What counts as a false match in AIS-optical correlation

A daily correlation run between optical detections and AIS broadcasts produces two kinds of output a watch desk has to sort through: matches that shouldn't have happened, and unmatched detections that aren't dark vessels at all. Both get lumped together as "false matches" in casual shop talk, but they're different failure modes and they need different checks before anyone logs a track as suspect.

A false positive match: wrong pairing, both vessels real

This is the case where a detection gets linked to an AIS track that belongs to a different vessel. It happens most often in three situations.

First, timing lag. A satellite pass and the nearest AIS position report are rarely simultaneous. If a vessel is making way at 10-12 knots and the AIS fix is six or eight minutes off from the image timestamp, the reported position can sit several hundred meters from where the optical sensor caught it. In open water that's usually fine. In a transit lane with three or four vessels converging, that drift is enough to pair a detection with the wrong hull.

Second, vessel density. Anchorages, transshipment points, and narrow approach channels put multiple AIS-broadcasting vessels inside one matching radius. A correlation pass that uses a fixed distance threshold will sometimes grab the nearer AIS track instead of the correct one, especially when two vessels of similar size are holding close station.

Third, AIS position interpolation. Some AIS feeds fill gaps between reports with a straight-line estimate of where the vessel "should" be. If a vessel changed course or stopped between reports, the interpolated point can land somewhere the vessel never occupied, and a detection near that estimated point gets matched on bad data.

None of these mean a vessel is hiding anything. It means the correlation logic paired a real optical detection with the wrong AIS point, and a reviewer needs to separate that from an actual AIS gap.

An unmatched detection that isn't a dark vessel

The flip side is a detection that comes back unmatched for a reason that has nothing to do with a vessel running dark on purpose.

Small craft under the AIS carriage threshold for their flag state won't broadcast at all, not because they're evading anything but because the rule doesn't apply to them. A skiff or a small day boat working inshore can generate an unmatched detection every single pass.

Satellite AIS coverage itself has known gaps, particularly in dense coastal zones where terrestrial receivers are saturated and satellite pickup is patchy. A vessel can be broadcasting fine and still not show up in the AIS record your correlation pulls from.

And on the optical side, not every detection is a vessel. Sun glint, breaking whitecaps in a swell, and cloud shadow at the edge of a scene all produce returns that look like a small hard target at 0.5-2 m resolution. These get flagged as unmatched because there's no AIS broadcast to compare them to, since there's nothing there to broadcast.

What to check before you log it as dark

A reasonable review order, before a track goes on the suspect list:

• Pull the AIS timestamp closest to the image time and check the gap. Anything under a couple of minutes is low risk for a mismatch.
• Check vessel density within the matching radius. One AIS track nearby is a clean case; three is worth a second look.
• Check size class against the detection. A 40 m longliner detection matched to a 90 m reefer's AIS point is a red flag for a wrong pairing, not a confirmation.
• Check the location against known AIS blackspots or non-carriage zones for small craft before assuming the vessel is concealing itself.

Running this by hand, pass after pass, is most of what eats a watch desk's time on a correlation workflow. Illegal Fishing Detection builds the daily unmatched-detection list with position, size class, and time already attached, so the review starts from a shorter, better-qualified set instead of a raw gap search across the whole AOI. Take a look at the daily unmatched detection list to see what that review queue looks like before a patrol asset gets tasked.