There is a small island in Marguerite Bay, on the southwest side of the Antarctic Peninsula, that is named after the birds that used to breed on it. They do not any more. Emperor Island, in the Dion Islands at 67.86 degrees south, 68.71 degrees west, was first surveyed in 1948 and was empty by about 2009. Of the 68 emperor penguin colonies this site tracks, it is the only one known to have gone outright.
We cannot tell you why, and neither can anyone else with certainty. What the satellite record can tell you is what the sea around that island did over the same years, and the answer is not the one most people would guess. The ice there is not leaving earlier. It is arriving later.
The season closed from the wrong end
Every year the sea around a colony freezes, holds, and opens again. The length of that season at Emperor Island, fitted across 38 usable ice years from 1982 to 2023, has fallen from about 235 days a year to about 160. That is 18.2 days per decade, plus or minus 6.8, and it survives the strictest test this site applies: it still clears its own error bars when the whole pre-1988 satellite era is thrown away.
A season has two ends, and this one is being lost almost entirely at one of them.
| end of the season | days per decade | clears its error bars |
|---|---|---|
| freeze-up, the ice arriving | 10.6 later | yes, and on 1988 onwards too |
| break-up, the ice leaving | 7.6 earlier | no |
| season length | 18.2 shorter | yes, and on 1988 onwards too |
Almost everything published about emperor penguins and sea ice watches the ice going out, because that is what strands a chick that cannot yet swim. Emperor Island is invisible to that measurement. On our own break-up map it is drawn as a hollow dot, correctly, because the date that map colours is the other one. It shows up only when you measure the freeze-up, and then it is one of the steepest sites on the continent.
It is not alone. Pfrogner Point, on the Ellsworth Land coast, has the same shape: no detectable trend in the break-up date, a freeze-up running 14.1 days per decade later, and a season 17.6 days per decade shorter. Two colonies, both in West Antarctica, both of which a retreat-date map cannot see at all. That is why a page on this site measures season length rather than a single date.
How much ice, as opposed to how long
The other way to ask the question is not when but how much. Averaged over April to November, the months a colony needs a platform, the concentration of sea ice around Emperor Island fits from 76% in 1979 to 51% in 2024. That is the steepest decline of all 67 colonies measured, by a clear margin.
And it does not clear this site's strictest gate, which is worth saying out loud because the temptation to quote the biggest version is real. Fitted across the whole record it is 5.6% per decade, four standard errors clear of zero. With an explicit step term for the pre-1988 instrument it is 4.1% per decade and still significant. With that era dropped outright it is 3.9% per decade and comes to 1.93 times its own error, just under the bar of two. Two of the three versions clear it and the conservative one does not. The site says exactly that rather than picking a favourite, and the reason that early era has to be handled separately is its own story.
The year the whole neighbourhood failed
In 2022, five emperor colonies on the Bellingshausen coast suffered catastrophic breeding failure, a result published independently of anything here. Rank all 67 colonies by how extreme their 2022 fledging-window ice was against their own record and four of those five land in the top six. Emperor Island, already empty by then, had the lowest fledging-window ice of its entire 45-year record that same year, 2.65 standard deviations below its own mean. On the separate measure of how early the sea opened, it ranks second of 62 at 3.44 standard deviations early.
None of that was tuned. The metric was built with no knowledge of the paper, and it reproduces the event. That is the strongest external check anything on this site has passed, and it is why we trust the shape of these numbers at all.
What this measurement is not
Three limits, all load-bearing.
This is the sea around the colony, not the ice under it. The satellite grid is 25 km per cell. Emperors breed on land-attached fast ice, which at most sites is a band far narrower than one cell. Thirty-three cells of sea within 100 km of the island go into this number, and none of them is the ground the birds stood on. Calling it the breeding platform would be a false statement, and there is a documented case where the distinction bites: the 2016 failure at Halley Bay was a fast ice break-up inside one inlet and this method ranks it a forgettable 13th of 54.
The archive ends on 31 December 2024. This is a climate record, not a live feed. Nothing here describes the sea around Emperor Island this winter.
There is no count, and there never will be from this site. We do not publish penguin numbers. The stain-detection work that would let us is not good enough, and a number we cannot defend is worse than no number. What we publish is where the 729 breeding sites are, and what the ice around them is doing.
The coincidence, stated and not claimed
The sea ice series at Emperor Island drops into the lowest values of its whole record right around the years the colony stopped breeding. It would be easy, and wrong, to run that as a headline. A 25 km cell cannot see why a colony failed, and Marguerite Bay in the 1990s and 2000s was changing in several ways at once. What is defensible is narrower and still worth knowing: the one emperor colony we know has gone is also the site with the steepest measured decline in surrounding sea ice, and the site whose sea now freezes latest relative to where it started.
Its page here carries every ice year of the record, both fitted ends of the season, and what the archive's own gap filling is worth against each. The source is the NOAA and NSIDC passive microwave sea ice concentration climate data record, free, keyless and daily since 1978. The continent-wide view is at /emperor-ice, where 57 of 67 colonies show no measurable change at all.