Introduction

Comprehensive and sustained observations of Arctic sea ice concentration (SIC) began in late 1978 with the launch of the passive microwave radiometer onboard the Nimbus 7 satellite1. Over the passive microwave record, the pan-Arctic minimum sea ice extent, which occurs during September, has decreased by ~14% per decade2. In addition, there have been recent trends toward a younger3, thinner4 and more mobile5 Arctic ice pack. These changes have resulted in stresses on already fragile ice-dependent ecosystems across the Arctic6,7. This retreat is also opening up new shipping routes across the Arctic Ocean that offer reduced transit times and costs8 as well as increased opportunities for resource extraction9. However, the increase in mobility of the ice pack may, in a counter-intuitive sense, lead to increased marine hazards10.

The most significant changes in summer Arctic ice cover have occurred in the Western Arctic, i.e. the Canada Basin along with the Beaufort and Chukchi Seas, where the September SIC trend has exceeded -20% per decade11. Within the Chukchi Sea, situated just north of Bering Strait (Fig. 1), the open water period has lengthened by ~80 days since 197912. This extension is primarily the result of a delay in the advance of sea ice in the fall, as opposed to an earlier retreat in the spring12. The ice-albedo feedback process and enhanced oceanic heat transport into the region through the Bering Strait have contributed to this loss12,13.

Fig. 1: Spatial and temporal variability in SIC (%) over the Chukchi Sea 1979–2024 as represented in the NOAA/NSIDC Climate Data Record.
figure 1

Monthly mean SIC during September: a 1979–2001; b 2002–2023 and c 2024. d Time series of the monthly mean SIC during September 1979–2024 as averaged over the Eastern (red curve) and Western (blue curve) Polygons. e Time series of the monthly mean SIC for the Eastern Polygon from September 2023 to September 2024 (red curve) with the black curve representing median ice concentrations for the period 2002–2023 and the shading representing values spanning from the 5th to 95th percentile. In (a) and (b), place names of interest are indicated as well as the locations of the Eastern (EP) and western (WP) Polygons with the 150 m isobath indicated by the dashed curve.

Full size image

The surface wind field plays an important role in the distribution of sea ice in the Western Arctic14,15. Climatological winds during the ice season are anti-cyclonic and are associated with an area of higher sea-level pressure known as the Beaufort High16. As a result, the Chukchi Sea typically experiences easterly surface winds and ice motion17. These winds can be enhanced if a low-pressure system is present to the south over the northern Bering Sea, resulting in an enhanced meridional pressure gradient across the southern Chukchi Sea18. Recently, there have been winters where the Beaufort High collapsed due to the intrusion of cyclonic low-pressure systems from either the subpolar Atlantic or Pacific Oceans, which has resulted in a reversal of the direction of the winds and ice motion, from easterly to westerly19,20, that can lead to localized thickening of sea ice along the coast of the Beaufort Sea21.

The climatological easterly winds over the Chukchi Sea form latent heat (a.k.a wind-driven) polynyas along its Alaskan coast22 and on the downwind side of Wrangel Island23. In addition to their role in ice production, these polynyas also help maintain the Arctic halocline by producing cold and salty water from brine rejection and air-sea interaction24,25. An ice thickness dipole has also been identified near Wrangel Island, with an area of higher ice thickness on the upwind side of the island and an area of lower ice thickness on the downwind side26.

Previous work on understanding the processes responsible for anomalous ice conditions throughout the Arctic has emphasized the important role played by the ice thickness budget27,28,29,30. Changes in ice thickness (see Methods) are a function of both ice “advection” (which here is used as a shorthand term for convergence or divergence associated with ice motion), and ice “production” (i.e., growth or melt).

In September 2024, sea ice was present along the Siberian coast of the Chukchi Sea—the first time since the turn of the century that sea ice had been observed in the region31. In this paper, we will use space-based passive microwave SIC observations, a coupled ice-ocean model, and an atmospheric reanalysis (See Methods) to characterize the processes responsible for the return of summer sea ice to the Chukchi Sea after a 25-year hiatus.

