Container port throughput estimation from stacked-box counting
Very-high-resolution optical imagery lets analysts count ISO containers on quaysides and in storage yards, then combine those counts with AIS vessel-call data to estimate weekly or monthly throughput without relying on port authority disclosures.
Sensors
- WorldView-2 / WorldView-3 (Maxar): Panchromatic ground sampling distance of 31 cm (WV-3) and 46 cm (WV-2) gives enough resolution to distinguish individual container footprints and estimate stack height from shadow length. Revisit at a given point is roughly 1 to 4.5 days depending on latitude and tasking priority.
- Pleiades 1A / 1B (Airbus): 50 cm panchromatic GSD with a daily combined revisit for the stereo pair. Stereo mode allows direct stack-height estimation from parallax, reducing reliance on shadow-based inference. Well-documented in published port-monitoring literature.
- Planet SuperDove: 3 m multispectral GSD is insufficient to resolve individual containers, but the near-daily global revisit makes it useful for detecting yard-occupancy change between higher-resolution collections, flagging days worth tasking with a commercial sensor.
- AIS vessel-call records: Automatic Identification System messages, aggregated by providers such as Spire or commercial maritime data feeds, supply vessel arrival and departure timestamps, vessel type and deadweight tonnage. These anchor the turnover-rate prior that converts a static box count into a throughput estimate.
What a 31-centimetre pixel actually resolves
An ISO standard twenty-foot-equivalent unit is 2.438 m wide and 6.058 m long. At WorldView-3's 31 cm panchromatic GSD, each container occupies roughly eight pixels across its short axis. That is enough to detect, localise and count individual boxes in a single storage row, and to distinguish a two-high stack from a four-high one by measuring the shadow cast at a known solar elevation angle.
The geometry is straightforward in principle. Shadow length divided by the tangent of the solar elevation angle gives apparent object height. A standard container is 2.591 m tall, so each additional tier adds a predictable shadow increment. In practice, adjacent stacks cast overlapping shadows, and cranes, reefer power columns and lighting gantries introduce confounding features. Published work on WorldView imagery at major Asian and European ports reports count accuracy in the range of 70 to 85 per cent of ground-truth TEU counts before correction, rising with better shadow-separation algorithms and cleaner imagery geometry.
Turning a snapshot count into a throughput estimate
A single image gives a yard-occupancy figure, not a throughput figure. The conversion requires a dwell-time or turnover-rate prior: how many days does a container typically sit in this yard before it moves? That figure varies enormously. A transshipment hub such as Port Klang or Colombo turns boxes over in two to four days on average. An inland feeder port serving slow-moving import cargo may hold containers for two to three weeks.
AIS vessel-call data supplies the empirical anchor. By matching vessel arrivals and departures to known vessel capacity and load factors from published fleet databases, analysts can estimate the volume of TEUs flowing through a port per vessel call. Combining that with the yard-occupancy snapshot and a dwell-time prior derived from the same AIS time series produces a throughput estimate with a quantifiable confidence interval. The method is not a black box: every assumption about dwell time and load factor is explicit and auditable.
Seasonal variation is a genuine complication. Pre-Lunar New Year surges at Chinese ports, or pre-Christmas peaks at European import hubs, compress dwell times and inflate yard occupancy simultaneously. A model calibrated on average conditions will overestimate throughput during peak periods unless the prior is updated. This is one reason why monthly estimates are more reliable than weekly ones, and why multi-year AIS archives matter.
Occlusion, cloud and the honest count ceiling
Stacking is the fundamental limit. A five-high stack of containers presents only its top layer to a nadir-viewing sensor. The bottom four tiers are invisible. Stereo imagery from Pleiades can recover stack height from parallax, but even stereo does not see through the top box. Analysts must apply a height-distribution prior derived from port operating rules and historical imagery to estimate total TEU volume from visible surface area.
Cloud cover is the other hard constraint. Optical sensors cannot see through cloud, and major ports in South-East Asia, West Africa and the Gulf of Mexico experience cloud cover on 40 to 70 per cent of days during wet seasons. A tasking strategy that combines high-resolution optical collection on clear days with Planet SuperDove's near-daily change detection to flag yard-state shifts on cloudy days partially mitigates this, but gaps in the time series remain. SAR-based yard-occupancy detection is a complementary approach covered separately in the grain-silo and bulk-storage page in this library.
Shadow accuracy degrades at solar elevations below about 30 degrees, which affects high-latitude ports in winter. At Helsinki or Vancouver in December, the sun angle is low enough that shadows from a three-high stack extend beyond the adjacent row, making automated height inference unreliable without manual correction.
