Anomalous container dwell-time detection at ports for smuggling risk
Very-high-resolution optical time series can track individual container positions across port yards, flagging dwell-time outliers and atypical inter-yard transfers as behavioural risk signals for customs prioritisation.
Sensors
- Planet SkySat: Delivers 0.5 m native resolution panchromatic and 0.8 m multispectral imagery. Can be tasked to revisit the same site multiple times per day (up to 12 times daily in published Planet specifications), which is the revisit cadence needed to resolve individual container movements within a single shift.
- Maxar WorldView-3: 0.31 m panchromatic, 1.24 m multispectral, 3.7 m SWIR. At 0.31 m ground sample distance a standard 20-foot ISO container (6.1 m × 2.4 m footprint) occupies roughly 20 × 8 pixels, sufficient for individual unit identification by colour and position. Revisit at mid-latitudes is roughly 1 to 4.5 days without tasking; tasked revisit can be same-day.
- Airbus Pléiades Neo: 0.3 m native panchromatic resolution, 1.2 m multispectral, with a constellation of four satellites giving sub-daily tasking capacity over most latitudes. Pléiades Neo's very high geolocation accuracy (CE90 better than 0.5 m without ground control in published Airbus specifications) reduces co-registration error when differencing sequential images, which matters when a container has moved only one slot in a yard.
- PlanetScope SuperDove: 3 m resolution, eight spectral bands, near-daily global coverage. Too coarse to resolve individual containers at most terminals, but useful for monitoring overall yard occupancy density and identifying periods of unusual congestion or clearance that prompt tasking of a higher-resolution asset.
What a port yard reveals from above
A container terminal is one of the most geometrically regular environments on earth. Containers are stacked in defined rows and bays, moved by equipment that follows predictable paths, and cleared according to vessel schedules that are publicly filed. That regularity is exactly what makes anomalies legible in a time series of satellite images.
At 0.3 to 0.5 m resolution, a 20-foot container occupies enough pixels to be tracked by colour, position and shadow height across sequential acquisitions. A container that does not move across multiple vessel calls, or that migrates between remote yard sections at night, departs from the statistical baseline of its neighbours. That departure is the signal. The satellite cannot determine what is inside the box; it records only that the box is behaving differently from the population around it.
The customs logic behind dwell-time risk scoring
The World Customs Organisation's SAFE Framework and the UNODC Container Control Programme both treat abnormal dwell time as a primary non-intrusive risk indicator. A container that sits in a terminal for weeks beyond the average clearance time for its declared commodity and origin corridor is, by published WCO methodology, a candidate for physical examination. The same logic applies to containers that cycle between bonded yards without apparent commercial purpose, a pattern associated with concealment during document re-processing.
Customs agencies have historically relied on terminal operator data feeds to compute dwell time. Those feeds are only as reliable as the operator's own record-keeping, and in some jurisdictions the records are manipulated. Independent satellite observation provides a cross-check that does not pass through the same data chain. It is not a replacement for manifest analysis or scanner data; it is a corroborating layer.
Building a dwell-time baseline from image time series
The analytic workflow begins with co-registered image stacks over the terminal of interest. At SkySat or WorldView-3 resolution, individual container positions can be extracted using object-based image analysis. Published methods in remote sensing literature treat the problem as a multi-object tracking task: each container is a candidate object, and its centroid position is logged per acquisition. A dwell-time distribution is then estimated from the population of all tracked objects over a baseline period, typically 60 to 90 days.
Anomaly flags are raised when a specific object's dwell exceeds a configurable percentile threshold, or when its trajectory between acquisitions is inconsistent with normal yard-management patterns. Night acquisitions, where available from tasked commercial satellites, add a second dimension: legitimate terminal operations follow shift patterns, and movement outside those windows is itself a flag.
The honest limits here are worth stating plainly. Cloud cover interrupts optical time series; a persistent overcast period of several days can create a gap large enough to miss a container's departure and return. Roofed or covered storage areas are opaque to optical sensors entirely. And at terminals where containers are stacked four or five high, only the top layer is visible. A container buried in a stack is invisible until it surfaces.
Which ports are worth watching, and why that question matters
Not all terminals warrant the cost of daily high-resolution tasking. The UNODC Container Control Programme has published its operational focus on specific port corridors in West Africa, South-East Asia and Latin America, which provides an open-source basis for prioritising tasking budgets. Throughput data from Lloyd's List Intelligence and UNCTAD's Review of Maritime Transport identify terminals where container volumes are high relative to declared customs staffing, a proxy for inspection-rate constraints.
Transshipment hubs warrant particular attention. A container moving from an origin port to a transshipment hub and then to a destination can acquire new documentation at each step. The satellite record is indifferent to document changes; it tracks physical presence. If the same physical object, identified by colour and position continuity, dwells at a transshipment terminal for an anomalous period before onward movement, that is visible regardless of what the manifest says at each leg.
