Container yard and port logistics area utilisation monitoring
VHR optical satellites count containers, estimate stack heights, and track yard occupancy at weekly or better cadence, giving port operators and trade analysts throughput proxies that bypass restricted or delayed customs reporting.
Sensors
- SkySat (Planet Labs): 50 cm native resolution, up to 12 tasked passes per day over a single site when the full constellation of 21 satellites is coordinated. Sufficient to resolve individual 20 ft and 40 ft containers and cast measurable shadows for stack-height estimation.
- WorldView Legion (Maxar): 30 cm panchromatic resolution, designed for high-frequency revisit over priority areas with up to 15 passes per day at mid-latitudes. Offers the sharpest shadow geometry of the commercial options listed here.
- Pleiades Neo (Airbus): 30 cm panchromatic, 1.2 m multispectral, revisit of roughly 1 day at equatorial latitudes when both Pleiades Neo satellites are tasked together. Colour bands aid container-colour classification and change detection.
- WorldView-3 (Maxar): 31 cm panchromatic, 1.24 m multispectral, 16-band SWIR available. Established archive since 2014 provides baseline imagery for multi-year occupancy trend analysis at major terminals.
What a shadow tells you about a stack
A standard ISO shipping container is 2.59 m tall. When a VHR satellite images a yard at a known sun elevation angle, the shadow cast by a stack of containers has a predictable length. Simple trigonometry converts that length into a stack count. At 30 cm resolution, shadows from a two-high stack are distinguishable from a four-high stack with reasonable confidence, provided the sun elevation is between roughly 30 and 60 degrees. Below 20 degrees the shadows overlap neighbouring stacks and the geometry becomes ambiguous; above 65 degrees the shadows are too short to measure reliably.
This method is not new. It has been documented in peer-reviewed remote sensing literature and applied in trade-intelligence contexts since at least the mid-2010s. The honest limit is that shadow-based height estimation carries an uncertainty of roughly plus or minus one container tier when shadows are partially occluded by adjacent stacks or yard equipment. Automated detection pipelines trained on labelled VHR imagery can count individual container footprints in plan view, then cross-reference the shadow-derived tier count to produce a volume estimate for each yard block.
Why weekly cadence matters more than single-image accuracy
A single image gives you occupancy at one moment. A weekly time series gives you dwell time, throughput rhythm, and the early signal of congestion building before it becomes a crisis. Ports operating near capacity show a characteristic pattern: yard density climbs, dwell time extends, and the ratio of containers in motion to containers sitting still drops. All three signals are visible in a consistent image series without access to any port management system.
SkySat's tasking flexibility is particularly useful here. Because individual satellites in the constellation can be redirected within hours, an analyst can request same-week imagery of a terminal that appeared congested in the previous pass. WorldView Legion is designed explicitly for this kind of high-frequency targeting. The practical constraint is cloud cover. Tropical and monsoon-affected ports, including many of the world's fastest-growing terminals in South and South-East Asia, can lose two to four weeks of optical coverage during peak rainy season. SAR is the obvious complement for those gaps, though SAR container counting is less mature than optical methods and currently requires more manual validation.
Object detection pipelines: what the published literature supports
Automated container detection on VHR imagery typically uses convolutional neural networks trained on labelled datasets of port imagery. Published studies in journals such as Remote Sensing (MDPI) have demonstrated mean average precision scores above 0.85 for container detection on WorldView and Pleiades imagery under good illumination conditions. Performance degrades in three documented circumstances: when containers are covered by tarpaulins or canvas (colour and edge cues are lost), when yard lighting creates artificial shadows in night-time imagery, and when containers are stacked directly against warehouse walls, which breaks the shadow geometry.
A practical pipeline combines a detector for individual container footprints in plan view with a separate shadow-measurement module. The footprint count gives a lower bound on occupancy; the shadow module gives a tier-weighted volume. Dividing the occupied area by the total yard area defined in a baseline polygon gives a utilisation percentage. That percentage, tracked weekly, is the core deliverable. It is a proxy for throughput, not a direct measurement of TEUs moved. Anyone using it as a trade signal should understand that distinction.
