Importance of combined winter and summer Arctic Oscillation (AO) on September sea ice extent

We examine the influence of winter and summer Arctic Oscillation (AO) on variations in the September Arctic sea ice extent (SIE). The winter and summer atmospheric patterns associated with year-to-year variations and detrended September SIE correlate with the positive winter AO and the negative summer AO, respectively. However, the interannual variations of winter and summer AO indices after 2007 are more weakly connected with year-to-year variations in the September SIE. Since 2007, the surface air temperatures over the Beaufort, Chukchi and East Siberian Seas are related to the interannual variations of the September SIE. Recent summer atmospheric patterns associated with the September SIE correlate with the summer AO pattern, but the summer anticyclonic circulation over the Arctic favours the recent low September SIE more than the seesaw pattern between mid- and high- latitudes. Recent winters’ positive AO have not contributed to the recent low September SIE because winter anticyclonic circulation over northern Eurasia is more directly connected with recent September SIE.


Introduction
Several studies have investigated and discussed the influence of important components in atmospheric circulation patterns over the Northern Hemisphere, such as the Arctic Oscillation (AO) and the North Atlantic Oscillation, and their influence on Arctic sea ice extent (SIE) (e.g. Deser et al 2000, Rigor et al 2002, Deser and Teng 2008, Stroeve et al 2011. The AO (Thompson andWallace 1998, 2000) represents the variability in pressure patterns over the Northern Hemisphere and is an important factor in the wintertime atmospheric conditions of the Northern Hemisphere. Rigor et al (2002) and Rigor and Wallace (2004) showed that a positive mode of the winter AO could induce negative anomalies in sea ice concentration during the following summer. Since the early 1990s, however, the trend of the winter AO index has been in a more neutral state and Arctic SIE has continued to decline (e.g. The dominant summer mode of the atmospheric circulation, namely, the summer AO was defined by Ogi et al (2004) using a separate empirical orthogonal function (EOF) analysis for each individual calendar month. In contrast, Thompson and Wallace (1998) used a single EOF analysis for all calendar months. The summer AO as defined by Ogi et al (2004) has a smaller meridional scale than the conventional AO presented by Thompson andWallace (1998, 2000). The conventional AO of Thompson and Wallace (2000)  September SIE may be correlated with the summer AO index defined by Ogi et al (2004). Ogi and Yamazaki (2010) showed that the seasonal sea ice retreat from May to September revealed a strong correlation with the summer AO. The summer AO also appear to play a role in summer weather and climate such as anomalous weather in mid-latitudes of the Northern Hemisphere (Ogi et al 2005, Tachibana et al 2010, Otomi et al 2012.
A number of previous studies have focused on either the effect of the preceding winter AO or the concurrent summer AO on September SIE. Within this study we consider the effect of the winter AO and the summer AO on the September SIE simultaneously. We think that it is important to consider the influence of both the winter AO and the summer AO on September SIE and its interannual variation due to the control which both dynamic and thermodynamic processes have on coupled ice motion, melt and decay. In this study, we examine whether wintertime and/or summertime AOs play an important role for year-toyear variations of Arctic September SIE.
Both year-to-year variations and multi-decadal trends of the September SIE were highly correlated with variations in sea-ice concentration over the marginal seas to the north of Alaska and Siberia (Ogi and Wallace 2007). Arctic sea ice has retreated rapidly in the recent decade and we consider separately the yearto-year variations and multi-decadal trends of the relationship between Arctic sea ice and atmospheric circulation. Finally we will discuss the influence of atmospheric circulations associated with AO (both winter and summer) on the September SIE with respect to both inter-and intra-annual variability.

