Open Water Is Where Vessels Come Closest. 16.7% of Open-Sea Encounters Pass Within 0.1nm — 49% More Than Anchorage.

The Setup

In 45 days of AIS data, 555,679 pairwise vessel encounters were detected across three navigational contexts: open sea, approach channels, and anchorages. The conventional assumption — shared by most maritime risk models — is that confined waters produce the closest encounters. Anchorages are crowded. Approach channels are tight. Open water is where vessels spread out.

The data does not support that assumption.

Open-sea encounters achieve an average minimum range of 0.464 nautical miles — shorter than approach channels (0.526nm) and shorter than anchorages (0.555nm). More significantly, 16.7% of open-sea encounters pass within 0.1nm of each other. The equivalent share in approach channels is 12.7%. In anchorages, 11.2%.

Open water is where vessels come closest.

The Chain

The geometry follows from detection physics, not hazard probability. Approach channels and anchorages impose structure: pilotage, berth queuing, vessel traffic services, tidal windows. Those constraints force vessels onto known tracks, which creates routine encounters at predictable ranges. The encounters are close but expected.

Open sea removes those constraints. Vessels self-navigate on free courses. When two free-course tracks converge, there is no institutional forcing function to enforce separation. The result is a different encounter distribution: fewer events overall, but a heavier tail at the extremes.

The 5th-percentile minimum range in open sea is 0.031nm — 57 meters. In approach channels it is 0.039nm (72m). In anchorages, 0.045nm (83m). The closest encounters happen where there is the least structure to prevent them.

A related signal amplifies the concern: the average time-to-CPA (TCPA) at the moment of record classification is negative in all three contexts, meaning encounters are classified after the closest-approach moment has already passed. In open sea, the average TCPA at classification is -2,777 seconds — approximately 46 minutes post-CPA. In approach channels and anchorages, the lag is approximately 12 minutes.

The system is observing open-sea encounters nearly four times later in the encounter arc than confined-water events.

The Implication

Detection timing matters because it determines what interventions are still possible. A 46-minute post-CPA classification window means the encounter is over before the alert fires. Vessels have already resolved — or not — by the time the record is created. For open-sea encounters that resolve at sub-0.1nm ranges, the detection system is a historian, not a watchkeeper.

That creates a specific problem for risk assessment pipelines that use encounter records to score vessels or routes. A vessel that repeatedly achieves minimum ranges below 0.1nm in open water will not accumulate real-time risk indicators. The records arrive later, in batch, after the navigational decision has already been made.

The 55,664 sub-0.1nm open-sea encounters in this period — 16.7% of the total — are events where vessels passed within the length of a large container ship of each other, in unstructured water, classified approximately 46 minutes after the fact.

Rule 17 flags (encounters where one vessel deviated from its required COLREGS course) appear in 49.5% of open-sea records and 53.0% of anchorage records. Flag rates near 50% in all three contexts warrant scrutiny: a rate this high may indicate the classifier is operating near its precision ceiling rather than that half of all encounters involve genuine deviations.

What to Watch

Two questions worth tracking against this baseline: whether the open-sea sub-0.1nm rate is stable across seasons or spikes during low-visibility periods, and whether specific vessel types or flag states are overrepresented in the open-sea tail. Neither can be answered from aggregate counts alone.

The context-tag distribution also matters structurally. Open sea accounts for 333,082 of 555,679 total encounters (59.9%). Any encounter-based risk model calibrated on the assumption that confined waters drive the extremes is working from the wrong base rate.

Limitations

TCPA values are computed at the moment of record classification, not at detection initiation. The 46-minute figure is an average; individual events range widely. Context tags are assigned algorithmically from AIS positions against zone polygons — reclassification errors at zone boundaries are possible. Minimum range values below 0.05nm fall within AIS positional uncertainty for some vessel classes, so the 55,664 sub-0.1nm count should be read as an upper bound on truly close encounters.

Data as of 2026-05-12. Source: pairwise_encounter, 45-day window, n=555,679 classified encounters.