THE BIG MELT
A data-driven story of sea-ice decline by Lukas Kreibig
The Arctic is warming nearly four times as fast as the rest of the world
Sea-ice melt is no longer a forecast
The Place
How the Ice Is Vanishing
In the Arctic, sea ice is not just scenery. For many places it is route, workplace, and risk. What changes when that season becomes shorter and harder to trust?
Greenland
The world's largest island is mostly ice sheet. Along the rocky coast, deep fjords cut inland and create sheltered places where people have lived with the Arctic sea for millennia.
Uummannaq Bay
600 kilometers north of the Arctic Circle, this bay opens into one of Greenland's most dramatic fjord systems. In winter, the sea itself becomes the road.
Uummannaq Island
About 1,300 people live here beneath a distinctive heart-shaped mountain. In Uummannaq, ice is not scenery. It is mobility, timing, and safety.
The question is how that reliability is changing.
The Place
Heart of a Seal
Uummannaq takes its name from the mountain known as the Heart of a Seal. Beneath it, the town sits between harbor, fjord, and winter routes.
On the Ground
When the fjord freezes, the map changes. Routes that do not exist in summer become possible by sled, snowmobile, or on foot, but only while the ice can be trusted.
On the Ground
Fishing still shapes Uummannaq. But the story is not only climate loss: ice conditions, harbor access, markets, boats, and deliberate ice breakup all affect daily life.
The Voices
Voices from Uummannaq
In the Life on Thin Ice study, residents describe how winter ice, travel, hunting, and fishing are changing. Their observations are where this measurement begins.
The Measurement
Only a Feeling?
But how do you test an observation like that? From above. For years, ESA's Sentinel-2 satellites have passed over Greenland and captured the fjord: image by image, winter by winter, until they add up to a comparable record.
First, the view from above: this is Uummannaq as the satellite sees it.
Now the raw winter image. Scenes like this are the basis for the measurement.
Since 2017, ESA's Sentinel program has provided regular images. Thousands of scenes have to become one consistent record.
A computer-vision pipeline reads each image. It masks land and dense cloud, then separates ice from open water in what is left.
The overlay is not the conclusion. What comes out is a single number per day: the share of the area that could be judged at all that day which the pipeline reads as ice. Land, cloud and data gaps are not in the denominator. A fjord frozen shore to shore under a clear sky therefore comes out at 100 percent, and a cloudy day is not marked down simply because there was cloud in the frame. That is the number every chart from here on shows.
The Measurement
The satellite image is still a photograph of light. It has texture, shadows, snow, water, and uncertainty.
The Mask Is a Reading
The overlay turns that image into classes. It does not replace judgment; it makes the judgment visible.
Pixel by Pixel
Each classified pixel is small. Together, thousands of them become an estimate of ice cover for that scene.
Then It Becomes a Time Series
Only after the same procedure is repeated across many scenes can one winter be compared with another.
The Measurement
One Winter, Day by Day
Watch one season fill in: the ice forms in February, holds through spring, then breaks apart by June. Each mark is one day, most of them one satellite scene.
Memory Meets Measurement
Now all measured seasons are in view. Each row is one winter in Uummannaq; each small mark is one day.
The Earlier Measured Winters
The winters of 2017 to 2020 are the comparison point, not a perfect past. They are the earliest stretch this chain can measure without gaps.
The Later Measured Winters
The winters since 2021 carry the summary value, measured against those first four years. It is a guide to the shift, not the whole explanation.
What the Chart Cannot Decide
Clouds, missing scenes, and local use remain part of the uncertainty. Weather, fishing economics, harbor access, deliberate ice breakup, and local knowledge all shape what ice means in practice.
The Pattern
Season by Season
Each small chart shows one Uummannaq ice season from February to June. The point is not one dramatic year, but the rhythm across years.
What the Record Shows
The highlighted seasons from 2021 on sit lower in the same calendar window than the four winters before them.
Less ice in the measured season
The satellite record and local memory point the same way: earlier breakup, while leaving room for local causes and year-to-year variability.
The Pattern
When "Normal" Shifts
To make the shift legible, we compare the first four measured seasons with the later measured years.
The early yardstick
The blue line shows the 2017 to 2020 mean. The light band shows how much those years varied.
A Changed Season
The red line shows the mean from 2021 on. In this comparison, the later years sit lower across much of the season.
That does not explain every local decision, and it is not yet proof: ten winters are too few to establish this difference statistically. The direction is the same under every defensible analysis. The certainty is not.
The Pattern
Back to the Opening Memory
The story began with a memory that spans a lifetime: as a child, the ice held into June or July. Ten winters of satellite data is far too short to test a childhood recollection. But it can ask the same question in numbers. Each dot is the day one Uummannaq winter's ice broke up.
Earlier, on Average
Across the measured years, breakup moved from a late-May average toward mid-May, about ten days earlier. Ten winters is a short window, so read it as a direction, not a fixed date.
And Harder to Predict
The bigger change is the swing: recent winters break up as early as late April or as late as June. That matches the other thing residents stressed: the season has become less stable and harder to trust. The satellite and the people on the ice describe one change from two sides: one measures the area, the other lives the risk.
The Consequences
The Arctic
From Uummannaq to the Arctic
Uummannaq shows one local shift. The Arctic-wide record gives the scale: winter maximums, summer minimums, old ice, and rapid regional warming.
