Video-from-space vessel dwell time and port throughput measurement
Sub-minute electro-optical video clips from LEO can time vessel arrivals, departures and berth occupancy at ports where AIS is unreliable or suppressed, building statistical throughput models that single frames cannot.
Sensors
- Planet SkySat video: Captures roughly 90-second clips at approximately 0.9 m native resolution and 30 frames per second. Sufficient to resolve vessel class and read gross movement direction. Revisit is opportunistic and demand-driven; a single site may receive one to three video passes per day across the SkySat fleet of around 21 satellites, though this is not guaranteed.
- Satellogic EarthView video: Delivers video at approximately 1 m resolution in the panchromatic band. Clip durations of up to 60 seconds. Constellation size constrains revisit; commercial tasking is required for repeat coverage of a specific port.
- BlackSky Global: Sub-metre resolution (approximately 0.9 m) electro-optical imagery with a constellation designed for high-revisit tasking. BlackSky publishes revisit rates of up to 15 times per day at mid-latitudes for its full constellation, though video clip mode availability varies by operational configuration.
- Umbra SAR video mode: Umbra's synthetic aperture radar can produce coherent multi-look sequences that approximate video-like temporal cadence, operating at spotlight resolutions down to approximately 0.25 m. All-weather, day-night capable. Vessel movement and berth state are detectable regardless of cloud, which matters greatly for tropical ports.
What a single frame cannot tell you
A snapshot confirms a vessel is present. It cannot confirm whether that vessel arrived two hours ago or two days ago, whether it is loading or discharging, or how long it will remain. Berth-occupancy dwell time, the metric port authorities and commodity traders actually need, requires at least two observations of the same berth: one showing the vessel absent, one showing it present, and ideally a third showing it gone again. With revisit intervals measured in hours rather than seconds, single-frame optical systems produce wide uncertainty windows around arrival and departure times.
Video clips collapse that uncertainty substantially. A 90-second clip at 30 frames per second gives an analyst 2,700 frames in which to observe bow-thruster wash, mooring-line activity and tug engagement, all reliable proxies for imminent departure or arrival. The movement direction of a vessel in a turning basin is unambiguous in video; in a pair of stills separated by hours, it is often not.
Resolution and frame-rate thresholds that actually matter
For vessel-class discrimination, 1 m ground sample distance is broadly sufficient to distinguish a Panamax container ship from a Handysize bulk carrier by length and beam ratio. Identifying a specific vessel by hull number requires 0.3 m or better, which only a handful of commercial platforms approach in video mode. Umbra's SAR spotlight mode at 0.25 m is the current commercial benchmark for fine-detail identification, though SAR backscatter interpretation differs from optical and requires trained analysts.
Frame rate matters less than resolution for dwell-time measurement. Even 1 frame per second is adequate to timestamp arrival and departure events, provided the clip overlaps with the event. The practical constraint is not frame rate but clip duration and revisit frequency. A 90-second clip that arrives three hours after a vessel has already berthed tells you nothing about arrival time. This is why multi-day stacking is the operative method: no single pass is sufficient, but a statistical ensemble of passes over weeks builds a distribution of likely dwell times per vessel class and berth.
Cloud cover is not a minor inconvenience
Many of the world's highest-throughput ports sit in the tropics: Singapore, Port Klang, Colombo, Santos. Persistent convective cloud cover means that electro-optical video is unavailable for a significant fraction of any given day. Published cloud-fraction climatologies for equatorial regions routinely show 70 to 80 percent cloud cover on a daily average. A port analytics programme that relies solely on optical video will have systematic gaps in its record during monsoon periods.
Two partial mitigations exist. First, SAR video-mode sequences from platforms such as Umbra penetrate cloud entirely, producing coherent vessel-detection data regardless of weather. Second, combining optical clips with AIS data, even knowing AIS can be spoofed or switched off, fills some gaps: if AIS shows a vessel departing and the next clear optical pass confirms an empty berth, the departure time can be bounded. Neither mitigation eliminates the gap; they reduce it. Any honest throughput model must carry a cloud-induced uncertainty term, particularly for tropical sites.
Building throughput models from stacked clips
A single video pass over a port is an anecdote. Thirty passes over thirty days, cross-referenced by berth identifier and vessel class, begin to constitute a throughput model. The method is straightforward in principle: for each berth, record the occupancy state (occupied or empty) at each pass timestamp. Fit a hidden Markov model or a simpler occupancy-rate estimator to the binary time series. The output is a probability distribution over dwell times and a berth-utilisation rate, both of which proxy cargo throughput when combined with vessel-class deadweight assumptions from Lloyd's or similar registries.
The statistical quality of the model improves non-linearly with revisit frequency. At one pass per day, dwell-time uncertainty for a 24-hour vessel call is essentially total. At six passes per day, uncertainty drops to a few hours for most vessel classes. Published work using Planet SkySat and similar constellations has demonstrated that berth-utilisation rates derived from multi-day optical stacks correlate well with official port statistics at major hubs, though the correlation degrades at smaller ports with fewer berths and lower call frequencies. Satellite-derived throughput is a leading indicator, not a certified customs record.
