Vessel dwell time and berth utilisation from optical imagery
Repeated high-resolution optical passes over commercial ports let analysts measure how long individual vessels occupy berths, revealing congestion days before it surfaces in official port statistics or freight indices.
Sensors
- Planet SuperDove (PlanetScope): 3-metre native resolution, 8 spectral bands (coastal blue through NIR), near-daily global revisit. Sufficient to detect vessels above roughly 50 metres in length and to distinguish occupied from empty berths. Intra-day tasking is not available; one pass per day per location is typical at mid-latitudes.
- Maxar WorldView-3: 30-cm panchromatic, 1.24-m multispectral. Resolves hull markings, fender positions and mooring lines clearly enough to support vessel-class identification. Revisit is 1–4.5 days depending on latitude and tasking priority; not a daily monitoring asset but the reference standard for detection confidence.
- Airbus Pléiades Neo: 30-cm panchromatic, 1.2-m multispectral, stereo and tri-stereo capable. Useful for shadow-geometry height estimation of superstructures, which aids vessel-type disambiguation. Revisit roughly 1 day for a single satellite; the two-satellite constellation improves this marginally.
- Sentinel-2 (ESA, open access): 10-metre multispectral resolution, 5-day revisit at the equator (2–3 days at higher latitudes with both satellites). Free and archival to 2015. Adequate for berth occupancy state (occupied or empty) on large vessels; hull-level detail is not achievable. Cloud cover is the dominant constraint at many port locations.
What a hull gives away from 400 kilometres up
A large container ship or bulk carrier is between 180 and 400 metres long, and its hull presents a high-contrast edge against the water surface regardless of paint scheme. At 30-cm resolution, you can read the waterline, count deck tiers, and in many cases match the silhouette against Lloyd's Register vessel classes. At 3 metres, you lose that granularity but retain enough signal to confirm presence, estimate length to within 10–15 metres, and distinguish a laden vessel (lower freeboard, wider waterline footprint) from one riding high in ballast.
Shadow geometry adds a second channel of information. A vessel moored at a quay casts a shadow whose length is a function of solar elevation angle and the height of the superstructure above water. With known acquisition time and sun angle, analysts can estimate bridge height to within a few metres, which narrows vessel-type candidates substantially. Wake patterns are less useful in berth analysis than in open-water detection, but residual turbidity around a recently arrived vessel can indicate arrival timing when compared against a prior clean-water baseline.
Turning a sequence of snapshots into a dwell-time series
A single image tells you whether a berth is occupied. A time series tells you for how long. The method is straightforward: register consecutive images to a common berth grid, run an object detector (template matching or a trained convolutional classifier) over each berth polygon, and record occupancy state with a timestamp. Dwell time is then the interval between first detection of arrival and first detection of departure, bounded by the revisit cadence.
This is where the honest limit bites. Planet's near-daily revisit means you can resolve dwell time to plus or minus roughly 24 hours under clear skies. If a vessel arrives and departs within a single inter-pass interval, it is invisible to the record entirely. Maxar WorldView-3 can be tasked to close that gap for specific ports of interest, but at higher cost and with no guarantee of cloud-free acquisition. For ports in persistently cloudy regions such as the Gulf of Guinea or South-East Asian archipelagos, optical methods alone will produce gaps of several days during monsoon seasons. Fusing optical detections with AIS position reports narrows the ambiguity, though AIS manipulation is a separate problem addressed elsewhere in this library.
Across a fleet of berths, the aggregate picture is more reliable than any individual dwell estimate. Berth utilisation rate, defined as the fraction of berth-hours in a period during which at least one vessel is present, smooths over individual detection gaps. A port operating at 85% berth utilisation for three consecutive weeks is a congestion signal that optical time-series analysis can surface well before port authority throughput figures are published.
Resolution floors and the minimum detectable vessel
At 3-metre resolution, a vessel needs to span at least 3–4 pixels in its shortest dimension to register reliably above background clutter. That sets a practical lower bound of roughly 40–50 metres in beam width, which excludes most coastal ferries, fishing vessels and small tankers. Harbour tugs and pilot boats are invisible. This matters for utilisation accounting: a berth occupied by a small feeder vessel will be undercounted relative to one occupied by a ULCC.
WorldView-3 at 30 cm has no meaningful minimum-length constraint for commercial shipping, but its infrequent revisit makes it a spot-check instrument rather than a monitoring backbone. The practical architecture for most port-monitoring programmes combines daily Planet coverage for occupancy state with periodic WorldView or Pléiades Neo tasking for vessel identification and length measurement.