Results

Characterization of the Chukchi Sea’s September 2024 anomalous ice cover

Figure 1 characterizes the spatial and temporal variability of mean September SIC in the Chukchi Sea as represented in the NOAA/NSIDC Climate Data Record1 (See Methods). Based on the results shown in this figure, we have chosen to use two different climatological periods, 1979–2001 and 2002–2023 to characterize interannual variability in this area. The Pettitt breakpoint test32 confirmed our choice of breakpoint with p < 0.01. The mean conditions for 1979–2001 (Fig. 1a) indicate that the region between Bering Strait and 71°N and between 175°W and 160oW was ice-free, with SIC < 15%. Farther west, SIC in Long Strait was between 15% and 50%, while north of 71oN, SIC was greater than 50%. On the other hand, for 2002–2023 (Fig. 1b), the situation is remarkably different with open water, i.e. SIC < 15%, present throughout the region. And finally, contrary to recent climatological conditions, September 2024 (Fig. 1c) was characterized by SIC > 50% extending northwards from the Siberian coast of the Chukchi Sea to the eastern tip of Wrangel Island.

To understand the anomalous sea ice conditions during 2024 in the context of interannual variability, area averaged SIC during September was calculated for both an Eastern Polygon (EP), co-located with the 2024 anomaly, and an adjacent Western Polygon (WP), where summer sea ice was present during the earlier climatological period. Before 2001, there were 5 Septembers in the years 1983, 1988, 1994, 1998, and 2000 where SIC in the WP exceeded 50%, with lower concentrations in the EP (Fig. 1d). After 2001, both polygons were ice-free in September, as was the broader Western Arctic Ocean until 2024, when ice remained in both areas with the much higher concentration in the EP. Indeed, 2024 had the highest September-average SIC in the EP over the entire record (Fig. 1d).

As noted above, there were summers throughout the satellite record during which ice was present along the Siberian coast of the Chukchi Sea. However, the spatial distribution of the ice cover was different before and after 2001. Before 2001, the WP always had the higher SIC during these heavy ice summers. In contrast, the WP was mostly ice-free in 2024. Supplementary Fig. 1 shows that September monthly mean SIC for the 5 heavy ice summers during the period 1979–2001 have a similar spatial distribution to the mean over that period (Fig. 1a). Both show a higher ice concentration region extending southeastwards along the Chukchi Sea’s Long Strait, with higher concentrations to the west. This distribution is very different from that observed in September 2024.

The SIC in the winter of 2023/2024 in the EP was typical of the last two decades up until May, when ice concentrations usually begin to decrease to less than 15% by July (Fig. 1e). In contrast, May and June 2024 were characterized by 100% ice cover, after which the ice concentration did gradually decrease, but only to a value of 50% by September. From May 2024 onwards, the ice concentration exceeded the 95th percentile threshold value for the 2002–2023 period.

To understand the anomalous SIC in the EP during September of 2024, it is necessary to examine the ice thickness budget of the region. Figure 2 shows the monthly mean sea ice thickness from the PIOMAS coupled ice-ocean model (See Methods) for January through September 2024. Focusing on the EP, one can see that during January (Fig. 2a), the maximum ice thickness was 1.5 m, increasing to 2.5 m during February (Figs. 2b) and 3.5 m during March (Fig. 2c). Ice thickness in this area also underwent less dramatic increases during April, May and June (Fig. 2d–f) before finally starting to decrease during July and August (Fig. 2g, h). Ice with thickness as large as 0.5 m remained through September (Fig. 2i).

Fig. 2: Evolution of the sea ice thickness (m) over the Chukchi Sea as represented in the PIOMAS 3SST dataset.
figure 2

Monthly mean sea ice thickness (m) during: a January, b February, c March, d April, e May, f June, g July, h August, and i September 2024. The 150 m isobath is indicated by the dashed curve.

Full size image

Figure 3 shows the time series of daily ice thickness, ice advection and ice production from PIOMAS (See Methods) averaged over the EP for both the 2002–2023 climatology and during 2023/2024. For the advection and production terms, the convention used is positive values represent increases in ice thickness. During the period 2002–2023 (Fig. 3a), ice in the EP tended to begin forming around 1 November, with the largest increases occurring through March before reaching a maximum value of 2.8 m in mid-May. At this time, thinning commenced with open water typically present by early August. In contrast during 2023/2024, ice formation was delayed by approximately one month, although by 1 January 2024 ice thickness was still close to the climatological value. During January, some thinning occurred before a rapid increase, from 2 m to 2.9 m in late February. From 1 March through 1 May, there was a more modest increase in ice thickness compared to climatology which reduced the difference between the two curves. For example, on 1 March, the 2024 ice thickness was 0.4 m thicker than climatology, while on 1 May, it was only 0.2 m thicker. As we will show, this reduction resulted from weak ice growth during this period in response to the advective thickening in February.

