A low headland of dark volcanic rock on the north side of an Antarctic island, gray water on three sides.
The ground is covered in shallow rings of stone, each one a nest scrape, each one built by a bird carrying pebbles one at a time.
Most of the rings are empty, and they have been empty long enough that the stones have settled into the grit.
On a January morning two small aircraft with propellers fly a grid pattern 220 feet overhead.
They are back on the gravel within hours.
The counting takes far longer.
Why a photograph counts better than a person can
Counting a penguin colony on foot means walking through it, which moves the birds and changes what there is to count.
It also means a single observer holding a number across hours of distraction, wind and cold, on ground where sight lines break every few yards.
A drone does none of that. It flies a fixed grid, overlaps every frame, and the images stitch into one flat map of the whole site at one instant.
An occupied nest on that map is a distinct shape with a distinct shadow, and it can be counted by several people independently and checked against each other.
Here that agreement between observers came out above 0.99, and the median counting error against ground checks was about 5 percent.
The gain is not speed. It is a fixed record anyone can recount.
What the flight found on the ground
Two drones flew at 220 feet and produced 1,802 photographs, stitched into a single image of roughly a square mile.
The raw count came to 19,886 occupied chinstrap nests, adjusted upward to 20,471 to allow for nests abandoned before the flight.
The last full survey of this site was made on foot in 1980 and recorded 49,410 nests.
That is a fall of about 59 percent across 45 years, at a site inside a protected area with no settlement and no fishing vessel tied up alongside.
Two species turned up that nobody had ever recorded here, 314 gentoo nests and five king penguins standing among them.
The site lost three fifths of its breeding birds.
The shape of what is left
The headline number is the decline, and the more useful finding is the geometry.
Historically this colony held roughly ten breeding groups. The drone map resolved 108 separate subcolonies scattered across the same rock.
The birds have not retreated into a smaller dense block. They have stayed spread across the original footprint in thinning fragments.
That is what site fidelity looks like from above. A chinstrap returns to the scrape it used before, whether or not its neighbors came back.
Dense colonies are defended collectively against skuas and other predators, so a scrap of twenty nests is far more exposed than a block of two thousand. Fragmentation is therefore not only a symptom but a mechanism.
The area is unchanged. The density is collapsing.
Why 45 years is the real problem
A single interval this long cannot separate a steady slide from a cliff edge somewhere in the middle.
The colony may have lost its birds gradually since 1980, or held steady for thirty years and fallen away in ten, and the data cannot say which.
That matters because the two shapes point at different causes, and the region has had krill fishing, warming water and changing sea ice across the same decades.
Regular surveys are what would settle it, and regular surveys did not happen because sending people to this headland is expensive and slow.
Software that flags what matters in an image is the other half of this shift, as a student in Ontario showed with telescope frames.
The nest counts, the earlier census and the subcolony figure are set out in a polar journal.
One measurement across 45 years is a line, not a curve.
What changes now that this is cheap
The reason this site went unmeasured for so long was cost, and that constraint has largely gone.
Two operators and a battery case can now produce in one morning what once took a field party a season to attempt badly.
The realistic consequence is annual coverage of sites that previously got one visit a generation, which turns a single number into a trend.
The unglamorous part is that somebody still has to count the nests in the images, and at 19,886 of them that is real human labor.
Polar fieldwork has always been mostly walking and looking, the way crews once searched 3 square miles of Greenland ice shoulder to shoulder.
The flight method, the image count and the site coordinates are recorded in an abstract service.
The birds are countable again, and the next real number arrives next January rather than in 2070.
