This paper presents results of a combined analysis of cloud observations made at the Antarctic base Faraday/Vernadsky between 1960 and 2005 and sea ice concentration from the HadISST1 data set.
The annual total cloud cover has increased significantly during this period with the strongest and most significant positive trend found in winter, and positive tendencies observable in all seasons. This trend is associated with a decrease in sea ice concentration in the area of the Western Antarctic Peninsula. Though the observed sea ice reduction is actually larger and more significant in summer and autumn, there is actually a significant relation between total cloud cover and sea ice concentration only in winter.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
Clouds play an important role in the radiation budget and hence the energy balance of the Earth's atmosphere. In addition to their direct impact, they also have an influence through a variety of feedback mechanisms. Clouds are very efficient absorbers of infrared radiation. They therefore have a strong natural greenhouse effect and can contribute to the warming of the atmosphere. At the same time, clouds efficiently reflect incoming solar radiation and thereby cool down the Earth's atmosphere (Kiehl and Trenberth, 1997). Changing practically any aspect of clouds, be it cloud type, location, liquid water content, base height, phase, habit, spectrum or life time, influences whether the cooling or the warming aspects of their properties dominate. External influencing factors are the solar elevation and the albedo of the underlying surface. Generally a low cloud cover would be seen as warming in winter (due to the low solar elevation longwave warming outweighs shortwave cooling) and cooling in summer (vice versa). However, over a high albedo surface, such as the ice or snow cover of Antarctica, low clouds can be warming at any time of year. The study by Town et al. (2005) describes the radiative fluxes and cloud radiative forcing over the South Pole in detail. Studies in the Arctic though have found that over a brief period in summer the cooling effect due to cloud shading is stronger than the warming caused by the clouds' greenhouse effect (Shupe and Intrieri, 2004). Model experiments by Lubin et al. (1998) have shown that changes in cloud particle phase and size may change the atmospheric circulation in the Southern Hemisphere significantly.
It is therefore a vital topic of current research to better understand how clouds will change as a consequence of global warming, and how this in turn, via feedback mechanisms, may influence future climate conditions.
Over the past 50 years the Western Antarctic Peninsula has been one of the most rapidly warming parts of the planet, with the largest warming occurring in the winter season (Vaughan et al., 2001; King et al., 2004; Turner et al., 2005). These different studies find warming rates that are 1 order of magnitude bigger than the mean rate of global warming as reported by the Intergovernmental Panel on Climate Change (IPCC, 2007). Meredith and King (2005) found a significant warming of ocean surface waters west of the Antarctic Peninsula in austral summer. Owing to a strong seasonal bias in the data collection, their study is restricted to the summer. It can be assumed though that this observed warming of the sea surface is present all year round not the least because changes in oceanic characteristics happen on a far longer time span than atmospheric changes. An increase in the sea surface temperature increases the latent and sensible heat fluxes from the ocean to the atmosphere and thereby potentially encourages cloud formation. According to IPCC AR4 uncertainties in the radiative forcing in processes and observations are limiting the confidence in attributing some observed climate change phenomena to anthropogenic or natural processes. Clouds have been confirmed as a primary source of uncertainties as they play an important role in the radiation budget and hence the energy balance of the Earth's atmosphere. Hence, it is of special importance to investigate the role clouds play, particularly in this region.
The objective of this paper is to investigate how the decreased sea ice cover and rising temperatures have influenced and may further influence the climate conditions of the Western Antarctic Peninsula, in particular with regard to cloud cover and cloud type. Currently, the only data sets available over a climatological time scale in this region are synoptic observations, and those carried out at the base Faraday/Vernadsky provide the longest and most consistent data set. An important aspect of the data analysis is to ensure that presented results are not artefacts due to observer bias or caused by the prolonged periods of darkness in this region.