Fig. 3: Temporal variability in sea ice characteristics averaged over the EP from September 1 2023 to October 1 2024 as represented in the PIOMAS 3SST dataset.
figure 3

Time series (red curves) of the: a sea ice thickness (m); b ice advection (m/d) and c ice production (m/d) during 2023/2024. The black curves represent median values for the period 2002–2023 with the shading representing values from the 5th to 95th percentile.

Full size image

The ice in the EP continued to thicken through May and June 2024 via convergence, reaching a maximum value of 3.3 m on July 1. After this time, it began to thin at a rate similar to the climatology; however, because it started anomalously thick, it ended up with non-zero thickness by September. Indeed, after mid-July, ice thicknesses in 2024 exceeded the 95th percentile value for the period 2002–2023.

The cumulative ice advection and ice production (Supplementary Fig. 2) are directly related to ice thickness (See Methods). They had roughly equal roles during 2002–2023 in the EP ice thickness budget, with ice advection contributing 1.35 m and ice production contributing 1.45 m to the maximum ice thickness by mid-May. However, in 2024 ice advection contributed much more to changes in thickness than ice production; in fact the latter was lower than climatology due to thicker sea ice.

The climatological ice advection term (Fig. 3b) was slightly positive from November through May, indicating convergence, before turning negative, indicating divergence. Extreme ice advection events of either sign are possible throughout the year, with the largest positive events typically between January and March and the largest negative events between May and July. During 2024, several positive events exceeded the 95th percentile, with two occurring during February. The event in late February was the largest in magnitude and duration during the 2023/2024 period and coincided with the large increase in ice thickness noted above. There were several smaller magnitude advection events of both signs between 1 March and 1 July.

The 2002–2023 climatological ice production (Fig. 3c) has its maximum around 1 December, after which it is positive but small until mid-May, when it becomes negative, reaching a minimum approximately one month later. Extreme ice production events are typically smaller than extreme ice advection events and tend to be one-sided, with large positive/negative events occurring only during the early/later part of the season. Relatively large values of ice production (i.e., ice growth) occurred in December 2023, with the onset of negative values (i.e., ice melt) in spring 2024 delayed by approximately one month compared to climatology. Notably, there were two large melt events that occurred in mid-June and early July 2024. As will be discussed in Section “Presence of sea ice in the Chukchi Sea during the 2024 melt season”, these events were primarily forced by the atmosphere.

The February 2024 ice advection event

Figure 3 indicates a large increase in ice thickness over the EP in late February 2024 resulting from ice convergence. The event occurred over 18–24 February with a maximum ice advection of 0.18 m/d. Figure 4 shows selected daily ice thickness and ice motion fields from PIOMAS that provide additional information on the nature of this event. Before the event started on 16 February 2024 (Fig. 4a), ice over the EP was 1.5–2 m thick. On 18 February (Fig. 4b), southwestward ice motion caused coastal convergence and an increase in ice thickness to values up to 3 m right at the coast. Over the next 6 days (Fig. 4c–e), this situation intensified, resulting in ice thicknesses over 3.5 m in the central and southern portions of the EP. By the conclusion of this 7-day (18–24 February) advection event (Fig. 4f), the maximum ice thickness over the EP had increased by l.5 m. During this period, ice motion was on the order of 15 km/d and so over the 7 day period, the ice traveled approximately 100 km where not blocked by the coast.

Fig. 4: Evolution of sea ice thickness as represented in the PIOMAS 3SST dataset during the February 2024 ice advection event.
figure 4

Daily mean sea ice thickness (m) and ice motion (km/d) over the Chukchi Sea in the PIOMAS 3SST dataset for February: a 16; b 18; c 20; d 22; e 24 and f 26 2024. The 150 m isobath is indicated by the dashed curve.

Full size image

The broad southwestward ice motion over the Chukchi Sea during this period also resulted in divergence along the Alaskan coast, which reduced ice thickness there. Such divergence is not unusual24, although its magnitude in 2024 may have been. The southwestward ice motion also resulted in a thickening of sea ice along the eastern (upwind) coast of Wrangel Island and a thinning of sea ice on the island’s western (downwind) coast. This is an example of the ice thickness dipole26.

