Asymmetry variations in Arctic summer onset and ending: Role of sea-ice melting

Previous studies found that in Arctic regions with severe sea ice melting, summer ending (SE) variations are significantly larger than summer onset (SO) variations in the past few decades. Based on short-term observations, researchers preliminarily suggested that radiation variations caused by an earlier melting onset could be the possible reason for asymmetric Arctic SO/ending variations (AASV). Based on observations and National Centers for Environmental Prediction/National Center for Atmospheric Research (NCEP/NCAR) Reanalysis I dataset, here we quantitatively characterize AASV by calculating the difference between the 11 year sliding trend of Arctic SO and SE. The results show that AASV positively correlates with sea ice melting in summer. The increased summer sea ice melting increases the area to absorb short-wave radiation in summer and then release more long-wave radiation to heat the lower atmosphere and delay the peak time of long-wave radiation releasing. The variations in radiation lead to a significant delay of the Arctic SE, with no significant variations in SO. We introduce CMIP6 historical and future simulations of 15 models to verify further the relationship between AASV and summer sea ice melting. Historical run reproduces the observed asymmetry, and future simulations under various warming levels show that AASV will vanish with disappeared melting variations or be strengthened with increased melting. The latter could delay freeze-up and further exacerbate the following years’ melting, which will enhance AASV. Furthermore, AASV will delay the onset and peak time of Arctic amplification.


Introduction
The sustained decline of sea ice has caused widespread concerns in recent years. The sea ice loss occurred in all months, with the strongest in late summer and the weakest in winter (Stroeve et al 2014). Since satellite records began in 1979, the Arctic sea-ice extent (SIE) in September has decreased by ∼50% (Stroeve and Notz 2018), accompanied by a significant decline in thickness (Kwok et al 2009), which may be linked to the prolonged Arctic summer melting season Polashenski 2012, Stroeve et al 2014).
Pinpointing the timing of the Arctic melting season is challenging (Markus et al 2009, Stroeve et al 2014, Huang et al 2017. Previous studies determined the onset of the melting and freezing season with satellite passive microwave data or other instrument data by calculating the daily melting (Markus et al 2009, Biss and Anderson 2014, Stroeve et al 2014, Serreze et al 2016. Based on these data, researchers found that variations of Arctic summer melting season onset is about twice that of the ending (Serreze and Stroeve 2015), which we summarized as the asymmetric Arctic summer onset (SO)/ending variations (AASV). The asymmetry is not invariable, but strengthened after the mid-1990s (Stroeve and Notz 2018 ). Notably, this asymmetry between SO variations and summer ending (SE) variations occurs only in the Arctic, especially in areas with significant summer sea ice decline, including Barents, Kara, Laptev, East Siberian, Chukchi, and Beaufort seas (Stroeve et al 2014).
The increase in air temperature under global warming may be an important cause of the sea ice loss and the prolonged summer (Perovich and Polashenski 2012, Notz and Stroeve 2016, Ding et al 2017. However, when the decline of SIE occurs, a series of feedback processes may further aggravate the sea ice loss Simmonds et al 2010, Taylor et al 2013) and amplify AASV. (Perovich et al (2008), Perovich et al (2011)) found that the total amount of solar energy absorbed during the summer was closely related to the melting onset. The earlier melting onset leads to the earlier development of ice-free areas that, in turn, enhance the ice-albedo feedback. As a result, the sea areas with melting sea ice can further release more long-wave radiation to heat the lower atmosphere during the subsequent cool season, which may be considered as an important reason for Arctic amplification effect appearing in the cool season (Dai et al 2019). Studies suggested that the amount of solar energy absorbed in summer could largely explain the observed delays in SE, and attribute the variations of the former to an earlier melting onset (Markus et al 2009, Stroeve et al 2014. Limited by a single trend of sea ice melting variations and the difficulty in capturing the variations of AASV through short-term observations, the relationship between summer melting trend and AASV is less discussed. Thus, the following key research questions are still open (a) why does the strong AASV only appear in the Arctic? (b) how is it related to the melting Arctic sea ice? and (c) what are the most probable consequences? We need to find an alternative dataset that could overcome the melting's limited short-term observations to answer these questions. Therefore, we use surface air temperature (SAT) to determine the Arctic SO and SE, and introduce coupled model intercomparison project phase 6 (CMIP6) simulations to explore the long-term relationship between AASV and Arctic sea ice melting variations and the influence of AASV. The rest of this paper is structured as follows. Section 2 introduces the data and methods. Section 3 shows the obtained results from observations and simulations, and section 4 presents the conclusions and discussion.

