Vessel anchorage queue density and waiting-time estimation
SAR imagery detects every vessel at anchor regardless of AIS status, while historical AIS archives supply individual dwell durations. Together they turn anchorage queues into quantified congestion signals for ports, traders and freight forwarders.
Sensors
- Sentinel-1 SAR (ESA/Copernicus): C-band (5.405 GHz) synthetic aperture radar. IW mode delivers 5 × 20 m ground resolution across a 250 km swath. Revisit over most ocean anchorages is 6 days with one satellite, 3 days with two. Free and open archive from 2014. Reliable vessel detection down to roughly 50 m vessel length in calm to moderate sea states; smaller craft and detection in high sea states are less reliable.
- ICEYE SAR constellation: X-band commercial SAR. Spot mode reaches 25 cm range resolution (1 m × 1 m products); Strip mode covers wider areas at roughly 3 m resolution. Revisit to a specific anchorage can be less than 24 hours with tasking. Better clutter suppression at X-band improves small-vessel discrimination versus C-band, though rain attenuation is higher.
- Capella Space SAR: X-band commercial constellation. Spotlight products at 50 cm resolution; Sliding Spotlight at roughly 1 m. Sub-daily revisit achievable through tasking. Particularly useful for resolving vessel type and approximate length when queue counts need to be broken down by ship class.
- AIS historical archive: Automatic Identification System position records, typically at 2 to 10 minute polling intervals for Class A transponders. Terrestrial AIS covers coastal anchorages within roughly 40 to 60 km of shore; satellite AIS (S-AIS) extends coverage globally. Archive depth from commercial providers reaches back to 2009 or earlier. Used here to compute per-vessel dwell time and to flag vessels absent from the AIS record.
Why a queue outside the gate is harder to measure than the queue inside
Port operators count berth occupancy from their own systems. The anchorage outside is nobody's yard. Ships arrive, hold position for hours or weeks, and the only official record is whatever their AIS transponder broadcasts. That record has two well-documented failure modes: transponders switched off deliberately, and position reports that stop simply because a vessel is stationary and some systems reduce polling frequency for anchored ships. Neither failure is random. Both introduce bias into any estimate built on AIS alone.
SAR solves the detection problem. Radar backscatter from a steel hull is strong, consistent and independent of whether the vessel's crew has chosen to broadcast. Sentinel-1's C-band returns a bright point target for any vessel above roughly 50 m in length under typical sea conditions. ICEYE and Capella at X-band push that detection floor lower and add enough resolution to distinguish a Panamax bulk carrier from a Handysize. The count you get from a SAR pass is a physical census, not a self-reported list.
Building a dwell-time estimate from two imperfect sources
A single SAR image tells you how many ships are present at the moment of acquisition. It does not tell you how long they have been there. That is where the AIS archive earns its keep. For vessels that are broadcasting, matching a SAR detection to an AIS track gives an arrival timestamp and, once the vessel moves, a departure timestamp. Dwell time follows directly. Across a large population of port calls, the distribution of dwell times at a given anchorage builds a baseline: median wait, 90th-percentile wait, seasonal variation.
For dark vessels, the approach is necessarily probabilistic. If a ship appears in a SAR image but has no corresponding AIS position within a reasonable spatial and temporal window, it can be assigned to the dwell-time distribution of vessels of similar size detected at the same anchorage. That is an estimate with real uncertainty, not a measurement. The honest way to report it is as a range, weighted by how many dark detections are in the queue. Anchorages where 5 per cent of vessels are dark behave differently from anchorages where 30 per cent are dark, and the analytics should reflect that difference explicitly.
Revisit cadence sets the temporal resolution of the congestion signal. At Sentinel-1's 3 to 6 day revisit, you can track weekly trends in queue length. With ICEYE or Capella tasked daily or sub-daily, you can watch a queue build and clear within a single weather window. The appropriate sensor choice depends on whether the buyer needs a market intelligence signal updated weekly or an operational alert updated hourly.
What the backscatter actually measures, and where it lies
SAR detects the normalised radar cross-section of a target. A large steel vessel at broadside can return 30 to 40 dBsm or more; a small wooden fishing vessel might return less than 10 dBsm and fall below the detection threshold entirely. Sea clutter, the radar return from wave surfaces, sets the noise floor. In high sea states, Beaufort 5 and above, that noise floor rises and small-vessel detection degrades meaningfully. Published validation studies on Sentinel-1 vessel detection in open literature report detection rates of 85 to 95 per cent for vessels above 100 m in calm to moderate conditions, dropping to 60 to 70 per cent for vessels below 50 m.
Vessel classification from SAR alone is approximate. Length can be estimated from the bright target's extent in a high-resolution product, but beam and draught are not directly observable. ICEYE Spot or Capella Spotlight products at sub-metre resolution allow a more confident assignment to ship-type categories (tanker, bulk carrier, container ship, general cargo) based on deck structure signatures, but this is still a probabilistic classification, not a registry lookup. Cross-referencing with AIS MMSI numbers resolves type unambiguously for broadcasting vessels.
