Winter sea ice in the Arctic could shrink by roughly a million square kilometers more than climate models project, according to a Cambridge-led study that tested the models against ocean sediments laid down about 6,000 years ago, when the Arctic was warmer than it is today. The finding is a projection built from past evidence and model weighting, and it describes how far the models may be off, not a loss that satellites have measured. The lead author, Crystal Fu of the University of Cambridge’s Scott Polar Research Institute, argues that the models have been too generous about how much ice survives the winter in a warmer climate.
Save more, waste less: The Cool Down’s newsletter sends smart, simple changes that help you and the planet in your everyday life.
We may earn a commission if you sign up.
Sediment cores from the mid-Holocene set the benchmark
The University of Cambridge’s release, dated October 16, 2026, says the team compiled the largest database yet of IP25, a biomarker produced by algae that live in sea ice, drawn from ocean sediment cores from the mid-Holocene, a warm interval about 6,000 years ago. Where the biomarker turns up in a core, sea ice was present at that spot in that season, which turns seabed mud into a record of ancient ice cover. The study, by C.Y. Fu, M.B. Osman and M.A. Aquino-López, appeared in Nature Geoscience, and co-author Marco Aquino-López works at CIMAT, the Centro de Investigación en Matemáticas in Mexico.
The choice of period is deliberate. About 6,000 years ago the Arctic was warmer than in the centuries since, with less sea ice, which makes the mid-Holocene one of the few intervals where models can be checked against real ice cover under conditions that resemble a warmer future. A model that cannot reproduce that older, warmer Arctic has less claim on authority when it projects the next one.
Weighting 27 climate models by how well they match the past
The researchers compared the sediment-based estimates with simulations from 27 climate models, then used Bayesian weighting to give more influence to the models that reproduced the mid-Holocene better. According to the release, the models as a group had overestimated winter sea ice, and the weighting removed that tendency. Fu said that applying the model weights to historical simulations “completely eliminated the positive bias.”
The method does not make any model run itself again. It changes how much each model’s output counts when the projections are averaged, so the figure is a shift in the weighted ensemble: future winter ice that comes out about one million square kilometers smaller than the unweighted models suggest. For scale, that is roughly the area of Egypt, and it is the gap between two sets of projections rather than the total amount of ice at stake.
The weighting also tightened the spread of outcomes. The release says the range of uncertainty in some projections narrowed by as much as 85%, and a narrower range is useful to anyone who has to plan around a projection as much as a lower central estimate is.
Winter ice versus the September minimum
Most public attention goes to the September minimum, when the ice cover is smallest. This study concerns the winter, when the ice is at its widest. Fu made the point that sea ice responds to warming whatever its cause: “if it warms, sea ice is going to melt.” The mid-Holocene serves as a test of how sensitive the models are rather than as a forecast for a particular year.
The National Snow and Ice Data Center tracks the observed ice. Its most recent analysis, dated October 7, 2026, reported that the Arctic minimum extent tied for tenth lowest on the satellite record with 2008, 2010 and 2025, and that the post-minimum freeze-up had been rapid. Those observations describe the present, and each year’s minimum is a separate matter from the winter maximum the study addresses; the Cambridge study asks what the models imply for a warmer future and whether they should be trusted as they stand.
The release adds that early signs suggest the same models may be underestimating snowfall over southeast Greenland and rainfall across the Mediterranean, with possible implications for the Greenland ice sheet and for flood risk in southern Europe. Those results are described only as early hints, and the release does not put numbers on either.
Funding and the ScienceDaily relay
The research was supported in part by the U.S. National Science Foundation and the Cambridge Trust, the release says. The study reached a wider audience through ScienceDaily, which republished the Cambridge material on October 26, ten days after the university’s own date.
Aquino-López, quoted in the release, said the work allows “more reliable maps of where Arctic winter sea ice is likely to disappear.” The open question for the field is whether other proxies of mid-Holocene conditions, weighted the same way, give the same million-square-kilometer gap, and whether the 85% narrowing holds when the weighting is applied to projections for other seasons and regions. The release does not report such tests, and the paper itself could not be opened for this article, so its methods section remains the place to check the details.
This article was produced with the assistance of AI and reviewed by Morning Overview editors prior to publication.
Save more, waste less: Subscribe to The Cool Down →
We may earn a commission if you sign up.
More from Morning Overview