2. Data and methods
2.1. Synoptic observations at Vernadsky
The first permanent research base on the Argentine Islands (65° 15′S and 64° 16′W) was built in 1947 on the site previously occupied by a research base during the British Graham Land Expedition. On 6 February, 1996, the station was officially handed over to the Ukraine and is since then known under its present name ‘Vernadsky’. Figure 1 shows the location of the base in Antarctica, and its position on Galindez Island in more detail. Though this is not the only data set of cloud observations from this part of Antarctica, its length and quality recommend it for this kind of analysis. A comprehensive atlas with global information on cloud cover and other cloud characteristics based on ground-based observations has been compiled by Warren et al. (1986). Owing to different periods of investigation and owing to their data assimilation scheme, results are only comparable to a limited extent.
With the onset of the satellite era, passive remote-sensing techniques have started to offer a reliable source of information on clouds with a high spatial coverage. They are by now widely used in other regions, though the determination of cloud cover is extremely difficult over the poles owing to the small difference between both albedo and temperature of the snow surface and clouds (Rossow and Schiffer, 1999). Several studies have tried to overcome these difficulties by using active remote-sensing methods (Noël et al., 2008; Sassen et al., 2008; Wang et al., 2008), but as these space-borne instruments have not yet been in operation for long, ground-based observations are still the only source of reliable data on climatological time scales from these remote regions.
Ground-based synoptic observations of cloud parameters are recorded according to the World Meteorological Organization's Manual on Codes (WMO, 1995).
Parameters used in this paper are total cloud cover (TCC), low cloud amount (LCA), as well as low cloud type (LCT), medium cloud type (MCT) and high cloud type (HCT).
TCC embraces the total fraction of the celestial dome covered by clouds irrespective of their genus and is expressed in okta.
LCA originally describes the amount of all low clouds present or, if no low clouds are present, the amount of all medium-level clouds. For this study, the observations have been restricted to only cases when low clouds were present, i.e. when there was a recorded observation of LCT. The same coding system (in okta) is used for observations of LCA as for TCC; the terms ‘total cloud cover’ and ‘low cloud amount’ are used for these parameters in the observational records as well as in the guidelines (WMO, 1995). Equivalent observations for medium-level or high clouds are not required by the WMO and are consequently not available. Coding of 0 okta for TCC (LCA) refers to absolutely clear sky (no low cloud); as soon as there is any (low) cloud visible the code 1 okta has to be used. Equivalently, 8 okta are recorded in cases when the sky is completely overcast (with low clouds). Any gap in the (low) cloud cover requires 7 okta to be recorded.
The parameters that are called LCT, MCT and HCT, respectively, in the observational records and in the present study refer to the parameters CL, CM and CH as used in the guidelines by the WMO (1995). Records of LCT, MCT and HCT allow conclusions of the frequency with which such clouds occur.
The period under investigation covers the years from 1960 to 2005. Cloud observations in general were begun at Faraday in 1947, but only since 1960 has the whole range of cloud parameters—as recorded presently—been observed. Data used for this study consists of observations of various cloud parameters, such as TCC and LCA. These parameters are usually observed and recorded at 3-h intervals, although only 6-h observations were used for this study as the observations at 00, 06, 12 and 18UTC were recorded most consistently.
The data set covers 16 802 days, so—looking at 6-h data—this means records potentially comprise 67 208 recordings per parameter. Of these 67 208 potential observations, 191 are missing, the data set is therefore 99.7% complete. Until 1987, no observation was missed; in 1992, the data set shows the highest number of missed observations (50, which is equivalent to 3.4% of potential observations during that year). From then on records show around 10 missed observations per year. The reduced number of records in the second half of the 1980s is most likely connected to the installation of an Automatic Weather Station (AWS). The high number of missed observations in 1992 occurred during the replacement of this first AWS with a newer model in March of that year. The change from personal to automatic recording of meteorological parameters meant that night-time synoptic observations no longer required a trained meteorologist to be on duty. Thus, from then on nocturnal cloud observations were made by the night watch rather than by a fully trained meteorologist.
Studies of cloud observations at the South Pole station indicate that cloud cover there is underestimated during winter, i.e. when cloud observations have to be carried out during darkness. At Vernadsky, which is located just North of the Polar circle, all year round there is enough day light around local noon (∼17UTC) to perform cloud observations, as even on a midwinter day for the observations at 18UTC the sun is less than 1° below the horizon. According to Hahn et al. (1995), twilight from the sun less than 9 degrees below the horizon is sufficient for an adequate detection of cloud cover. For comparison, in all tables, values are given on the basis of all observations (00, 06, 12 and 18UTC) and on the basis of 18UTC observations only.
