2.1 Data

SLP and wind data used in this study were obtained from the NASA Modern-Era Retrospective Analysis for Research and Applications, version 2 (MERRA-2), atmospheric reanalysis (Gelaro et al., 2017) available from 1980 to present. MERRA-2 has a 1∕2^∘ (lat) and 2∕3^∘ (long) horizontal resolution, with 72 vertical levels topped at 0.01 hPa. Surface wind reanalysis is defined at 2 m above the surface.

The National Snow and Ice Data Center (NSIDC) provides the monthly-mean SIE data as the Sea Ice Index, version 3 (Fetterer et al., 2017). SIE is defined as the total area covered with at least 15 % concentration of sea ice for the given area. NSIDC also provides the sea ice motion (SIM) data (Polar Pathfinder Daily 25 km EASE-Grid Sea Ice Motion Vectors version 4; Tschudi et al., 2019) and sea ice age (SIA) data (EASE-Grid Sea Ice Age version 3; Tschudi et al., 2016). These data were produced from various sources of satellites: the Advanced Microwave Scanning Radiometer for the Earth Observing System (AMSR-E), Advanced Very High Resolution Radiometer (AVHRR), Scanning Multi-channel Microwave Radiometer (SMMR), Special Sensor Microwave/Imager (SSM/I), Special Sensor Microwave Imager/Sounder (SSMIS), and National Centers for Environmental Prediction/National Center for Atmospheric Research (NCEP/NCAR) and the buoy observations from the International Arctic Buoy Programme (IABP). The horizontal resolution of SIM and SIA is 25 and 12.5 km, respectively, and the data cover the area from 48.4 to 90^∘ N with Equal Area Scalable Earth Grid (EASE-Grid) for the period of 1978 to 2016 and 1984 to 2016, respectively.

Sea ice concentration (SIC) data were obtained from the National Oceanic and Atmospheric Administration (NOAA) Optimum Interpolation Sea Surface Temperature version 2 (OISST v2; Reynolds et al., 2002), which has a horizontal resolution of 1^∘ with the data period of 1982–2017.

This study also used the sea ice thickness data from the Pan-Arctic Ice Ocean Modeling and Assimilation System (PIOMAS) reanalysis produced by the Polar Science Center (Zhang and Rothrock, 2003). The original data with a horizontal resolution of approximately 22 km were regridded onto the $\mathrm{1}{}^{\circ }\text{lat}\times \mathrm{1}{}^{\circ }\text{long}$ grids for the analysis.

2.2 Methodology

The EOF analysis was conducted to obtain leading modes of atmospheric variability in SLP over the Arctic region (70–90^∘ N) during summer (June–August), which is same method used to define the AD (Wu et al., 2006; Watanabe et al., 2006; Wang et al., 2009; Overland and Wang, 2010). The analysis was first conducted for the entire analysis period of 1980–2017 and then for the two separate periods of 1982–1997 and 1998–2017 to examine the decadal changes in the leading modes of atmospheric circulation. The separation of the analysis period before and after 1998 is identical to Yang and Yuan (2014). There was a robust decadal change in the relationship between the atmospheric circulation and the sea ice extent in the mid-1990s. The leading three modes were analyzed in this study, which accounts for more than 80 % of the total Arctic SLP variability in summer. Due to the data availability, sea ice motion was analyzed just for 1982–2016 and sea ice age was analyzed for 1984–2016.

The convergence of sea ice motion is calculated as

where V is ice motion vector and each u and v indicate the zonal and meridional direction, respectively.

3.1 Atmospheric circulation in the Arctic

Figure 1 shows the interannual variation of the Arctic (70–90^∘ N) surface temperature (Fig. 1a) and SLP (Fig. 1b) anomalies during JJA, as well as SIE anomalies in September (Fig. 1c). Under global warming, there is an increasing trend in surface temperature since 1998, which is about 3 times larger (0.16 ^∘C per decade) than that for the period before 1997 (0.6 ^∘C per decade). A more rapid decrease in SIE is also observed in the recent period ($-\mathrm{1.1}\times \mathrm{106}$ km² per decade), compared with the earlier period ($-\mathrm{0.4}\times \mathrm{106}$ km² per decade). This accelerated melting trend is reported by Comiso et al. (2008). On the other hand, the negative trend of SLP anomalies before 1997 turns into weakly positive after 1998.

