On 9 August the Southern Ocean lost 207,000 square kilometres of sea ice in one day. It is the middle of the Antarctic winter, the sun has not risen over most of the continent for weeks, and the ice is supposed to be growing. It went backwards instead, by about the area of Belarus.
The reading that morning was 16.085 million square kilometres. That is the 4th lowest 9 August in the satellite record, and 1.274 million square kilometres below the average for the date. The missing ice is about one and a quarter times the area of Egypt.
What the number actually counts
Sea ice extent is not a measurement of ice. It is a measurement of sea. A satellite passes over, splits the Southern Ocean into cells 25 kilometres on a side, and works out what fraction of each cell is covered. Every cell that comes back at 15 per cent or more is added to the total in full. A cell that is 16 per cent ice counts the same as a cell that is solid.
That sounds like a flaw and it is the opposite. The 15 per cent line sits at the edge of the pack, where concentration falls from near total to open water within a cell or two. A threshold at a cliff edge is stable. Shift every concentration reading in the archive by three points, which is roughly the size of the worst known instrument problem in this record, and extent moves by at most 0.20 million square kilometres and usually by nothing you could measure. It is one of the most robust numbers in climate science, which is exactly why it is the one on our front page.
Forty eight years of record, forty three readings
The continuous record starts on 26 October 1978. For 9 August that gives 48 years of calendar coverage, but only 43 actual readings, and the gap is worth knowing about because most reporting quietly rounds it away.
The first satellite in the series, Nimbus-7 SMMR, ran every other day. It did not break, it was designed that way. So the early years of the record have holes in them, and a rank against "every previous reading on this date" is a rank against 43 numbers, not 48. Drop that early sensor entirely, which is the stricter test this site applies to anything it publishes, and today's figure is the 4th lowest of 39. Same answer, smaller denominator, and now nothing in it depends on an instrument that stopped flying in 1987.
| Reading | Value |
|---|---|
| Extent, 9 August 2026 | 16.085 million km2 |
| Average for 9 August | 17.359 million km2 |
| Anomaly | -1.274 million km2 |
| Rank, all readings | 4th lowest of 43 |
| Rank, 1988 onward | 4th lowest of 39 |
A one day fall in midwinter is weather, not melt
Nothing melted on 9 August. At that latitude in August there is no sunlight to melt it with. What moves the number on a one day timescale is wind: a storm pushes the pack together, floes that were spread across four cells at 20 per cent concentration crowd into two cells at 40, and half of them drop below the 15 per cent line and stop being counted. The ice is still there. The sea it was spread across is not covered any more.
This is why a daily figure is a reading and not a trend, and why we publish the rank against the same calendar date rather than the raw value. The seasonal swing in Antarctic sea ice is enormous. In the 46 year archive the median year runs from a minimum of 2.79 million square kilometres in late February to a maximum of 19.09 in September, a sevenfold breath in and out. Against that, a 200,000 square kilometre daily wobble is noise. Against the 43 ninth of Augusts on record, being fourth from the bottom is not.
The annual maximum is still ahead. Across the 46 complete years from 1979 to 2024 it landed on 19 September at the median, and never outside the window from 29 August to 2 October. Six weeks of growth left, and this year starts them 1.27 million square kilometres down.
So is Antarctic sea ice shrinking?
The honest answer is that a straight line cannot describe what this record has done, and we will not draw one.
Fit a trend across the whole homogeneous archive, 1979 to 2024, and Antarctic sea ice extent falls by 0.063 million square kilometres per decade with an uncertainty of 0.058. That is 1.1 times its own error bar, which is nothing. Fit the same trend from 1988 onward, dropping the early sensor, and it falls by 0.171 with an uncertainty of 0.085, which clears at 2.0 times. Two periods, two answers, one of them significant and one of them not. When your two defensible gates disagree, you have no number.
What the record says without any fitting at all is far more interesting. Ask, for each of the 366 calendar dates in the year, which of the 46 years holds its record low and which holds its record high. No period is chosen. Nothing is fitted. The answer separates almost perfectly: 345 of the 366 record lows fall in 2016 to 2024, and 361 of the 366 record highs fall in 2007 to 2015. Two adjacent windows out of forty six years. Drop the pre-1988 sensor and it becomes 366 and 366 exactly.
2023 alone holds 253 of those record lows, including every date from 1 May to 24 August, 116 consecutive days on which no other year in the record has ever been lower. Antarctic sea ice climbed to a maximum around 2014 and fell off a cliff after 2016. A single regression line through a rise and then a fall describes neither half, and reports a small number with a large error bar as though it were the finding. The shape is the finding.
The whole 16,730 day record is drawn out on our sea ice page, including both fitted lines across a shape neither of them fits. The live daily number and the rest of the continent are on the map, and if you want to know why we built any of this, start here.
Where the number comes from
The daily figure is the NSIDC Sea Ice Index, product G02135, published as one open CSV file with no key, no account and no paywall, unbroken since October 1978. The 46 year archive behind the trend figures is the NOAA and NSIDC passive microwave climate data record, version 4, read daily and ending 2024-12-31. The two are different products with different land masks, the archive reads about 2 per cent high against the index, and we never splice them.