Data
For the period from 1979 to 2014 we use the winter AO index as defined by Thompson andWallace (1998, 2000), which is available from the Climate Prediction Centre, and the summer AO index as defined by Ogi et al (2004).
The passive microwave record of daily sea ice concentration fields from 1979 to 2014 was obtained from the National Snow and Ice Data Center and used to determine September SIE. Sea ice concentration is derived from passive microwave brightness temperatures using the NASA Team sea ice algorithm (http:// nsidc.org/data/docs/noaa/g02135_seaice_index/ index.html). Within this study we analyze the raw and detrended September SIE data, while also calculating an index of the one-year difference in September SIE (DSIE). The ΔSIE index is derived from DSIE=SIE (Yr1)−SIE(Yr0), the change in September SIE from one year to the next.
Monthly fields of sea level pressure (SLP) and near surface (2 m) air temperature (T2m) were retrieved from the National Centers for Environmental Prediction/National Center for Atmospheric Research reanalysis data set from 1979 to 2014 (Kalnay et al 1996, http://esrl.noaa.gov/psd/data/gridded/ data.ncep.reanalysis.derived.html). Within this study 'winter' 'summer' and 'autumn' are defined as December-January-February, June-July-August, and September-October-November, respectively. As a result the winter of 1980 represents the mean of December 1979, and January and February 1980. Figure 1 shows the patterns in winter (top row) and summer (bottom row) SLP anomalies regressed on inverted DSIE (left column), raw (middle column) and detrended (right column) September SIE time series for the period of 1979-2014. The winter SLP anomalies regressed on −DSIE exhibit negative anomalies over the Arctic and the northern North Atlantic, especially strong negative anomalies are apparent over the northern North Atlantic. This winter pattern associated with −DSIE suggests an enhanced sea ice export out of the Barents Sea toward the North Atlantic Ocean during winter, which may be contributing to the significant negative trends in sea ice concentration during winter in the Barents Sea (Deser and Teng 2008, Ogi and Rigor 2013). The SLP anomalies based on the variations in the inverted detrended September SIE (figure 1(c)) are also similar to the SLP anomalies in relation to the −DSIE ( figure 1(a)). The positive SLP anomalies in relation to the inverted raw September SIE (figure 1(b)) are strongly apparent over northern Eurasia.

Results
All patterns of summer SLP regressed on the inverted September SIE fields are characterized by anticyclonic circulation SLP anomalies over the Arctic Ocean (figure 1 bottom row). Positive SLP anomalies across the Pole suggest a surface wind flow from the Beaufort and Chukchi Seas toward Fram Strait, and have favoured enhanced sea ice export from the Arctic toward Fram Strait. Figure 2 shows the SLP anomalies regressed on (a) winter AO and (b) inverted summer AO indices. Both winter and summer AO patterns are characterized by a seesaw between mid-and high-latitudes, though the seesaw is reversed between the two seasons. However, a smaller area of SLP anomalies is observed during summer than during winter, and the area of greatest SLP anomalies is displaced poleward during summer as compared to during winter.
The winter AO index shown in figure 2(a) is characterized by cyclonic SLP anomalies over the high-latitudes, especially over the Norwegian and Greenland Seas. These negative anomalies would have a strong cyclonic circulation with the flow directed toward and through Fram Strait, and southward along the East coast of Greenland. The pattern based on the winter AO resembles the winter atmospheric pattern regressed on the −DSIE and the inverted detrended September SIE shown in figures 1(a) and (c).
The summer SLP anomalies regressed on all September SIE indices (figures 1(d)-(f)) show similar patterns to the negative summer AO with a strong anticyclonic circulation over the Arctic Ocean. Ogi and Wallace (2007) showed that the summer anticyclonic circulation over the Arctic favoured low September SIE by producing an anomalous Ekman drift of ice from the marginal seas toward the central Arctic. The present study shows clearly that the dominant pattern of summer anticyclonic circulation in relation to low September SIE is similar to the negative phase of the summer AO.
To assess the relationship between the regression patterns in relation to September SIE and winter-  summer AO, the correlation coefficients are shown in table 1. All correlation coefficients are high, except for the coefficient between the raw September SIE and winter AO. These results are consistent with figures 1 and 2 where winter SLP anomalies dominate the negative anomalies over East Greenland and the summer positive anomalies over the Arctic influence the decrease of September SIE. In addition, these SLP anomalies are quite similar to the winter and summer AO patterns (figure 2), except for the raw September SIE and winter AO ( figure 1(b)).
To assess the influences of both winter and summer AO on the September SIE, using the winter and summer AO indices  4(b)). Since the 2007 September sea ice minimum open water has become more prevalent in the Beaufort, Chukchi and East Siberian Seas. As a result surface waters in these Seas are capable of absorbing much more solar heat during summer, thereby accelerating the ice albedo feedback loop (e.g. Babb et al 2016). The increased role of this process has altered the climatological state of these Seas and may be responsible for the weakened relationship between the AO and September SIE since 2007 even if the AO has the same pattern and same forcing both before and after 2007. This result is consistent with sea-ice concentration assessments summarized across the Arctic marginal ice zones in a recent study of open water fractions, and trends, in different areas of the Arctic as a function of season (Barber et al 2015).
Concerning the September SIE decrease, figure 5 shows winter and summer linear trends in SLP for comparison with the regression patterns in figures 1 and 2. The winter SLP has a strong positive trend over northern Eurasia that extends over the marginal seas north of Eurasia. The summer SLP has a strong positive trend over the Arctic Ocean and Greenland, and negative trend over northern Eurasia and Alaska. These atmospheric trends are similar to the regression patterns of the raw September SIE (figures 1(b) and (e)); therefore the large negative trend in September SIE appears to be connected with the positive anomalies during winter over northern Eurasia, and the positive anomalies during summer over the Arctic Ocean. During the winter of 2009/2010 the AO was in an extreme negative phase, yet the Arctic ice pack melted considerably during the summer of 2010. Stroeve et al (2011) pointed out that this reflects a pronounced difference in atmospheric circulation during the winter of 2009/2010 compared to the mean anomaly pattern based on past negative AO events. The summer atmospheric pattern in recent years has a persistent change relative to previous decades that suggests an enhancement of the Arctic Dipole (AD, Overland et al 2012). Wang et al (2009) argued that the summer AD pattern is actually different than the summer AO pattern. The AD anomaly produced a meridional wind anomaly that drove sea ice out of the Arctic from the Western to the Eastern Arctic toward Fram Strait. Recent record lows of Arctic sea ice were found to be triggered by the AD pattern (Overland et al 2012). The persistent changes in recent atmospheric patterns both in winter and summer might be the reason that the connection between the AO and September SIE appears to have weakened since 2007.