Turning a count into a congestion index
Raw vessel counts become useful only when normalised against a baseline. An anchorage holding 40 ships might be normal for Singapore's western approaches and alarming for the outer roads of a mid-sized bulk terminal. The congestion index Satellize produces for this use case expresses current queue density as a percentile of the historical distribution for that specific anchorage, computed from the same SAR and AIS archive that supplies the dwell-time baseline. A reading at the 90th percentile means the queue is larger than it has been on 90 per cent of comparable observation dates in the archive.
Waiting-time estimates derived from the index carry explicit confidence intervals. Where AIS coverage is dense and dark-vessel rates are low, the interval is narrow. Where coverage is sparse or dark rates are high, the interval widens, and the deliverable says so. Commodity traders using the signal to infer vessel availability, and freight forwarders quoting transit times, both benefit from knowing the uncertainty rather than receiving a falsely precise single number.
Practical limits a buyer should understand before commissioning
Cloud cover does not affect SAR. That is one genuine advantage over optical methods. But SAR has its own acquisition constraints: orbital geometry means a given anchorage may be imaged from ascending and descending passes at fixed local times, which can alias tidal or diurnal patterns in vessel behaviour if not accounted for in the analysis.
Sentinel-1C, launched in late 2023, maintains the free Copernicus archive. Sentinel-1B was lost in 2021, which reduced the constellation to a single satellite until 1C arrived, temporarily lengthening revisit intervals. Users relying on historical Sentinel-1 data should be aware that the 2021 to 2023 period has sparser coverage than the periods before and after. Commercial SAR from ICEYE and Capella fills that gap but at cost and with tasking lead times that vary by orbit availability.
Satellize runs anchorage queue analytics on Sentinel-1 open data as the baseline layer, with commercial SAR tasked on client licence for anchorages requiring sub-daily updates. The Tonga crop-estimation programme is the company's only publicly named analytics engagement to date, but the SAR vessel-detection pipeline draws on the same open-data infrastructure.
Typical figures
| Spatial resolution (Sentinel-1 IW mode) | 5 × 20 m (range × azimuth) |
| Spatial resolution (ICEYE / Capella Spotlight) | 25 cm to 1 m, depending on product tier |
| Revisit cadence | 3 to 6 days (Sentinel-1 two-satellite); sub-daily with commercial tasking |
| Radar frequency | C-band 5.405 GHz (Sentinel-1); X-band ~9.6 GHz (ICEYE, Capella) |
| Minimum detectable vessel length | ~50 m in calm conditions (Sentinel-1); ~10 m in Spotlight mode (ICEYE/Capella) |
| AIS archive depth | 2009 to present (commercial S-AIS providers); terrestrial AIS varies by port |
| AIS position polling interval | 2 to 10 minutes for Class A transponders under way; up to 3 minutes at anchor |
| Detection probability (vessels >100 m, calm sea) | 85 to 95 per cent (published Sentinel-1 validation literature) |
| Latency from SAR acquisition to processed output | Sentinel-1 NRT products: 1 to 3 hours post-acquisition; commercial SAR: varies by tasking contract |
| Congestion index archive depth | 2014 to present using Sentinel-1 open archive |
Analytics Satellize can run
| Anchorage vessel count | SAR constant false alarm rate (CFAR) detection applied to Sentinel-1 or commercial SAR imagery; standard published method for maritime target detection | GIS point layer of detected vessels per anchorage zone, updated per satellite pass |
| Dark-vessel fraction | Spatial join of SAR detections against AIS position records within a configurable time and distance window; unmatched detections flagged as non-broadcasting | Per-anchorage dark-vessel count and percentage, delivered as tabular report or API feed |
| Individual vessel dwell time | AIS track segmentation to identify arrival and departure timestamps at defined anchorage polygons; dwell computed as elapsed time between first and last position within polygon | Per-MMSI dwell record appended to AIS history; exportable as CSV or GeoJSON |
| Anchorage congestion index | Current queue count expressed as a percentile of the historical distribution for that anchorage, derived from the Sentinel-1 archive from 2014 onward | Weekly or daily index score with confidence interval, delivered as time-series chart and data feed |
| Estimated pre-berth waiting time | Regression of current queue length against historical dwell-time distributions for vessels of matching type; probabilistic output with stated uncertainty range | Waiting-time estimate in hours (median and 90th percentile) per anchorage, in scheduled report |
| Vessel-type classification | SAR target length estimation combined with AIS vessel-type field for broadcasting ships; probabilistic type assignment for dark vessels based on radar cross-section and deck-structure signatures in high-resolution products | Queue breakdown by ship category (tanker, bulk, container, general cargo) as stacked time-series |
| Congestion alert | Threshold trigger when congestion index exceeds a client-defined percentile; computed on each new SAR acquisition | Email or webhook alert with anchorage name, current count, index score and comparison to 30-day average |
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.