Container ship load estimation from waterline draught proxies
High-resolution optical and SAR imagery can resolve a container vessel's freeboard at berth or anchor, letting analysts infer approximate load state without touching a cargo manifest. Accuracy depends on knowing the tide, the hull class, and the pixel.
Sensors
- Planet SuperDove: 3 m native pixel at nadir, 8 spectral bands including coastal blue and NIR. Daily revisit over most ports. The 3 m resolution is sufficient to resolve freeboard on Panamax and larger vessels (beam typically 32 m or more), but marginal on feeder ships below 150 m length.
- Capella Space SAR (X-band): Spotlight mode delivers 0.35–0.5 m resolution, usable day or night through cloud. X-band backscatter from the hull-water interface is geometrically sharp; the waterline appears as a bright return boundary. Tasked on demand, latency typically under 12 hours from order to delivery.
- Sentinel-2 MSI: 10 m resolution in visible bands. Adequate for rough load-state classification (laden vs. light ballast) on very large vessels such as ULCVs with beam above 50 m, but insufficient for precise freeboard measurement on standard Panamax or smaller. Free and open; 5-day revisit at mid-latitudes.
- AIS vessel registry feeds: Not an imaging sensor, but essential context. MMSI-linked AIS position fixes identify the vessel; cross-referencing against Lloyd's Register or IHS Markit hull databases supplies the class-specific Plimsoll marks, deadweight tonnage and design draught needed to convert a measured freeboard into a load estimate.
What a freeboard measurement actually tells you
Every commercial vessel has a load line, colloquially the Plimsoll mark, stamped on the hull amidships. It encodes the minimum legal freeboard for a given water density and season zone. The distance between that mark and the actual waterline at any moment is a direct function of how much cargo, fuel and ballast water the ship is carrying. For a modern Ultra Large Container Vessel with a design deadweight of around 200,000 tonnes and a summer draught near 16 m, each additional metre of draught corresponds to roughly 12,000–15,000 tonnes of displacement, depending on hull form.
Satellite imagery does not read the Plimsoll mark directly. What it captures is the visible freeboard: the height of painted hull above the waterline. If you know the vessel class and can find its published hydrostatic tables or design draught, you can translate that freeboard into an approximate load fraction. The word approximate matters. Ballast water can substitute for cargo tonne-for-tonne in displacement terms, so a ship sitting low may be heavily laden or simply carrying ballast. Context from AIS voyage history, port call sequence and cargo-type databases is needed to resolve the ambiguity.
Geometry of freeboard measurement in satellite imagery
Freeboard measurement from nadir-looking imagery is a parallax problem. The waterline is at sea level; the top of the visible hull freeboard is a metre or more above it. At nadir the two are co-located in the image plane, so measuring their vertical separation requires either off-nadir geometry (where the hull face is partially visible) or inference from shadow length. Planet SuperDove collects at off-nadir angles that vary by pass; Capella's spotlight SAR, collected at incidence angles typically between 25° and 50°, produces a layover geometry in which the hull face is visible as a bright strip, and the width of that strip in slant-range is a function of freeboard and incidence angle.
The practical resolution floor for freeboard measurement is approximately 0.5 m per pixel. Below that, the hull-water boundary blurs into two or three pixels and sub-metre freeboard changes become undetectable. Planet SuperDove at 3 m native resolution can distinguish broad load states, perhaps three bins: light ballast, partial load and deep laden. Capella at 0.35 m can in principle resolve freeboard to within 0.3–0.5 m under good geometry, which on a large vessel translates to a load uncertainty of roughly 4,000–7,000 tonnes. Sentinel-2 at 10 m is useful only as a screening layer to flag vessels worth tasking at higher resolution.
The tide problem, and how to handle it
Freeboard as measured in an image is freeboard relative to the sea surface at the moment of collection. Tidal range at major ports varies from under 0.3 m (Singapore, Mediterranean) to over 10 m (Liverpool, Bay of Fundy). An uncompensated tidal error of 1 m on a Panamax hull produces a load estimation error of roughly 8,000–10,000 tonnes, which can exceed the entire cargo of a feeder vessel.
Correction requires a tidal model referenced to chart datum at the port in question. The UK National Tidal and Sea Level Facility, NOAA's tidal prediction service and the FES2014 global tidal model (published by LEGOS/CNES) all provide predictions accurate to 0.05–0.15 m RMS at well-gauged ports. Applying the correction is straightforward if the image timestamp is precise, which it is for both Planet and Capella products. Residual error after tidal correction is dominated by local storm surge, wind setup and, at anchorages, wave orbital motion. Analysts should flag any image collected during wind speeds above roughly 15 knots as having elevated waterline uncertainty.
Hull fouling introduces a secondary bias. A vessel with heavy biofouling sits slightly deeper than its clean-hull hydrostatics predict, typically by 0.1–0.3 m for a vessel overdue for drydock. This is small but not negligible for high-precision applications.
Linking the image to the hull specification
A freeboard measurement without a hull identity is nearly useless. AIS provides the MMSI number, which links to IMO number, which links to Lloyd's Register or IHS Sea-web records containing the vessel's class, design draught, deadweight and Plimsoll-mark positions. For vessels broadcasting AIS, this chain is reliable. The complication arises when a vessel is at anchor in an uncrowded roadstead with several ships nearby: image-to-AIS matching requires correlating the vessel's position in the image with the AIS fix at the closest timestamp, accounting for AIS reporting intervals (Class A transponders report every 2–10 seconds under way, but may report every 3 minutes at anchor).