Figure 1(a) Surface temperature and (b) sea level pressure anomalies in the Arctic region north of 70^∘ N during June–July–August (JJA). (c) is the sea ice extent (SIE) in September in the same area. The anomalies are the departures from the average of 1981–2010. The dashed line shows the trend before and after 1998.

Download

We investigate if these changes in the trend of surface temperature, SLP, and SIE were caused by any changes in the large-scale atmospheric circulation pattern in the mid-1990s. Figure 2a presents the climatological-mean surface wind pattern during JJA (vector) and the interannual variability of sea ice in September (shaded) as measured by the standard deviation of sea ice extent for the period 1982–2017. The secular negative trend in sea ice concentration was removed from each grid to consider only the interannual variation in 1982–2017. In summer, most of the sea ice variability is along the edge of the climatological-mean sea ice distribution (black line, denoted as the border of 15 % of sea ice concentration), which is particularly large in the Pacific section such as in the Beaufort Sea and the East Siberian Sea. The time-mean surface wind shows the cyclonic (counterclockwise) circumpolar circulation around the North Pole with an anticyclonic (clockwise) circulation over the Beaufort Sea. Figure 2b shows the difference in the interannual variability between the early and the recent period. Because of more sea ice loss in the recent period, the boundary of sea ice extent shifts poleward. This makes the region of interannual sea ice variability move northward and closer to the North Pole. The difference map of surface wind between the two periods shows a basin-wide anticyclonic motion with a strong southward outflow to the North Atlantic. This surface wind pattern is consistent with the finding by Ogi and Rigor (2013), who attributed the rapid decreasing trend of SIE after 1996 to the anticyclonic surface wind pattern in the Arctic. Zhang et al. (2016) also suggested the intensification and stabilization of the Beaufort Gyre in the recent period. Comparing the boundary of the sea ice area between the two periods, sea ice loss in the recent period is much larger in the Pacific section than in the Atlantic. As the sea ice area retreats poleward, it moves the center of action in sea ice variability to the north, showing a significant increase in the variability in the Beaufort Sea and in the East Siberian Sea. It is accompanied by sea ice variability decrease in the coastal sea in the Pacific section due to the overall retreat of sea ice. Noticeably, sea ice concentration over the Odden, a tongue of sea ice over the northeastern coast of Greenland (Shuchman et al., 1998), does not show a significant difference between the two periods. This spatial difference in sea ice melting is also shown in previous studies (e.g., Parkinson, 2014a; Parkinson and Cavalieri, 2012). It is intriguing what causes this asymmetric sea ice melting pattern, and this study further examines the change in the atmospheric circulation with a particular focus on the impact of the TDS on sea ice variation.

Figure 2(a) Climatological-mean (1982–2017) pattern of surface wind at 10 m during JJA (vector, m s⁻¹) and the SIE in September (solid line). The shaded area shows the standard deviation from the detrended yearly variation of September SIE in each location. (b) The difference of time-mean surface wind (vector, m s⁻¹) and the standard deviation of September SIE between the recent and the early period (1998–2017 minus 1982–1997; shaded). The solid lines indicate the SIE for the early (black) and the recent (red) period, respectively.

To obtain dominant variability modes of atmospheric circulation, the EOF analysis was conducted for seasonal-mean SLP anomalies in the Arctic. Figure 3 shows the three leading EOF modes. The first mode (EOF1) represents the AO, explaining ∼50 % of the total variance. The corresponding principal component time series (PC1; Fig. 3d) is highly correlated with the AO index produced by the NOAA Climate Prediction Center (CPC) (r=0.95). The center of the negative SLP anomalies is located over the Arctic and this anomaly extends to Greenland. The surface wind vector regressed onto the PC1 time series (Fig. 3a) shows a cyclonic circulation over the Arctic, corresponding to the positive phase of AO, with a large-scale southerly flow from the Atlantic, via the Fram Strait, to the Arctic Ocean. The second mode (EOF2) shows a dipole structure, with one center over the Laptev Sea and the Kara Sea and the other over the Beaufort Sea. Variation of this dipole pattern could enhance or reduce meridional flow crossing the Arctic and affect the TDS. This low-level flow may play a significant role in sea ice motion, which will be discussed later. The third mode (EOF3), which shows a comparable magnitude in variance compared to that of the AD (17.2 % vs. 11.5 %), has not been addressed much in previous studies. This study defines this mode as A3, which is characterized by a cyclonic circulation over the East Siberian Sea and an anticyclonic circulation occupying a broad area from over the Barents Sea to Greenland. This dipole SLP pattern is associated with the atmospheric flow squared to the TDS as it penetrates through the North Pole. The remaining modes from the EOF analysis account for a much small variance – lower than 6 %. Hereafter, this study focuses on the three leading modes (AO, AD, and A3) that explain almost 80 % of the total SLP variability, and the normalized PC time series are used for representing a temporal variation of each mode and the subsequent regression analysis. The positive phase is more dominant prior to 2007 in the AD index (PC2), and then it exhibits negative values successively, including the years of sea ice minima in 2007 and 2012. This feature is also supported by Overland et al. (2012). On the contrary, the A3 index does not seem to have a noticeable long-term trend.

