Charles Rotter
A new paper, A recent stabilization in the lengthening of the Arctic sea ice melt season into a highly variable regime, from NASA Goddard and the University of Maryland, published in Communications Earth & Environment, reports that the Arctic sea ice melt season stopped getting longer around 2010. For the fourteen years from 2010 through 2023, it actually got slightly shorter.
From the abstract:
However, since 2010 the melt season length has stabilized (~ 108 days), showing no consistent change over the years, instead becoming highly variable (+/− 11 days), largely driven by a loss of multi-year ice in 2000–2009 and a small change in the freeze onset (~ 2 days). There is a stark difference between the decades, where the largest changes in the melt season occurred between 2000–2009 (+ 25 days) and the smallest occurred between 2010–2023 (−2 days).
Here is their chart of it, with a spelling error:
The melt season grew by about 40 days from 1979 to 2023, and most of that came in a single decade, 2000–2009, the blue line. The purple 2010s bounce around and go nowhere. There is no trend line on them because there’s no trend. Table 1 of the paper labels 2010–2023 the “Stabilized Era.”
Here is the next sentence of the abstract:
This leads us to believe that, while there might be some periodicity in the processes that control the decadal variations in the melt season length, anthropogenic forcing has altered the Arctic background state and led to a new Arctic melt season that is much longer with a much thinner ice pack that is more susceptible to external forcings.
“Leads us to believe.” Nothing in the paper tests anthropogenic forcing. There is no model run, no attribution analysis, no forcing dataset. When the authors go looking for what drove the changes, they find clouds:
Between 2000–2009 there was a large increase in net shortwave (SW) radiation (+ 12.6 Wm−2) and therefore the SEB (+ 10 W m−2, driven by decreasing surface albedo (−0.022%) and cloud cover (−2.3%)). During 2010–2023, an increase in cloud cover (+ 7.2%) and only minimal changes in albedo (0.011%) drove a net decrease in SW (−3.14 Wm−2) and summer SEB (−2 W m−2).
And what moved the clouds:
The significant decrease in cloud cover in the Chukchi and Beaufort seas and the Canadian Archipelago (western basins) with increasing cloud cover in the Barents Sea (and other eastern basins) reflects a positive AD in 2000–2009 (Fig. 7h).
This is starkly contrasted with significantly increasing cloud coverage in the Chukchi and Beaufort seas, the Canadian Archipelago, Central Arctic and E. Greenland Sea (western basins) in 2010–2023 but declining cloud cover over the Barents, Kara, and E. Siberian Seas (eastern basins) (Fig. 7l), corresponding to a negative AD.
AD is the Arctic Dipole, a pressure pattern that is part of the Arctic’s natural circulation. Compare the middle map with the right one. The same stretch of the western Arctic clears out in 2000–2009 and clouds up in 2010–2023. A positive Dipole let in more sun during the decade of rapid lengthening, and a negative Dipole brought the clouds back. The energy reaching the surface followed:
Hatching marks significant trends. Fig. 6a, e, i from Boisvert et al. 2026. Only a few small patches there are hatched as significant, and they’re outnumbered by the hatched blue elsewhere on the same map
So did the ice:
Whereas in 2010–2023, there is an overall decrease in the ROC of net daily SEB energy entering the system (−1.26 W m−2), which could result in a growth of up to 7 cm of ice over the 14 years. Indeed, CryoSat-2 observations suggest that there were areas with slight thickening of the ice in October/November over this decade (3.3 cm average increase in October/November SIT in CryoSat-2 over the record (2011–2021))
So the stall goes down as internal variability, and the lengthening before it goes down as a changed background state caused by people. Clouds moved around by the Dipole explain both decades in the paper’s own analysis, but only the earlier one gets the anthropogenic label.
The paper ends on the one place it could find going the other way, and the authors concede the overall decline in the same breath:
Despite the overall net JJAS SEB decrease in 2010–2023, there is a notable increase of summer net JJAS SEB over portions of the Central Arctic where the remaining MYI resides (Fig. 6i). Between 2010 and 2023, the sea ice in the Central Arctic has received roughly +0.5 W m−2 more energy per year (~ 7 W m−2 overall). This energy could melt ~18.4 cm more ice by the end of the record, significantly thinning and weakening the MYI SIT in this area (Fig. S3l). With this increase in energy, this extra JJAS net SEB energy has the potential to melt 6% of the thickest sea ice north of Greenland, which averages around 3 m. Crucially, if this trend continues and exceeds winter regrowth, the MYI could become thin enough to be vulnerable to melt out in the future, resulting in an ice-free Arctic in the summer.
That is panel i of the energy maps above. Look north of Greenland. The orange there is pale and patchy and carries little of the hatching that marks a significant trend. From that, plus an “if this trend continues,” they get to an ice-free summer.
The introduction does its part before the reader reaches any data. It brings in Arctic warming at nearly four times the global rate, ice-free summers projected for the 2030s to 2060s, bigger ocean swells, coastal erosion, seal reproduction and polar bear predation, and the jet stream. The paper looks at none of it. “Imperative” shows up three times.
Then there is the proofreading. The paper was received in May 2025 and accepted in April 2026, and nobody caught the following. The abstract and Table 1 put the stabilized melt season at about 108 days. The Results section says:
In the last decade, however, the later melt onset together with the large interannual variability in both the melt and freeze onset led to a stabilization of the Arctic melt season to an average ~120 days.
The abstract gives the albedo changes as −0.022% and 0.011%. The paper’s own albedo maps in Fig. 6 are labeled in fraction per year, which makes the abstract off by a factor of a hundred. The standard deviation is said to rise to “10.6 days in 2010–2013.” The headline chart is titled “Melt Seaon Length.” Table 1 defines the Arctic Ocean as regions 9–13 and 15, Table 2 uses 9–15 excluding 14, and the Methods say:
The Arctic Ocean region defined in this study comprises regions 2–15 defined in Fig. 1a.
None of this overturns the result, but a summary going out to the press ought to get its own numbers straight.
If less clouds under a positive Arctic Dipole drove the 2000s lengthening, where does anthropogenic forcing enter the calculation?
Why is the 2010–2023 stall internal variability while the 2000–2009 jump is a new background state?
If summer energy went down over the Arctic Ocean, why end the paper on the one area where it went up?
And is the stabilized melt season 108 days or 120?
==========================================================
Quotes and figures from Boisvert et al. 2026, https://doi.org/10.1038/s43247-026-03534-8, used under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). Figures cropped from the originals.
