Antarctic Sea Ice Variability and Trends, 1979-2010

. In sharp contrast to the decreasing sea ice coverage of the Arctic, in the Antarctic the sea ice cover has, on average, expanded since the late 1970s. More specifically, satellite passive-microwave data for the period November 1978 – December 2010 reveal an overall positive trend in ice extents of 17,100 ± 2,300 km 2 /yr. Much of the increase, at 13,700 ± 1,500 km 2 /yr, has occurred in the region of the Ross Sea, with lesser contributions from the Weddell Sea and Indian Ocean. One region, that of the Bellingshausen/Amundsen Seas, has, like the Arctic, instead experienced significant sea ice decreases, with an overall ice extent trend of -8,200 ± 1,200 km 2 /yr. When examined through the annual cycle over the 32-year period 1979-2010, the Southern Hemisphere sea ice cover as a whole experienced positive ice extent trends in every month, ranging in magnitude from a low of 9,100 ± 6,300 km 2 /yr in February to a high of 24,700 ± 10,000 km 2 /yr in May. The Ross Sea and Indian Ocean also had positive trends in each month, while the Bellingshausen/Amundsen Seas had negative trends in each month, and the Weddell Sea and Western Pacific Ocean had a mixture of positive and negative trends. Comparing ice-area results to ice-extent results, in each case the ice-area trend has the same sign as the ice-extent trend, but differences in the magnitudes of the two trends identify regions with overall increasing ice concentrations and others with overall decreasing ice concentrations. The strong pattern of decreasing ice coverage in the Bellingshausen/Amundsen Seas region and increasing ice coverage in the Ross Sea region is suggestive of changes in atmospheric circulation. This is a key topic for future research.


Introduction
Satellite passive microwave observations of the Earth's polar sea ice cover over the 20 last three decades have provided the basis for studying its regional, seasonal, and interannual variabilities. The Arctic sea ice cover has exhibited a significant decline over this period (e.g., Vinnikov et al., 2006;Meier et al., 2007;Parkinson and Cavalieri, 2008;Stroeve et al., 2011a). In contrast to the Arctic, the total Antarctic sea ice cover has increased somewhat from 1979 through 2010. Analyses of the Antarctic sea ice cover Introduction this study we extend the 28-yr (1979-2006) time series presented by Parkinson and Cavalieri (2008) and discuss the differences in the Arctic sea ice variabilities and trends observed between the 28-yr and 32-yr periods. Figure 1 shows the 32-yr averaged sea ice concentration maps for March, the month of maximum sea ice extent on average, and for September, the month of minimum 5 sea ice extent on average. Figure 1 also shows the location of each of the nine Arctic regions analyzed in this study. Much background material about Arctic sea ice and the characteristics of each of the nine regions used in the analyses is given in Parkinson et al. (1999) and in Parkinson and Cavalieri (2008) and is not repeated here. The emphasis in this study is on the changes that occurred from the previous 28-yr time 10 series results to the current 32-yr results. In Sect. 2 we present a brief description of the data sets used and the methods employed to derive the sea ice time series. In Sect. 3 we present the results of our analyses for sea ice extents and sea ice areas. In Sect. 4, a summary of results and conclusions are presented. 15 The data set from which the sea ice extents and areas are calculated consists of sea ice concentration maps derived from the radiances obtained from the following satellite microwave radiometers: the Nimbus 7 Scanning Multichannel Microwave Radiometer (SMMR), which operated from 26 October 1978 through 20 August 1987, and  which may differ in their measurement wavelengths, fields-of-view, angles of incidence, ascending node times, and calibrations. In a study of the consistency of long-term observations of oceans and ice from space, Zabel and Jezek (1994) found that ocean observations must be matched at the geophysical product level rather than at the sensor radiance level, an approach that was already in use by the sea ice community. Details 5 of the approach, including filling data gaps, reducing land-to-ocean spillover effects, reducing weather effects over ice-free ocean, and finally matching sea ice extents and areas for each pair of overlapping sensors, are discussed by Cavalieri et al. (1999). Starting with the Nimbus 7 SMMR each subsequent sensor was matched with the previous sensor using the available period of overlap to minimize the differences. Although we did not have the desired complete year of overlap for each pair of sensors, we were able to reduce ice extent differences during periods of sensor overlap to 0.1 % or less and ice area differences to 0.6 % or less (Cavalieri et al., 1999). Most recently, we did have a complete year of overlap (2007) between the F13 SSMI and the F17 SSMIS sensors with which to match the F13 SSMI and F17 SSMIS sea ice extents 15 and areas. A description of the methods used and the level of agreement obtained are provided by Cavalieri et al. (2012).

