Air cargo apron ULD and pallet inventory from very-high-resolution imagery
Sub-metre optical satellites can count unit load devices and pallets staged on cargo aprons, producing a throughput signal independent of airline or airport disclosure. This page covers sensor physics, detection limits, and how ULD counts relate to published freight tonne-kilometre data.
Sensors
- Maxar WorldView-Legion: 30 cm panchromatic resolution, revisit of up to 15 times per day over a single target at mid-latitudes. The combination of resolution and revisit frequency makes it the primary sensor for ULD counting and change detection within a single operational day.
- Airbus Pléiades Neo: 30 cm panchromatic, 1.2 m multispectral in four bands (blue, green, red, near-infrared). Multispectral capability aids discrimination of aluminium ULD panels from concrete apron, and of tarpaulin-covered pallets from bare ones. Revisit up to twice daily per site with the two-satellite constellation.
- Planet SkySat: 50 cm panchromatic, 1 m multispectral. Slightly coarser than WorldView-Legion or Pléiades Neo but capable of same-day tasking and useful for monitoring mid-sized cargo aprons where ULD clusters span several metres.
- Sentinel-2 MSI: 10 m multispectral resolution, free and open, 5-day revisit at equator. Cannot resolve individual ULDs but provides a consistent multi-year archive for detecting gross changes in apron utilisation patterns that contextualise higher-resolution counts.
What a ULD looks like from 500 km up
A standard IATA LD3 container is roughly 1.53 m wide and 1.56 m tall. An LD11 pallet stretches to 3.17 m by 2.24 m. At 30 cm ground sample distance, the LD3 occupies approximately five by five pixels in panchromatic imagery and the LD11 roughly ten by seven. That is enough geometric signal to detect and count objects, but not enough to read markings or assess structural condition.
The spectral signature is distinctive. Aluminium ULD panels have high reflectance across visible and near-infrared wavelengths, producing bright returns against the grey concrete of a cargo apron. Tarpaulin-shrouded pallets are darker and more spectrally variable, depending on cover colour. The near-infrared band on Pléiades Neo is particularly useful here: bare aluminium stays bright in NIR while most tarpaulin materials absorb more, giving a modest but measurable contrast. Shadow geometry at low sun angles adds a height cue, though apron geometry often limits shadow length.
The loaded-versus-empty problem
Counting ULDs is tractable. Knowing whether they are loaded is harder, and honesty about that limit matters for any throughput inference.
Loaded ULDs typically sit higher, with cargo building above the container lip or pallet edge. At 30 cm resolution this height difference is visible as a shadow extension and a slightly larger footprint, but the signal is noisy. A loaded LD3 with a 1.6 m build-up casts a longer shadow than an empty shell, but shadow length also depends on sun elevation, which varies by season and time of acquisition. Analysts can use the ratio of shadow length to known container width as a rough height proxy, a method well-documented in the remote sensing literature for shipping containers, but the uncertainty on any single object is high. Aggregate statistics across a cluster of twenty or more ULDs are more reliable than per-unit estimates.
Tarpaulin colour and surface irregularity provide a secondary cue. A flat, smooth tarpaulin surface suggests an empty or lightly loaded pallet; an irregular, bulging profile suggests cargo. Again, this is a probabilistic signal, not a deterministic one. Buyers should treat loaded-fraction estimates as indicative, with error margins that widen when sun elevation is above 60 degrees and shadows shorten.
Correlating counts with published freight data
IATA publishes monthly air freight tonne-kilometre statistics at the airline level, and ICAO publishes annual airport-level freight data. These are the natural validation anchors for apron-derived ULD counts.
The correlation is not direct. ULD inventory on an apron at a given moment reflects staging time, not just volume. A hub with fast turnarounds will show fewer ULDs at rest than a hub with longer dwell times, even at the same throughput. The useful signal is therefore the product of count and dwell duration, which requires at least two image acquisitions per day to estimate. WorldView-Legion's intra-day revisit makes this feasible for major cargo hubs. For airports with only one daily acquisition opportunity, analysts can use the ratio of loaded-to-empty ULDs as a proxy for utilisation intensity rather than absolute volume.
Published studies on container yard occupancy at seaports, which face analogous counting problems, show that object-detection models trained on very-high-resolution imagery achieve precision and recall rates above 90 percent for clearly separated objects under good illumination. Cargo aprons introduce additional complexity: ULDs are sometimes stacked two high in staging areas, and aircraft fuselages occlude portions of the apron. Both effects cause undercounting, and the degree of occlusion varies with the acquisition angle.
