AIS Integrity Watch
Where the world's positioning signal stops being trustworthy, week by week. Five heuristics over public AIS, aggregated by zone.
0
Anomalies flagged · W29
0
Eastern Mediterranean · W29
0
Vessels observed · W29
Zones × weeks
Flagged anomalies per zone per week, all heuristics combined. The brighter the cell, the noisier the zone. The watch is young: columns accumulate one per week.
| Zone | W29 |
|---|---|
| Strait of Hormuz | 0 |
| Black Sea | 4 |
| Gulf of Finland | 7 |
| Eastern Mediterranean | 23 |
| Southern Red Sea | 0 |
| Gulf of Guinea | 0 |
What kind of noise
Weekly totals by heuristic, from the same sampled sessions. Interference clusters and displaced positions concentrate where the shooting is; the rest is the long tail of a degraded signal environment.
The five heuristics
- Jamming cluster
- Many vessels in one area simultaneously reporting displaced or circular positions: the signature of area GPS interference, not of any single ship.
- Impossible jump
- A position sequence implying a speed no hull can reach. Flags teleporting tracks caused by spoofed or corrupted broadcasts.
- Speed anomaly
- Reported speed over ground far outside a vessel class's envelope, sustained across several messages.
- Dark gap
- A prolonged silence from a transponder inside a sensitive zone, bounded by normal broadcasts before and after. Counted only above a duration threshold.
- Position on land
- Broadcast coordinates that place a vessel solidly inland. Almost always interference or spoofing; occasionally a misconfigured transponder.
How the watch samples
Each week the pipeline opens one live aisstream.io connection covering all six zones and listens for a bounded session (~12 minutes), applying the five heuristics to the raw position stream in memory. Only zone-level counts are kept. Sampling means the numbers are a floor, not a census — comparable week to week because the method never changes, and biased low everywhere shore-receiver coverage is thin (Hormuz notably, where jamming also suppresses reception).
This module never names a ship. Anomaly detection points at zones and signal environments, not at vessels: attribution from open AIS alone is unreliable, and this observatory does not do name-and-shame.