The ice around the emperors

A third gate, and it costs this page a number

The two gates above both assume the record is a straight line with noise on it. On 13 August every fitted series on this site was made to nominate its own worst break year — the year at which the level of the record changed rather than its slope — against a Monte Carlo null, with no event supplied. 5 of the April–November trends published here came back flagged, and 5 of them are trends this page calls robust.

They are not all the same kind of problem, and the differences decide what may still be quoted.

1987 is the satellite. The breaks at Gould and Rupert Coast land on the instrument change this page's own second gate already exists for: their robust verdict is fitted 1988 onward and survives, while the 46-year slope in the table below is the one carrying the step. That is the third independent route to a conclusion this site reached in July.

1994 is one place, not two. Barrier Bay and West Ice Shelf are 75.2 km apart, inside the radius at which two sea-ice records stop being independent, so they are one measurement made twice — and it steps down. The year is not an instrument, not an iceberg anybody knows of, and turns up in scattered places a thousand kilometres away, so no cause may be named for it.

2011 is the one that changes a published number. At Mertz Glacier the April–November ice fits a rise of +1.31 points per decade from 1988 and this page published it as robust. Allow the series one level shift and the slope is -2.04 with a step of +8.78 — sign reversed, and the step is seven times the slope it replaces. The scan picked 2011 out of 21 candidate years knowing nothing about icebergs; it is the first full ice year after B09B grounded in Commonwealth Bay. So the count of colonies gaining ice on this page is 4 and not 5.

Which year did the ice change? — all 5,686 series asked the same question →

Where the ice is going, and where it is coming

Grouped into the five standard sea-ice sectors. The Bellingshausen–Amundsen sector is the only one whose colonies average a loss; the Ross Sea sector averages +1.38 points a decade — a gain, at 8 colonies out of 11. A colony sitting near a sector bound falls on one side of a line drawn by convention rather than by geography, which is why the map above is the better read of this.

The steepest fall of all 67 is at the one colony known to have been lost

Emperor Island is the Dion Islands site in Marguerite Bay, first surveyed in 1948 and gone by about 2009. Its April–November sea ice fits 76% in 1979 down to 51% in 2024 — 4.0 standard errors clear of zero, the largest decline of any emperor colony on the continent — and the series drops into the low forties right around the years the colony stopped breeding.

That is a coincidence worth stating and not a cause worth claiming. A 25 km grid cell cannot see why a colony failed, and this measurement is the sea around the site rather than the ice under the birds. What it does establish is that the metric points at the right place without being told where to look.

The same coast, measured a second way

Everything above is how much ice a colony has. Read daily rather than monthly, the same archive answers a different question: when that ice goes. What ends a breeding season is not a low annual average — it is the sea opening before a chick can swim, and that is a date. Measured at 61 colonies, the retreat now runs significantly earlier at 9 and later at 6, and the geography is the one on the map above, arrived at independently. At Bear Peninsula the sea used to clear around 20 Jan and now clears around 22 Oct — 90 days.

See when the ice goes at every colony →

A check nobody tuned for

Five Bellingshausen colonies were published as catastrophic 2022 breeding failures. Ranking all 67 colonies by how far their 2022 fledging-window ice fell below their own record — a measurement built without reference to that work — puts every one of those five in the top seven. The first 9 places on the ranking are one continuous stretch of coast, from the Antarctic Peninsula round to Marie Byrd Land, before a colony from anywhere else appears. Rothschild Island read 0.2% against its own 80% average, 5.3 standard deviations down.

5 colonies had their lowest fledging-window ice of the entire 45-year record in that one year. An independently reported event, reproduced by a metric that was not fitted to it — the strongest external check anything on this site has passed.

All 8 colonies losing ice

Every one of these has an April–November slope more than twice its own standard error below zero. The solid line is the April–November mean, the dashed grey the November–December fledging window, and the orange rule the fitted trend. Same fixed 0–100% axis in each, so the panels can be read against one another.

Every colony, ranked

Slopes are quoted ± two standard errors. A trend is called significant only above twice its standard error.

How this is measured, and what it is not

What it is. Mean sea-ice concentration over a fixed set of ocean cells within 100 km of the colony. Two windows are reported: April–November, the months a colony needs a platform, and November–December, when an early break-out takes the chicks before they can swim. A trend is called significant only above twice its standard error.

It is not fast ice. At 25 km the land-attached ice emperors breed on is narrower than one cell and cannot be resolved. This is the sea around the colony, not the platform under it.

It is not current. The record ends 2024-12-31. It is an archive, not a feed, and must never be presented as the current state of the ice. The live daily figure on the wire is a different product and the two are never blended.

Two sensitivity tests, reported rather than tuned away. Passive microwave suffers land spillover in coastal cells. Repeating every fit with all land-touching cells dropped keeps the sign at 31 of 34 significant colonies; the ones that move are weak positives, and no significant decline changes sign.

A straight line over 46 years is a poor model of a record that changed regime around 2016. Refitting on 1996-2024 alone keeps the sign at roughly three quarters of colonies, so single-colony slopes are weaker evidence than the regional pattern they form.

1 colony is left out entirely: Gipps Ice Rise has 7 ocean cells within 100 km, fewer than 8 ocean cells within 100 km, so no figure is published for it at all.

And still no count of the birds themselves

This is the first number a colony page on this site has ever been able to carry, and it is a number about ice rather than about penguins. There are no population figures anywhere on this site: detecting colony size from satellite imagery does not work yet — our own detector correlates at r = −0.16 against ground counts across eight colonies. All all 729 breeding sites are mapped, and none of them are counted.

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 →