In August 1984, a team of NASA scientists sat down to process ten-month-old satellite data and found something odd about a portion of it. The previous October, the ozone-mapping instrument aboard the Nimbus-7 satellite had recorded so little ozone over Antarctica that the processing software, built to catch instrument glitches, had flagged most of the readings as implausible and excluded them from the standard analysis, without erasing the underlying data. The team, led by A. Fleig and including P.K. Bhartia and Donald Heath, faced an ordinary calibration puzzle. A sudden spike in flagged data almost always means a broken sensor, not a broken sky.
Their first move was standard practice: check the suspect satellite numbers against a ground station. The only Antarctic ozone readings reaching them in real time came from the American station at the South Pole, and that instrument showed ordinary values, close to 300 Dobson units, the standard measure of how much ozone sits in the column of air overhead. Against that reading, the satellite's numbers looked like what the software had already flagged them as: a fault. By December, four months after first noticing the anomaly, the team was confident enough in its own data to submit an abstract to a conference eight months away. Farman's group published first.
At the British Antarctic Survey's Halley Research Station, a young physicist named Jonathan Shanklin was digitizing a backlog of hand-recorded ozone readings ahead of a 1983 public open day. The station had been taking the same measurement, almost without interruption, since 1957. Shanklin noticed the spring numbers looked lower each year, though a single year proves nothing; weather varies. What convinced the team was a plot of the minimum eleven-day average across two and a half decades: a decline too steady to blame on any one storm. Farman, Gardiner and Shanklin published it in Nature that May, five months after NASA's own team had privately reached the same conclusion and filed it as a conference abstract for that August.
The following year explained NASA's hesitation. The South Pole station, whose ordinary-looking numbers had reassured the Ozone Processing Team, turned out to have been reading the wrong instrument channels throughout that period: its own data was the broken kind. When NASA reprocessed the archived Nimbus-7 record with that flag turned off, the hole was there in the satellite data too, confirmed and formally published the next year. The instrument built to catch errors had caught a real one. The instrument sent to check it had an error of its own.
A sensor reporting an impossible number and a sensor reporting a true one produce the same symptom: a reading nobody expected. Telling them apart takes a second check, and a second check is only as good as the instrument behind it. The South Pole station had been reading the wrong channels that whole time, and nobody had caught it yet. Halley's team did not have a better satellite. They had the same ground record, the same measurement, every spring, for a quarter century before anyone needed to defend it. That is what an anomaly costs to confirm: not a better hunch, but a longer memory.