Table 4 but for MMS
As in Table 4 but for MMD