Meteorological conditions as represented in ERA5 during this late February 2024 ice advection event are shown in Fig. 5. The sea-level pressure, 10 m wind vectors and 10 m wind speed averaged over 18–24 February 2024 (Fig. 5a), indicate that there was a large pressure gradient across the Chukchi Sea associated with the presence of the Beaufort High to the north and the Aleutian Low over the Bering Sea. This resulted in northeasterly surface wind speeds across the southern Chukchi Sea in excess of 16 m/s, consistent33 with the ice motion observed during the event (Fig. 4). The time series of the normal component of the 10 m wind along the northern and eastern boundaries of the EP (Fig. 5b) indicates two periods of enhanced winds from the north, one centered on 9 February and the other on 21 February that correspond to the two large ice convergence events during that month (Fig. 3b). The latter had wind speeds near the 99th percentile value during 19–22 February. This was the longest period with sustained windspeeds above this threshold during February over the entire record from 1979 to 2024. In addition, 6 days during the event had wind speeds above the 95th percentile threshold. The point here is that the area experienced both anomalously strong and anomalously sustained northeasterly winds during this event.

Fig. 5: Meteorological conditions during the late February 2024 ice advection event as represented in ERA5.
figure 5

a Sea-level pressure (mb-contours), 10 m winds (m/s-vectors) and 10 m windspeed (m/s-shading) averaged over the period February 18–24 2024. b Time series of the normal component of the 10 m wind across the northern and eastern sides of the EP during February 2024 with the period February 18–24 indicated by the light shading. The black dashed line represents the median, with the red and blue lines representing the 95th and 99th percentile values, all based on data from February 1979–2024.

Full size image

During February 2024, the Beaufort High’s location and maximum sea-level pressure were similar to climatology, while the location of the minimum sea-level pressure associated with the Aleutian Low was shifted northwards and was 6 mb lower than climatology (Supplementary Fig. 3), which together shaped the enhanced pressure gradient responsible for the gale force winds over the southern Chukchi Sea. The 2–6 day band pass filtered variance of the sea-level pressure field provides information on synoptic-scale variability of atmospheric weather systems34,35. This field during February 2024 had minimum values in the center of the Beaufort High, suggesting that it was relatively stable with respect to variability on synoptic time scales. Thus the northward migration of a deeper than normal Aleutian Low along with a stationary Beaufort High pattern of average intensity most likely contributed to the sustained winds observed during the late February 2024 ice advection event.

Presence of sea ice in the Chukchi Sea during the 2024 melt season

During 2024, and unlike the 2002–2023 climatology (Fig. 3a), sea ice thickened within the EP during May and early June before beginning to thin in July. To investigate the processes responsible for the continued thickening and delay in melt onset, we show time series of ice thickness, ice advection and ice production (See Methods) for the period 1 May–1 October 2024 as well the climatology for 1979–2001 and 2002–2023 (Fig. 6).

Fig. 6: Temporal variability in sea ice mass budget terms from the PIOMAS 3SST dataset averaged over the EP from May 1 to October 1.
figure 6

Time series (black curves) of the: a ice thickness (m); b ice advection (m/d); and c ice production(m/d) during 2024. The red/blue curves represent mean values for the period 1979–2001 and 2002–2023.

Full size image

Comparing the evolution of ice thickness (Fig. 6a), we note that ice thinning starts about one month earlier in the more recent climatology relative to the 1979–2001 period. In contrast, from mid-June 2024 onwards, the thickness evolution tracks that of the earlier climatological period.

The thickening that occurred in May and early June 2024 resulted from multiple, large-amplitude convergent ice advection events (Fig. 6b) that were larger than the subsequent, less frequent convergent events that occurred in mid-June and early July. There were also some large divergent ice advection events during May to early July.

During the earlier climatological period, ice melt starts around mid-May with peak melt in late June, while during 2002–2023, it occurs in mid-June with peak melt in early August (Fig. 6c). In addition, there is a local maximum in ice melt in mid-July for the recent climatological period. No such secondary maximum was present for the earlier climatological period. During 2024, ice production generally tracks the behavior of 1979–2021, with some large episodic melt events during June and July that contributed to a large decrease in ice thickness (Fig. 6a).

Figure 7 shows the time series for the two components of ice production, (See Methods) for the period 1 May–1 October 2024 as well the climatology for 1979–2001 and 2002–2023 within the EP. Surface melt generally starts earlier than bottom melt, with both processes starting earlier in recent years. Furthermore during the recent climatological period 2002–2023, the lag between the maximum surface melt and bottom melt has declined, from ~1 month in the past to ~2 weeks, owing more to earlier bottom melt. During 2023–2024, the two terms show multiple minima, with both melting processes generally tracking that seen in 1979–2001 (with the exception of large bottom melt events in July 2024).