Results
Sea ice extents and areas are presented for the entire Northern Hemisphere as monthly averaged values, monthly deviations, and yearly and seasonally averaged val-Introduction For the purpose of providing a relative measure of the "significance" of the trends, we calculate a ratio R of the estimated trend to its standard deviation. R provides a continuous measure of "significance" and is less arbitrary than the standard 95 % or 99 % levels of statistical significance generally used. Nonetheless, these levels of significance are indicated in the trend tables to provide a reference for comparison 5 with other published values. These levels of statistical significance are obtained using the Student's t-test with the null hypothesis of a zero trend and assuming 30 (32-2) degrees of freedom. The threshold values of R corresponding to the 95 % and 99 % levels of statistical significance are 2.04 and 2.75, respectively. Although the use of this statistical significance test has been criticized both for the use of the null hypothesis 10 and the arbitrary levels of significance (e.g., Nicholls, 2001) and for issues related to the autocorrelation of the data (e.g., Santer et al., 2000), we still indicate these levels of statistical significance in the tables to provide a relative measure of the robustness of the trend. 15 Sea ice extent is defined as the cumulative area of all grid cells having at least 15 % ice concentration. The reasons for using a 15 % cutoff have been discussed previously by Parkinson and Cavalieri (2008). We first examine sea ice extent variabilities and trends for the Northern Hemisphere as a whole and then for each of the nine Arctic regions. 20 A comparison of the 32-yr March and September maps of Fig. 1 with the corresponding maps for the 28-yr period ( Fig. 1 in Parkinson and Cavalieri, 2008) Figure 2a shows the monthly average sea ice extents for the total Northern Hemisphere as well as the average seasonal cycle (Fig. 2a inset) based on monthly averages. Generally, the month of maximum extent is March, whereas the month of minimum extent is September. The three years with the largest March sea ice extents for the Northern Hemisphere were 1979, 1983, and 1982 having extents of 16.1, 15.9, and 5 15.8 million square kilometers, respectively. The three years with the smallest September sea ice extents were 2007, 2008, and 2010 with extents of 4.3, 4.7, and 4.9 million square kilometers, respectively. Figure 2b shows the sea ice extent monthly deviations for the November 1978-December 2010 (32 1/6 yr) period with an overall trend of −51.4 ± 1.9 × 10 3 km 2 yr −1 . 10 The largest negative excursion from the trend line occurs in September 2007, which had the smallest sea ice extent (4.32 × 10 6 km 2 ) for the 32-yr period. The overall trend is 14 % more negative than the corresponding trend for the 28-yr period (Parkinson and Cavalieri, 2008). Figure 2c shows both the yearly and seasonal trends for the Northern Hemisphere 15 for the 32-yr period 1979-2010. The trend values are provided in Table 1. All the trends are negative with increasingly larger negative trends from spring, winter, autumn, and summer. A comparison of seasonal trends from the 28-yr to 32-yr period shows that for the 32-yr period the magnitude of the negative trend for summer has increased by 31 %, for autumn has increased by 20 %, for winter has increased by 1 %, but for spring, 20 it has decreased by 4 %. While the Northern Hemisphere yearly trend of −51.5 ± 4.1 × 10 3 km 2 yr −1 for the 32yr period   (Table 1) is 14 % more negative than the corresponding trend of −45.1 ± 4.6 km 2 yr −1 for the 28-yr period  reported by Parkinson and Cavalieri (2008), the changes from the 28-yr period to the 32-yr period on a regional 25 basis were far from uniform.  Figure 3 shows the sea ice extent monthly deviations and their trends for each of the nine Arctic regions and for the hemisphere as a whole. Examination of the figure illustrates the very large differences in regional sea ice extent variabilities over the 32-yr period. The three regions contributing most to the overall negative trend of 5 the Northern Hemisphere are the Arctic Ocean (−14.4 ± 1.5 × 10 3 km 2 yr −1 ), the Kara and Barents Seas (−13.5 ± 0.9 × 10 3 km 2 yr −1 ), and the Baffin Bay/Labrador Sea region (−7.6 ± 0.6 × 10 3 km 2 yr −1 ) (Fig. 3). These three regions account for over half of the total area of the nine Arctic regions combined. The yearly trends for the Arctic Ocean and Kara and Barents Seas regions (Ta-10 ble 1) represent increases of 45 % and 27 % in the magnitude of the negative yearly trends, respectively, compared with the corresponding yearly trends for the 28-yr period  reported by Parkinson and Cavalieri (2008). Other regions showing more negative trends for the longer period are the Seas of Okhotsk and Japan, the Gulf of St. Lawrence, and the Canadian Archipelago. 15 Some regions had less negative yearly trends for the 32-yr period compared to the 28-yr period. These included the Bering Sea, Hudson Bay, Baffin Bay/Labrador Sea, and the Greenland Sea regions (Table 1 versus Parkinson and Cavalieri, 2008). In contrast to the other eight regions, the Bering Sea had a positive trend (+0.3 ± 0.8 × 10 3 km 2 yr −1 ) for the 32-yr period, although not statistically significant (Ta-20 ble 1). For the 28-yr period, the Bering Sea trend was −0.5 ± 1.0 × 10 3 km 2 yr −1 (Parkinson and Cavalieri, 2008). far the largest regional negative trend occurs for the Arctic Ocean region in the month of September (Fig. 4h). The Seas of Okhotsk and Japan and the Gulf of St. Lawrence are completely free of ice during July, August, and September, whereas the Bering Sea is totally ice free only in August. The Northern Hemisphere (Fig. 4j) shows an interesting seasonal pattern in the A comparison of Fig. 12 in Parkinson and Cavalieri (2008) with our Fig. 4 shows that 15 the Bering Sea had positive trends only for the months of January and February for the 1979-2006 period, in contrast to the four months of positive trends for the 32-yr period (Fig. 4b). For the Northern Hemisphere as a whole the largest negative trend occurs in September for both the 28-yr and the 32-yr periods, but for the 28-yr period the September negative trend was approximately −57 × 10 3 km 2 yr −1 , versus the much 20 higher magnitude −80 × 10 3 km 2 yr −1 trend for the 32-yr period.