Where this fits alongside AIS, and where it does not
AIS, when honest, is excellent. It provides continuous position, speed and heading at update intervals of seconds, with vessel MMSI and IMO identifiers attached. The problem is not that AIS is wrong; it is that AIS is voluntary at berth in many jurisdictions, is trivially spoofed, and is sometimes deliberately switched off by operators who do not want their port calls recorded. Vessel dwell time derived from AIS alone is therefore unreliable for exactly the calls that matter most to sanctions monitors, commodity intelligence analysts and port-capacity planners.
Video-from-space does not replace AIS. It provides an independent, physics-based observation that cannot be spoofed by a transponder setting. A vessel cannot make itself optically invisible by switching off a radio. The two data streams are complementary: AIS supplies identity and continuous track; video supplies ground-truth occupancy timestamps and movement evidence for calls where AIS is absent or suspect. Satellize uses this combined approach in its analytics work, including the kind of port-monitoring configurations it has discussed with sovereign clients building their own maritime-awareness capabilities.
Current limits and what they mean for procurement
The honest position is that no commercial video constellation currently offers guaranteed sub-hourly revisit over a specific port on demand. SkySat and BlackSky come closest at mid-latitudes, but tropical ports with persistent cloud will see far fewer usable passes. Umbra SAR video mode adds weather independence but at higher tasking cost and with a smaller fleet. Dwell-time uncertainty at low revisit rates can exceed the dwell time itself for short port calls, which are common for tankers and ro-ro ferries.
Buyers should specify their minimum acceptable dwell-time uncertainty before selecting a sensor mix. For commodity intelligence at major bulk terminals, where vessel calls typically last 48 to 96 hours, even two optical passes per day may be adequate. For short-call ferry terminals or naval logistics facilities, SAR is the more reliable foundation. The analytics architecture matters as much as the sensor: a well-designed occupancy model with honest uncertainty bounds is more useful than a high-confidence figure built on insufficient passes.
Typical figures
| Typical spatial resolution (optical video) | 0.9 to 1.0 m GSD (SkySat, Satellogic, BlackSky); finer detail requires tasking at highest-quality modes |
| Typical spatial resolution (SAR video mode) | 0.25 to 1.0 m (Umbra spotlight); all-weather, day-night |
| Clip duration | 30 to 120 seconds per pass, depending on platform and tasking mode |
| Frame rate (optical) | Typically 30 fps (SkySat); adequate for movement-direction and event-timestamping at any frame rate above 1 fps |
| Revisit frequency | 1 to 15 passes per day at mid-latitudes (BlackSky published figure for full constellation); 1 to 3 per day more typical for a single site with SkySat or Satellogic; cloud reduces usable optical passes significantly in tropics |
| Minimum detectable vessel length | Approximately 20 to 30 m at 1 m GSD; reliable class discrimination above roughly 50 m |
| Spectral bands | Panchromatic plus RGB (optical platforms); X-band SAR (Umbra) |
| Latency (tasked collection to delivery) | Typically 2 to 24 hours from collection; near-real-time pipelines possible with direct-downlink ground stations |
| Archive depth | SkySat video archive from approximately 2017; BlackSky from approximately 2019; Umbra SAR from approximately 2021 |
| Cloud-cover impact | Optical unusable under opaque cloud; SAR unaffected. Tropical ports: expect 50 to 80% of optical passes to be cloud-contaminated in wet season |
Analytics Satellize can run
| Berth occupancy time series | Binary occupancy classification per pass timestamp, per berth polygon; interpolated with uncertainty bounds between passes | GIS layer (GeoJSON or Shapefile) showing per-berth occupancy state and confidence interval, updated per new collection |
| Vessel dwell-time distribution | Hidden Markov or parametric occupancy model fitted to multi-day stacked clip observations; vessel class assigned by length-beam ratio at 1 m GSD | Statistical report: median, 10th and 90th percentile dwell times by vessel class and berth, with sample-size and cloud-gap caveats |
| Berth-utilisation rate | Fraction of observation windows in which berth is occupied, aggregated weekly or monthly; normalised for cloud-gap days | Time-series chart and CSV export; updated weekly or on tasking cadence |
| AIS-dark vessel call detection | Cross-reference video-derived occupancy events against AIS message log; flag berth-occupancy intervals with no corresponding AIS transmissions | Alert feed: timestamped dark-call events with vessel-length estimate and berth identifier |
| Port throughput index | Berth-utilisation rate multiplied by vessel-class deadweight assumptions from public registry data; aggregated across all berths at a facility | Monthly throughput index report with uncertainty range; comparable across facilities using consistent methodology |
| Movement-event log (arrivals and departures) | Frame-differencing or optical-flow analysis within video clips to detect vessel motion in turning basin; timestamped against clip metadata | Structured event log (CSV or JSON) with event type, timestamp, berth, and estimated vessel class |
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.