How VPM works
VPM predicts which port a vessel will call at next — and, from that, when it will arrive, how long it will stay, and what it will need when it does. This is the whole pipeline, end to end.
What VPM predicts
VPM answers one question in advance: which port will this vessel call at next. The next call is the anchor. Everything a supplier actually cares about is derived from it — bunker demand, the arrival and departure windows, voyage cost, CO₂.
It also predicts the second and third calls ahead, so a plan can look further than the next few days.
What goes in
- Live vessel positions from terrestrial AIS — AISStream and the Finnish Digitraffic network — with a paid feed held in reserve for outages.
- Each vessel's history of port calls, going back to 2020–2021.
- Port coordinates from UN/LOCODE, so a position resolves to a named port.
- Real shipping-lane density (HELCOM, EMODnet) so distances follow the sea route a ship actually takes, not a straight line across land.
- Vessel particulars — engine power, service speed, fuel type, tank capacity — collected by hand from class records and owners' fleet specifications.
No proprietary AIS is resold. VPM's output is a decision — "this vessel, this port, this need, this window" — never the underlying position feed.
How a prediction is made
Two signals, weighed against each other for every vessel individually:
1. Pattern
A statistical model built from the vessel's own port history that captures how this ship actually moves — the routes it repeats, the rotations it runs, the ports that reliably follow one another. For a liner on a fixed service, this alone is a strong predictor.
2. Declaration
The destination the vessel broadcasts over AIS, normalised against a dictionary of real ports and known transit waypoints so that "FIHMN", "Hamina Fi" and "Hamina" all resolve to the same place.
Which one to trust
VPM measures, per vessel, which of the two signals has been more reliable across that vessel's history, and leans on that one. A predictable liner is carried by its pattern; a tramp vessel working the spot market is carried by what it declares. The choice is made per ship, not once for the whole fleet.
From "next port" to "what they'll need"
Once the next call is known, the rest follows:
- Fuel. A consumption model calibrated against real reported burn — sensitive above all to speed, and corrected for emission-control zones and for dual-fuel / LNG machinery — estimates the burn rate, the fuel grade, and how much is needed to reach the next port.
- Timing. ETA along the real sea route, adjusted for whether the vessel is laden or in ballast; the expected time alongside; and the departure window that follows.
- Cost & CO₂. The same fuel figure converts directly into a voyage cost and an emissions estimate.
The picture, flipped
Asked the other way round, the same engine becomes a sales tool. The Port View takes one port and lists the vessels heading toward it over the coming days — each with its load state, its likely need, its departure window, and a contact for the operator. That is the product a local supplier buys: not a database of ships, but a short list of who to call, and when.
Measured, not guessed
Every accuracy figure VPM publishes comes from a chronological back-test. The model is only ever shown history from before the call it is predicting, and it is scored against what actually happened next — never against a simulation.
Accuracy is not uniform. When a vessel declares a destination that resolves cleanly, the next-port call is right about 87% of the time. When it doesn't, the prediction falls back to pattern alone and is right closer to 4 times in 10. VPM shows which case you are looking at rather than averaging the two into a single comfortable number. The full method is in Accuracy & methodology.
What it doesn't do yet
- Coverage is regional and growing. The engine is geography-independent — it predicts any vessel with a repeating pattern, in any sea — but a port is only useful once enough of its traffic is tracked.
- Some vessels resist prediction. A ship with no repeating route and no declared destination will predict poorly, and VPM says so rather than dressing a weak guess as a confident one.
- Layers beyond fuel and oil — spare parts, scheduled maintenance, provisioning — run on the same engine but are on the roadmap, not live.