Maritime observation desk
Radio positions, encounters and Sentinel-2 vessel observations at the main chokepoints. Each carries its own clock.
Satellite scenes
Browse dated 10 m Sentinel-2 passes. This published snapshot is read-only. Review ships, reject false detections, or correct boxes in Image Lab.
A scene is a timestamp.
Choose a watch tile on the map to inspect its dated image. Enlarging the image does not increase its native resolution.
Oil-market context
Experimental oil-equivalent estimates cover only observed vessels and selected satellite tiles. Vessel type and imagery do not verify petroleum cargo; these are not global supply figures or oil-price forecasts.
Estimated oil at sea
Observed tanker oil-equivalent, central estimate—Waiting for the first estimate- Change, 24 h
- —
- Change, 7 days
- —
- Brent, latest EIA
- —
- Recording since
- —
What the estimate contains
| Component | Vessels | Barrels (low / central / high) |
|---|
By cargo
Oil at sea against Brent
The chart starts after the first hourly estimate.
Does it move the price?
Correlation of day-on-day change in AIS-observed oil at sea with Brent's change the same day and 1, 7 and 30 days later. The AIS component is used because its receiver network is fixed; satellite passes come and go. Correlation is not causation, and it needs weeks of history before it means anything.
Physical balance
| Laden oil (AIS, with draught) | Tankers | Barrels (low / central / high) |
|---|
Price projection from volumes
A scenario model, not a validated forecast. When laden arrivals expected over the horizon run above their trailing average, more oil is reaching buyers, and the model moves Brent down by the supply shock divided by the combined short-run elasticity of supply and demand (0.24 central, from Caldara, Cavallo & Iacoviello 2019; up to 0.50 in the band). The observed network is assumed to sample seaborne flows, which carry 76% of world oil (EIA).
Where laden cargo says it is going
Shading and circles show each destination's share of laden and part-laden AIS barrels (central estimate). Destinations come from the crew-typed AIS field; "for orders" and unrecognised entries are listed but cannot be mapped.
By region
| Region | Source | Tankers | Central barrels |
|---|
Where laden cargo says it is going
| Reported destination | Tankers | Central barrels |
|---|
From the AIS destination field, typed by the crew. "For orders" means the buyer or discharge port is not fixed yet, which is common for traded cargoes.
Monthly coastal satellite detections of large hulls
Global Fishing Watch publishes Sentinel-2 vessel detections across most exclusive economic zones by month (CC0). Tanker Watch streams the file and keeps hulls of 150 m or more that GFW does not flag as infrastructure, ice or non-vessel. "Matched to AIS" means GFW linked the detection to a transponder; unmatched hulls are dark-activity candidates, not proof. Hull type and load are not given. These counts do not enter the oil-equivalent estimate above.
| Region | Large hulls | 200 m + | Matched to AIS | Stopped |
|---|
US crude inventories (EIA weekly)
Waiting for EIA weekly data.
Vessels in the estimate (largest first)
| Vessel | Class | Region | State | Destination | Central barrels | Source |
|---|
How the estimate is built
Length gives a size class (Handysize to ULCC), each with an industry deadweight range. Cargo is taken as 95% of deadweight, converted at about 7.33 barrels per tonne for crude-class ships, 7.9 for product carriers and 7.6 for LR1/LR2/Aframax, which carry either. Loaded share comes from the crew-reported draught against the class's laden and ballast draught, with a 15-point allowance for stale entries. Satellite hulls have no draught, so they count 0 to 100% loaded, 50% central. Satellite tiles where AIS already sees three or more vessels are left out, and hulls matched to AIS are never counted twice.
Documented cargo (highest confidence)
Tankers can carry different liquids, travel in ballast, wait in storage or change course. Public vessel positions alone cannot establish the world's free-flowing oil.
Physical balance
- In transit
- Not measured
- Floating storage
- Not measured
- Arrivals in selected window
- Not measured
Enter a comparable reference flow after importing cargo evidence. This is a physical comparison, not a price forecast.
Global coverage is unknown. No extrapolation from the Baltic to the world's fleet.
Published spot prices
Price history has not been retrieved.
Source: U.S. Energy Information Administration. Publication time is separate from the observation date.
Cargo evidence ledger
One cargo ID per parcel. Reassigning a parcel between vessels does not add to inventory.
| Parcel / vessel | Route | State | Volume range (bbl) | Observed / ETA | Evidence |
|---|---|---|---|---|---|
| No cargo evidence imported. A blank global estimate is more useful than an invented one. | |||||
Record a confirmed whole-parcel transfer
This changes the carrying vessel only. Partial transfers require documented parcel splitting outside this tool; an encounter lead is insufficient evidence.
Long-horizon research
30- and 90-day expanding-window tests compare a fixed model against a no-change price benchmark. Target windows are purged from training.
No price prediction is published from an empty or unqualified cargo history. Test results remain research candidates until independently validated.
Connections & coverage
Access, observation age and interpretation belong beside the data, not buried beneath it.
Connections
Communications evidence
Radio-frequency detections can complement AIS, but no free global maritime satcom-traffic feed is connected here. LRIT is restricted; commercial RF geolocation needs separately authorised access.
Required: id, lat, lon, timestamp, source, evidence_url. RF also requires uncertainty_m. No private communications content is collected.
Methods & boundaries
What does the imagery actually support?
Sentinel-2 visible bands have 10 m native pixels. Large vessels can occupy multiple pixels; identity, cargo class, draught and oil volume generally require other evidence. Zoom is display magnification, not a resolution upgrade.
Object proposals use per-channel local contrast against a scene-adaptive noise floor, a sea mask that keeps hulls the scene classifier mislabels, and shape and clutter checks. On a labelled Fujairah crop it found 21 of 21 visible vessels; on a rough-sea crop it proposed none. That is two scenes, not a calibrated rate. It does not classify tankers, and review labels are the local analyst's judgement.
Why isn't an unmatched object called a dark tanker?
A missing match can reflect stale or absent reception, timing, an incorrect position, or a non-ship object. Only short, time-aligned tracks are considered; interpolation does not cross a gap longer than 20 minutes. One AIS identity is assigned to at most one proposal, and ambiguous spatial matches are marked.
How does oil-in-transit become a price signal?
This release does not equate more oil at sea with lower prices. Longer voyages or storage can increase that stock while reducing prompt arrivals. Cargo volumes, arrival timing, coverage, inventory and demand context must be comparable before estimating a relationship. The arrival comparison is a conditional physical measure, not a forecast in dollars.
What survives a restart?
The local SQLite ledger stores positions, encounter evidence, cargo revisions, image metadata and reviews. Positions are retained for 30 days; cargo and reviewed evidence are retained. The recorder operates while the server is running, even with the browser closed. Keys are kept in server memory or supplied through environment variables.