Two distinct declining trend of autumn Arctic sea ice concentration before and after 2002

This study investigates the Arctic sea ice concentration trend during 1979–2021 and explores why the autumn Arctic sea ice loss is accelerated after 2002 and its trend declining center shifts from the Chukchi Sea to the Barents-Kara-Laptev Seas. Attribution analysis reveals that the enhanced summer sea ice concentration negative trend in large part explains the autumn sea ice concentration accelerating reduction, whereas it is the trend center shift of increased downward longwave radiation that accounts for mostly of the autumn sea ice concentration decline center shift. Further analysis suggests the downward longwave radiation trend is closely related to large-scale atmospheric circulation changes. A tendency towards a dipole structure with an anticyclonic circulation over Greenland and the Arctic Ocean and a cyclonic circulation over Barents-Kara Seas enhances (suppresses) the downward longwave radiation over Western (Eastern) Arctic by warming and moistening (cooling and drying) the lower troposphere during 1979–2001. In comparison, a tendency towards a stronger Ural anticyclone combined with positive phase of the North Atlantic Oscillation pattern significantly promotes the increase of downward longwave radiation over Barents-Kara-Laptev Seas during 2002–2021. Our results set new insights into the Arctic sea ice variability and deepen our understanding of the climate change.


Introduction
An evident negative trend in Arctic sea ice extent has been observed in all seasons and on a yearly average basis during the period of satellite observations (Parkinson et al 1999, Simmonds 2015).This sea ice loss not only impacts the local climate but also is assumed to affect the weather and climate in northern middle latitude (Deser et al 2004, Honda et al 2009, Inoue et al 2012, Wu et al 2013, Liptak and Strong 2014, Kug et al 2015, Sorokina et al 2016, Yao et al 2022).
The rising concentrations of atmospheric greenhouse gases are viewed as a main driving factor of this sea ice decline trend (Serreze et al 2007, Stroeve et al 2007).With the preferential loss of multi-year ice in comparison with relatively thin first-year ice, the downward trend of Arctic sea ice extent steepens from the end of 20 century (Comiso et al 2008, Serreze and Stroeve 2015, Lee et al 2017).However, most of the current climate models do not correctly capture this significant decrease (Stroeve et al 2007(Stroeve et al , 2012)), which indicates that our understanding of the controlling factors of sea ice melt may still be insufficient, especially in a warmer climate.
Recent rapid loss of Arctic sea ice is in part due to the enhanced ice-albedo feedback (Perovich andPolashenski 2012, Stroeve et al 2012) and strong positive ice-temperature feedback (Screen and Simmonds 2010).More and more studies emphasize the role of large-scale atmospheric circulation changes on this rapid reduction (Overland and Wang 2010, Ding et al 2017, Luo et al 2017, Chen et al 2018, Wang et al 2023).
The increased occurrence of Ural blocking is concurrent with the winter sea ice loss over the Barents-Kara Seas through poleward energy transports (Gong and Luo 2017, Chen et al 2018, Kim et al 2022).The positive phase of the North Atlantic Oscillation combined with Ural blocking is an optimal circulation pattern that transports warm and moisture air into the Barents-Kara Seas (Luo et al 2017).The anomalous poleward transport of atmospheric energy, by strengthening downward longwave radiation (DLR), can lead to sea ice loss during boreal winter over the Arctic, especially in the northern Barents-Kara Seas (Francis and Hunter 2006, Park et al 2015, Tian et al 2022).Liu et al (2021) indicated that the Pacific North American pattern is important for projections of Arctic climate changes, and that greenhouse warming and the resultant persistent positive Pacific North American trend is likely to increase Arctic sea-ice loss in summer by increasing heat and moisture fluxes from local processes and from advection of North Pacific air masses into the western Arctic.The more frequently occurring blocking over Beaufort Sea contributes to the local warming trend in summer by increasing shortwave radiation (You et al 2022).The negative North Atlantic Oscillation (equivalent to the intensified Greenland high) is suggested to reduce cloud cover over the Beaufort Sea in summer, promoting ice-albedo feedback due to increased downward shortwave radiation (Wang et al 2023).In addition, sea ice drift related to atmospheric circulation patterns may do contributions to the sea ice decline (Comiso 2006, Francis and Hunter 2007, Kwok et al 2005, Bi et al 2021).
The enhanced warm Atlantic water flowing into the Arctic region has also been recognized to be related to the Arctic sea ice reduction (Spielhagen et al 2011).It is likely the phase of the Atlantic multidecadal Oscillation can significantly modulate the Arctic sea ice extent and Ural blocking variability.The joint effects of the high sea surface temperature over Barents-Kara Seas and the quasi-stationary and long-lived Ural blocking events can explain the dramatic decline of the winter sea ice during recent decade (Luo and Yao 2018).On the other hand, Li et al (2018) considered the Arctic Atlantic sea ice loss during a cold Atlantic multidecadal Oscillation phase favors increased Ural blockings.These features are almost absent during the warm Atlantic multidecadal Oscillation phase.Kim et al (2020) revealed that the atmospheric circulation associated with Pacific decadal oscillation differently influences the variability of sea ice extent in the Pacific Arctic sector during summer by modulating poleward moisture transport.The impacts of atmospheric circulation associated with a positive Arctic dipole pattern on SIC change over different regions of the Pacific Arctic sector are dependent on the phase shifts of Pacific decadal oscillation (Bi et al 2021).Overall, sea ice trend cannot be simply attributed to changes of single climate variability index.
Previous studies pay much attention to the rapid pan Arctic sea ice extent decline or SIC reduction over a specific region.Actually, from a hemispheric viewpoint, since around 2000, not only the sea ice extent decline accelerates, but the pattern of Arctic SIC trend, especially in autumn, has changed greatly, which is less investigated.Therefore, in this work, based on the exploration of dominant factors influencing the autumn SIC trend, we attempt to reveal the atmospheric circulation changes before and after 21st century.

