Port congestion measurement from anchorage vessel queues
Repeat-pass optical and SAR imagery provides an independent physical count of vessels queuing at anchor outside major ports, giving a congestion signal that leads official throughput statistics by days to weeks.
Sensors
- Planet PlanetScope: 3 m optical resolution, daily revisit over most of the globe. Sufficient to resolve individual vessels larger than roughly 20 m and to distinguish hull from wake. Cloud cover is the operative constraint; persistent overcast anchorages (e.g. Pearl River Delta in summer) can drop usable passes to two or three per week.
- ESA Sentinel-1 SAR (C-band, 5.405 GHz): Interferometric Wide Swath mode delivers 10 m ground range resolution across a 250 km swath, with a 6-day repeat at the equator (3-day with both satellites). Cloud-independent. Vessel backscatter at C-band is strong enough to detect ships above roughly 20–30 m length in moderate sea states, though very small wooden or fibreglass vessels can fall below the noise floor.
- ESA Sentinel-2 MSI: 10 m resolution in visible and near-infrared bands, 5-day revisit with both satellites. Useful for optical confirmation of SAR detections and for colour-based discrimination between vessel classes (hull paint, wake colour). Not a primary detection layer given the resolution limit on smaller vessels, but free and globally archived.
- Maxar WorldView-3: 31 cm panchromatic, 1.24 m multispectral. Tasked on demand. At this resolution, deck equipment, hull markings and freeboard are legible, which is useful for resolving the bunkering-versus-queuing ambiguity on specific vessels of interest. Revisit is opportunity-dependent and cost is non-trivial; this sensor is best used for targeted confirmation rather than routine counting.
Why anchorage counts lead port statistics
Port authorities publish throughput figures in arrears, typically monthly, and those figures reflect what has already moved. An anchorage queue is the congestion forming in real time. When a vessel anchors to wait for a berth, it appears in the next satellite pass. When the queue clears, it disappears. The lag between the physical event and the official statistic can run from a few days to several weeks depending on the port and the reporting regime.
This is not a subtle signal. During the 2021 Southern California port congestion episode, the number of vessels at anchor off Los Angeles and Long Beach was visible in public Sentinel-1 imagery weeks before the disruption appeared clearly in published container dwell statistics. The anchorage count is, in effect, a leading indicator built from physics rather than paperwork.
How a vessel gets counted, and what can go wrong
Detection works differently across sensor types. In SAR imagery, a vessel's metal superstructure returns a strong radar echo relative to the sea surface. Automated constant false alarm rate (CFAR) detectors, well-documented in the published remote sensing literature, scan for bright pixels above a locally adaptive threshold. The output is a point or polygon per detected vessel. In optical imagery, the process relies on contrast between hull and water, often supplemented by wake geometry.
The honest limits matter. Sentinel-1 at 10 m resolution will miss vessels shorter than roughly 20–30 m in anything above a moderate sea state, because wave clutter competes with small-vessel backscatter. PlanetScope at 3 m will miss vessels obscured by cloud or haze, which is not rare in monsoon-affected anchorages. Vessel density is also a problem: when ships anchor very close together, detections merge and the count undershoots reality. Maxar WorldView-3 resolves this at the cost of tasking time and budget.
Computing dwell time across multiple passes
A single image gives a headcount. A time series gives dwell time, which is the more useful metric. The method is straightforward in principle: a vessel detected at position P in pass T1 and again at position P (within a small tolerance for anchor swing, typically 200–500 m) in pass T2 has been at anchor for at least the interval T2 minus T1. Across a dense pass schedule, the cumulative distribution of dwell times for the anchorage population describes how long vessels are waiting.
In practice, the correlation step requires care. Anchor swing means a vessel may shift its position by several hundred metres between passes. A vessel that departs and is replaced by another of similar size at the same spot will be miscounted as a single continuous presence unless the imagery resolution is sufficient to distinguish hull markings. This is where WorldView-3 targeted tasking earns its cost: on vessels of specific interest, hull identification resolves the ambiguity that lower-resolution multi-pass correlation cannot.
Bunkering vessels, ship-to-ship transfers, and other confounds
Not every anchored vessel is queuing for a berth. Bunkering (taking on fuel) is routinely conducted at anchor, often from a small tanker alongside. A vessel bunkering typically completes the operation in 12–36 hours and departs without ever calling at the terminal. If it is counted as a berth-queue vessel, the congestion metric is overstated.
