There is a stretch of the Amundsen Sea coast where the sea used to stay frozen until January and now opens in October. It is called Bear Peninsula, it is at 74°24'S 110°12'W in Marie Byrd Land, and the nearest research facility is 568 km away and is shut for the winter. An emperor penguin colony breeds there.
Of the 61 emperor penguin colonies where this site can measure a full sea ice season, Bear Peninsula has lost more of its season than any other, and it loses it in a way none of the other sixty do: at both ends of the year at once.
The numbers
Across 41 usable ice years, 1979 to 2023, the sea ice season within 100 km of the colony has shortened by 26.0 ± 7.0 days per decade. Fitted, that is 322 days of sea ice a year at the start of the record and 207 at the end. A hundred and fifteen days gone, out of a year that used to be frozen for eleven months of twelve.
| measurement | per decade | start of record | end |
|---|---|---|---|
| season length | −26.0 ± 7.0 d | 322 days | 207 days |
| ice leaves | −20.5 ± 6.2 d | 20 Jan | 22 Oct |
| ice arrives | +5.5 ± 4.3 d | 5 Mar | 29 Mar |
| Apr to Nov ice | −1.12 ± 0.96 pts | 77% | 72% |
The two dates are the interesting pair. Most colonies that are losing sea ice are losing it at the break-up end only, which is what you would expect from a warming ocean pulling the end of the season forward. Bear Peninsula is losing it at both: the ice arrives 25 days later in the year than it used to and departs 90 days earlier. Both clear twice their own error bars, and the two add to the season change exactly, because within one ice year the length of the season is the day the ice leaves minus the day it arrives.
That exactness is not a coincidence, it is a rule this project had to learn. Two slopes fitted over different sets of years do not decompose, and subtracting them anyway invents a number. All three fits here run over the same 41 ice years for that reason.
What 22 October means to a bird
An emperor penguin lays in May or June on land-attached sea ice, incubates through the winter dark, and the chick fledges in December or January. A chick that goes into the water before it has grown waterproof feathers drowns. The date on this site's colony pages that matters is 15 December, the rough point at which a chick still cannot swim.
Since 2000, the sea around Bear Peninsula has opened before 15 December in 23 of 24 ice years. At the start of the record it was opening in the third week of January.
Now the caveat, and it is a load-bearing one rather than a legal disclaimer. What is measured here is the sea within 100 km of the colony, averaged over 17 ocean cells on a 25 km grid. It is not the fast ice the birds actually stand on. Fast ice is the sea ice held fast to the coast, it is usually a fringe a few kilometres wide, and it is narrower than a single cell of this product, so it cannot be resolved. We know exactly how much that matters because we tested it against a published colony collapse and the test failed. An early opening of the sea here is a plausible companion to a bad season, never a demonstrated cause of one.
And there is no count on this page or on the colony page, because counting penguins from orbit does not work yet. This site publishes where colonies are and what the ice around them is doing, and says nothing at all about how many birds are there.
The amount of ice is the least interesting number here
Sea ice concentration in the April to November window has fallen 1.12 ± 0.96 points per decade, from a fitted 77% in 1979 to 72% in 2024. That clears its error bar, barely, and it is the sort of change that would be unremarkable on its own. Five points of concentration over 46 years is not a story.
The timing is the story. The same water holds nearly as much ice as it used to and holds it for three and a half months less of the year. That is the general lesson from this archive: when the ice goes is a different measurement from how much there is, and at 15 of 61 colonies the two disagree, sometimes in sign. Cruzen Island's sea ice is increasing while its retreat date runs 13 days per decade earlier.
Two things this record admits about itself
First, the archive's own guesswork. Before 1988 the satellite flew every other day and the data producer filled in the missing days, so 1,557 of the 3,287 daily readings at this colony in that era, 47%, were interpolated rather than observed, and nothing in the data says which. Measured against a modern ground truth and propagated through this particular fit, that is worth ±1.7 days per decade against the fit's own ±7.0. Small, real, printed separately rather than folded in.
Second, and newer: when we asked every fitted series on this site to nominate its own worst break year, Bear Peninsula's retreat date came back flagged. It picks 1991, and with a step allowed the slope softens from 20.5 days per decade to 9.0. The direction survives, so what is in doubt is the size rather than the sign, and the flag does not clear the correction for having looked everywhere, so it may not be quoted as a break on its own. The 1988-onward refit of the same colony does not flag at all. It is on the colony page because we would rather print a question mark than hide one.
Where this sits on the continent
Bear Peninsula is not an isolated case, it is the extreme end of a pattern with a geography. Fifteen of 61 emperor colonies carry a significant trend in when their sea ice leaves, nine earlier and six later, and the earlier ones cluster on the Amundsen and Bellingshausen coasts and the Antarctic Peninsula while the later ones sit in Victoria Land. Bear Peninsula is on the Amundsen coast, in the middle of the falling half, at the far end of it.
The full 46 year record, both dates, the season decomposed into its two ends and every gate each of those trends passed or failed is on the Bear Peninsula colony page, and all 61 colonies are drawn together on the season page. The underlying record is the free NOAA and NSIDC passive microwave sea ice climate data record, 47 files, and it reads end to end in about half a minute.