Discussion and conclusions
We have provided evidence that both winter and summer AO are correlated with year-to-year differences of the September Arctic SIE (DSIE). The atmospheric patterns are characterized by winter cyclonic circulation anomalies over East Greenland and summer anticyclonic circulation anomalies over the Arctic that are similar to the AO during both winter and summer. The winter and summer AO indices proved to be a good predictor of DSIE from 1979 to 2007. However, the predicted DSIE using the winter and summer AO indices was poor after 2007. Results from our study point out that autumn surface temperature over the Beaufort, Chukchi and East Siberia Seas are closely related to September SIE since 2007. Zhang et al (2013 showed that preconditioning toward thinner sea ice, a∼40% reduction in sea ice volume over 2007-2011, and the presence of ocean heat in the near surface created the 2012 record low Arctic SIE. Kwok and Cunningham (2010) showed that the loss of multiyear sea ice associated with melt in the Beaufort Sea during summer contributed directly to the depletion of Arctic multiyear sea ice coverage, especially after 2000. Serreze et al (2009) andScreen andSimmonds (2010) pointed out that the recent Arctic warming signal is linked to open water areas while the recent sea ice reduction during summer and early fall Winter and summer SLP trends anomalies (figure 5) are associated with the large negative trend of September SIE (figures 1(b) and (e)). Ogi and Wallace (2012) pointed out that the pressure circulation over the Arctic in the summer from 2007 to 2011 has been much more anticyclonic than during previous years and that change in the anticyclonic circulation has contributed to the lower trend in recent September SIE. In addition, wind forcing is more direct than the SLP, and the flow of sea ice from the Arctic toward Fram Strait both in winter and summer would act to reduce the Arctic sea ice. The AO patterns are consistent with the wind patterns, but the statistical correlations between sea ice and wind are higher than those between sea ice and SLP because the physical relationship is more direct. Actually, the summer winds over the Arctic during 2007, 2010 and 2011 contributed to the low Arctic SIE due to their strong anticyclonic wind anomalies over the Arctic and an anomalous ice export toward Fram Strait (Ogi and Wallace 2012).
The winter positive SLP trend over northern Eurasia is large and the positive anomaly is connected with the September SIE decline. The winter trend pattern is not similar to the winter AO pattern. Honda et al (2009) showed that the Siberian coast ice, especially the change of ice concentration over the Barents-Kara Seas is connected with anticyclonic anomalies extending from Eastern Europe to the West Siberian Plain. The winter sea ice concentrations over the Barents-Kara Seas have large negative trends (Ogi and Rigor 2013). This recent decrease may be connected with recent positive SLP anomalies. Holland and Stroeve (2011) also pointed out that the winter AO and summer ice extent correlation had weakened in recent years, and suggested that a weaker winter AO might contribute to the September ice cover extent. After the record low September SIE in 2007, it appears that ocean behaviour changed significantly from the preceding climatology.