The time series of observations carried out at different times during the day (Figure 2) as well as those of all observations combined (Figure 4) show differences, but none that suggest a bias due to changes in observational practice. Differences between the time series of 06UTC and 18UTC observations may originate in an underestimation of the cloud cover during darkness, but this does not explain the trend seen in both data sets (Figure 2).
The annual course of TCC shown in Figure 3 based on 18UTC (daylight) and 06UTC (darkness) observations shows that there is actually less cloud cover in winter at Vernadsky than in summer. While the more distinct minimum in winter for 06UTC observations may reflect the underestimation of cloud cover during darkness reported by Town et al. (2007), this reasoning is not valid for the 18UTC observations. Given the seasonality e.g. of the sea ice cover, the albedo and the surface temperature of the ocean in the vicinity of the station that potentially influence the formation of clouds, an actual annual cycle seems absolutely plausible. Figure 3 also shows, as grey-shaded area, the variability of monthly mean TCC throughout the year. This emphasizes that the differences between observations at different times of the day are actually small compared to this variability. As an example, the monthly mean TCC for 2003 is also shown.
HadISST1 is the Met Office Hadley Centre's observational data set of sea ice and sea surface temperature. It is a global data set on a 1° latitude–longitude grid comprising sea ice information and sea surface temperature from 1870 to date, based on a variety of sources including digitized sea ice charts and passive microwave retrievals. The sea ice fields are made more homogeneous by compensating satellite microwave-based sea ice concentrations for algorithm deficiencies in the Antarctic. This study considers data from 1979 onwards, as sea ice data has been shown to become reliable with the onset of the satellite era. Before that, sea ice data is based on two climatologies, i.e. contains neither seasonal nor inter-annual variability (Rayner et al., 2003; Lachlan-Cope, 2005). Monthly sea ice concentration data was extracted from the data set for the grid point closest to Vernadsky and for an area to the west of the Antarctic Peninsula covering the region between 60°S and 70°S and from 60°W to 70°W. Sea ice concentration is defined as the fraction of an area, which is covered by sea ice. The area in this case is defined by the model's grid size.
All trends presented in this study are calculated using a standard least squares method. The methodology used to calculate the significance levels is based upon Santer et al. (2000).
3. Total cloud cover
Over the period under investigation, the average TCC at Vernadsky was 6.5 oktas with a standard deviation of 0.2. Given the differences in the data base, this agrees remarkably well with results by Warren et al. (1986) of 85% (6.8 okta) in their equivalent grid box.
When averaging monthly mean values of cloud cover over the complete period, a mean annual cycle becomes apparent. Its main feature is a minimum in June with 6.0 oktas TCC, which is imbedded in otherwise quite homogeneous cloud cover conditions of up to 6.8 okta (February and November), but there is no distinct secondary minimum. The inter-annual variability is very small with values ranging from 6.1 (1985) to 6.9 oktas (1972). The amplitude of monthly mean cloud cover values within one individual year in contrast, varies over a huge range with values from less than 1 okta (0.9 okta in 1968) to more than 3 oktas (3.1 oktas in 1987). As a consequence, the annual cycle is not necessarily visible in individual years. In 2003, for example, this leads almost to a reversal of the mean annual cycle, so that a main maximum is observed in winter (August) and the minimum is observed in summer (December) (Figure 3).
Figure 4 shows the annual and seasonal mean TCC for the period under investigation together with linear trends. The TCC has increased during the years from 1960 to 2005, and this positive trend is present in all seasons. The exact trends and their significance level are listed in Table I. The trend is strongest during winter months (0.014 oktas/year) and about half as strong annually. The annual trend is significant at the 5% level; the trend for the winter months is significant at the 1% level.
