In Simple Terms
A new study suggests that the Arctic may have less winter sea ice than climate models have led us to expect. By comparing the models with chemical clues preserved on the seafloor from about 6,000 years ago, researchers found that some models may overestimate how much ice remains. The findings could also help scientists better understand possible changes in snowfall and rain far from the Arctic.
Winter in the Arctic Matters Far Beyond the Arctic
The disappearance of summer sea ice gets plenty of attention, but winter matters just as much. During the cold months, sea ice in the Arctic Ocean reaches its annual maximum. How much of the ocean is covered affects the amount of heat and moisture that can pass from the water into the air. Estimating winter ice is therefore more than a seasonal detail on climate maps: it is part of understanding how changes in the Arctic might influence weather patterns elsewhere.
Sea ice acts as a barrier between the ocean and the atmosphere. When the ice cover shrinks, more of the sea surface is exposed, changing the transfer of heat and moisture into the air. Scientists are interested in these changes because they may be linked to weather shifts far away. The relationship between Arctic conditions and local weather, however, is complex and cannot be summed up in a single rule.
Earlier studies have linked declining Arctic ice with colder winter conditions and heavier snowfall in the United Kingdom, northern Europe, and North America. Improving estimates of sea ice is therefore not just about picturing what the Arctic Ocean might look like in the future. It can also help build a clearer understanding of climate risks that matter to populated regions. The new findings offer a way to investigate those connections—not to make a definite forecast for the weather in any particular year.
The question has become especially timely as winter sea ice has reached notable lows. In March 2025, its annual extent hit the lowest level since modern observations began. In 2026, it reached that record low again. Maximum extent means the area covered by ice at the height of the cold season; it is not a measure of summer ice or of the ice’s total volume.
Even so, predicting changes in winter sea ice remains harder than the number of climate models might suggest. The simulations can differ widely when researchers estimate how much ice will cover the Arctic during the cold months. If models broadly agree on the direction of change but disagree about the area of ice that will remain, the range of possible futures stays wide. That makes it harder to compare results and determine how much change is most likely.
Clues Preserved Beneath the Ocean Floor
Rather than relying only on modern observations, a team from the University of Cambridge and Mexico’s Center for Mathematical Research (CIMAT) looked for traces of ancient climate preserved on the ocean floor. Their work drew on sediments from the mid-Holocene, about 6,000 years ago. At that time, small changes in Earth’s orbit helped make the Arctic warmer and less ice-covered than it had been under colder conditions.
Ancient sediments do not offer a photograph of the ocean, but they preserve chemical traces left by organisms that lived there. One such trace comes from a biomarker called IP25, produced by algae associated with sea ice. When the remains of these algae become buried in sediment, the marker can survive and help researchers infer that ice was present. The team assembled the largest database of this indicator to date and added other chemical traces left by organisms that live in open water.
Comparing indicators of ice with those of open water gives researchers a more balanced picture than relying on a single marker. Traces from ice-associated algae point to icy conditions, while the other markers help estimate how much open-water habitat was present. Using both kinds of evidence, the researchers reconstructed past ice conditions in parts of the Arctic and then used that ancient climate record to test how well climate simulations performed.
The mid-Holocene is valuable because it provides a real-world example of a warmer Arctic with less ice, rather than a result generated only by a computer. The warming then was driven by natural changes in Earth’s orbit, while today’s warming is associated with human activity. The team focused on how the ice responded to warmth. The basic physical processes that cause ice to form or melt can still be studied under different climate conditions.
How the Team Tested 27 Climate Models
After reconstructing ancient ice conditions, the researchers compared their evidence with simulations of the same period produced by 27 climate models. A climate model is a computational tool that represents interactions among parts of the climate system, such as the ocean, atmosphere, and ice. It uses those interactions to simulate the past or explore possible futures. But models do not all perform equally well: some may represent historical conditions more accurately than others.
The team used a statistical approach called Bayesian weighting. Put simply, the researchers gave more weight to simulations that came closer to the evidence from the sediments and less weight to those that differed more. They did not select one model and discard the rest. Instead, they adjusted how much each model contributed to the overall picture according to how well it matched the ancient evidence.
This kind of test connects what scientists know about the past with what they are trying to estimate about the future. Rather than treating every simulation as equally reliable, the method uses the ancient climate record to identify which models are better at representing the ice’s response. After calculating the appropriate weights, the researchers applied the same approach to simulations of recent periods and future scenarios.
The approach rests on the idea that the physical processes governing sea ice do not change completely just because the cause of warming or the time period is different. The team therefore reasoned that weights based on testing models against the past could also help improve how those models represent the recent past and the future. This does not eliminate uncertainty; it redistributes confidence among the models using an independent line of evidence.
A Bias Toward Too Much Ice
The test showed that, taken together, climate models tended to calculate a larger area of winter sea ice than the ancient evidence supports. In other words, they had a positive bias in their estimates of ice cover: the simulations retained more ice than the mid-Holocene climate reconstruction indicated was present.
When the researchers applied the new weights to simulations of historical periods, that bias disappeared, according to lead researcher Crystal Fu of the Scott Polar Research Institute at the University of Cambridge. In projections of the future, the estimated ice area was about one million square kilometers lower than in previous estimates. That figure does not mean a specific area will disappear in a single year. It describes the difference between the adjusted future results and earlier estimates.