Data and method
2.1.Data Daily mean SIC for the period of 1979-2021 is obtained from the National Snow and Ice Data Center, which is with a spatial resolution of 25 km (Cavalieri et al 2012).We also use the daily 25-km EASE-Grid sea ice motion data set from the National Snow and Ice Data Center (Tschudi et al 2019) and sea-ice thickness data from the coupled Pan-Arctic Ice Ocean Modeling and Assimilation System (Zhang and Rothrock 2003).Daily mean dataset of the fifth generation atmospheric reanalysis from European Centre for Medium-Range Weather Forecasts with 1°× 1°latitude/longitude is adopted to analyze the large-scale meteorological conditions in this study (Hersbach et al 2020).The variables that we use include the downward shortwave radiation, DLR, sea surface temperature, total column water, temperature and 500-hPa geopotential height.Daily anomalies are obtained by subtracting the first four harmonics of the calendar day mean annual cycle, and then preprocessed to monthly anomaly for the following calculations, in which September-November represents the autumn and so on.

Method
In order to show the temporal evolution of Arctic sea ice, a SIC index is defined, which is by averaging the monthly SIC anomalies weighted by cosine (latitude) over the Arctic Ocean poleward of 52°N.We estimate linear trends based on a least squares linear regression (Strang and Freund (1986)).The Mann-Kendall test is adopted to statistically assess if there is a monotonic upward or downward trend of the SIC index over time (Mann 1945, Kendall 1975).
In the attribution analysis, a simple linear model x is used to quantitatively estimate the component of sea ice concentration decline forced by the variability of variable x.Firstly, the coefficient e is calculated based on the linear regression between the detrended SIC and x.Then, the SIC trend caused by the variable x can be estimated by multiplying the linear trend in x by this regression coefficient.The approach is based on the perspective that assumes the long-term trend may be realized through the same processes that cause its interannual variability (Park et al 2015).
To explore the atmospheric regime changes related to the trend of Arctic DLR, a DLR projection index ( ) is the DLR anomalies in every autumn, R ij is the observed DLR trend in the Arctic, and q is the latitude.An estimate of the interdecadal trend in various variables can be obtained based on their linear regression against the DLR projection index by multiplying the DLR index trend.