Data
This study uses the daily National Centers for Environmental Prediction/National Center for Atmospheric Research (NCEP-NCAR) Reanalysis-I dataset (Kalnay et al 1996), including the SAT, net short-wave radiation (NSWR), and clear sky upward longwave radiation (ULWR) from 1979 to 2020. We also use daily SAT data of 31 meteorological stations in the Arctic region obtained from the World Meteorological Organization (WMO) website (http://climexp. knmi.nl/showmap.cgi), monthly satellite data of NSWR and ULWR from the Clouds and the Earth's To explore the long-term feedback between AASV and sea ice melting, we use daily SAT and monthly sea ice concentration (SIC) data from historical  and future (2015-2100) simulations from the CMIP6 under various Shared Socioeconomic Pathways scenario. The scenarios considered here are low (SSP1-2.6), middle (SSP2-4.5), and high (SSP5-8.5). SSP5-8.5 consists high enough emission SSP scenario (SSP5) to generate a radiative forcing of 8.5 Wm −2 in 2100. SSP2-4.5 combines medium societal vulnerability (SSP2) with a forcing of 4.5 Wm −2 . SSP1-2.6 combines low vulnerability with low challenges for mitigation (SSP1) as well as a low forcing of 2.6 Wm −2 . 15 models, which include the required historical and future data (daily SAT and SIC data), were selected (table 1). Other models are not selected because they cannot meet the required daily data. The SIC was remapped onto 1 • × 1 • , and all the other data were interpolated to a 2.5 • grid using a bilinear interpolation scheme. In addition, we performed an independent analysis of the first simulation results for each model and then averaged all models with equal weighting to get the multi-model ensemble mean results.

Definition
Following previous studies (Huang et al 2017, Francis andWu 2020), the Arctic SIE is defined as the area with >15% ice concentration. However, there are challenges in determining the division of the Arctic seasons (Stroeve et al 2014, Serreze et al 2016. Since the methods used in previous studies to determine the onset of arctic melting by satellite passive microwave data are difficult to reproduce in CMIP6 simulations, we identify the Arctic summer through SAT. Referring to the seasonal division in mid-latitudes (Wang et al 2021), the Arctic SO and SE in this paper are defined as the day when the daily SAT exceeds (below) the 75th percentile of the 1979-2014 mean daily temperature. The rest of the year is the nonsummer days. It should be noted that, in the observation results, the summer SO/SE identified by using SAT is close to that identified by using microwave data (about 10 d) in previous studies (Markus et al 2009, Stroeve et al 2014.
We calculate the 11 year sliding trend of nonsummer days in the first half year and non-summer days in the second half year and then multiply the sliding trend by −1 to represent the SO and SE variations, respectively. A positive trend means the summer tends to be prolonged. The higher the trend, the earlier (later) the SO (SE). Therefore, the AASV can be quantified by calculating the difference between SE and SO variations. This approach allows us to capture the AASV variations and further investigate their relationship with sea ice melting. The summer sea ice melting is obtained by taking the difference between the SIE at SE and that at SO. The Arctic Amplification index is calculated as the 21 years sliding ratio of surface temperature trends between the Arctic and the rest of the northern hemisphere (Johannessen et al 2016, Davy et al 2018. It should be noted that the main results of this paper are not sensitive to the threshold.