Table I. Annual and seasonal trends in TCC at Vernadsky during the period 1960–2005
In numbers, these trends have led to an increase in the annual mean TCC at Vernadsky by almost 5% during the time from 1960 to 2005. The trend in winter corresponds to an increase in TCC by more than 8%. Regarding individual calendar months, the strongest significant trend (<1%) was observed for August with an increase in TCC by 0.022 oktas/year. Other months with significant trends are February (0.012 oktas/year) and July (0.019 oktas/year). Warren et al. (1986) find a reduction in the annual TCC but increases in the autumn to the west of the Antarctic Peninsula and in winter to the east of the Antarctic Peninsula, though no significance level is given.
The seasonality of the trend means that the seasonal cycle described above showing a minimum TCC in June becomes less and less distinct and might in future even lead to a rearrangement of the annual cycle as already observed in 2003.
The annual number of observations recording 0 or 1 okta has decreased significantly, a trend that is also visible (and significant) in summer, autumn and is especially distinct in winter (Table II). While generally these categories do not contain many observations, during winter months on average 12% and up to 21% of observations record either 0 or 1 okta TCC. Other categories of TCC do not show any significant trends. Hence, the increase in TCC described above is a result of a decrease of observations recording 0 or 1 okta. Warren et al. (1986) find in their analysis of ‘clear sky frequency’ also a decrease throughout all seasons. This shows again a qualitative agreement.
Table II. Annual and seasonal trends in observations recording 0 or 1 okta TCC at Vernadsky during the period 1960 to 2005
Low clouds are the only cloud group for which the WMO guidelines require the observation of the cloud amount.
The long-term annual average of the LCA at Vernadsky is 4.3 ( ± 0.7) oktas and thus more than 2 oktas smaller than the TCC. The annual mean LCA shows a much larger inter-annual variability than the TCC with values ranging from 3.3 oktas in 1998 to 5.9 oktas in 1972. The inter-seasonal variability in the LCA is also slightly larger than that seen in the TCC with amplitudes in individual years ranging from 1.3 oktas in 1998 to 4.3 oktas in 1997.
The long-term average over the whole period reveals an annual cycle, which is not generally visible in individual years. It is formed of a minimum of 3.8 oktas in July and a maximum in April (4.7 oktas).
No significant trends were observed during the period of interest for the annual or seasonal mean LCA. On a monthly basis, a significant negative trend was observed for the month of December though. The average LCA in December has decreased by 11.5% over the entire period, which corresponds to almost 1 okta. This trend is significant at the 10% level.
Figure 5 shows the distribution of LCA observations to the different categories from 0 to 8 oktas based on 6-h annual and seasonal records as well as for annual records of 18UTC observations. It can be seen that for total cloud amount—including observations carried out during darkness—the characteristics of the data set do not change. The percentages of observations allocated to the categories of 0 and 1 okta are significantly larger than those in records of TCC.
Studying the development of occurrences of the different possible observational categories, the picture is less distinct than that found in TCC. More detailed analysis shows that the number of extreme cases recorded (0, 1, 7 or 8 oktas) has decreased significantly throughout the year, while observations of moderate LCA have increased. The significant decrease in extreme categories (0, 1, 7 or 8 oktas) is only present in the sum of these categories. The numbers of observations in the individual categories all show negative tendencies, but none of these individual trends is significant.
The decrease in observations of ‘extreme’ categories is balanced by significant increases in the frequency of moderate cases of cloud observations. A significant increase in the annual number of observations recording 2, 3 and 4 oktas of LCA was found. This shift can be observed during all seasons. Results of the trend analysis for extreme and moderate cases of LCA are compiled in Tables IIIa and IIIb.
Table IIIA. Annual and seasonal trends in observations of extreme cases of LCA at Vernadsky from 1960 to 2005
The TCC has increased over the period, especially during winter months, while the overall LCA does not show any corresponding trends. Indeed, the only significant trend in overall LCA is the observed decrease in December. This encourages the conclusion that any increase observed in the TCC must be caused by an increase in either medium-level or high cloud amount or by an increase in observations of either or both of these cloud groups.