Trend transition of Arctic sea ice concentration
A linear downward trend of Arctic SIC in each season can be clearly seen in figure 1, as well as its annual mean, in which the autumn Arctic sea ice has the largest retreat (−0.28%/a) and then the summer (−0.27%/a).The downward trend of winter Arctic sea ice (−0.16%/a) is a little larger than that during the spring (−0.14%/a).In addition, a significant trend transition can be seen in all seasons around 2000, which is verified by the Mann-Kendall test.Thus, we divide the satellite record into two periods, labelled as the first period (P1) and the second period (P2).Note that there are slightly different turning points in different seasons.To be specific, P1 and P2 are 1979-1998 and 1999-2021 in spring, 1979-1999 and 2000-2021 in summer, 1979-2001 and 2002-2021 in autumn, 1979-2000 and 2001-2021 in winter.For an annual mean, P1 and P2 correspond to 1979-1999 and 2000-2021 respectively.We find that in each season the SIC declining trend during P2 accelerates compared with P1.Especially, the autumn SIC during P2 has the most dramatic declining rate, which goes near double the rate during P1.The Arctic sea ice declining trend and its acceleration can also be reflected in figure 2. In spring, the maximum sea ice retreat during P1 concentrates in the Greenland Sea, Barents Sea and Sea of Okhotsk.During P2, only the negative trend center over the Barents Sea deepens.In addition, a new significantly negative trend appears over the Bering Sea during P2, however, which shows a weak positive trend during P1.For summer, the negative trend centers move poleward, which concentrate in the Beaufort-Chukchi-Kara-northern Barents Seas during P1, strengthening and extending to the Beaufort-Laptev-Kara-northern Barents Seas during P2.In autumn, the negative trend still appears along the northern borders of the continents during P1, but it concentrates in the Chukchi Sea.As well, an evident accelerating decline during P2 compared with P1 can be seen, however, the minimum center is located over the northern Barents, Kara and Laptev Seas.This shifting of sea ice retreat center from P1 to P2 means that the autumn sea ice trend can not be simply considered as the sea ice retreat persisting of the preceding summer.The changes of linear trend in winter are very similar to those in spring, except that the negative trend centers over Greenland Sea, Barents Sea and Sea of Okhotsk all deepens during P2.
From the above analysis, it suggests that the sea ice retreat in different seasons is possibly driven by different mechanisms or strengths of the mechanisms.The most dramatic sea ice declining rate and spatial trend center shifts from the Chukchi Sea to Barents-Kara-Laptev Seas in autumn trigger our interests.Therefore, in the following, we attempt to answer the following questions: 1) Why does the negative linear trend of SIC accelerate in autumn?2) What makes the negative linear trend center of SIC shift from the Chukchi Sea to the Barents-Kara-Laptev Seas? 3) What changes occur to the associated mid-high latitude atmospheric circulations?