Fig. 7: Temporal variability in the ice production budget from the PIOMAS 3SST dataset averaged over the EP from May 1 to October 1.
figure 7

During this time, ice production is always negative, i.e., melting. Time series (black curves) of the ice production (m/d) due to: a surface processes and b bottom processes during 2024. The red/blue curves represent mean values for the period 1979–2001 and 2002–2023.

Full size image

Supplementary Fig. 4 shows the time series of positive degree days at 925 mb from ERA5 for the period 1 May–1 October 2024, as well as the climatologies for 1979–2001 and 2002–2023 within the EP. Similar behavior also occurred for the 2 m air temperature although the signal is muted due to a tighter coupling with the sea ice. One can see that the earlier onset of surface melt during 2002–2023, as compared to 1979–2001, is the result of an earlier onset of above-freezing temperatures. In general, the time series for 2024 tracks that of the earlier climatological period with two exceptions, mid-June and early July, during which there was a rapid increase in positive degree days resulting from periods of higher air temperatures. These correspond to two periods of rapid surface ice melt (Fig. 7a).

Supplementary Fig. 5 shows time series of the two components of bottom melt: (i) surface fluxes into the ocean surface layer and (ii) ocean dynamics (i.e., mixing and advection) (see Methods) for the period from May 1 to October 1, 2024, as well as the climatologies for 1979–2001 and 2002–2023 within the EP. Surface processes contribute more significantly than ocean dynamics. Both terms exhibit a similar delay of approximately one month in the timing of the minima, with melting during the 2002–2023 period and 2024 extending through the end of September.

Changing nature of ice advection in the Chukchi Sea

We conclude with an assessment of the changing nature of ice advection in the region and its impact on the 2024 sea ice season. We focus on February as this is a month during which convergent ice motion events are common along the Siberian coast of the Chukchi Sea (Fig. 3b), as well as the month during 2024 that had the largest convergent advection event that contributed to the persistence of ice through the following summer. The ice advection field is episodic, and so to focus on extreme events, we chose to look at changes over a 7-day period that corresponds to the length of the 2024 event. We compared the magnitudes of the change in ice thickness, cumulative ice area and ice volume fluxes (See Methods) to all 7-day periods during February 1979–2023. Figure 8 shows the probability distributions for the 7-day changes in these fields within the EP for the period 1979–2001 and 2002–2023 with values from the 2024 event superimposed.

Fig. 8: Evolution of the ice thickness and ice motion probability distributions averaged over the EP during February 1979–2024 as represented in the PIOMAS 3SST dataset.
figure 8

Probability distribution of the: a 7-day differences in sea ice thickness; b cumulative 7-day ice area flux, and c cumulative 7-day ice volume flux for the period 1979–2001 (blue) and 2002–2023 (red). The vertical dashed lines represent the respective 95th percentile values while the thick black vertical line represents the values during the February 2024 ice advection event.

Full size image

For all three fields, there has been a shift to the right in the distributions with an increase in mean, median and 95th percentile values for the 2001–2023 period compared to the 1979–2001 period. The changes in the ice volume flux between the two periods are more muted than the ice area flux. This is because the ice volume flux is a function of ice thickness, and the recent thinning has reduced the magnitude of the changes. For the change in ice thickness and the ice volume flux, the maximum during the February 2024 event was the second largest during the entire record, while for the ice area flux, it was the largest. The bottom line here is that synoptic scale (i.e., 7-day average) events with large increases in ice thickness as well as in the ice area and ice volume fluxes are more frequent since the turn of the 21st century. For example, during the period 2002–2023, the 95th percentile values for the period 1979–2001 would occur 132% more often for the ice thickness change, 270% more often for the ice area flux, and 134% more often for the ice volume flux (Table 1).