Honest limits of the method
Spatial resolution is the first constraint. At 0.3 m, individual containers are trackable in principle, but geolocation accuracy and image-to-image co-registration error can be of the same order as a single container width, especially over terminals with minimal ground-control infrastructure. This means that a container that has moved one slot laterally may not be distinguishable from one that has not moved at all. The practical minimum detectable displacement is roughly two to three container widths, or about 5 to 7 metres, under good co-registration conditions.
Revisit is the second constraint. Even with SkySat's published multi-daily tasking capacity, cloud, satellite scheduling conflicts and cost all reduce actual delivery. A realistic operational cadence for a single terminal might be two to four clear acquisitions per day at best, dropping to zero during persistent cloud events. In tropical port environments, this is a significant operational limitation.
The method produces a behavioural anomaly flag, not evidence of contraband. Every flag requires human review against manifest data, vessel call records and other intelligence. The false-positive rate depends heavily on how well the baseline distribution captures legitimate seasonal variation in terminal operations, such as pre-holiday cargo surges or port-strike backlogs.
Satellize's analytics team has applied similar object-persistence methods in agricultural settings, including the Tonga crop-estimation programme, and the same statistical baseline approach transfers to port environments with appropriate domain adaptation.
Integrating satellite flags into a customs workflow
The output of a satellite-based dwell-time analysis is most useful when it feeds into an existing risk-scoring system rather than operating as a standalone channel. Published customs risk-management frameworks, including the WCO's Columbus Programme guidance, describe layered scoring in which multiple independent indicators are combined before a physical examination decision is made. A satellite-derived dwell-time anomaly flag is one such indicator, weighted alongside manifest anomalies, shipper history and route-risk scores.
Practically, this means the deliverable from a satellite analytics engagement is a structured data feed, updated on each clear acquisition, that lists container-position objects flagged as outliers with their yard location, estimated dwell duration and the statistical basis for the flag. That feed can be ingested directly into a customs risk-management platform. The decision to examine remains entirely with the customs authority.
Typical figures
| Best available spatial resolution | 0.31 m panchromatic (Maxar WorldView-3); 0.3 m (Pléiades Neo); 0.5 m (Planet SkySat) |
| Minimum detectable container displacement | Approximately 5 to 7 m under good co-registration conditions (2 to 3 container widths) |
| Tasked revisit (single site) | Up to 12 times per day (SkySat, published Planet specification); sub-daily (Pléiades Neo 4-satellite constellation); 1 to same-day (WorldView-3 tasked) |
| Spectral bands used | Panchromatic for position tracking; RGB and NIR multispectral for container colour classification and change detection |
| Cloud impact | Optical sensors only; persistent cloud can create gaps of several days; no SAR substitute at this resolution for container-scale tracking |
| Coverage per acquisition | SkySat: approximately 21 km² per scene strip; WorldView-3: approximately 13.1 km × variable strip; Pléiades Neo: 14 km × 14 km standard scene |
| Archive depth | Maxar archive from 2014 (WorldView-3 launch); Planet SkySat archive from 2016; Pléiades Neo from 2021 |
| Delivery formats | GeoTIFF image stacks; GeoJSON object-track layers; CSV dwell-time anomaly tables; risk-score feed for customs platform ingestion |
| Visible container layers | Top layer of stacked containers only; containers in covered or roofed storage are not detectable |
Analytics Satellize can run
| Container position time series | Object-based image analysis and multi-object tracking applied to co-registered high-resolution image stacks; published in remote sensing literature as a standard geospatial object-persistence method | GeoJSON layer per acquisition showing detected container objects with centroid coordinates, estimated dimensions and acquisition timestamp |
| Dwell-time anomaly flags | Statistical baseline modelling of dwell-time distribution across all tracked objects over a 60 to 90 day reference period; flags raised at configurable percentile threshold (e.g. 95th percentile of commodity-stratified baseline) | Structured CSV or JSON feed listing flagged container objects with yard location, estimated dwell duration, deviation from baseline and confidence band |
| Night-movement anomaly detection | Comparison of container positions between last daytime acquisition and first morning acquisition to identify objects moved outside normal shift hours; requires at least two acquisitions bracketing the night window | Alert layer of objects with anomalous nocturnal displacement, delivered within hours of morning acquisition |
| Yard occupancy density trend | PlanetScope SuperDove near-daily 3 m imagery used to compute fractional occupied area per yard zone via spectral segmentation; used to trigger high-resolution tasking when occupancy anomalies appear | Weekly occupancy trend chart per yard zone; tasking trigger recommendation |
| Inter-yard transfer pattern analysis | Graph-based analysis of container object trajectories across yard zones over multi-week time series; identification of transfer sequences inconsistent with normal terminal logistics flow, cross-referenced against published terminal operating patterns | Risk-scored transfer-event report, suitable for direct input into a WCO-aligned customs risk-management platform |
| Historical baseline audit | Retrospective analysis of archived imagery (WorldView-3 from 2014, SkySat from 2016) to establish long-run dwell-time norms and identify past anomaly clusters at a named terminal | PDF baseline audit report with annotated image evidence and statistical summary; supports customs programme design |
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.