Anchorage dwell-time analysis for commodity flow intelligence
How long a named ship sits in a roadstead reveals loading rates, demand shocks, and supply-chain stress before any price feed catches up. This page explains the sensors, methods, and honest limits of satellite-derived dwell-time analysis.
Sensors
- Planet PlanetScope: 3 m native resolution (resampled to 3.125 m product), daily global revisit across the full constellation of ~180 Dove satellites. Sufficient to resolve vessel outlines down to roughly 20 m length and to confirm anchor position within a single pixel cluster. The daily cadence is the key asset for sub-24-hour dwell resolution.
- Airbus Pléiades Neo: 30 cm panchromatic, 1.2 m multispectral. Tasked on demand; not a daily-revisit sensor. Used for hull-geometry confirmation and vessel identity cross-checks when AIS is absent or spoofed. Revisit at a given point is 1–2 days without off-nadir tasking, better with it.
- ESA Sentinel-2 MSI: 10 m in visible and NIR bands, 5-day revisit at the equator with both satellites (2A and 2B). Free and open archive from 2015. Adequate for counting vessels and detecting gross dwell changes in large anchorages; too coarse for reliable individual-vessel identity matching in dense clusters.
- Spire Global spaceborne AIS: Spire operates more than 100 LEO satellites carrying AIS receivers. Provides near-global vessel position reports with latency of minutes to low single-digit hours depending on orbit geometry. Used to anchor (literally) the optical detections to IMO numbers, flag states, and voyage histories.
Why dwell time carries more signal than vessel count
Port congestion pages count ships. Dwell-time analysis tracks individual named ships across consecutive satellite passes and measures how many hours or days each spends at anchor before moving to berth or departing. The difference matters because a roadstead with twenty vessels all turning over in six hours looks identical to one with twenty vessels stuck for a week. The commodity signal is entirely in the duration.
A bulk carrier waiting nine days at the Pilbara iron-ore anchorages before loading tells you something about berth availability, vessel scheduling, or export-quota pressure that a static headcount cannot. A VLCC sitting at a West African offshore transfer point for three days, then departing in ballast, tells you the cargo never materialised. These inferences require identity continuity across passes, not just detection.
Matching hulls across passes when AIS goes quiet
The cleanest method correlates optical detections with AIS. Spire's spaceborne AIS constellation covers most major anchorages with report intervals short enough to assign an MMSI to each optical detection. Once you have the MMSI, voyage history, cargo type, and deadweight tonnage follow from public registries.
AIS correlation breaks down in three situations: vessels operating in AIS-shadow (reception gaps in high-traffic or remote areas), vessels with malfunctioning transponders, and vessels that have deliberately switched off. In those cases, hull-geometry matching takes over. At PlanetScope's 3 m resolution, a vessel's length-to-beam ratio, superstructure position, and bow shape are measurable to within a few metres. Across passes separated by 24 hours or less, a ship at anchor changes heading with the tide but not position. Matching on georeferenced centroid plus hull-length class reduces false-match rates substantially, though it cannot deliver the certainty of an MMSI. Pléiades Neo imagery, tasked for confirmation, resolves hull markings and funnel colours at 30 cm and can close most remaining ambiguities.
The revisit problem: why daily coverage is the floor, not the ceiling
A sensor with a five-day revisit can tell you a vessel was present on day one and absent on day six. It cannot tell you whether the ship waited four days or left after eighteen hours. For commodity-flow inference, that ambiguity is often the entire question. Daily optical coverage, as PlanetScope provides, resolves dwell times to roughly plus-or-minus twelve hours once you account for the time-of-day offset between consecutive passes.
Cloud cover is the honest constraint nobody should understate. Tropical anchorages, which include many of the world's most commercially significant roadsteads, can be obscured for days or weeks at a time. The practical response is to fuse optical passes with spaceborne AIS: AIS provides position continuity through cloud, while optical provides the geometry confirmation and detects vessels running dark. Neither source alone is sufficient for high-confidence dwell-time attribution in persistently cloudy regions.
Sub-daily revisit is achievable in principle through multi-constellation tasking, but commercial tasking costs scale quickly and the marginal value of two-hourly optical passes is limited for vessels that typically spend twelve hours to several days at anchor. The sensible architecture for most clients is daily optical plus continuous AIS, with very-high-resolution tasking reserved for specific vessels of interest.
From dwell time to commodity inference: the logic chain
The inference from dwell time to commodity signal runs through a chain of public information. A vessel's cargo type is declared in AIS voyage data and port-state records. Its deadweight tonnage is in the Lloyd's or Equasis registry. Loading and discharge rates for bulk commodities are published or estimable from berth specifications. If a Capesize bulk carrier of known deadweight sits at a coal anchorage for four days before moving to berth, and the berth's published loading rate is around 50,000 tonnes per day, the implied cargo volume is bounded within a useful range.