Several signals help distinguish the two populations. Bunkering vessels are frequently accompanied by a small tanker in close proximity, visible in WorldView-3 imagery and sometimes in PlanetScope. AIS data, where broadcast honestly, will show a vessel's declared destination and activity status, though this is operator-reported and not always reliable. Dwell-time distribution itself is informative: bunkering dwell clusters below 48 hours, while berth-queue dwell in a congested port can extend to 10–20 days. Separating the two populations requires combining the imagery count with at least one independent signal, whether AIS, vessel type registry, or tasked high-resolution confirmation.
What the data actually delivers to a buyer
The primary output is a time-series count of vessels at anchor in a defined anchorage polygon, updated on each usable satellite pass. From this, analysts derive: the queue length on any given day, the median and 90th-percentile dwell time over a rolling window, and the rate of change in queue length (which distinguishes a congestion event that is building from one that is resolving).
Commodity traders use anchorage counts for dry bulk and container ports as a proxy for near-term freight rate pressure. Port authorities use them to benchmark their own published statistics against an independent physical measure. Insurers and logistics operators use dwell-time distributions to price delays and plan contingencies. Satellize runs this analysis on open constellations (Sentinel-1, Sentinel-2) with commercial tasking added where clients need sub-metre confirmation on specific vessels. The Overhead column has covered anchorage queue dynamics at several major ports using exactly this method.
Archive depth and what it enables
Sentinel-1 data extends back to April 2014 for areas covered by the initial satellite. Sentinel-2 archive begins in mid-2015. Planet's commercial archive depth varies by area of interest but commonly runs to 2016 or earlier for high-priority regions. This means an analyst can reconstruct the anchorage queue history for a major port across nearly a decade, establishing baseline seasonality, identifying anomalous congestion events, and correlating queue behaviour with commodity price movements or geopolitical disruptions.
That historical depth is often underused. A buyer who commissions only a current snapshot misses the context that distinguishes a structurally congested port from one experiencing a transient disruption. The baseline matters as much as the current reading.
Typical figures
| Primary SAR resolution (Sentinel-1 IW mode) | 10 m ground range, 250 km swath |
| Primary optical resolution (PlanetScope) | 3 m, daily revisit (cloud-permitting) |
| High-resolution confirmation (WorldView-3) | 31 cm panchromatic, tasked on demand |
| Sentinel-1 revisit at equator | 6 days single satellite, 3 days with Sentinel-1A and 1B combined |
| Minimum detectable vessel (SAR, moderate sea state) | Approximately 20–30 m length; smaller vessels may fall below CFAR threshold |
| Typical dwell-time resolution | ±6 hours at 3-day revisit; ±12 hours at 6-day revisit |
| Archive depth | Sentinel-1 from April 2014; Planet from approximately 2016 for priority regions |
| Delivery formats | GeoJSON vessel detections, GeoTIFF imagery chips, CSV time-series counts, PDF summary reports |
| Latency (open constellation) | Sentinel-1 NRT products available within 1–3 hours of acquisition via Copernicus Data Space |
Analytics Satellize can run
| Anchorage vessel count time series | CFAR detection on Sentinel-1 SAR; contrast-based detection on PlanetScope optical | Daily or per-pass CSV and GeoJSON layer with vessel count, anchorage polygon, and acquisition timestamp |
| Dwell-time distribution per anchorage | Multi-pass position correlation with anchor-swing tolerance; kernel density estimation of dwell intervals | Rolling 30-day dwell-time histogram (median, 90th percentile) delivered as chart and structured data feed |
| Congestion trend alert | Statistical process control on rolling queue-length series; alert triggered when count exceeds historical seasonal baseline by a defined threshold | Email or API alert with anchorage name, current count, baseline, and percentage deviation |
| Bunkering versus berth-queue classification | Dwell-time clustering combined with proximity detection of small tanker alongside; cross-referenced against AIS declared activity where available | Annotated vessel list distinguishing queue population from bunkering population, with confidence flag |
| High-resolution vessel confirmation | Tasked WorldView-3 acquisition; hull marking and freeboard extraction at 31 cm resolution | Imagery chip with vessel identifier annotation and analyst note on load status or activity |
| Historical congestion baseline report | Retrospective CFAR detection across Sentinel-1 archive; seasonal decomposition of queue-length series | PDF report covering up to 10 years of anchorage queue history for a named port, with annotated anomaly events |
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.