Figure 3Leading EOFs of JJA-mean SLP anomalies (shaded) for (a) AO, (b) AD, and (c) A3. (d, e, f) are the corresponding time series, respectively. The regression patterns of surface wind onto the corresponding EOF time series are also given in (a, b, c) (vector).

Figure 4The three leading EOFs of JJA-mean SLP (shaded) in the early (1982–1997, a, c, e) and the recent (1998–2017, b, d, f) period. (a, b) for the first mode, (c, d) for the second, and (e, f) the third mode. The regression pattern of surface wind anomalies (vector) is also shown in each map.

3.2 Decadal change in atmospheric circulation and sea ice

This study further examines if the atmospheric circulation pattern has changed significantly on a decadal scale, as well as its effects on the sea ice distribution change. The dominant atmospheric circulation patterns in the Arctic were again obtained from the EOF analysis, applied for the two different periods, before and after 1998. It should be noted that the results were not sensitive when any year between 1996 and 2000 was chosen for the period separation. Figure 4 compares the leading EOF modes of monthly SLP anomalies for the two periods. The AO mode is captured as the first mode for both periods, followed by AD and A3, as in the sequence from the analysis to the entire period. Each corresponding PC time series shown in Fig. 5a–c (dashed lines) has a high temporal correlation with those in Fig. 3, suggesting that each EOF modes are robustly identified as internal modes, regardless of analysis time. The significant difference is in the EOF spatial pattern of the AD. In the early period, one of the dipoles is located in the Kara and the Laptev Sea and the other in the Beaufort Sea so that the surface wind across the Arctic is parallel to the dateline. On the other hand, the center of variability tends to shift eastward in the recent period. In particular, the variability maximum in the western hemisphere shifted from Queen Elizabeth Islands to Greenland. The pattern correlation over the western hemisphere (60–90^∘ N, 0–180^∘ W) is estimated to test the statistical difference between the two patterns. Estimating the statistical significance with the effective sample size according to Bretherton et al. (1999; see the Supplement), the pattern correlation between the early and recent AD shows 0.58, which suggests the early and the recent AD patterns are not statistically same. On the other hand, the pattern correlation between the early and the recent AO patterns is around 0.99, suggesting no statistical difference. This change in the SLP pattern accordingly changes surface wind direction. In particular, it shows a clear development of the southerly flow in the recent period, blowing from the North Atlantic to the Arctic Sea via the Greenland Sea. This southerly flow does not exist evidently in the early period.

Figure 5Time series of the three SLP EOFs (solid black) for (a) AO, (b) AD, and (c) A3. Time series of September SIE are also presented in each figure (solid red). In (a, b, c), the EOF time series from the separate analysis for the period before and after 1998 are also compared (dashed lines in black). (d) is the correlation coefficient between each EOF time series and the September SIE for the entire (1980–2017), early (1982–1997), and recent (1998–2017) period, respectively. A linear trend in each EOF time series was eliminated before calculating correlation. Filled color indicates statistically significant at the 95 % confidence level.

