The iceberg that shut Cape Denison for a decade.

For thirty-one summers the sea off Cape Denison was open water. Mean summer sea ice concentration over the 2001 to 2010 ice years: 4.02%. Then, over 2011 to 2020, 53.51%. Nothing else on this continent did that. We scored every one of the 728 measurable penguin breeding sites in Antarctica on exactly that change, blind, with no idea in advance which would come out on top, and Cape Denison came 1st of 728 at nine times its own year to year variability.

The cause was an iceberg with a name, a date, and two separate collisions.

What B09B did

In February 2010 the iceberg B09B collided with the Mertz Glacier Tongue in George V Land, a floating ribbon of glacier about 78 km long sticking out into the Southern Ocean, and broke it off. The tongue became an iceberg in its own right. That December, B09B ran aground in Commonwealth Bay, immediately next to Cape Denison at 67.01 south, 142.66 east.

Two things went at once. The floating tongue that had shaped how sea ice moved along that stretch of coast was gone, and a large grounded berg was now sitting in the bay. The sea ice that used to clear each summer stopped clearing. The Adelie colony at Cape Denison, whose route to open water ran across that ice, collapsed afterwards. The published attribution is to the ice inside the bay, and that distinction turns out to matter more than it sounds, which is the second half of this post.

The number, in the site's own record

Our sea ice archive reads the ocean cells within 100 km of every known breeding site, one value per ice year, from 1979 to 2023. Cape Denison has 41 usable ice years out of 45. Here is what the record does across the event.

Read the right hand column again. The typical Cape Denison ice year before the event carried 284 days of sea ice. In 2012, 2013 and 2019 the sea around the site was ice covered on every single day of the ice year. Not a shorter summer. No summer.

The nine neighbours, which is how we know it is real

A single site ranking first out of 728 could be a broken cell set or a coastline artefact. It is not, because the sites nearest to it fill the rest of the leaderboard. Ranks 2 through 10 are MacKellar Islands, Cape Hunter, Way Archipelago, Cape Pigeon Rocks, Watt Bay, Curzon Islands, Port Martin, Cape Jules and Cape Bienvenue. All Adelie sites, all strung along about 200 km of the George V Land coast, all measured independently, all showing the same jump in the same year. That coherence is the check.

And then it turns into a warning

This is where the story stops being a story and becomes a rule. Every one of those ten sites appears, on a plain reading, to have a strong rising trend in summer sea ice. At Cape Denison a straight line through all 41 ice years gives +9.9 points per decade against a standard error of 2.5, which is nearly four sigma. Textbook significant. Publish it and you would be reporting that summer sea ice off George V Land has been climbing steadily for four decades.

Allow that same fit a single level shift at 2010 and the slope becomes minus 5.1 points per decade, does not clear its own error bars, and the shift comes out at 50.0 points. There is no trend. There is one week in 2010 and a flat line either side of it. All nine neighbours reverse sign the same way.

This site already refuses to fit a line across two known discontinuities: the 1988 satellite changeover, which invented sea ice trends at emperor colonies, and the 2013 arrival of Landsat 8 in the ice velocity record. The Mertz event is the third, it is larger than either, and it is the only one that is real ice rather than an instrument. Nothing in the data file announces it. A satellite changeover is documented somewhere. An iceberg is not.

What this measurement is entitled to say

The step control we run is not a Mertz detector and must not be sold as one. It fires at 49 of the 728 sites across 8 regions. What marks George V Land out is size and consequence: everywhere else the shift runs between 0.4 and 20.3 points and the sign reverses at none of the other 40 sites. The gap between 20.3 and 35.6 points is empty. Where the flag fires elsewhere it means the weaker and still disqualifying thing, that a straight line is a poor description of a series with a regime change in it.

And there is a boundary here that has to be stated carefully. The colony at Cape Denison failed because of the ice inside Commonwealth Bay. Commonwealth Bay is one bay, and at 25 km per cell it is smaller than a single measurement in this archive. We did not see it and we cannot. What we detected is the regional summer sea ice along 200 km of coast, which is what removing a 78 km floating tongue and grounding a large berg did to that coast. The bay killed the colony. The measurement is entitled to the coast, and that is the whole claim.

There are no penguin counts here, at Cape Denison or anywhere else on this site. Counting colonies from orbit is a problem we have not solved and will not pretend to have solved. What we can measure is the ice, and at this one cape the ice did something in a single December that four decades of climate did not.

Cape Denison and the other 728 breeding sites are on the breeding atlas, the thinning that turns 729 sites into 56 independent measurements is on the places page, and the underlying sea ice record is the free NOAA and NSIDC passive microwave climate data record.

One last thing, and it is live rather than historical. B09B is still on the US National Ice Center's tracked iceberg list, which this site refreshes every day. Its last fix, on 7 August 2026, puts it at 66.08 south, 143.19 east, 106 km north of Cape Denison and clear of the bay, carrying 508.94 square kilometres. It has its own page here, at /iceberg/b09b, alongside the other 32 bergs currently under track at /icebergs. Most Antarctic news arrives as a number with no object attached to it. This one has a name, a size, a position, and a decade of consequences behind it.