MCT and HCT are the only two parameters the WMO stipulate in their guidelines regarding these cloud groups. The amount of medium-level and high clouds is not part of the synoptic observations' routine. Thus, there is no direct way of gaining information on the amount of medium or high clouds to verify the above hypothesis. One possibility to overcome this problem is to use the parameters ‘MCT’ and ‘HCT’ under the assumption that these parameters can only be recorded if medium-level respectively high clouds are present. Obviously, this method only allows conclusions about the presence of medium-level and high clouds but does not allow deducing the actual amount of the medium-level or high clouds that were observed.
Neither the occurrence of medium level clouds nor that of high clouds shows any significant trends on the annual scale. Especially the frequency with which medium-level clouds are recorded shows no trend or tendency in any season. The frequency with which high clouds are observed shows a negative tendency throughout the year. For autumn months, this trend is significant within the 10% level.
As mentioned above, the increase in total cloud amount cannot be explained by an increase in LCA or an increase in the number of low cloud events observed. As medium level clouds do not show any tendency to occur more or less frequent and high clouds even show a tendency to occur less frequently than in the past, on occasions when medium or high clouds do occur their amount must have increased. High clouds—i.e. their presence and type—can obviously only be recorded if they are not obscured by underlying cloud layers. The fact that the number of observations recording high cloud (type) is decreasing may indirectly indicate that it is actually the medium level cloud amount that has increased. Unfortunately, observational records do not allow for this effect to be studied directly.
6. Sea ice
It is likely that the extent and concentration of sea ice around the Antarctic Peninsula largely affects the TCC at Vernadsky. Areas of open water in the sea ice facilitate the turbulent heat fluxes from the relatively warm ocean to the atmosphere and hence are a source for energy and moisture.
The following section therefore looks at the sea ice concentration around the Antarctic Peninsula. Monthly sea ice concentration data was extracted from the HadISST1 data set for the grid point closest to Vernadsky and for an area to the west of the Antarctic Peninsula covering the region between 60°S and 70°S and from 60°W to 70°W. Although the data set extends back to 1870, only data from 1979 onwards is considered, as sea ice data has been shown to become noticeably more reliable with the onset of the satellite era.
For the area as well as for the grid point, a significant decrease in the annual mean sea ice concentration for the period under investigation is found. This trend is persistent throughout the year (Table IV). This is in accordance with recently published studies of sea ice changes around the Antarctic (e.g. Zwally et al., 2002) that indicate a decline in sea ice towards the west of the Antarctic Peninsula.
Table IV. Mean sea ice concentration and trends near Vernadsky (1979–2005, HadISST1)
Mean sea ice concentration (60°S - 70°S, 60°W - 70°W)
Accounting for natural spatial variability, the comparison between the grid point data and the areal data show that the grid point extracted sea ice concentration generally represents the characteristics of the area well. The correlation between the areal and the grid point data for the annual mean sea ice concentration is 0.75; for individual seasons, the correlation ranges from 0.58 in autumn to 0.86 in summer. The stronger trends at the grid point compared to the spatial trends reflect that the strongest decrease in sea ice concentration is observed close to the coast.
The time series of seasonal mean sea ice concentration and TCC show a highly significant correlation for the winter season (Figure 6). The correlations of − 0.58 and − 0.53 for grid point and area sea ice concentration respectively are significant at the 1% level. Correlations between annual mean values and for other seasons are not significant though. This is in agreement with results published by King (1994) who demonstrated a significant positive correlation between air temperature at Faraday in winter and cloud cover, and a significant negative correlation between winter temperature and sea ice extent.
During winter, when the temperature difference between the air and the ocean can easily reach 30K and high wind speeds are frequently observed, turbulent heat fluxes over open water may be as large as some 100 W/m2, while at the same time over thick sea ice they often amount to only a few watts per square metre (Brümmer et al., 2005). Under such conditions, even little changes in sea ice cover lead to huge changes in the turbulent fluxes. Thus, the sea ice decrease observed in summer and autumn—though larger at the grid point and more significant for the area than winter trends—does not show a significant correlation to the TCC.
Observational data from Vernadsky show a significant trend towards a higher TCC. Annually, the TCC has increased by 0.008 okta, which corresponds to an increase by almost 5% over the complete period from 1960 to 2005. This trend is especially present in winter when the TCC has increased by more than 8% over 46 years, a trend that is significant at the 1% level. A tendency towards more clouds can be seen throughout the year, but during spring, summer and autumn the trend is not significant.