Download

The change in the spatial pattern of the AD can impact the sea ice variability significantly. As shown in Fig. 5d, the correlation between the AD and the September SIE time series dramatically increased in the recent period ($r=-\mathrm{0.05}$ in the early period vs. r=0.57 in the recent). Although we remove the linear trend in each PC time series and SIE before the correlation analysis to avoid any spurious relationship influenced by the trend, the result is not qualitatively different when including the trend. It is also noted that, when we eliminate the SIE trend in each period for considering the rapidly declining trend in the recent period, the correlation between the AD and the September SIE increases significantly in the recent period (r=0.57). This feature suggests that the SIE variability becomes more connected with the AD in the recent period. The strong relationship between the AD in boreal summer and the September SIE is consistent with the results from Wang et al. (2009). It is also noted that the correlation between AO and SIE shows a statistically significant correlation at the 95 % confidence level for the entire analysis period (r=0.33). Rigor et al. (2002) showed the AO exhibited a strong relationship with the Arctic sea ice in winter, and Park et al. (2018) further highlighted the connection between wintertime AO circulation anomalies on the following summer sea ice extent. Williams et al. (2016) also suggested a lagged impact of the AO in winter affecting sea ice concentration in summer. This study hypothesizes that, although both the AO and the AD contribute to the interannual variability of the September SIE, the strong relationship with SIE is largely due to the change in the spatial pattern of AD in the recent period. Responsible physical and dynamical mechanisms will be discussed in more detail in the next subsection.

Although our analysis focuses on the interannual variability, the result provides an implication on the recent negative SIE trend as well. The positive trend of SLP in the recent period (Fig. 1b) corresponds with the negative trend both in AO and AD (Fig. 3d and e, respectively). The decline of SIE caused by a more frequent occurrence of the negative AO seems to be related to the warming and positive SLP anomalies in the pan-Arctic. This decline of SIE can be accelerated by a more frequent occurrence of the negative AD through enhancing TDS, which is discussed more in the next section.

We also note that the relationship of the sea ice with A3 is negligibly small for 1980–2017, where the correlation tends to decrease rapidly in the recent period. We focus more on the AD to account for sea ice variability in the next section.

3.3 Mechanisms of sea ice response

We next analyze the September sea ice concentration (SIC) to explain the spatial distribution of sea ice variability. Figure 6 shows the regression pattern of September SIC onto the AD index (PC2) in each period. The regressed patterns are plotted with respect to the negative AD phase in order to better represent the southerly wind-induced ice loss over the Pacific sector of the Arctic. During the negative phase, the September SIC seems to increase over the Fram Strait in the early period with a decrease in the Pacific sector over the East Siberian Sea, the Chukchi Sea, and the Beaufort Sea in the recent period. Comparing SIC between the early and the recent period, sea ice over the Fram Strait remains almost the same in September, suggesting less contribution to the entire Arctic SIE. In contrast, the SIC reduction in the Pacific sector is more dominantly contributing to the decrease in the entire Arctic SIE.

Figure 6Regression patterns of sea ice concentration (shaded,  %) and the surface wind anomalies (vector, m s⁻¹) onto the AD index for (a) the early (1982–1997) and (b) the recent (1998–2017) period. The AD index was reversed in sign for the melting phase of sea ice. Colored lines indicate the time-mean sea ice concentration of 15 % (red dotted), 30 % (red dashed), 50 % (red solid), and 80 % (blue solid) in each period.

Atmospheric circulation associated with the AD is closely linked to the 850 hPa temperature change (see Fig. S1 in the Supplement). As the center of SLP anomalies shifts eastward in the recent period, the warm area over the Beaufort Sea moves to Greenland. Considering the surface wind anomalies in Fig. 6, the Greenland region seems to be affected by warm temperature advection from the southerlies along the Baffin Bay (see Fig. S2). However, these pattern changes cannot explain increased sea ice melting over the Pacific sector of the Arctic in the recent period as shown in Fig. 6. Therefore, this study further examined the decadal changes in the net downward surface heat flux associated with the AD. The surface heat flux change driven by AD (Fig. 7) corresponds well with the sea ice concentration change pattern shown in Fig. 6. The net downward surface heat flux anomalies are negative in the sea-ice-increasing region, such as in the Atlantic sector of the Arctic Sea, and positive for the rest of the regions where the sea ice decreases. The surface heat flux anomalies are mostly contributed by the changes in the shortwave radiation terms (see Fig. S3). In particular, the upward shortwave radiation tends to decrease significantly in the East Siberia, Chukchi, and Beaufort seas where the sea ice melting signal by AD is pronounced in the recent period (Fig. 6b). This process is also reflected in the reduction of surface albedo due to the sea ice decrease (Fig. 8). The response of surface albedo becomes larger and much clearer in the recent period while the spatial pattern of surface albedo regression onto the AD index is essentially the same. Although the response in the surface heat flux suggests the positive sea-ice–albedo feedback process engaged in the Arctic sea ice variability associated with the AD, the question on what mechanisms are primarily responsible for triggering this feedback process remains.