What berth utilisation data actually tells a buyer
Port authorities publish throughput in TEUs or tonnes, but those figures lag by weeks or months and reveal nothing about the distribution of waiting time across berths or vessel classes. Satellite-derived dwell time fills a specific gap: it shows which berths are chronically long-dwell (indicating slow cargo operations, equipment failure or customs hold), which are cycling rapidly, and how the pattern shifts after a policy change or infrastructure investment.
For commodity traders and freight-rate analysts, a sustained rise in average dwell time at a major iron ore or grain terminal is a leading indicator of supply-chain stress. The Baltic Dry Index and similar freight benchmarks are market-priced and therefore already partially forward-looking; satellite dwell data is observational and independent of market sentiment. That independence is the value. It cannot be managed, seasonally adjusted or delayed for political reasons.
Sovereign port authorities have a different use case: benchmarking their own terminal against regional competitors using a common measurement methodology that does not rely on self-reported statistics. Satellize has run similar independent-measurement programmes in the Pacific, including the Tonga crop-estimation work, and the principle transfers directly to port monitoring: a small state gains an external audit capability it could not otherwise afford.
Archive depth and the baseline problem
Planet's commercial archive runs back to approximately 2016 for most ports. Sentinel-2 covers from mid-2015. That gives analysts a six-to-nine-year baseline against which current utilisation can be compared, which is long enough to capture a full infrastructure investment cycle at most ports. Maxar's archive is deeper still, with WorldView-1 imagery available from 2007 at selected locations.
Building a reliable baseline requires handling heterogeneous image quality across the archive: varying sun angles, atmospheric conditions, sensor generations and ground-sample distances. Normalising these is standard practice in the remote-sensing literature but adds processing overhead. Analysts should expect that the earliest years of any archive will carry wider uncertainty bands on occupancy estimates than more recent, denser coverage. Stating that uncertainty explicitly, rather than presenting a smooth time series, is the honest approach.
Typical figures
| Spatial resolution (primary source) | 3 m (Planet SuperDove); 30 cm panchromatic (WorldView-3, Pléiades Neo) |
| Revisit cadence | Near-daily (Planet); 1–4.5 days tasked (WorldView-3); 5-day free (Sentinel-2) |
| Dwell-time resolution | ±24 hours under clear skies with daily Planet coverage; worse during cloud events |
| Minimum detectable vessel length | ~50 m at 3 m GSD; no practical commercial-vessel floor at 30 cm |
| Spectral bands used | Visible (RGB), NIR for water/hull contrast; panchromatic for shadow geometry |
| Archive depth | 2015–present (Sentinel-2, open); 2016–present (Planet commercial); 2007–present (Maxar, selected sites) |
| Cloud-cover constraint | Optical only; persistent cloud (e.g. monsoon seasons) can cause multi-day gaps |
| Delivery latency | Typically 12–48 hours from satellite pass to processed output, depending on tasking pipeline |
| Coverage | Any commercial port globally; near-daily coverage improves above 30° latitude with Planet |
| Delivery formats | GeoTIFF chips, GeoJSON berth-state records, CSV dwell-time tables, dashboard API feed |
Analytics Satellize can run
| Berth occupancy state log | Object detection on registered image time series; template matching or CNN classifier over predefined berth polygons | Timestamped GeoJSON record per berth, updated each clear-sky pass |
| Vessel dwell-time estimate | Interval analysis between first arrival detection and first departure detection, with uncertainty bounds derived from inter-pass gap | CSV table of vessel events per berth: estimated arrival window, departure window, dwell duration with confidence interval |
| Berth utilisation rate | Fraction of berth-hours occupied, aggregated weekly or monthly across the berth fleet; smooths individual detection gaps | Time-series chart and tabular report, exportable to BI tools |
| Vessel-class identification | Shadow-geometry height estimation combined with hull-length measurement at high resolution (WorldView-3 or Pléiades Neo); matched against published vessel-class dimension ranges | Annotated image chips with vessel-class label and confidence score |
| Congestion alert | Rule-based trigger when rolling average dwell time or utilisation rate exceeds a defined threshold relative to the historical baseline | Email or API alert with supporting image and statistical context |
| Port-to-port utilisation benchmark | Parallel time-series analysis across a defined set of comparable terminals using a common berth-polygon schema and normalised detection methodology | Comparative dashboard showing ranked utilisation rates and dwell distributions across selected ports |
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.