A difference of about one million square kilometers is no minor adjustment to an Arctic map. If earlier estimates tended to retain a broader ice cover, lowering the projections points to a different possibility for where winter ice may retreat. More accurate maps can help researchers identify the areas most exposed to change and describe the likely scale of that change using a method tested against the historical record.
Still, the findings should not be read as certainty about how much ice will exist in any particular future year. The study improved the way researchers handle a group of models and reduced some sources of uncertainty, but it did not turn climate projections into daily or annual forecasts. The results remain estimates based on evidence and mathematical methods, with limits imposed by both the ancient record and the simulations.
Narrowing the Range of Uncertainty
Studies of ancient climate face a double challenge: reconstructions of the past are uncertain, and model-based estimates of the future are uncertain too. That can make it difficult to place old geological evidence alongside modern projections and compare them directly. Fu noted that records of past climates have accumulated over recent decades, but connecting them to future projections has remained challenging because both sides of the comparison come with their own uncertainties.
The weighting method offers a way to bring the two closer together. It uses ancient evidence to assess how well simulations perform, then carries that assessment forward when building projections. According to the study, the range of possible results narrowed by as much as 85% in some projections. That percentage applies to some of the cases the team tested; it does not mean uncertainty in every aspect of the Arctic climate fell by the same amount.
Narrowing the range of possibilities is useful even when it does not remove uncertainty altogether. The smaller the gap between possible outcomes, the more clearly scientists can discuss where ice is likely to retreat and how confident they can be in that picture. Study co-author Marco Aquino-Lopez of CIMAT said that drawing on evidence from the past makes it possible to create clearer maps of areas where winter ice could disappear—and to show more clearly what remains unknown.
The findings also show that the geological record is more than a record of events that are over. When scientists interpret the remains of algae and other chemical traces in sediment with statistical tools and climate simulations, those traces become a practical test of the models used to study today’s climate. It is a way to learn from past climate conditions while recognizing that their causes and circumstances differed from those of the present.
What Could This Mean for Europe?
The study looked beyond the area covered by sea ice. When the researchers applied the adjusted model weights to precipitation projections, they found signs that current models may also underestimate future snowfall in southeastern Greenland and rainfall in the Mediterranean region. These are possibilities indicated by simulations, not proof that every location will see a specific increase in every season.
Snowfall in Greenland matters because it affects whether the island’s ice sheet gains or loses ice. More snow accumulating on the ice sheet could affect its stability, so scientists need a better understanding of how precipitation there may change—not just measurements of the sea ice around the Arctic. The results suggest that improving the way winter sea ice is represented could also prompt a reassessment of some climate estimates related to snowfall in the region.
The possibility of more rain in the Mediterranean basin matters to parts of southern Europe that face flood risks. The study alone cannot say when rain will fall or how intense it will be in any particular city. It does, however, offer another reason to improve models that examine how changes in the Arctic may affect precipitation farther away. Regional weather is shaped by many factors, but more accurate estimates of the Arctic’s influence can help complete the picture.
So the question reaches beyond the Arctic without becoming a separate issue: knowing how much winter ice there is forms part of understanding changes that may accompany warming elsewhere. The new findings do not prove that every change in European snowfall or rainfall is caused by declining sea ice. They do suggest that some current projections may not capture the full range of possible changes.
Using the Past to Improve Future Projections
One of the study’s most important contributions is not simply its lower estimate of future ice. It also shows a way to test projections. The process starts with natural material preserved on the seafloor, uses it to reconstruct an ancient environment, and then compares that environment with simulations from 27 climate models. The results of the test are then used to adjust the models’ relative weights when estimating what may happen in the modern climate and the future.
The different sources of warming in the mid-Holocene and today make it important to distinguish between a warming’s cause and the ice’s response. Ancient warming was linked to changes in Earth’s orbit, while current warming is associated with human activities. But Fu emphasized that ice melts in response to rising temperatures regardless of what causes the warming. This does not mean every detail of the ancient climate can be transferred directly to the present; rather, the ancient record helps test how well models represent the ice’s response.
The study was published in Nature Geoscience and received partial funding from the U.S. National Science Foundation and a Cambridge fund. Crystal Fu is a member of Girton College at the University of Cambridge. The work puts ancient sediment data at the heart of the discussion about how reliable estimates of Arctic ice are, rather than treating those data as separate from climate projections.
The Bottom Line
By comparing evidence preserved on the ocean floor with simulations from 27 climate models, researchers found that models may estimate more Arctic winter ice than the evidence supports. Chemical markers, including IP25 produced by ice-associated algae, let the team test the models against conditions from about 6,000 years ago, when changes in Earth’s orbit made the Arctic warmer and less icy.
After giving greater weight to the models that best matched the ancient evidence, the researchers removed the bias that had caused historical simulations to overestimate ice cover. Future projections fell by about one million square kilometers compared with earlier estimates, and the range of some possible results narrowed by as much as 85%. The improved estimates may also affect projections of snowfall in southeastern Greenland and rain in the Mediterranean, with potential implications for the ice sheet and flood risks. These are signals from models, not certainty about local conditions, but they show how the ancient record can help make climate projections more balanced.