What Summer Leaves Behind
At the end of every summer the sea ice pulls back to its minimum. What is left has been measured without a gap since 1979.
This is what September looked like in the 1980s: a closed sheet reaching far off the coasts of Canada and Siberia.
Around the turn of the century the edge starts to give way. Where ice used to lie, the end of summer leaves open water.
Today an area larger than Greenland is missing by the end of summer. The pale outline shows where the ice still stood in 1980.
The Arctic
The Arctic Seasonal Record
Each line traces one year of Arctic sea-ice extent, from autumn minimum to spring maximum and back again. Blue lines show older years, red lines show recent years.
1979 to 2000: The Cold(er) Half
The first half of the record sits visibly higher. The ice had more room to recover through the year.
2000 onward: The Recent Half
In the more recent years, the lines sit lower through much of the season.
Latest Available Year
Finally, isolate and compare it with the rest of the record.
Below the Earlier Range
Recent years often sit below the older envelope, especially around the winter maximum and late-summer minimum.
The Arctic
Five Decades, One Direction
Each colored band represents a decade of Arctic ice measurements, against the 1981 to 2010 average. Watch them appear one by one and notice the unmistakable staircase downward.
The 1980s: The Old Normal
This soft blue band is our starting point, and it stays above the line all year. The 1980s run 0.34 to 0.97 million km² above the 1981 to 2010 average, with the widest surplus in late September.
The 1990s: First Cracks
The band drops but never crosses over. At 0.09 to 0.40 million km² above the average, the 1990s are the last decade that spends the whole year on the ice-rich side of the line.
The 2000s: A Lower Baseline
The curve moves below the line and stays there, every day of the year, 0.27 to 0.79 million km² short of the average. On 68 of those days the shortfall passes 0.6 million km².
The 2010s: Acceleration
The decline steepens. On 345 of 365 days the gap now exceeds 0.6 million km², the decade runs 1.07 million short across the year, and in early October it reaches 1.87 million.
The 2020s: A Lower Range
The current decade sits lowest of all: 1.21 million km² below the 1981 to 2010 average across the year, and 2.21 million short in mid October.
The data shows no short-term return to the older range.
The Arctic
Below the Average
Each bar shows how far a year finished above (blue) or below (red) the 1981 to 2010 average. Zero is that fixed reference, not the average of every year on the chart.
The 1980s sit on average 0.47 million km² above the line, the 1990s only 0.18. Nine of ten years finish above it in both decades, so what shrinks is the distance, not the frequency. From 2000 both give way at once: not a single year above the reference, and the 2010s average 1.11 million km² below it.
The reference line does not drift with the data. It stays on the 1981 to 2010 average, the same line the decade chart uses. The bars are what moved.
The Drivers
The Drivers
The Warming Signal
Three lines, one story: global CO₂ emissions (green), global temperature (blue), and Arctic temperature (red) have climbed for decades.
The physical link is well established: more greenhouse gases trap more heat. In the Arctic, that warming shows up especially strongly.
The Arctic is an early warning system because regional change there connects back to weather, oceans, and climate elsewhere.
The Drivers
Apples to Apples
How do you compare CO₂ emissions (measured in tonnes) with temperature (degrees) and ice extent (square kilometers)?
By converting each series to the same kind of distance from its own historical average. Different units become comparable without pretending they are the same.
The Mirror Image
When we flip the ice-loss data upside down, the pattern becomes easier to compare: rising CO₂ and temperature move against falling ice.
Three different measurements do not become the same thing. They become readable on the same scale.
The Drivers
: The Latest Comparison
In the latest available year, , the Arctic temperature anomaly sits well above the global mean.
The red bar is more than a number. It shows how unevenly warming is distributed.
The comparison is a reminder that global averages can hide sharp regional change.
Scientists expect ice-free Arctic summers before 2050
The Arctic is the planet's canary
Will we listen?
Listening is not the same as measuring.
The satellites count the ice. The people of Uummannaq live with what its absence takes: routes, timing, trust, a way of being on the sea. The record here is ten winters; the memory is a lifetime. Both point the same way.
The future is unwritten. What we make of the measurement and the memory is the open question.
Lukas Kreibig
Data journalism, coding & visual design.
© 2026, all rights reserved
Contact
Phone: +49 (0) 176 444 69 498
Email: lukas.kreibig@posteo.de
Sources & Methodology
The state of the records could not be retrieved.
Chat with Knud Rasmussen
This chatbot is not Knud Rasmussen, and it was not trained on him. For every question it retrieves matching passages from “Eskimo Folk-Tales”, the collection of Greenlandic legends gathered by Danish-Greenlandic polar explorer Knud Rasmussen (1879 to 1933), and lets a language model (Claude Haiku 4.5 via OpenRouter) answer from them. Nothing is invented: what is not in the collection, it cannot tell.
Known as the 'Father of Eskimology,' Knud Rasmussen conducted groundbreaking expeditions across the Arctic from 1912 to 1924. Fluent in Greenlandic, he collected over 20,000 pages of Inuit oral traditions and stories of spirits, shamans, hunting, and Arctic life.
The 'Eskimo Folk-Tales' contain hundreds of stories: from Tulugaq the great raven who brought light to the world, to Sedna the sea goddess. These tales offer unique insights into the spiritual world of the Inuit.