Where the signal is most useful to financial analysts
Port throughput data published by port authorities is typically released with a one to three month lag, and some ports in emerging markets publish annually or not at all. Satellite-derived estimates fill that gap for trade-finance lenders assessing collateral quality, commodity traders monitoring import and export flows, and sovereign analysts tracking industrial activity in economies where customs data is opaque.
The method is particularly sharp for detecting step-change events: a port going from 70 per cent yard occupancy to 20 per cent in a fortnight is a signal worth investigating regardless of the absolute TEU count. Sanctions-related cargo diversion, labour disputes, infrastructure damage after a weather event, or a shift in shipping alliances can all produce that kind of signature. The relative change is often more actionable than the absolute estimate.
Satellize applies this method as part of its broader financial-intelligence analytics suite. For clients requiring recurring port monitoring, the workflow integrates commercial tasking on WorldView or Pleiades with open AIS feeds and automated change-detection on Planet's daily archive.
What to expect from a delivered product
A realistic deliverable for a single major port is a monthly TEU-occupancy estimate with a stated uncertainty band of plus or minus 15 to 25 per cent, depending on cloud frequency, image geometry and the quality of the AIS prior for that port. For well-documented ports with dense AIS coverage and frequent clear-sky opportunities, the uncertainty narrows. For smaller ports with sparse vessel-call data and frequent cloud, it widens, and the analyst should say so explicitly rather than presenting a false point estimate.
Throughput estimates are most credible when presented as an index relative to a baseline period rather than as an absolute TEU count. Saying that yard occupancy at Port X is 35 per cent below its 24-month average is a more defensible statement than claiming the port handled 180,000 TEUs last month. Buyers of this analysis should ask any provider to show their calibration methodology and the AIS data sources behind their dwell-time priors.
Typical figures
| Typical spatial resolution (primary sensor) | 31 cm pan (WorldView-3); 50 cm pan (Pleiades 1A/1B) |
| Minimum resolvable target | Single ISO container (2.4 m × 6 m footprint) at 30–50 cm GSD |
| Revisit frequency (commercial tasking) | 1–4.5 days (WorldView-3, latitude-dependent); ~1 day combined (Pleiades pair) |
| Change-detection cadence (Planet SuperDove) | Near-daily at 3 m GSD for yard-state flagging |
| Spectral bands used | Panchromatic (count, shadow); multispectral RGB (container colour classification, quality control) |
| Stack-height estimation method | Shadow-length geometry (solar elevation known) or stereo parallax (Pleiades stereo mode) |
| AIS data latency | Near-real-time to 24-hour delay depending on provider and vessel AIS transmission rate |
| Imagery archive depth | WorldView-2 from 2009; Pleiades from 2012; Planet from 2016 (varies by port location) |
| Throughput estimate uncertainty (typical) | ±15–25% of ground-truth TEU count; narrows with dense AIS priors and frequent clear-sky collections |
| Delivery formats | GeoTIFF annotated count layers, GeoJSON yard-occupancy polygons, CSV throughput time series, PDF analytical report |
Analytics Satellize can run
| Monthly TEU yard-occupancy estimate | Object detection on VHR panchromatic imagery using known ISO container dimensions as a size prior; shadow-length stack-height inference | GeoJSON polygon layer with per-zone TEU count and stack-height distribution; monthly PDF summary |
| Throughput index (relative to baseline) | Yard-occupancy count combined with AIS-derived dwell-time prior and vessel-call volume; Bayesian update on seasonal turnover model | CSV time series of indexed throughput with stated confidence intervals, updated monthly |
| Yard-state change alert | Automated change detection on Planet SuperDove daily mosaic; threshold trigger on occupancy delta exceeding one standard deviation from rolling mean | Email or API alert with flagged date, occupancy delta and recommended tasking window for VHR follow-up |
| Vessel-call to cargo-volume reconciliation | AIS arrival/departure log matched to vessel deadweight and published load-factor distributions; cross-checked against yard-occupancy imagery | Spreadsheet reconciliation table per vessel call; anomaly flags where imagery and AIS imply inconsistent cargo volumes |
| Seasonal dwell-time prior calibration | Multi-year AIS archive regression against historical imagery occupancy series to derive port-specific, month-of-year dwell-time distributions | Port-specific prior parameter table, updated annually, supplied as model input documentation |
| Multi-port comparative activity dashboard | Standardised occupancy index across a defined portfolio of ports, normalised by yard capacity derived from imagery-based footprint mapping | Interactive GIS dashboard or static monthly PDF comparing relative activity across client-specified port list |
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.