Observed results
With the significant rise of SAT, the length of the worldwide summer has prolonged in recent decades, as it is in the Arctic. However, only in the Arctic do the SO and SE reflect the asymmetry, which means the delay of the SE is much more apparent than the early arrival of the SO in the Arctic (figures 1(a) and (b)), unlike anywhere else in Northern Hemisphere (NH). The observed and reanalyzed variations of SAT (figure 1(d)) and the daily melting of satellite records (Stroeve et al 2014) can capture this asymmetry. This asymmetry occurred especially in areas where Arctic sea ice melted significantly after the mid-1990s, including Barents, Kara, Laptev, East Siberian, Chukchi, and Beaufort seas. Therefore, we set these sea areas as key regions (Abbreviated as KR, ranging from 20 • W, 70 • N east to 140 • W, 80 • N) to explore the relationship between AASV and sea ice melting. Figure 1(e) shows little change in the SO in KR, while the SE is significantly delayed after the mid-1990s, contributing to the AASV. The asymmetry can also be found in most meteorological observation stations in KR ( figure 1(d)). In addition, we note that before the mid-1990s, the SE variations is negative in KR ( figure 1(e)), especially in the Barents Sea ( figure   S1(b)). This feature is related to the negative phase of Atlantic Multidecadal Oscillation (AMO), which can influence the anomalous oceanic heat transport toward the Arctic (Miles et al 2014, Zhang 2015 and regulate the sea ice melting in the sector of the Arctic by affecting the atmospheric blocking frequency over the Euro-Atlantic sector (Peings andMagnusdottir 2014, Omrani et al 2016). Similar variations could be found in Arctic SIE at the SO and SE, which may be related to the total freshwater inflow affected by AMO (Sun et al 2015).
The AASV is significantly correlated with the summer sea ice melting from SO to SE, and the correlation coefficient is as high as 0.48 ( figure 2(a)). It should be noted that the correlation coefficient between the two is 0.37 (0.44) after removing the linear trend during 1979-2020 (1996-2020). It indicates that their relationship may be affected by global warming and AMO. The shrinking SIE increases the ice-free sea areas with low albedo, enhancing the ability to absorb short-wave radiation during the summer. The energy is stored in the upper oceans and could be released through longwave radiation to heat the lower atmosphere in subsequent days (figures 2(b) and (c)). It can be found that, especially after the mid-1990s, the shortwave radiation absorbed in KR during June-July-August has increased significantly due to the increase of sea ice melting. Accordingly, more longwave radiation has been released to heat the lower-level atmosphere after September, which is an important reason for the delay of the SO. On the other hand, the lagged response time of the longwave radiation release to shortwave radiation absorption becomes longer, which can be found by the lag time variations in the maximum correlation coefficient between ULWR and NSWR (figure 2(d)). These variations in radiation well matched the spatial distribution and could be captured from reanalysis data and satellite observations (figure S2). Limited by short-term observations and monotonic trends in AASV and summer sea ice melting, it is hard to capture the variations of AASV, and here we introduce various scenario simulations from CMIP6.