Mechanism of changes of sea ice concentration trend in autumn 4.1. Dominant factors influencing autumn sea ice concentration trend
To reveal the dominant factors influencing the autumn SIC variability, the following variables, such as the last season's SIC, sea ice shift, sea ice thickness, sea surface temperature, DLR and downward shortwave radiation, are considered.We note that downward shortwave radiation is not a main factor accelerating the sea ice melting trend because it is positively correlated with the SIC variation (figure S1).The autumn SIC trend contributed by other variables during P1 and P2 are shown in figures 3 and 4 respectively.We find that the autumn SIC decline during P1 is mostly contributed by the summer SIC, which emphasizes the particular importance of the last season's sea ice state (figure 3(a)).The significant summer SIC negative trend and its strong positive correlation with autumn SIC both do contributions (figures 3(f) and (k)).The next contributor to the autumn SIC decline is the enhanced DLR (figure 3(b)), which is partly due to its strong positive trend over the Beaufort and Chukchi Seas (figure 3(l)).The increased sea surface temperature and thinning sea ice contribute a little to the negative SIC trend over Chukchi Sea (figures 3(c) and (d)), in part because their trends are weak (figures 3(m) and (n)).Owing to a weak correlation with the SIC variation (figure 3(j)), the influence of sea ice convergence appears to be negligible to the autumn sea ice trend (figure 3(e)).As well as during P1, the negative trend of the autumn SIC during P2 is still dominated by the summer SIC (figure 4(a)).The accelerated retreat may partly owe to the enhanced summer SIC negative trend (figure 4(k)), which further confirms the importance of early summer sea ice state in explaining the abrupt declines of September sea ice extents (Zhan and Davies 2017).When we pay attention to DLR, it is found that due to the more enhanced DLR, especially over Barents-Kara Seas (figure 4 Our results suggest that though the enhanced summer SIC negative trend can in large part explain the autumn SIC accelerating reduction, however, it is the trend center shift of increased DLR that accounts for mostly of the autumn SIC trend center shift.Therefore, in the following subsection, we attempt to explore the atmospheric regime changes those are closely associated with these different DLR trends in P1 and P2, respectively, so as to reveal the possible reasons underlying the different trends of autumn SIC in these two periods.In this work because we focus on the SIC trend before and after 2002 separately, due to the limitation of linear attribution analysis method we used, the role of oceanic process may not be fully evaluated, such as the Atlantic multidecadal oscillation    A comparison of the estimated DLR trend pattern during P1 with the observed DLR trend over Arctic during P1 shows that these two patterns are indeed similar, and so does during P2 (figure 6).This implies that a large fraction of the interdecadal trend in the DLR arises from interannual fluctuations in the DLR fields as represented by the DLR index.Considering that the effective emitters of Arctic DLR are cloud liquid water and cloud frozen water, we firstly investigate the connection between the DLR and the total column water.As seen in the estimated trend of total column water against the projection indices during P1 and P2 (figures 7(a) and (d)), it indicates that the total column water pattern is similar to the DLR over the Arctic, with positive (negative) anomalies over the Western (Eastern) Arctic during P1 and stronger positive anomalies over Barents-Kara Seas during P2.In addition, we calculate the estimated trend of vertically averaged temperature between 1000-875 hPa.As can be seen, the trend patterns of vertically averaged temperature against the projection indices closely resemble those for the DLR, respectively (figures 7(b) and (e)), which implies the increased DLR is a direct result of the warming atmosphere, as well as an increase in total column water.The DLR, humidity and temperature are all affected by atmospheric circulation.From the estimated trend pattern of 500-hPa geopotential height, a tendency towards an anticyclonic circulation over Greenland and the Arctic Ocean is clearly seen during P1 (figure 7(c)).This anticyclonic circulation is also found by Ding et al (2017), who explored the summer atmospheric circulation related to September sea ice trend from 1979 to 2014.A difference is that another cyclonic circulation is found over Barents-Kara Seas in this research, which is probably due to the different time period.Associated with this atmospheric circulation trend, the DLR over Western (Eastern) Arctic increases (decreases) owing to the warm and moist (cold and dry) air in the lower troposphere.During P2, a stronger positive geopotential height anomaly dominates over the Barents-Kara Seas resembling a Ural blocking pattern.What is more, a positive phase of the North Atlantic Oscillation-like structure appears over its upstream (figure 7(f)), which is more favorable for the generation and maintenance of a downstream blocking (Luo et al 2016).The combined Ural anticyclone with the positive phase of the North Atlantic Oscillation structure during P2 is similar to that found in Luo et al (2017) and Chen et al (2018), who examined the winter rapid sea ice decline over Barents-Kara Seas during 2000-2015.This structure is the optimal pattern for the warm and moist air being transported into the Barents-Kara Seas, and thus leading to the local enhanced DLR trend.

Conclusions
In this work, we have explored the interdecadal trend of Arctic SIC during 1979-2021.It is found that Arctic SIC has decreased substantially in every season since the satellite record, and its declining trend becomes steeper since around 2000, in which, the autumn Arctic sea ice has the largest retreat, whose declining rate in recent two decades goes near double the rate before.A noteworthy phenomenon is that besides the sea ice decline acceleration, a trend center shift in the autumn SIC occurs, which focuses on the Chukchi Sea during 1979-2001 moving to the Barents-Kara-Laptev Seas during 2002-2021.By applying the attribution analysis, we reveal that the autumn SIC decline is mostly contributed by the summer SIC, which emphasizes the particular importance of the last season's sea ice state (Stroeve et al 2007).This point is quite different from winter Arctic sea ice decline revealed by Park et al (2015), who found that the last season's SIC contributes small to the SIC decline.The DLR plays a second role in the sea ice retreat, whereas, it determines the SIC trend center shift.The increased sea surface temperature and thinning sea ice contribute small to the sea ice decline and pattern shift, and the contribution by sea ice shift can almost be ignored.
Further analysis indicates the DLR trend is closely related to the humidity and temperature in the lower troposphere associated with the large-scale atmospheric circulation changes.This has been shown in the form of a schematic diagram in figure 8.During P1, a tendency towards a dipole structure with an anticyclonic circulation over Greenland and the Arctic Ocean and a cyclonic circulation over Barents-Kara Seas enhances (suppresses) the DLR over Western (Eastern) Arctic by warming and moistening (cooling and drying) the lower troposphere.During P2, a stronger Ural anticyclone combined with positive phase of the North Atlantic Oscillation pattern enhances more evidently the DLR over Barents-Kara-Laptev Seas.Though the causality cannot be confirmed from regression analysis, the anomalous atmospheric patterns associated with the DLR index suggest a substantial changes of mid-high latitude weather patterns from 1979-2001 to 2002-2021.It is known that the changes of mean atmospheric modes may be realized through changes in the frequency of occurrence of sub-seasonal weather events, which may also have an effect on the sea ice variation on the intraseasonal time scale (Wang et al 2020, Jiang et al 2021).Therefore, it is worthwhile to explore the changes of Arctic sea ice variation on the intraseasonal time scale and its possible mechanism in the future.