Figure 7Regression patterns of net surface heat flux (sensible + latent + net shortwave + net longwave, W m⁻²) onto the AD index for (a) the early (1982–1997) and (b) the recent (1998–2017) period. The heat flux is defined as positive for the downward trend, and the AD index is reversed in sign to represent the melting phase of sea ice. The dotted area indicates the statistically significant region at the 95 % confidence level.

Figure 8Surface albedo changes associated with the summer AD in (a) the past (1982–1997) and (b) the recent (1997–2017) periods, respectively. The AD index is reversed in sign before regression.

The difference in the SIC response in the Pacific sector seems to be related to the change in sea ice dynamics. In summer, sea ice tends to move along the geostrophic wind (Thorndike and Colony, 1982; Ogi et al., 2008). The sea ice motion associated with the AD (cf. Fig. 9a and b) becomes faster in the central Arctic, particularly around the boundary of the sea ice extent (Fig. 9c) in the recent period. Through this, the sea ice is drifted more toward the Norwegian Sea and discharged to the North Atlantic. This sea ice motion change is consistent with the change in the surface wind driven by the AD (cf. Fig. 9d and e), suggesting that the sea ice is drifted toward the direction of the surface wind to a good approximation. Although the sea ice motion can be depicted only over the sea-ice-covered area, the sea ice motion can be inferred from the pattern of surface wind in its final stages in discharging and melting in the North Atlantic. This feature is also evident in the sea ice motion divergence patterns, in which the divergence anomalies become larger in the Chukchi and the Beaufort seas due to more enhanced transpolar sea ice drift motion toward the Atlantic sector (Fig. 10).

Figure 9Regression pattern of sea ice motion onto the AD index and its difference (a, b, c) and regression pattern of surface wind onto the AD index and its difference (d, e, f) in the early (1982–1997, a, d) and the recent (1998–2017, b, e) period, respectively. The shaded areas denote the sea ice age (a, b, c) and the sea ice thickness (d, e, f) in September averaged over each period and its difference.

Figure 10Regressed sea ice motion vector (arrow, cm s⁻¹) and its convergence (shaded, 100 per day) onto the AD index in (a) the early (1982–1997) and (b) the recent (1998–2017) period, respectively.

Comparing the surface wind pattern associated with the AD in the past and the recent periods (Fig. 9f), the largest difference is shown over the Fram Strait. In the past, the transpolar wind has backed into the Atlantic side of the Arctic Sea (i.e., Barents and the Kara Sea, Fig. 9d), whereas it has extended further south to the Atlantic recently (Fig. 9e). This enhanced outflow should be responsible for the sea ice drift and large sea ice discharge through the Fram Strait (Fig. 9b). From these figures, we can see that northerly winds have been strengthened from the central Arctic to the North Atlantic in the recent period to provide a more favorable condition for sea ice to be discharged to the North Atlantic.

Another contribution may come from the change in oceanic heat transport. The surface wind over the Chukchi Sea is meridional in the past period, while it becomes zonal in the recent period (Fig. 9d and f). The sea ice over the Pacific sector may be affected by relatively warm water inflow from the North Pacific through the Bering Strait (Shimada et al., 2006; Woodgate et al., 2010; Carmack et al., 2015; Serreze et al., 2016b), and this process can be enhanced in the recent period by the increased zonal component of surface wind blowing over the Chukchi Sea. The zonal surface wind anomalies can provide a more favorable condition for the warm water transport in a meridional direction toward the Arctic sea ice regions via Ekman transport. The enhanced zonal wind component in the Pacific sector is connected with the intensification of the anticyclonic circulation anomalies over the Beaufort Sea as shown in Fig. 6. Wu et al. (2014) also indicated a significant intensification of the Beaufort High during summer and autumn and attributed it to the rapid decline of sea ice over this region in the recent decade. This mechanism seems to work during 2007–2012 when the Beaufort High was persistently strong (Overland and Wang, 2010). It also seems to explain relatively weak sea ice melting during the recent three summers (2015–2017), when the cyclonic circulation anomalies were likely to prevent sea ice from melting.

One of the remaining factors to be considered is the changed property of Arctic sea ice. The sea ice becomes younger (Fig. 9c) and presumably thinner in thickness. This thickness change is also confirmed well by the PIOMAS reanalysis data (Fig. 9f). It suggests the Arctic sea ice becomes more vulnerable to dynamical forcing such as atmospheric wind and oceanic heat transport, especially in the melting phase.