Simulated results
Historical simulations of 15 models consistently reproduce the observed asymmetry. Although the intensity of asymmetry varies between the models, the averaged AASV in historical simulations also increased since the mid-1990s ( figure S3). Under SSP1-2.6, SSP2-4.5, and SSP5-8.5 scenarios, summers have lengthened significantly in both the NH and KR as expected (figures 3(a)-(c)). The earlier SO and the delayed SE both contributed. It can be found that the summer variations are more obvious in the Arctic than that in the NH the warming scenario intensifies (figures 3(d)-(f)). Under the SSP1-2.6 scenario, the SO variations in the KR region is close to zero, and the SE variations gradually decreased, and disappeared after the middle of the 21st century ( figure 3(g)).
Under the SSP2-4.5 scenario, the SO variations are slightly higher than 0 in the 21st century, and the positive SE variations decrease in the late 21st century ( figure 3(h)). While under the SSP5-8.5 scenario, the SO and SE variations continued to increase (figure 3(i)).
To further investigate the correlation and causal relationship between the AASV and SIE in KR, we calculate the AASV (brown lines in figures 4(a)-(c)) and the summer sea ice melting in KR (green lines in figures 4(a)-(c)). Around the middle of the 21st century, with the vanish, slowdown, or continuous intensification of sea ice melting variations, AASV also disappears under SSP1-2.6, decreases under SSP2-4.5 and maintains under SSP5-8.5 accordingly. This suggests that the intensity of AASV is closely related to the summer SIE variations. Under SSP1-2.6 scenario ( figure 4(d)), in the first half of the 21st century, AASV and SIE variations show a significant positive correlation, while in the second half of the 21st century, when AASV and SIE change close to zero, the correlation between them almost disappeared. A similar phenomenon can be observed under SSP2-4.5 scenario ( figure 4(e)). When global warming tends   to stabilize, the sea ice melted in summer will no longer increase (the SIE variations tend to zero), and the correlation between AASV and summer sea ice melting begins to decline significantly. Under the SSP5-8.5 scenario (figure 4(f)), due to the continuous intensification of global warming, the summer sea ice melting is maintained at a high level, and the sliding correlation between AASV and SIE changes is also maintained around 0.37. The sliding correlation coefficients between AASV and SIE show different changes under the three scenarios. This does not mean that their relationship is sensitive to future radiative forcing, but it indicates the coexistence relationship between the two. In other words, AASV occurs in the area where sea ice melts violently in summer. When the melting is no longer intensified, AASV disappears. Based on this relationship, especially under the SSP5-8.5 scenario, when the region with severe sea ice melting moves from the edge of the Arctic to the center, the region with the most significant occurrence of AASV also moves to the center of the Arctic ( figure S4). It is suggested that future work should consider capturing AASV dynamically rather than exploring AASV variations in fixed areas. In general, the shortening of the previous sea ice recovery period (delayed SE of the previous year and the advanced SO of the current year) exacerbates the summer sea ice melting, which may further strengthens the AASV.
In terms of the sliding correlation coefficient between AASV and summer sea ice melting, they influence each other. It is clear that the longer summers lead to an increase in the summer sea ice melting, while it is worth noting that the increased sea ice melting also further lengthens the summer, particularly delaying the SE rather than causing the SO to start earlier. Although it is difficult to capture the effects of AASV in short-term observations, it can be found that AASV is likely to have severe impacts on the arctic climate and ecology based on future simulations. The continuous strengthening of AASV will significantly reduce the length of the SIE recovery period, especially under the SSP2-4.5 and SSP5-8.5. Figures 5(a)-(c) show the daily SIE variations in KR (the value of each grid is the difference between the SIE on that day and the SIE of the previous day in KR). It can be found that the most intense melting will almost occur in July-August, with no significant advance. However, the time of maximum sea ice recovery will significantly delay, especially under SSP5-8.5. It is projected to delay from November to February of the next year. On the one hand, AASV makes the timing of SIE entering a rapid recovery period be delayed (delay about 1 month under SSP2-4.5 ( figure 5(b)), and about 2-3 months under ). On the other hand, AASV also results in a shorter time for SIE to remain at a higher level, leaving less time for the sea ice to grow thick and making it more likely to disappear in the coming year. Furthermore, the relationship between AASV and summer melting could greatly affect the Arctic amplification effect (figures 5(d)-(f)). Regardless of the intensity variations, the appearance and peak time of Arctic amplification will be significantly delayed by almost the same time with the delay of sea ice recovery. While the weakest Arctic amplification still occurs in June-August, hardly in advance. Although only the temperature gradient in the lower troposphere between the Arctic and mid-high latitudes of the NH during the occurrence of the Arctic amplification effect is reduced Simmonds 2013, Dai et al 2019), it is worth further investigating whether its effect on the weather in the middle latitudes has a seasonal change.

Conclusion and discussion
This paper investigates the feedback between AASV and summer sea ice melting based on reanalysis, observations and CMIP6 simulations. Unlike previous studies using daily melting, here we take the 75th percentile of the daily SAT as the threshold to determine the SO and SE. It could be found that the rate of delayed SE is significantly higher than earlier SO in areas with significant sea ice melting. Enhanced sea ice melting, the main contributor to AASV, increases the ability of the oceans to absorb more shortwave radiation in summer and then release more energy to heat the lower atmosphere later. To overcome the temporal limitations of observations, we used 21st century simulations of CMIP6 to capture AASV variability and investigate the role of sea ice melting in it. In CMIP6 simulations, AASV gets stronger (weaker) as sea ice melt increases (decreases) and disappears when the sea ice melting trend stalls. Especially under the scenario of SSP5-8.5 with the aggravation of global warming, the enhanced AASV contributes to the shortening of the sea ice recovery period and further promotes the acceleration of sea ice melting, which reflects the positive feedback between AASV and sea ice melting.
AASV significantly contributes to the Arctic summer extension and determines its direction, providing a reference for the schedule of the Arctic route (Wei et al 2020) as well as causing more damage to the Arctic ecology. The positive feedback between AASV and sea ice melting not only affects local summer variations in the Arctic but also may affect the weather and climate on a larger scale (Overland et al 2011, Screen and Simmonds 2013, Cohen et al 2014, Luo et al 2018. For example, AASV is likely to delay the onset and peak time of the Arctic amplification effect by 1-2 months in the late 21st century. Therefore, further research is needed to determine how such changes might affect the winter-spring climate in mid-high latitudes. In addition, key issue for future studies is to design more sensitivity experiments and use the sea ice data with a higher temporal resolution to quantify the relationship between AASV and sea ice melting.