Landsat's guano record backdates 18 emperor penguin colonies by 17 years on average

Eighteen colonies backdated by an average of seventeen years is roughly 306 colony-years of occupancy that happened and went unrecorded ^[1]. That revises the survey record, not the population itself: the birds were there all along, the paperwork wasn't. The seventeen years are measured from the date a human first identified each colony ^[12], so the quantity being estimated is survey effort.

The missing piece is the denominator. Satellite surveys put the current total at 66 colonies ^[1], which makes 18 about 27% of the known sites ^[2], but the University of Freiburg team's sample size is not given in the account, and backdating only works on colonies you already know are there. A colony that formed in 1985 and dissolved in 1995 leaves a stain in the archive that nobody has a reason to search for.

Smyley Island is the clearest example. About 6,000 birds in the Bellingshausen Sea, traced now to at least 1989, two decades before its previously known start, which puts the old date near 2009 ^[13][5]. In 2022 the colony split up, some penguins moving onto a large iceberg grounded near the coast, in a season it may have failed completely ^[15]. It has kept appearing since, generally re-establishing near icebergs along the ice shelf edge, most recently in December 2025 ^[16], giving a record of about 36 years ^[3].

The thing a stain does not tell you is how many chicks left it. Landsat cannot resolve individual penguins; the signal is guano ^[10]. Presence means adults gathered on a surface, which is not the same as fledging. Michelle LaRue of the University of Canterbury phrases it accordingly, describing "a degree of resilience" and birds that "seem to be doing OK now" ^[17], and her stated threshold is frequency: one bad year, even a total failure, does not end a colony, whereas failures year after year mean the birds cannot keep up ^[18][19].

The input side, sea ice data, is harder to pin down than the output side, breeding outcomes. Total Antarctic sea ice has been declining since 2016 ^[3], and the 2022 record low preceded the breeding failures British Antarctic Survey researchers called catastrophic in the Bellingshausen ^[8]. Landfast ice, the surface breeding actually requires ^[2], is tracked poorly ^[4], and one study has it trending upward in the Bellingshausen and East Antarctica while declining in West Antarctica and the Weddell Sea ^[5]. A decadal regional trend pointing the right way did not prevent the bad season in that same region.

Some models project the species could disappear by 2100 ^[6]. The phys.org account does not say what those models assume about where the birds breed ^[23], and the second 2026 study is summarised as finding more flexibility than expected without its figures appearing in the text available ^[21][24]. My read, with its conditions attached: this archive widens the error bars more than it moves the centre, and the statistic that would settle the question is failures per colony per decade, which continuous imaging since the early 1980s is uniquely positioned to reconstruct ^[11].