An archive under the headlines

Every dispatch above resolves to a page with the working shown. There are 889 of them: one per research station, one per tracked iceberg, one per measured glacier, one per known penguin breeding site. All of it regenerated from open data, none of it hand-typed.

And one archive with no headline in it

Antarctic sea ice across the whole satellite record — 16,730 daily readings, 1979–2024 — trends -0.063 ± 0.057 million km² per decade, which is 1.1 times its own standard error and therefore nothing — and 1.9 on the 1988 gate, which is nothing either. That null result is the most quoted number on this continent and almost nobody publishes it as one. What the record does carry is a shape: 345 of the 366 calendar dates in the year have their record low since 2016, and 361 have their record high between 2007 and 2015. The whole record →

And one number nobody publishes

There are 114 research facilities in Antarctica run by 31 national programmes, and every map of them is a map of the dots. Measured on the ice grid, the space between them is the better number: the furthest point on the ice sheet is 924 km from the nearest facility in midsummer, and 1,815 km in winter once the 59 seasonal ones have shut. Half the mapped ice is more than 515 km from anywhere that stays open. How far it is from anybody →

And one map with a finding in it

Plot all 729 known penguin breeding sites in Antarctica and they are not spread around Antarctica. Cut the continent into 36 ten-degree wedges of longitude and three adjacent wedges hold 510 of them — and 4 of the six species never leave that window. The two that do, the emperor and the Adélie, are the two whose habitat this site can measure: the emperor breeds in 34 of the 36 sectors. No counts, and there will not be any until they can be defended. The whole map →

And one archive that counts itself

Every one of those 728 breeding sites now has a 46-year sea-ice record, and the honest headline is how few measurements that is. The patch of sea each record reads is 100 km across, so two sites within 200 km are not two readings of anything: thinned until none overlap, 728 sites become 56 places. The species with the most sites have the fewest — 369 chinstrap sites stand in 8 places, while the emperor's 67 stand in 45. 729 sites, 56 places →

And one picture of all of it at once

Those 728 records drawn together — 29,848 readings, one pixel each, coloured by where that ice year ranks in its own site's record. Nothing fitted. What a page per site cannot show is which years moved everywhere at once, and the answer is the last two the archive holds: in 2023 the sea-ice season came in below its own median at 657 of the 728 sites and at 44 of the 56 independent places — the most in 41 years, against the 28 places any year gets by construction. 2022 is second. Every site, every ice year →

And one number this site got wrong

Every straight line published here — all 5,686 of them — was made to nominate its own worst break year against a null with no break in it, knowing nothing about satellites or icebergs. Almost all of it came back clean: 4 of 5,686 series carry a level shift that survives the correction, and every one is the 1988 instrument change this site already refuses to fit across. One did not. Mertz Glacier was published here as an emperor colony whose sea ice is robustly rising; allow the record a single step in 2011 and the slope reverses, from +1.31 to -2.04 points a decade. Which year did the ice change? →

Start somewhere

The biggest wintering stations, the biggest bergs, and the emperor colonies — the ones that breed on sea ice rather than on land. Or read the working: field notes →

How the ranking works

A feed sorted by recency tells you what happened last. A feed sorted by size tells you what is big. Neither tells you what is unexpected, which is the only thing worth a headline. Every dispatch here carries a stated statistical basis — where a reading falls in its own historical distribution, or how far a fitted trend exceeds its own uncertainty — and is ranked on that. A standing fact with no baseline behind it is capped well below anything measured, so it can never lead.

The same rule cuts the other way. Nine of the 13 glaciers we measure show no detectable change, and that null result is published as a dispatch rather than quietly dropped. There are no penguin population figures anywhere on this site, because counting colonies from orbit does not yet work — our detector correlates at r = −0.16 against ground counts. A number we cannot defend is worse than no number.