Vessel classification from imagery alone, without AIS, is possible using length-to-beam ratios and container stack geometry visible in very-high-resolution imagery. Published work using convolutional neural networks on Planet and Maxar imagery has demonstrated vessel-type classification at accuracies above 90% for broad categories (container, bulk, tanker). That is sufficient to select the right deadweight range but not to pin down a specific hull's hydrostatics. AIS linkage remains the preferred route.
Honest limits of the method
Several failure modes are worth stating plainly. Cloud cover blocks optical collection entirely; in the tropics, cloud-free imagery over a specific anchorage may be available on fewer than half of all days. SAR is cloud-independent but requires tasking, which costs money and introduces latency. Neither sensor can see through a vessel's hull to distinguish cargo from ballast water. Load estimation accuracy degrades for feeder vessels below about 150 m in length, where the freeboard change per tonne of cargo is small relative to pixel size. Vessels in dry dock or floating drydock have no meaningful waterline and should be excluded.
The method works best as a relative indicator rather than an absolute one. Comparing the same vessel across multiple port calls, or comparing a fleet of similar vessels at the same anchorage, produces more reliable signals than a single-image absolute load estimate. Satellize applies this approach in its analytics pipeline, drawing on open Sentinel-2 archives for historical context and commissioning Capella tasking for current-state precision. The methodology is analogous to, though distinct from, the crop-estimation work Satellize runs for the Kingdom of Tonga, where multi-date comparison similarly outperforms single-date inference.
What the output is actually good for
The primary buyers of this analysis are commodity traders and shipping analysts who want an independent check on reported cargo volumes at key loading or discharge ports, port authorities monitoring whether vessels are arriving at declared draught, and maritime insurers assessing overloading risk. A secondary use is sanctions monitoring: a vessel loading a sanctioned commodity at a restricted port may not broadcast AIS, but its draught change between arrival and departure is visible in tasked SAR imagery.
The output is not a cargo manifest. It is a load-state signal with quantified uncertainty, useful for narrowing the range of plausible cargo volumes and for flagging anomalies worth investigating through other means. Used that way, it is genuinely informative. Treated as a precise cargo measurement, it will mislead.
Typical figures
| Best achievable spatial resolution | 0.35 m (Capella Space spotlight SAR); 3 m (Planet SuperDove optical) |
| Practical freeboard measurement floor | ~0.5 m/pixel; below this the hull-water boundary is unresolvable |
| Revisit (optical, Planet SuperDove) | Daily over most ports; cloud-free availability varies by region and season |
| Revisit (SAR, Capella) | On-demand tasking; typical order-to-delivery latency under 12 hours |
| Spectral bands used | Visible (blue, green, red) and NIR for hull-water contrast; X-band SAR (9.6 GHz) for all-weather collection |
| Tidal correction accuracy | 0.05–0.15 m RMS at well-gauged ports using FES2014 or national tide models |
| Load estimation uncertainty (large vessel, good geometry) | ±4,000–7,000 t for ULCV class after tidal correction; larger for feeder vessels |
| Archive depth (Sentinel-2) | From 2015; free and open via Copernicus Data Space |
| Minimum useful vessel size | ~150 m LOA at 3 m resolution; smaller vessels require sub-metre imagery |
| Delivery formats | GeoTIFF (imagery), GeoJSON or CSV (freeboard measurements and load estimates), PDF report |
Analytics Satellize can run
| Single-vessel load-state classification | Freeboard extraction from off-nadir optical or SAR imagery, corrected for tidal state using FES2014 or port gauge data, compared against class hydrostatics from AIS-linked registry lookup | Per-observation report: vessel identity, image timestamp, measured freeboard, tidal correction applied, estimated load fraction and uncertainty band |
| Fleet-level load trend series | Multi-date Sentinel-2 archive screening to classify load state across a defined vessel fleet or port anchorage; high-confidence observations flagged for Capella tasking | Time-series CSV and chart showing load-state distribution across fleet, with anomaly flags |
| Draught-change event detection | Comparing pre-berth and post-berth freeboard measurements on the same vessel to infer net cargo loaded or discharged; requires two cloud-free or SAR observations bracketing the port call | Event-level GeoJSON with estimated cargo volume change and confidence interval |
| Sanctions or anomaly screening layer | Cross-referencing AIS gap events (vessel goes dark) with SAR-derived draught change at known loading terminals; method published in open maritime surveillance literature | Alert feed: vessel MMSI, location, draught-change magnitude, AIS gap duration, risk flag |
| Port-level aggregate load index | Averaging load-state classifications across all identifiable container vessels at anchor or berth in a defined port bounding box, repeated at each available image date | Weekly index value per port, delivered as time-series API feed or spreadsheet |
| Hull-class hydrostatics reference dataset | Structured compilation of published design draughts, deadweight figures and Plimsoll-mark positions for the 50 most common container vessel classes, sourced from Lloyd's Register and IMO documentation | Lookup table (CSV) used as the calibration layer for all freeboard-to-load conversions |
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.