Every site, every ice year

The two most widespread short seasons are the last two

A site sits below its own median in half of its ice years by construction, so the number to read every column against is 364 of the 728 sites and 28 of the 56 independent places. In the 2023 ice year it is 657 sites and 44 places. In 2022 it is 642 and 40. No other ice year in the record reaches 34 places, and the third-placed year, 2010, reaches 33.

This is a rank statistic and it is deliberately the same family as the statistic this site's hemispheric record leads on: it fits no line, chooses no period, and asks only which years hold which places in a record that is far too short to carry a slope. What it cannot do is say how much — see the refusals at the foot of this page.

The instrument change was checked rather than assumed

Five of the columns on the left of the sheet belong to a different satellite, and this project has already paid for treating that lightly: the pre-1988 sensor reads sea-ice concentration low, by more the closer to a coast you get, and it once cost this site a published headline about sea ice around the emperor colonies. But a concentration bias and a threshold count are not the same kind of number — the ice edge is a place where concentration falls from a hundred to nothing within a cell or two, so shifting every value by a few points moves a count of days above 15% hardly at all. That is an argument, not a result, so here is the result. Refit the medians and rerun the count on the 36 ice years from 1988 alone: 2023 is still first at 659 sites and 44 places, and 2022 is still second at 650 and 40. The headline does not depend on the older instrument, and the page says so because the check could have gone the other way.

Which is not the same as saying the left edge of the sheet may be read. All five of the pre-1988 columns sit below the fifty per cent line — the emptiest is 1981, where only 102 of the 728 sites and 18 of the 56 places rank below their own median — and in the atlas window they are a nearly solid block of long seasons. Two things are true of that block at once and this archive cannot separate them: it is the start of a record this site already publishes as a real shortening of the ice season at the rock-nesting sites, measured inside the homogeneous era at both ends; and it is also a different satellite. The count above is quoted because it holds either way. The block is not.

The tall panel is not the strong half of the evidence

The left panel holds 509 of the 728 sites and 12 independent places. The right panel holds 219 sites and 44 places. That is the whole of 729 sites, 56 places arriving as a picture: a site's record is averaged over the sea within 100 km, so two sites less than 200 km apart are largely reading the same water, and the left panel is mostly the same water drawn several hundred times. Every region below is printed with its independent places beside its sites for that reason — 12 regions have every one of their sites below its own median in 2023, and between them they hold 15 independent places.

And the thinning itself is arbitrary, so the count was shuffled before it was published. The rule — keep no two sites within 200 km — is not arbitrary, but which site stands for a stretch of coast is, because the thinning keeps whichever it reaches first. Shuffle the order 200 times and the kept set runs from 51 to 59 places, 2023's count of them runs from 38 to 47, and 2022's from 33 to 43. The count moves and the verdict does not: 2023 is the year with the most places below their own median in 197 of the 200 shuffles. That is the rule this archive already follows, applied to a new number. The count and the direction survive the arbitrariness; a magnitude read off a thinned list never does, which is why there is no number of days anywhere on this page.

The 56 independent places, one by one

The thinned set this page counts on, each with its own record. 44 of them measured a shorter sea-ice season in the 2023 ice year than their own median across the 41 years, and 12 did not.

What may not be read off this sheet

No trend, anywhere. Nothing on this page is fitted. 29,848 ranks are not 29,848 data points about a direction, and this record is not long enough to carry a slope through — which is exactly why the page is built out of ranks.

No size. The colour is a rank inside one site's record, so a deep orange cell says among the shortest seasons this site has had and never how many days shorter. How much shorter is a different question, it is answered per site rather than per place, and the place-weighted size of it cannot be quoted at all.

The four blank columns are a coverage gate, not four years without ice. The 1984–1987 ice years fail the archive's own 95% coverage rule at every one of the 728 sites, so they are drawn as gaps. A line through them would invent four years of record.

This is not the current state. The archive ends 2024-12-31, so the last complete ice year is 2023 and the sheet stops there. It is a climate record, not a live feed — the live sea-ice number on /sea-ice is a different measurement of a different thing.

It is the sea around a site, never the fast ice under it. At 100 km a cell is far larger than the ice a colony stands on, and this archive has already been checked against two published breeding failures to find out which class of event it can see: a regional collapse, yes; one bay's fast ice breaking up, no.

And there is no penguin count on this page, or anywhere on this site. Every row here is a measurement of sea ice at a place penguins are known to breed. It is not a measurement of the penguins. Detecting colony size from satellite imagery does not work yet, and a number we cannot defend is worse than no number.

Where this sits

Provenance