Sea ice areas
Sea ice areas are defined as the cumulative sum of the product of the grid cell ice concentration and grid cell area for all grid cells having at least 15 % ice concentration. Sea ice extents are usually greater (and never less) than sea ice areas, but the two 25 are equal for the case where all ice concentrations are 100 %. We first examine the variabilities and trends for the Northern Hemisphere as a whole and then for each of the nine Arctic regions.

Northern Hemisphere total
Similar to Fig. 2a for sea ice extents, Fig. 5a shows the monthly average sea ice areas for the total Northern Hemisphere as well as the average seasonal cycle (Fig. 5a inset) based on monthly averages. Overall, the month of maximum sea ice area is March, closely followed by February. The month of minimum sea ice area is September. The

5
year with the largest March sea ice area for the Northern Hemisphere is 1979, having an area of 14.2 million square kilometers, whereas the year with the smallest September sea ice area (3.1 million square kilometers) is 2007. The hemispheric yearly trend of −49.6 ± 4.0 × 10 3 km 2 yr −1 for the 32-yr period ( Table 2) is 21 % more negative than the corresponding trend of −41.0 ± 4.3 × 10 3 km 2 yr −1 for the 28-yr period  reported by Parkinson and Cavalieri (2008). Figure 5b shows sea ice area monthly deviations with an overall trend of −49.3 ± 1.9 × 10 3 km 2 yr −1 and the year-to-year variability is almost identical to that for the sea ice extents. Likewise, the sea ice area yearly and seasonal averages shown in Fig. 5c are similar to the corresponding plots for sea ice extent, although all the values 15 are reduced. The seasonal trends are more negative for winter, spring, autumn, and summer (Table 2 versus Parkinson and Cavalieri, 2008). We next examine the changes in trends by region. As with the sea ice extents, the changes from the 28-yr period to the 32-yr period were far from uniform. 20 Regionally and seasonally, the sea ice area variabilities are somewhat different than those for sea ice extents, reflecting the influence of sea ice concentration variability. The sea ice area trends are sometimes smaller than the corresponding sea ice extent trends, sometimes larger, and sometimes of opposite sign. Figure 6 shows the sea ice area monthly deviations and their trends for each of 25 the nine Arctic regions and for the hemisphere as a whole. As with the sea ice extent trends, the three regions contributing most to the overall negative sea ice area trend of the Northern Hemisphere are the Arctic Ocean (−18.4 ± 1.6 × 10 3 km 2 yr −1 ), the Kara and Barents seas (−12.3 ± 0.8 × 10 3 km 2 yr −1 ), and the Baffin Bay/Labrador Sea region (−5.9 ± 0.6 × 10 3 km 2 yr −1 ). When examined on a yearly average basis (Table 2), the trends for the Arctic Ocean and Kara and Barents seas regions represent increases in magnitude (more negative) 5 of 63 % and 27 %, respectively, compared with the corresponding trends for the 28-yr period  reported by Parkinson and Cavalieri (2008). Other regions showing more negative trends for the longer period are the Gulf of St. Lawrence and the Canadian Archipelago.

