Dry Bulk Carrier Load Estimation from Freeboard Measurement
Sub-metre optical satellites can read a bulk carrier's Plimsoll markings and freeboard height to estimate cargo load, giving commodity analysts a cross-check on declared manifests for iron ore, coal and grain flows.
Sensors
- Airbus Pleiades Neo: 30 cm native panchromatic resolution, enabling direct pixel-level measurement of Plimsoll line markings (which are typically 25 mm wide strokes on a 300 mm disc) and freeboard height estimation against hull reference features. Revisit at any given point roughly once per day with the four-satellite constellation.
- Maxar WorldView-3: 31 cm panchromatic, 1.24 m multispectral. Sufficient to distinguish the white Plimsoll circle and load-line letters from hull paint, and to measure freeboard in calm sea states. SWIR bands can separate hull colour from waterline foam in ambiguous lighting.
- Planet SkySat: 50 cm panchromatic, 1 m multispectral. Marginally adequate for freeboard estimation on larger vessels (Capesize LOA around 290 m gives a long hull reference baseline), less reliable on Handymax hulls where the freeboard range is smaller.
- ESA Sentinel-2 MSI: 10 m multispectral in visible and NIR bands. Cannot resolve Plimsoll markings, but at anchorages with dense vessel clusters it supports fleet-level screening: laden versus ballast status is detectable from hull colour contrast (red anti-fouling exposed in ballast condition) at this resolution, flagging which vessels warrant commercial tasking.
What the load line actually encodes
The International Load Line Convention (ILLC), administered through IMO, requires every ocean-going cargo vessel to carry a Plimsoll mark: a circle bisected by a horizontal line, flanked by letters denoting the certifying authority, and accompanied by a ladder of seasonal and zone load lines. On a Capesize bulk carrier the full mark assembly spans roughly 540 mm vertically. The distance from the relevant load line to the waterline is the freeboard, and freeboard is a direct proxy for displacement.
Displacement minus lightship weight equals deadweight, and deadweight for a known vessel class translates to cargo tonnes through published hydrostatic tables. The relationship is not perfectly linear across the full draught range, but for the middle 60 percent of the load range it is close enough that a freeboard measurement accurate to 10 cm corresponds to a displacement error of roughly 1,000 to 3,000 tonnes on a Capesize vessel with a deadweight of 170,000 to 180,000 tonnes. That is a 1 to 2 percent uncertainty, which is commercially meaningful but useful as a cross-check on declared cargo.
Why dry bulk is harder than crude tankers
Crude tankers have a relatively simple hull form and a single commodity with a known density. Dry bulk carriers carry iron ore (density around 2.5 t/m³), coal (0.8 to 0.9 t/m³) or grain (0.7 to 0.85 t/m³) in the same hull. Two vessels riding at identical freeboards can be carrying very different cargo masses depending on commodity. The analyst must resolve commodity identity before converting freeboard to tonnes.
Commodity identity comes from context: port of loading, AIS voyage data, known trade routes and seasonal patterns. A Capesize departing Tubarao (Brazil) or Port Hedland (Australia) is almost certainly carrying iron ore; one leaving Richards Bay (South Africa) is almost certainly carrying coal. Grain is typically loaded at smaller terminals in shallower water, which limits vessel size and shifts the problem toward Panamax and Supramax hulls. The freeboard measurement is necessary but not sufficient on its own.
There is also a structural complication. Dry bulk hatch covers sit proud of the main deck, and on a heavily laden vessel the deck may be only a metre or two above the waterline. Wave wash, foam and spray can obscure the hull-waterline boundary in imagery taken during any sea state above Beaufort 3. Calm anchorages, not open-ocean transits, are where this method works.
Extracting freeboard from sub-metre imagery
The measurement chain has three steps. First, vessel detection and georeferencing: the vessel is located in the image, its heading estimated from shadow or wake geometry, and the hull outline extracted. Second, waterline detection: the boundary between hull paint and water is found using a gradient or threshold operation on the panchromatic band. Third, Plimsoll mark localisation: the load-line disc and horizontal marks are identified by template matching or edge detection, giving a pixel-space ruler calibrated to the known physical dimensions of the marks.
Freeboard is then the pixel distance from the relevant load line to the waterline, converted to metres using the known mark dimensions as a scale reference. On Pleiades Neo at 30 cm, a 10 cm freeboard increment corresponds to roughly 0.33 pixels, which is at the limit of sub-pixel estimation methods. In practice, analysts report freeboard to the nearest 20 to 30 cm, which is honest. WorldView-3 performs similarly. Both sensors require that the vessel be reasonably broadside to the satellite look angle; end-on views compress the hull and make the marks unreadable.
Sea-state is the dominant error source. Published work on optical freeboard measurement (several studies in the MDPI Remote Sensing journal have examined tanker cases) consistently finds that calm anchorages yield standard deviations of 15 to 25 cm in freeboard retrieval, rising sharply in choppy conditions. This method is an anchorage tool.
Fleet screening before expensive tasking
Commercial tasking of Pleiades Neo or WorldView-3 costs money per square kilometre. Running it over every bulk carrier at every anchorage globally is not practical. Sentinel-2, free and with a five-day revisit at 10 m resolution, provides the triage layer.