This development manifests itself mainly in a significant decrease in observations with 0 and 1 okta cloud cover throughout the year. These trends are significant at the 1% level for annual observation numbers (−0.12 obs/year) and for the number of recordings in winter (−0.23 obs/year).
The increase in TCC is not reflected in the amount of low clouds. Observations of LCA show no significant trend either annually or in any of the seasons. There is evidence though that the average amount of low clouds in summer is decreasing, a trend which is significant in December. No indication was found that events of low clouds occur more frequently. Hence, it can be assumed that the increase in total cloud amount is not due to an increase in LCA but is caused by an increase in medium or high cloud amount. This is especially obvious in winter when the TCC shows its largest increase, while the amount of low clouds does not change, and in December (Southern Hemisphere summer) when LCA actually shows a significant decrease, while TCC shows a tendency to increase during that season.
As there is no data available on the actual coverage by medium and high clouds, the frequency with which they were recorded has been analysed. There is no evidence suggesting that the frequency with which medium level clouds occur has increased. With regard to high clouds, this study found that they show a tendency (and in autumn a significant trend) to occur less frequently. This could possibly be caused by an increase in medium-level cloud cover, as this would generally inhibit high cloud observations. These two findings strongly support the hypothesis that medium-level and/or high cloud cover has increased; it should be treated with caution, though, until stronger evidence can be presented.
The observed shift in LCA to more cases of moderate cloud amount may indicate that a decrease in sea ice leads to more cases of convective clouds. The fact that these trends show their highest significance outside winter reflects the normal inter-annual variability in sea ice concentration in winter, which leads to huge inter-annual changes in turbulent fluxes and therefore in the availability of heat and moisture essential for convective cloud formation. A preliminary analysis of the frequency and type of precipitation events at Vernadsky backs this hypothesis.
A phenomenon of importance to the northern Antarctic Peninsula is the Southern Annular Mode (SAM). Recent studies (Thompson et al., 2002; Thomspon and Solomon, 2000; Marshall, 2003; Marshall, 2007) have revealed a trend towards a more positive polarity in the SAM. Lubin et al. (2008) show that this intensifies the cyclones around the Antarctic and shifts the area of cyclonic activity polewards. This development could have brought the belt of cyclonic activity closer to Vernadsky and hence led to the increase in cloud cover. It has to be said though that largest trends in SAM are found in autumn and summer, so that this trend does not match the temporal patterns of the increase in cloud cover. An inter-comparison of the cyclone activity as it is represented in ERA-40 and NCAR-NCEP by Wang et al. (2006) finds an increase in strong-cyclone activity south of 60°S and a decrease in the belt between 40°S and 60°S. Whether the change in cloud parameters found at Vernadsky are connected to the change in SAM or generally in cyclonic activity can therefore not be stated with certainty, not the least because of the station's location close to the border between the opposing trends found by Wang et al. (2006).
The consequences of an increased cloud cover on sea ice are complex. Over ice-free areas, the cloud cover will increase the albedo, and thus the cooling effect may outweigh the natural greenhouse effect, while over areas that are permanently ice covered their greenhouse effect will dominate. Improved cloud-detecting schemes for satellite imagery of the polar regions are urgently needed to investigate these developments. CloudSat and CALIPSO, satellites launched in recent years, already deliver more accurate observations of clouds and ice with high temporal and spatial resolution. Though data sets are still too short for a climatological analysis, they provide promising opportunities for the investigation of clouds in the polar regions.
The coincidence of the largest increase in air temperature (as reported by Vaughan et al., 2001; King et al., 2004; Turner et al., 2005) and in TCC (as presented in this study), which are both observed during winter, is noteworthy. Though further investigation into the detailed processes is needed, it seems apparent that the increase in cloud cover contributes to the atmospheric warming, while in return the increase in air temperature may encourage the formation of clouds.
Future work will aim to utilize an archive of satellite imagery covering the Antarctic Peninsula and parts of East Antarctica to retrieve spatial information on cloud properties and their trends in that region.