6% of Active Hex Cells Account for 71% of Encounter Events
Of the 18,400 hex cells with at least one logged vessel-hour in the trailing 60 days, 1,104 cells — 6% — account for 71% of all recorded pairwise encounter events. The median cell logs 1.4 vessel-hours per day. The 99th percentile logs 89. That is a 63× gap.
The Setup
Vessel density in H3 space is the count of vessel-hours accumulating in a given hex cell over a rolling window. It is distinct from encounter events — density is the concentration metric, encounter is the output. High-density cells are encounter factories.
The distribution is heavily right-tailed. Half of all active cells carry fewer than 1.4 vessel-hours per day. The top 1% carry an average of 74 vessel-hours per day. Within the global top 1%:
- Singapore Strait region: 38 cells
- Malacca Strait approach: 31 cells
- Houston Ship Channel: 22 cells
- English Channel / Dover Strait: 19 cells
- Istanbul Strait approaches: 14 cells
These 124 cells collectively account for 29% of all encounter events across the tracked fleet.
The Chain
High-density cells produce encounters that are harder to resolve. In a cell with 4 vessels transiting simultaneously there are 6 potential pairwise encounter pairs. At 8 vessels, there are 28 pairs. At 12 vessels — not unusual in the Singapore Strait during peak hours — there are 66 pairs to evaluate simultaneously.
The encounter classification engine scores every pair for Rule 17 obligation, extremis threshold, and maneuver geometry. At low cell density, nearly every encounter resolves cleanly. In top-1% cells, the queue becomes recursive: a maneuver to resolve Pair A changes the geometry for Pairs B and C before either resolves.
Port dwell adds a confounding layer. Anchorage zones inflate raw density scores without contributing to underway encounter risk. Separating anchorage-stationary vessel-hours from underway-mobile ones improves signal quality, but the raw density metric remains a useful leading indicator in isolation.
The Implication
The 63× density ratio between median and p99 cells has a direct consequence for risk screening. Anomaly models calibrated on fleet-wide averages will systematically under-flag behavior in high-density cells and over-flag normal behavior in sparse cells.
A 30-degree course change in a 1.4 vessel-hour cell is unusual. The same maneuver in an 89 vessel-hour cell is almost certainly routine traffic separation. Density-normalization is necessary before any behavioral anomaly score is interpretable.
The 38 Singapore Strait cells in the global top 1% also carry the highest concentration of STS transfer events and dark event flags in the dataset — not because those behaviors are inherently more common per vessel in Singapore, but because the denominator is enormous. Absolute event counts mislead without a density denominator.
What to Watch
Cell density rankings are not static. The Istanbul Strait cells, currently holding 14 cells in the global top 1%, have been elevated since late February — consistent with documented rerouting away from Red Sea transit.
The Malacca approach is the leading indicator to monitor: those 31 cells have shown a 12% density increase over the trailing 30 days, outpacing the Singapore Strait (+4%) and the English Channel (-2%). If that trajectory holds, Malacca overtakes Singapore in total top-1% cell count within 6-8 weeks.
Limitations
Density metrics are only as good as AIS coverage. In cells with known transponder suppression — and the top 1% includes several that overlap with documented dark event clusters — vessel-hours are understated. The 89 vessel-hour p99 figure is a floor, not a ceiling.
Cell boundaries introduce edge effects in narrow straits where the H3 grid cuts perpendicular to the traffic lane. Vessels transiting along cell edges generate partial vessel-hours in two adjacent cells, diluting density scores for both. At resolution 8 this matters most in channels under 1.5 km wide.
Overwatch tracks vessel_density_h3 across 60-day rolling windows at H3 resolution 8. Encounter events are computed via pairwise_encounter at 0.5 nmi proximity threshold. Coverage includes vessels with at least one AIS ping per 30-minute window.