At 10 m, individual Plimsoll marks are invisible, but hull colour is detectable. A ballast-condition bulk carrier rides high, exposing a band of red anti-fouling paint above the waterline that is spectrally distinct in Sentinel-2's red and NIR bands. A laden vessel shows a much narrower red band or none at all. This binary laden/ballast discrimination, applied across a major anchorage, identifies which vessels are worth the cost of a commercial collect. It also builds a time-series record of loading events at a port without incurring tasking costs for every revisit.
Translating pixels to commodity intelligence
The output of this pipeline is a per-vessel load estimate, expressed as a percentage of deadweight, with an associated uncertainty range. To convert that to a commodity flow figure, the analyst combines it with vessel identity (from AIS or optical classification), vessel-specific deadweight from Lloyd's or similar registries, and commodity attribution from port context.
At the fleet level, aggregating load estimates across vessels departing a given port over a month produces a commodity export volume estimate that can be compared against official port statistics, which are typically published with a lag of several weeks to months. The comparison is imperfect: official statistics are usually more accurate, but they arrive late. Satellite-derived estimates arrive within hours of image acquisition and can be updated daily.
Satellize runs this kind of commodity-flow analytics on open and commercial imagery, with the Sentinel-2 screening layer running continuously and commercial tasking added on client licence. The methodology is the same family of photogrammetric measurement used in the Tonga crop-estimation programme, applied to a very different observable. The analytics output is a structured data feed, not a narrative report, because the buyer community here is quantitative: commodity trading desks and freight analysts who want numbers, not prose.
Honest limits of the method
Cloud cover is the obvious constraint. Major bulk loading ports in Brazil, West Africa and Southeast Asia experience frequent overcast periods. A single clouded collect is recoverable; a week of cloud during a rapid loading cycle is not, and the analyst must fall back on AIS departure timestamps and historical load-rate assumptions.
Vessel identity ambiguity matters too. If AIS is dark or spoofed at the moment of imaging, the analyst cannot reliably match the vessel to a registry entry and therefore cannot look up the lightship weight needed to convert freeboard to cargo tonnes. This method is most reliable when used in conjunction with AIS data, not as a replacement for it.
Finally, the method says nothing about cargo quality, moisture content or grade. A Capesize carrying 170,000 tonnes of low-grade iron ore fines and one carrying 170,000 tonnes of high-grade lump ore look identical from orbit. Grade intelligence requires other sources.
Typical figures
| Best panchromatic resolution | 30 cm (Pleiades Neo), 31 cm (WorldView-3) |
| Minimum freeboard measurement precision | ±20 to 30 cm in calm anchorage conditions |
| Fleet screening resolution | 10 m (Sentinel-2 MSI) |
| Sentinel-2 revisit | 5 days at equator (single satellite), 2 to 3 days at mid-latitudes with both satellites |
| Pleiades Neo revisit | Approximately 1 day at any point with 4-satellite constellation |
| Minimum vessel size for reliable mark detection | Panamax and above (LOA > 200 m) at 30 cm resolution; Capesize preferred |
| Sea-state limit | Beaufort 3 or below for reliable waterline detection |
| Spectral bands used | Panchromatic (freeboard); red and NIR (laden/ballast screening); SWIR (hull/waterline disambiguation) |
| Archive depth | Sentinel-2: from 2015; Pleiades Neo: from 2021; WorldView-3: from 2014 |
| Cargo mass uncertainty (Capesize) | Approximately 1 to 2% of deadweight at ±25 cm freeboard precision |
Analytics Satellize can run
| Per-vessel load status (laden / ballast / partial) | Sentinel-2 hull-colour spectral classification at 10 m; red anti-fouling band exposure index | GIS layer refreshed on each Sentinel-2 pass; vessel polygons attributed with load-status flag and confidence score |
| Freeboard height estimate (metres) | Sub-pixel waterline detection and Plimsoll mark template matching in Pleiades Neo or WorldView-3 panchromatic imagery | Per-vessel data record with freeboard value, uncertainty range and sea-state flag |
| Cargo mass estimate (tonnes, with uncertainty band) | Freeboard-to-displacement conversion using vessel-class hydrostatic assumptions; commodity attribution from port context and AIS voyage data | Structured JSON or CSV record per vessel; integrates with client freight or trading desk data feeds |
| Port-level commodity export volume (monthly) | Aggregation of per-vessel load estimates across all departures detected at a named port; cross-referenced against AIS departure events | Monthly time-series report by port and commodity class, delivered as tabular data with methodology notes |
| Loading event detection and timing | Change detection on sequential Sentinel-2 or commercial imagery at anchorage; freeboard decrease between two collects indicates loading in progress | Alert feed with vessel identifier, estimated load-rate (tonnes per hour) and expected departure window |
| Manifest cross-check flag | Comparison of satellite-derived cargo mass estimate against declared manifest tonnage where publicly available; statistical outlier flagging | Discrepancy report with confidence interval; intended as an investigative lead, not a definitive finding |
Who does the work
We can get this done for you. Satellize runs its own analyst desk and a strong science team. You do not buy a data feed and work out what it means; our people source the imagery, run the analysis described on this page, and hand you the answer with its confidence limits stated. Discuss this requirement.