Inland terminals and the data gap they fill
The use case extends well beyond coastal ports. Inland container depots, rail-connected logistics parks, and bonded warehouse yards often have no public reporting obligation at all. Customs data for these facilities, where it exists, may be aggregated at a national level and released quarterly. Satellite imagery is frequently the only timely, spatially specific source of occupancy information for these sites.
Inland yards also tend to have simpler geometry than major port terminals. Fewer cranes, less equipment clutter, and more regular stacking patterns mean that automated detection often performs better than at a busy quayside terminal. The trade-off is that inland yards are smaller, so the absolute container counts are lower and percentage occupancy estimates carry higher relative uncertainty. A yard holding 200 containers where the detector misses 20 has a 10 percent error. At a terminal holding 20,000 containers, the same miss rate is operationally acceptable.
Honest limits before you commission an analysis
Cloud cover is the most significant operational constraint and cannot be engineered away for optical sensors. A port in a persistently cloudy region may yield only six to eight usable images per quarter even with aggressive tasking. Tasking costs for WorldView Legion and Pleiades Neo at high revisit frequency are not trivial; a programme monitoring a dozen terminals weekly will require a commercial imagery licence sized accordingly.
Shadow-based stack counting assumes the sun angle is known precisely, which it is, but it also assumes containers are stacked vertically and flush with each other. Leaning stacks, angled rows, and partial loads all introduce error. Container colour classification from multispectral bands can help distinguish operator fleets and identify empty versus loaded units by colour convention, but this is an inference, not a direct measurement. Satellize runs this class of analytics on tasked commercial imagery for clients who need trade-intelligence proxies or port-capacity planning inputs. The Tonga crop-estimation programme uses a structurally similar object-counting approach on agricultural parcels, which informed some of the detection pipeline design. For port work, the honest starting point is a pilot over a terminal where ground-truth data are available, so that detection accuracy can be validated before the method is scaled.
Typical figures
| Best available spatial resolution | 30 cm panchromatic (WorldView Legion, Pleiades Neo, WorldView-3) |
| Typical revisit at high-priority sites | Daily to sub-daily with SkySat or WorldView Legion tasking; 1–2 days with Pleiades Neo |
| Minimum detectable target | Individual 20 ft ISO container (6.1 m × 2.44 m footprint) at 30–50 cm resolution under clear-sky conditions |
| Stack height estimation range | 1 to 5 tiers via shadow length; reliable at sun elevation 30–60°; uncertainty ±1 tier in cluttered yards |
| Spectral bands used | Panchromatic (shadow geometry); RGB and NIR multispectral (container colour, surface classification) |
| Cloud cover constraint | Optical only; persistent cloud can reduce usable acquisitions to 6–8 per quarter at tropical sites |
| Archive depth | WorldView-3 archive from 2014; SkySat from 2016; Pleiades Neo from 2021 |
| Delivery formats | GeoTIFF orthoimage, GeoJSON container polygons, CSV occupancy time series, PDF trend report |
| Typical analysis latency | 24–72 hours from image acquisition to delivered analytic, depending on pipeline automation level |
Analytics Satellize can run
| Yard occupancy percentage (weekly) | Automated container footprint detection via CNN on VHR panchromatic imagery; occupied area divided by total yard polygon | Time-series CSV and dashboard feed, updated per acquisition |
| Container stack-height map | Shadow-length measurement at known sun elevation angle; trigonometric conversion to tier count per yard block | GeoJSON polygon layer with tier-count attribute per yard zone |
| Throughput proxy index | Week-on-week change in container count and occupancy; ratio of containers in motion (visible equipment, open lanes) to static inventory | Weekly index report with trend chart and anomaly flags |
| Congestion alert | Threshold trigger when yard occupancy exceeds a client-defined level (e.g. 85%) across two consecutive acquisitions | Email or API alert with supporting image chip and occupancy figure |
| Multi-terminal benchmarking | Standardised occupancy and stack-volume metrics applied consistently across a portfolio of monitored terminals | Monthly comparative report in PDF and spreadsheet format |
| Historical baseline and trend analysis | Retrospective processing of WorldView-3 or SkySat archive imagery using the same detection pipeline | Multi-year occupancy trend dataset, GeoTIFF archive, summary report |
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.