Aggregate dwell times across an anchorage, tracked weekly, produce a queue-depth time series. Rising queue depth combined with stable inbound vessel arrivals implies a discharge-side constraint: berth congestion, labour action, or infrastructure failure. Falling queue depth with falling arrivals implies demand softening. These patterns precede official trade statistics by weeks. The method is not infallible. Vessel scheduling decisions, weather delays, and charter-party terms all introduce noise. The signal is probabilistic, not deterministic, and should be presented as such.
Honest limits of the method
Optical sensors cannot see through cloud, and they cannot measure cargo directly. Freeboard-based load estimation is a separate technique covered on a sibling page. Dwell-time analysis infers cargo volumes indirectly and carries compounding uncertainties: the vessel's actual cargo may differ from its declared type, loading rates vary by terminal and season, and AIS voyage data is self-reported and occasionally incorrect.
Hull-geometry matching at 3 m resolution works well for vessels above roughly 100 m length. Smaller coastal vessels, fishing boats, and harbour craft are detectable but not reliably identifiable by hull shape alone. Dense anchorages, where vessels lie close together, increase the risk of centroid-overlap errors in automated detection pipelines. Manual review of flagged cases is still necessary for high-stakes attribution. Satellize incorporates analyst review for individual vessel-of-interest tracking, as distinct from bulk statistical analysis of anchorage queues.
Archive depth is a genuine advantage of this method. Sentinel-2 provides a free optical archive from 2015, and Planet's archive extends to 2016 for many locations. AIS archives from providers like Spire go back several years. Reconstructing dwell-time histories for a specific anchorage over five or more years is operationally feasible and useful for establishing baseline seasonality before interpreting current anomalies.
Putting it to work
The practical workflow starts with a defined polygon around the anchorage of interest. Automated vessel detection runs on each available optical pass. AIS correlation assigns identities where possible. Hull-geometry matching fills gaps. Dwell times are computed per vessel per visit, aggregated into weekly queue-depth statistics, and flagged when they exceed historical percentile thresholds.
Satellize runs this architecture on open constellations and adds commercial tasking where clients need sub-metre confirmation of specific hulls. The Tonga crop-estimation programme is our most public analytics engagement, but the underlying pipeline, detect, identify, measure, contextualise, is the same logic applied to a different domain. Analysts who want to explore a specific anchorage or commodity corridor can request a scoped trial covering a defined roadstead and a defined historical window. That is a concrete starting point, not an open-ended commitment.
Typical figures
| Optical spatial resolution (routine monitoring) | 3 m (PlanetScope); 10 m (Sentinel-2) |
| Optical spatial resolution (identity confirmation) | 0.3 m panchromatic (Pléiades Neo, tasked) |
| Revisit cadence (optical) | Daily (PlanetScope constellation); 5 days at equator (Sentinel-2, both satellites); 1–2 days (Pléiades Neo, off-nadir) |
| AIS report latency (spaceborne) | Minutes to low single-digit hours depending on orbital geometry (Spire Global) |
| Minimum detectable vessel length (optical) | ~20 m at 3 m resolution; ~50 m at 10 m resolution (practical detection threshold in open water) |
| Dwell-time resolution | ±12 hours with daily optical; continuous with AIS fusion through cloud |
| Spectral bands used | Visible (RGB) and NIR for vessel-water contrast; panchromatic for hull-geometry matching |
| Archive depth | Sentinel-2 from 2015 (free); Planet from ~2016; Spire AIS several years depending on licence |
| Coverage | Global; cloud cover limits optical passes in tropical regions, mitigated by AIS fusion |
| Delivery formats | GeoJSON vessel tracks, CSV dwell-time tables, GeoTIFF detection layers, time-series dashboards |
Analytics Satellize can run
| Per-vessel dwell-time log | Optical detection with AIS correlation and hull-geometry matching across consecutive passes | CSV or GeoJSON table: vessel identity (MMSI or hull-class), entry timestamp, departure timestamp, computed dwell hours, anchorage zone |
| Anchorage queue-depth time series | Aggregated vessel-count and dwell-hour statistics per defined anchorage polygon, computed weekly | Time-series chart and underlying data feed; anomaly flags when queue depth exceeds historical 80th or 95th percentile |
| Commodity-flow index for a named route or terminal | Dwell-time aggregates cross-referenced with vessel cargo type from AIS voyage data and public registry records | Weekly index report with narrative interpretation; delivered as PDF and structured data |
| Dark-vessel dwell attribution | Hull-geometry matching for vessels absent from AIS during optical detection window; cross-checked against Pléiades Neo tasking where warranted | Alert report per identified dark-vessel event, with confidence rating and supporting imagery chips |
| Historical baseline reconstruction | Retrospective optical detection pipeline run over Sentinel-2 and Planet archive for defined anchorage and date range | Multi-year dwell-time database for seasonality analysis; delivered as GIS layer and tabular archive |
| Supply-chain stress indicator | Statistical comparison of current queue depth and dwell distribution against historical baseline; regression against published commodity price series | Monthly briefing note with indicator score and confidence interval; suitable for trading or procurement teams |
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.