Regional
Some regions had less negative trends for the 32-yr period compared to the 28-yr period. These included the Seas of Okhotsk and Japan, Hudson Bay, Baffin Bay/Labrador Sea, and the Greenland Sea regions. As with sea ice extent and in contrast to the other eight regions, the Bering Sea had a positive trend (+0.5 ± 0.6 × 10 3 km 2 yr −1 ) for the 32-yr period. For the 28-yr period, the Bering Sea trend was negative (−0.2 ± 0.7 × 10 3 km 2 yr −1 ). Figure 7 shows the sea ice area trends by month for each of the nine Arctic regions and for the Northern Hemisphere as a whole. While the seasonal patterns for the Northern Hemisphere and for each of the Arctic regions generally match those for the sea ice extents, there are two important differences. First, for the Northern Hemisphere, the 20 smallest sea ice area trend occurs in March (Fig. 7j), whereas for sea ice extent the smallest trend occurred in May (Fig. 4j). As with sea ice extent, the largest negative trend occurs in September, although the magnitude of the ice area trend for October is almost as large, in contrast to the sea ice extent October value (Fig. 4j). The second difference between the sea ice extent and area trends by month is that 25 during the months of January through April the sea ice area trends are slightly positive for the Arctic Ocean and the Canadian Archipelago regions in contrast to the zero Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | sea ice extent trends for the same months and regions. This implies that the sea ice concentration has been increasing over at least portions of these regions. These Arctic Ocean and the Canadian Archipelago ice area trends are somewhat less than the positive trends found by Parkinson and Cavalieri (2008) for the 28-yr period. These two regions are the only ones with statistically significant positive sea ice area trends for 5 the winter season as a whole, although the Bering Sea has a statistically insignificant positive trend (Table 2). For the 28-yr period (Parkinson and Cavalieri, 2008), the Bering Sea had positive trends in ice areas and ice extents only for the months of January and February, whereas for the 32-yr period, the ice areas and ice extents had positive trends from January through April (Figs. 4 and 7). Also, for the entire Northern Hemisphere, the maximum negative ice-area trends for the 28-yr period were between 50-55 × 10 3 km 2 yr −1 for the months of June, July, August, and September, whereas for the 32-yr period the maximum negative ice-area trend for the month of September, is almost −75 × 10 3 km 2 yr −1 .

Summary and conclusions
The decline of the Arctic sea ice cover continues to accelerate overall, but with significantly different regional and seasonal trends. Over the 32-yr period there was a 14 % more negative trend in yearly average sea ice extent for the entire Northern Hemisphere than for the previously analyzed 28-yr period (Parkinson and Cavalieri, 2008). 20 Seasonally, the sea ice extent trend for the Northern Hemisphere became even more negative by 31 % for summer and by 20 % for autumn. Regionally, the largest percent change occurred for the Bering Sea which changed from a small negative trend to a small positive trend in the yearly average, but neither trend was statistically significant. Hudson Bay, Baffin Bay/Labrador Sea, and the Greenland Sea regions had some- 25 what smaller negative yearly average trends for the 32-yr period, whereas the Seas of Okhotsk and Japan had a slightly greater negative trend. In contrast, the Gulf of St. Lawrence, the Kara and Barents seas, the Arctic Ocean, and the Canadian Archipelago all had considerably larger negative yearly average trends than they had for the 28-yr period.
The regional, seasonal, and even monthly sea ice trend variabilities presented here reflect the complex nature of the Arctic climate system. The principal driver of the Arc-5 tic's rapidly retreating ice cover has been the long-term upward trend in Arctic air temperatures, which has been reported to be more than twice the Northern Hemisphere trend (e.g., Johannessen et al., 2004), but this Arctic amplification is not constant (Chylek et al., 2009) and may even increase in coming decades (Serreze and Barry, 2011). Superimposed on this upward trend of Arctic air temperatures has been the in-10 fluence of atmospheric forcing on various spatial and temporal scales. The relationship between large-scale atmospheric circulation patterns such as the Arctic Oscillation and sea ice variability has been problematic (Stroeve et al., 2011b). Part of the problem is that even slight changes in the relative positions of low and high pressure systems and associated wind patterns will influence the growth or decay of sea ice (Maslanik et 15 al., 2007a). Atmosphere-sea ice relationships have been found to be regionally dependent, and for a full understanding of the observed regional sea ice trends, non-linear interactions among large-scale systems need to be considered (Liu et al., 2004).
The reduction of the summer sea ice cover and its thinning have resulted in greater ocean-to-atmosphere heat transfer, reinforcing the ice-albedo feedback mechanism 20 and leading to an accelerated loss of summer sea ice (Maslanik et al., 2007b). A complicating factor in understanding the processes driving the observed sea ice variability is that the loss of sea ice during summer has in itself had an effect on large-scale atmospheric circulation changes in the Arctic (Overland and Wang, 2010). These processes taken together with the observed rate of sea ice loss for the Arctic Ocean region during 25 summer portends an eventual seasonal ice-free Arctic. Introduction