Table 1 Number of days in excess of the 95th percentile value for 1979–2001 for both 1979–2001 and 2002–2023 and percent difference
Full size table

Discussion

As noted by Thoman31, sea ice was present along the Siberian coast of the Chukchi Sea during September 2024 (Fig. 1c). Summer sea ice in this region has occurred in the past, the last time being in 2000. However, the distribution of sea ice during September 2024 was different than what had occurred earlier. In particular, during the 1980s and 1990s, the climatological SIC during September and during high ice concentration summers was largest in the WP (Fig. 1a and Supp Fig. 1). In contrast, during 2024, it was highest in the EP, with very little ice present in the WP (Fig. 1c and d). This difference in the spatial extent of sea ice suggests that different mechanisms were responsible for the persistence of sea ice in the region during September 2024 compared to during the 1980s and 1990s.

Focusing on 2024, we have shown that a large convergent ice advection event occurred in late February, resulted in an approximate 0.9 m increase in average ice thickness within the EP (Fig. 3a, b). During this event, sustained southwestward ice motion across the Chukchi Sea resulted in ice convergence along the Siberian coast (Fig. 4). The ice motion was forced by sustained northeasterly gale force winds that were in excess of the 99th percentile threshold for 6 days during late February (Fig. 5). On a synoptic timescale, the large pressure gradient between a stationary and average Beaufort High and a northerly-positioned and deepened low-pressure system over the Bering Sea was responsible for the high and sustained winds (Supplementary Fig. 3).

Recently thinning of sea ice via melt along the Siberian coast of the Chukchi Sea typically begins in May and results in open water by late July (Fig. 6a). However, additional convergent advection events during May and early June 2024 led to enhanced ice thickness that approached values typically seen during the 1980s and 1990s (Fig. 6a, b). The onset of ice melt during 2024 was delayed to mid-June, with peak melting (Fig. 6c, d) occurring in mid-July, similar to the seasonal timing during the 1980s and 1990s (Fig. 6c). Bottom melt due to surface fluxes plays a dominant role in ice melt (Fig. 7 and Supplementary Fig. 5). This is consistent with observations that the transmission of shortwave radiation through the ice and in leads makes the more significant contribution to bottom ice melt13,36,37.

With respect to summer ice melt, we found that all terms had an earlier onset during the 2002–2023 period as compared to the 1979–2001 period (Figs. 6, 7 and Supplementary Fig. 5). For surface melting (Fig. 7a), the shift was approximately two weeks, while for bottom melting it was approximately 1 month. An advance in the onset of above-freezing temperatures during recent years was most likely responsible for the change in the timing of surface ice melt. The approximately 1 month shift in bottom melt mirrors that in ice thickness (Fig. 6a). As noted above, bottom melt in the EP is primarily the result of shortwave radiation transmission through thin ice and into leads, which results from a thinner and looser ice pack.

There is evidence that the increased mobility of the Arctic ice pack contributes to local maxima in sea ice thickness27,29,30. Probability distributions for ice thickness change, ice area and ice volume fluxes over the EP (Fig. 7) indicate shifts towards higher values of the mean, median and 95th percentile consistent with increases in sea ice mobility. These shifts are similar to what has been documented concerning changes in global temperature anomalies38, indicating that there has been an increase in the magnitude of extreme events in the region recently. Furthermore, values of these parameters during the late February 2024 event exceeded the 95th percentile values for both periods, and were in fact the largest or second largest observed during February 1979–2024.

As noted in the Introduction, the retreat of sea ice and its thinning provides opportunities for new shipping routes in the Arctic that provide shorter voyages and reduced costs, as well as opportunities for resource extraction8,9. However, paradoxically, the increased mobility of sea ice can increase maritime operational hazards10. For example, during September 2012, the period with the lowest recorded ice extent in the Arctic, Shell’s drilling operations in the Beaufort Sea were impacted when mobile pack ice approached the drill site39,40. It is also of interest to note that during August 1983, one of the other heavy ice summers in the region (Fig. 1 and Supp Fig. 1), over 40 ships were trapped by mobile pack ice along Long Strait, with at least one freighter sinking as well as significant damage to one of Russia’s nuclear icebreakers41,42.

The return of sea ice to the Chukchi Sea during the summer of 2024 after an absence of 25 years was the result of several anomalous convergent ice advection events, the largest of which occurred in late February 2024, that resulted in ice thicknesses in the region during June that were close to or exceeded what was typical during the 1980s and 1990s. The thicker ice took longer to melt, resulting in the persistence of sea ice through September 2024. We have also shown that events with large increases in ice thickness as well as in the ice area and ice volume fluxes are more frequent since the turn of the century, suggesting the somewhat counter-intuitive result that we may see more instances where ice persists through summer in the coming years.