Figure 1 .
Figure 1.Time series of the Arctic sea ice concentration index (%) during 1979-2021 (black line) for (a) spring, (b) summer, (c) autumn, (d) winter and (e) annual mean, in which the yellow line represents the linear trend during 1979-2021, and pink (purple) dashed line represents the linear trend during P1 (P2), respectively.

Figure 2 .
Figure 2. Linear trends of sea ice concentration in P1 (%/a) for (a) spring, (b) summer, (c) autumn and (d) winter.(e), (f), (g) and (h) correspond to (a), (b), (c) and (d) respectively, but during P2.The dots indicate values that are statistically significant at the p <0.1 level for the Student's t test.

Figure 3 .
Figure 3. (a)-(e) Sea ice trend caused by five variables, (f)-(j) linear regressions of the five variables against the autumn sea ice concentration, and (k)-(o) linear trends of the five variables for summer sea ice concentration (SuSIC, first row), downward longwave radiation (DLR, second row), sea surface temperature (SST, third row), sea ice thickness (SIT, fourth row), sea ice motion and its divergence (SIM, fifth row) during P1.Dots and vectors indicate statistical significance at p < 0.1 level for the Student's t test.
(l)), DLR contributes greatly to the negative SIC trend over this region (figure 4(b)).It seems that the DLR trend changes may explain mostly of the autumn SIC negative trend center shift from Chukchi Sea during P1 to Barents-Kara-Laptev Seas during P2 (figures 3(b) and 4(b)).Though the trends of increasing sea surface temperature and thinning sea ice thickness are enhanced (figures 4(m) and (n)), they still explain a little fraction to the negative autumn SIC over norther-Barents, Laptev, East Siberian and Chukchi Seas (figures 4(c) and (d)).The influence of sea ice convergence still appears to be negligible to the autumn sea ice trend (figure 4(e)), because of its weak correlation with the SIC variation (figure 4(j)).
(Chen et  al 2018, Li et al 2018) and the Pacific decadal oscillation (Kim et al 2020).

Figure 5 .
Figure 5.The time series of the DLR projection indices (a) during P1, (b) during P2.The dashed line indicates the linear trend of the index.

Figure 6 .
Figure 6.The estimated trends for DLR (W m −2 yr −1 ) obtained by multiplying the regression coefficients and the trend in the DLR index (a) during P1, (b) during P2.

4. 2 .
Atmospheric regime changes that drive the DLR trend To explore the atmospheric regime changes before 2001, a DLR index is constructed by projecting the seasonally detrended DLR anomalies in autumn during 1979-2001 onto the DLR trend during P1 (as shown in figure 3(l)) weighted by cosine (latitude) over the Arctic 65°N poleward.Similarly, a DLR index after 2002 is onto the DLR trend during P2 (as shown in figure 4(l)).The time series of the DLR projection indices are shown in figure 5.As can be seen, the DLR indices both during P1 and P2 undergo a positive trend, as indicated by the dashed line in figure 5.In comparison with P1, the trend of DLR index in P2 accelerates.

Figure 8 .
Figure 8. Schematic diagram of the autumn sea ice concentration (SIC) trend center shift.The bottom level represents the surface, in which the circles represent the position of the negative sea ice concentration trend.The top level refers to 500 hPa, on which the wave trains including the positive and negative geopotential height trend centers are shown.On the middle level, the circles are trend centers of downward longwave radiation (DLR).