Antarctica's 729 penguin sites are really 56 places.

The chinstrap penguin has 369 known breeding sites in Antarctica, more than any other species. Measured against the satellite sea ice record, those 369 sites are eight measurements. The emperor has 67 measurable sites, a fifth as many, and they are 45. The species with the most sites has the fewest places, and working out why is the most useful thing this archive has produced about itself.

Across all six species, 728 measurable breeding sites reduce to 56 independent places. That number is not a criticism of the site list. It is the error bar that belongs beside every count published off it, including our own.

Where the shrinkage comes from

Sea ice concentration in the polar satellite record is delivered on a 25 km grid. To give a breeding site a sea ice history you take the ocean cells within a radius of it and average them, one value per ice year. Our radius is 100 km, which is a compromise: small enough to be about that stretch of coast, large enough to hold at least eight ocean cells at almost every site.

Which means any two sites less than 200 km apart share cells. Not metaphorically, the same physical pixels of the same satellite pass, contributing to both records. Two sites 12 km apart are one measurement written down twice. The South Shetland Islands hold 144 breeding sites. Every one of them reads very nearly the same water.

So we thinned the list. Walk the 728 sites in a fixed order, keep a site only if it is at least 200 km from everything already kept, and see how many survive. Fifty-six do.

The pattern in that table is not statistical, it is geographic, and we wrote up the geography two days ago. Chinstrap, gentoo, macaroni and king breed on bare coastal rock, and almost the only bare coastal rock in Antarctica is on the Peninsula and the islands north of it. Those four species have 453 breeding sites between them, and those 453 sites stand in eight independent places. The emperor and the Adelie go right round the continent: 347 sites, 55 places.

One twelfth of the coastline holds most of the dots. It also holds most of the duplication.

What this changes about what we can say

It sets a rule this site now follows. For the four rock-nesting species, any figure counted across their sites is really a figure about the South Shetlands and the Peninsula, counted several hundred times, so it must be published per place or published with its thinned count beside it. Never per site alone.

Here is the rule doing work. The length of the sea ice season at rock-nester sites went from a median 112.9 days a year over the 1988 to 1997 ice years to 74.0 days over 2014 to 2023. A season roughly a third shorter in three decades. It is shortening at 104 of those 453 sites and lengthening at none of them, which sounds like an overwhelming result and is a real one.

Thin those 104 sites and they are three places: Barrientos Island in the Aitcho Islands at 16.7 days per decade, Aspland Island off Elephant Island at 16.0, and Cone Island in Marguerite Bay at 29.7. Three. Both statements are true and only one of them is a count of evidence.

Over the same two decades, the emperor's sites went from 339.9 days of sea ice to 335.5, and the Adelie's from 294.1 to 307.2. Nothing. The species that live where the sea ice is are not the ones whose sea ice season has moved.

The number we will not print

There is a figure this analysis produces that we refuse to publish, and refusing is the point of saying so out loud. If you thin the list and then average the change across the surviving places, you get a place-weighted size of the loss. Run the thinning 200 times with the walk order shuffled and that figure lands anywhere between 6.6 and 39.0 days per decade for the chinstrap, because the greedy thinning keeps whichever representative it happens to reach first.

What survives the shuffling is the count of independent places, which stays between 7 and 9, and the direction, which is a loss in 200 of 200 attempts for all four rock-nesters and crosses zero for the emperor and the Adelie. So: quote the per-site number as what it is, the typical breeding site, or quote the direction. Never quote a magnitude off a thinned list. Most analyses that thin a dataset for independence stop at the thinning and publish the average. We ran the thinning 200 times specifically to find out whether we could, and the answer was no.

Two things this is not

It is not a claim that 673 of the breeding sites are worthless. They are 673 real places with real birds on them, and for anything measured on the ground, a nest count, a diet sample, a tracking study, each one is its own observation. The thinning applies to one specific thing: a satellite sea ice record read at 25 km resolution. Match the independence of your dataset to the resolution of your instrument, not to the length of your list.

And it is not a penguin count. This site publishes positions, sea ice and season length, and no colony populations of any kind. Our own attempt at counting colonies from orbit by detecting guano stains returned a correlation of minus 0.16 against ground counts, which is worse than useless, so there are no counts here and there will not be until that changes.

All 729 sites are drawn on the breeding atlas, the thinning is drawn twice on the places page, and every site has its own page with its own 46 year sea ice record under the site directory. The underlying record is the free NOAA and NSIDC passive microwave climate data record, which runs from 1979 to the end of 2024.