Methods

Sea ice concentration

We use Version 5 of the NOAA/NSIDC Climate Data Record of SIC43. This record is based on data from a sequence of multi-channel passive microwave radiometers on various satellites and was created to meet standards for transparency and reproducibility1 The record is based on two retrievals of SIC from the multi-channel microwave data, the NASA Team and the Bootstrap algorithms44. It is available from October 1979 onwards on a polar stereographic grid with a resolution at 70oN of 25 km at a temporal resolution of 2 days up to August 1987 and daily afterwards.

PIOMAS

PIOMAS is a coupled sea ice and ocean model. The sea ice model is a 12-category thickness and enthalpy distribution model, which is coupled with the POP (Parallel Ocean Program) ocean model developed at the Los Alamos National Laboratory45. The PIOMAS model domain is based on a curvilinear grid with the north pole of the grid displaced into Greenland. It covers the area north of 49° N and is one-way nested into a similar, but global, ice-ocean model46. In the Chukchi Sea region, the model has a grid cell size of approximately 40 km. PIOMAS assimilates observations of sea surface temperature47. It can also assimilate ice concentration observations48. In this study, satellite ice concentrations are assimilated only near the ice edge (defined as 15% ice concentration). This means that the assimilation is allowed only in the areas where either model or satellite ice concentration is at or below 15%. In other words, no assimilation is conducted in the areas where both model and satellite ice concentrations are above 15%49. We refer to this model run as 3SST. The model is forced by daily mean NCEP/NCAR reanalysis data50, including 10 m surface winds, 2 m surface air temperature, specific humidity, precipitation, evaporation, and downwelling longwave and shortwave radiation. PIOMAS data is available from 1979 onwards with a temporal resolution of 1 day. PIOMAS ice thickness output has been extensively validated against in-situ and satellite observations4,51,52.

Ice thickness budget

The ice thickness budget can be expressed as30:

$$\frac{\partial {\rm{h}}}{\partial {\rm{t}}}={{\rm{F}}}_{{\rm{adv}}}+{{\rm{F}}}_{{\rm{prod}}}$$
(1)

where h is the ice thickness (m),

Fadv is the ice advection term (m/d) and

Fprod is the ice production term (m/d).

In addition, the ice production term can be further defined in terms of surface and bottom contributions30:

$${F}_{{prod}}={F}_{{surface}}+{F}_{{bottom}}$$
(2)

where Fsurface is surface ice growth or melt and Fbottom is bottom ice growth or melt. Furthermore, bottom ice production can be expressed as30:

$${F}_{{bottom}}={F}_{{bottom}/{surface}}+{F}_{{bottom}/{oceanic}}$$
(3)

where \({F}_{{bottom}/{surface}}\) is bottom ice growth (in winter) due to atmospheric cooling and melt (in summer) due to surface fluxes and \({F}_{{bottom}/{oceanic}}\) is bottom ice melt due to oceanic dynamics (i.e. heat transport by advection and diffusion).

Integrating (1) yields

$$h\left(t\right)=h\left({t}_{o}\right)+{{CF}}_{{adv}}\left(t\right)+{{CF}}_{{prod}}\left(t\right)$$
(4)

where: CFadv is the cumulative ice advection term (m) and CFprod is the cumulative ice production term (m).

Ice area and ice volume fluxes

The ice area flux provides a characterization of the transport of ice across a given flux gate53 and is defined as:

$${IAF}=\sum {i}_{{conc}}{u}_{n}\Delta x$$
(5)

where IAF is the ice area flux (m2/s), iconc is the ice concentration (%) and un is the normal component of the ice motion.

Similarly, the ice volume flux provides a characterization of the ice mass transport across a given flux gate53 and is defined as:

$${IVF}=\sum {{hu}}_{n}\Delta x$$
(6)

where IVF is the ice volume flux (m3/s).

ERA5 reanalysis

ERA5 is the fifth generation reanalysis from the ECMWF54. It is based on Cycle 41r2 of the ECMWF’s Integrated Forecast System. It has a horizontal resolution of 0.25o and temporal resolution of 1 h. Here we use data from 1979 onwards. A comparison with monthly mean wind data from Wrangel Island indicated that ERA5 does a good job in representing the wind climate at this site with an rms error on the order of 1 m/s and a correlation coefficient of 0.926 A number of studies have indicated that the ERA5 outperforms other reanalyses with respect to the representation of the climate throughout the Arctic55,56.