In November 2012 an Australian research vessel approached an island in the Coral Sea east of Australia. The charts on board had it, and so did Google Maps: an elongated patch some 24 kilometres long. The ship sailed there. Nothing was there, only open water, with the echo sounder reading about 1,400 metres of depth below it.
The crew had not discovered an island. They had undiscovered one.
How an island survives two hundred years
The origin probably lies in the 18th century. A chart connected with one of James Cook's expeditions records a sandy island in this area. What was actually sighted back then is an open question: shoals, floating pumice on the surface, or a position that slipped. Whoever made the entry could not verify it. Whoever copied it had no reason to.
From there on a mechanism takes over that still applies today: charts are not resurveyed, they are copied. Anyone drawing a nautical chart works from existing ones. An island once recorded needs no confirmation from anybody. It would have to be actively disproved, and that would mean sending a ship to a spot where there is nothing to gain. For navigation the entry was even useful: an island that does not exist is one you give a wide berth to anyway.
Charts are not resurveyed. They are copied.
— TerraTally · History desk
The French hydrographic service struck the island from its official charts as early as the 1970s, after the area had been overflown. That helped only so much. Other chart series kept it, and when chart holdings were digitised during the 1980s and 1990s, the entry migrated into global geodatabases.
1774
Cook expedition: a sandy island appears on the chart.
1970s
The French hydrographic service deletes the entry after an overflight.
2012
A research vessel sails through: open water, about 1,400 metres deep, the final “undiscovery”.
This is the part that goes beyond a curiosity. As long as charts existed on paper, an error stayed with one publisher and one print run. Once chart holdings were moved into databases, the same error gained a new property: it could be reused any number of times without anyone checking where it came from.
Sandy Island therefore showed up in services that had it not from a nautical chart but from a dataset that had it from another database, which had it at some point from a chart. The error was not examined, it was passed on. Repetition looked like confirmation: anyone comparing two sources found the same island and took it as established.
After 2012 the island was removed from the major mapping services and datasets. What it took was a ship with an echo sounder heading for that exact spot, and a crew unwilling to let the contradiction between chart and measurement rest.
What remains striking is the ratio of effort to effect. Putting an island on a chart cost one stroke of a pen. Getting it off again took two hundred years and a research voyage.