How long the ice lasts

Which end the season is lost at

The bar is the change in season length. It is split into the two ends that produce it: the break-up end, the ice going earlier or later, and the freeze-up end, the ice arriving later or sooner. These are not two separate measurements set beside one another — within one ice year the season length is the retreat date minus the advance date, so with both ends fitted over the same ice years the two segments add to the total exactly. That last condition is the whole trick: fitted over different years, as the two dates are elsewhere on this site, they do not add to anything.

At 5 of the 8 only the break-up end clears its own error bars, at 2 only the freeze-up end, and at Bear Peninsula both — the only one of the 61 colonies measured where the ice demonstrably arrives later and leaves earlier.

The 2 colonies the retreat date cannot see

On the retreat-date map Emperor Island and Pfrogner Point are hollow dots — no detectable change, which is the honest answer to the question that map asks. Their sea ice does not leave measurably earlier than it used to. It arrives 10.6 and 14.1 days per decade later, and their ice seasons are 18.2 and 17.6 days per decade shorter as a result. One of the two is Emperor Island — the Dion Islands in Marguerite Bay, the only emperor colony known to have been lost.

This is what the decomposition is for. A measurement that only watches the ice go will report nothing at a site that is losing its season at the other end.

The steepest of all is losing it at both ends

Bear Peninsula, in Marie Byrd Land, had 322 days of sea ice a year at the start of the record and has 207 now — -26.0 ± 7.0 days per decade, the steepest change in season length of the 61 colonies measured. The ice arrives 5.5 days per decade later and leaves 20.5 days per decade earlier, and it is the only colony where both ends clear the gate on their own, of the 61 measured.

And two colonies going the other way

The majority answer here is no change at all, and the minority is not all one direction. 2 of the 8 colonies with a defensible trend have a longer ice season than they did. They are shown because leaving them out would make the page a selection rather than a measurement — the same reason the map draws its 53 hollow dots at full size. The steepest of them, Stancomb, is the clearest case for the gate: its bigger end by far is the freeze-up one, at +21.1 days per decade, and that end does not survive dropping the pre-1988 sensor — so the page names the other one and the map draws the total.

Every colony with a measurable ice season

Mean season is the average number of days a year the sea around the colony sits above its own whole-record normal. The two end columns add to the days-per-decade column exactly, by construction. The verdict column uses the robust gate — twice its standard error, and still that without the pre-1988 sensor — so it is stricter than the one the retreat-date table uses, and 11 colonies that look significant on the plain fit read as no change here. 6 of the 67 emperor colonies on this site have too few usable ice years to fit a season length at all.

How this is measured, and what it is not

What it is. How many days of the year sea ice is present around an emperor penguin colony, and how much of any change in that belongs to the ice arriving later against the ice leaving earlier. The ice year runs 15 February to 14 February so one freeze-and-thaw cycle falls inside one label; the threshold is each colony's own whole-record mean concentration, fixed. A trend is called robust only where it clears twice its own standard error over the full record and still clears it, with the same sign, refitted on 1988 onwards alone.

The decomposition is an identity, not a model. duration slope = retreat slope − advance slope, over shared ice years. Checked at every colony before writing. Both ends are refitted here over exactly the ice years in which the season length exists, which is why the parts add to the whole. The figures for the two ends on this page will therefore differ slightly from the advance and retreat slopes quoted on the retreat-date page, which are fitted over every year each date exists in. Both are correct; only one pair can be added together. The check is run at all 61 colonies before this page is built.

The freeze-up end is the fragile one, and it is gated harder for it. The freeze-up end is the fragile one. The threshold each colony is measured against is a whole-record mean, so the later, higher-reading sensors set it; where a colony's early-era winter peak barely clears that threshold there is no freeze-up crossing to find and the date lands months late. The robust gate is what removes those, and it removes three quarters of the advance-date trends the plain fit calls significant.

The freeze-up date has a floor. The advance date is floor-censored: it is counted in days after 15 February and cannot be negative, so a colony still above its own mean when the ice year opens scores zero. Every colony carries the count of ice years this affects. Among the 8 colonies drawn as a change here, 7 carry any censored years at all; the worst affected site anywhere in the archive is Rupert Coast at 18 of 38 ice years, and it is not one of them.

It is not fast ice. At 25 km the land-attached ice emperors breed on is narrower than one cell. This is the sea around the colony.

It is not current. The record ends 2024-12-31, so the last complete ice year is 2023. It is an archive, not a feed.

Three questions, one archive

The same 25 km satellite record answers three different questions about the sea around an emperor colony, and they do not give the same answer. How much ice there is across the breeding months. When that ice goes. And, here, how long it is there at all — which is the only one of the three that can tell you which end of the year a colony is losing. Reading them against one another is the point: a colony can hold its ice and lose its season, and two of them do.

There are still no population figures anywhere on this site. Counting a colony from satellite imagery does not work yet — our own detector correlates at r = −0.16 against ground counts across eight colonies — and a number we cannot defend is worse than no number.

The continent behind these colonies

The same archive summed over the whole hemisphere rather than read at a colony says something that sounds like a contradiction and is not. Across all 46 years Antarctic sea-ice extent trends -0.063 ± 0.057 million km² per decade — 1.1 times its own standard error, and from 1988 onwards 1.9, so there is no trend this record can defend on either gate. And 345 of the 366 calendar dates in the year have their lowest reading of all 46 since 2016. A hemisphere with no trend in it is made of coasts moving in opposite directions, which is exactly what these colonies are. See the whole record →