Coverage
What VPM sees today — the seas it covers, the ports it resolves, how far its history reaches, how often it refreshes, and how coverage expands from here.
Where VPM works today
VPM runs live across the Baltic Sea and the North Sea, with individual vessels tracked well beyond that where their routes lead — the North Atlantic, the Mediterranean, Southeast Asia.
The engine is geography-independent. It does not model a region; it models a vessel's repetition. A ship with a consistent route predicts well wherever that route runs — one of the strongest vessels in the fleet works the North Sea, not the Baltic. What limits coverage is not the sea, but how much of a given port's traffic is being tracked.
The fleet
VPM tracks a live fleet across the Baltic and North Sea, and onboards new vessels continuously. There is no fixed ceiling on fleet size — adding vessels is a matter of ingestion and history, not of what the engine can carry.
Coverage is deliberately built where it is useful rather than wide for its own sake. For predicting a single vessel, a focused fleet is enough. For the port-side view — "who is arriving at this port" — coverage of a region's whole traffic is what matters, and that is the direction of expansion.
Ports
- 254 ports in the core geofence — a curated set plus UN/LOCODE coordinates — each with a name, position and detection radius.
- 461 ports appear as live destinations in the Port View: every port the tracked fleet has called at more than once.
- Around 550 ports carry country attribution, for flag and region filtering.
A vessel position resolves to the nearest port within radius. Port names are normalised — "Hamina (Fredrikshamn)", "Hamina Fi" and "FIHMN" all collapse to one place — so history and predictions line up.
How far back it reaches
Vessel histories reach back to 2020–2021, totalling more than 17,500 recorded port calls. The prediction model, though, uses at most 36 months of history per vessel: back-testing shows that for around nine vessels in ten, older history adds nothing, and for many the best window is shorter still. More past is available; the model has measured how much of it is actually predictive.
How often it updates
| Signal | Refresh |
|---|---|
| Live vessel positions (terrestrial AIS) | Continuous |
| Paid position feed (outage backup) | Every 12 hours |
| Predictions recomputed | Every 2 hours |
| Movement model retrained | Every 24 hours |
A prediction is locked once per voyage — made at departure, scored on arrival — so accuracy reflects a genuine forecast rather than a destination read off AIS mid-passage.
How coverage grows
Two things happen when a vessel is added:
- That vessel becomes predictable once it has enough history.
- Its track contributes to the shared route network — the straits, canals and roadsteads every other vessel also uses. Sea-route distances, ETAs and canal-transit logic get sharper for the whole fleet, not just the new ship.
A port becomes commercially useful at the point where enough of its arriving traffic is tracked to make the Port View complete rather than partial. That threshold, per port, is what expansion targets.
The size of the opportunity
Full coverage of the Baltic is a defined target, not an open-ended one. Published traffic figures for the region:
These describe the market, not VPM's current fleet. They set the ceiling of what full Baltic coverage means.
Other seas
The same engine runs anywhere there is AIS coverage and shipping-lane density data to route against — HELCOM density for the Baltic, EMODnet for the North Sea, the Channel and the Bay of Biscay, and a global lane grid beyond that. Extending to a new sea is a data-onboarding exercise, not a rebuild.