Detection limits and conditions that degrade the signal
Cloud cover is the primary operational constraint. Optical imagery is blind through cloud, and major cargo hubs in East Asia and South Asia face cloud fractions above 70 percent for extended periods. A programme relying solely on optical tasking will have gaps of days to weeks at affected airports during monsoon season. SAR can detect large metal objects on aprons but at the resolutions commercially available today, individual ULDs are at or below the detection threshold for most SAR systems. Fusion with SAR is therefore more useful for detecting gross apron activity than for counting equipment.
Apron clutter is the second constraint. Ground service equipment, aircraft stairs, fuel bowsers and baggage carts occupy the same spectral and size space as small ULDs. A well-trained object detector can distinguish a loaded LD3 from a baggage cart by aspect ratio and shadow profile, but errors increase in congested areas. Analysts should flag counts from heavily congested apron sections as lower-confidence.
Archive depth varies by sensor. Pléiades Neo launched in 2021; WorldView-Legion began full operations in 2023. Trend analysis stretching back more than three or four years requires stitching in data from earlier sensors such as WorldView-2 or WorldView-3, which have 31 cm and 31 cm panchromatic resolution respectively, at the cost of some consistency in the time series.
Turning counts into an analytic product
The practical output is a time-series index: ULD count per apron zone, sampled at whatever acquisition cadence the tasking budget allows, normalised for known occlusion zones and flagged for cloud-affected acquisitions. When stacked against published IATA monthly freight data, the index can be calibrated to produce a lead indicator, since apron inventory tends to build before reported volumes peak.
Satellize applies this counting methodology within its broader freight analytics framework, using open Sentinel-2 data for long-run context and commercial tasking for high-cadence counts at specific airports of interest. The Tonga crop-estimation programme established the organisation's object-counting pipeline on small, well-defined targets; the apron use case extends that approach to a noisier, more cluttered scene.
For clients who need a validated throughput signal rather than a raw count, the recommended workflow is: establish a baseline count during a period when IATA data is available for the same airport, fit a simple regression between count-times-dwell and reported tonnes, then apply that model forward. The model degrades if the airport changes its ULD fleet composition or apron layout, so periodic recalibration is necessary.
Typical figures
| Spatial resolution (primary sensors) | 30 cm panchromatic (WorldView-Legion, Pléiades Neo); 50 cm panchromatic (SkySat) |
| Multispectral resolution | 1.2 m (Pléiades Neo 4-band); 1 m (SkySat 4-band) |
| Intra-day revisit (WorldView-Legion) | Up to 15 acquisitions per day over a single target at mid-latitudes |
| Minimum detectable ULD footprint | Objects approximately 1.5 m × 1.5 m detectable at 30 cm GSD; reliable counting above ~2 m × 2 m |
| Spectral bands used | Panchromatic (detection and counting); blue, green, red, NIR (loaded/empty discrimination) |
| Cloud impact | Total optical blackout under cloud; no SAR substitute at individual ULD scale with current commercial SAR resolution |
| Archive depth | Pléiades Neo from 2021; WorldView-Legion full ops from 2023; WorldView-2/3 archive from 2009/2014 for longer baselines |
| Typical tasking latency | Same-day to next-day delivery for priority tasking on WorldView-Legion and Pléiades Neo |
| Delivery formats | GeoTIFF ortho imagery, GeoJSON object-count layer, CSV time-series index, PDF analytic report |
Analytics Satellize can run
| ULD and pallet count per apron zone | Object detection on sub-metre panchromatic imagery using convolutional neural network or template-matching; aspect ratio and shadow-length filters to reduce false positives from ground equipment | GeoJSON layer with per-object centroids, confidence score, and zone aggregate counts; updated per acquisition |
| Loaded-fraction estimate | Shadow-length-to-width ratio as height proxy, combined with NIR reflectance contrast between bare aluminium and tarpaulin; probabilistic classification, not deterministic | Per-zone loaded/empty ratio appended to count layer, with confidence interval based on sun elevation at acquisition time |
| Intra-day dwell and turnover index | Object persistence tracking across multiple same-day acquisitions (WorldView-Legion); ULDs present in image N but absent in image N+1 treated as departed | CSV time-series of dwell duration distribution per zone; requires minimum two acquisitions per day |
| Monthly throughput proxy index | Regression of count-times-estimated-dwell against published IATA or ICAO freight data for calibration period; forward application as lead indicator | Monthly index report with confidence bounds, flagged gaps from cloud cover, and comparison to latest available IATA published figures |
| Apron utilisation change alert | Statistical process control on rolling 30-day ULD count baseline; alert triggered when count exceeds two standard deviations from baseline | Email or API alert with annotated image chip and count deviation magnitude |
| Long-run trend baseline from Sentinel-2 | Spectral anomaly detection on 10 m NIR band over apron zones; not a ULD count but a gross utilisation signal for airports where sub-metre tasking is not yet active | Annual trend chart per airport, used to prioritise which sites warrant commercial tasking investment |
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.