Airport aircraft counting and stand utilisation monitoring

Very-high-resolution optical imagery at 0.3–0.5 m lets analysts count aircraft, infer type from geometry, and convert stand occupancy into throughput proxies. Published studies using WorldView and Pléiades data show the method works; the hard constraint is that no tasking system can guarantee overpass timing against a morning banking wave.

Sensors

Maxar WorldView-3: 0.31 m panchromatic, 1.24 m multispectral, 8 SWIR bands. Single-pass revisit of roughly 1 day at mid-latitudes under tasking. The panchromatic channel resolves engine nacelles and fuselage length well enough to separate narrow-body from wide-body types.
Airbus Pléiades Neo: 0.30 m panchromatic, 0.75 m multispectral. Stereo and tri-stereo collection in a single pass enables shadow-derived height estimates, useful for distinguishing double-deck aircraft such as the A380 from standard wide-bodies. Constellation of two satellites gives daily revisit potential over most airports.
Planet SkySat: 0.50 m panchromatic, 1.0 m multispectral. Up to 12 tasked passes per day over a target at low latitudes, though per-pass cloud risk accumulates. Useful for intra-day stand-turnover monitoring where timing of the overpass can be loosely controlled.
Airbus SPOT 7: 1.5 m panchromatic. Insufficient to resolve nacelle count reliably, but adequate for gross aircraft presence or absence on large remote stands. Useful as a low-cost archive source for historical occupancy baselines going back to 2014.

What a parked aircraft gives away from 500 km up

At 0.31 m ground sample distance, a Boeing 737-800 (fuselage length 39.5 m, wingspan 35.8 m) occupies roughly 125 by 115 pixels in a WorldView-3 panchromatic image. That is more than enough to measure fuselage length and wingspan from shadow geometry, count engine nacelles, and place the aircraft into a broad type class: single-aisle narrow-body, twin-aisle wide-body, or freighter with distinctive nose or tail door geometry. Published work in the remote-sensing literature using Pléiades and WorldView imagery has demonstrated type classification at accuracy rates above 85 percent for the major commercial categories under clear-sky conditions.

Shadow length is particularly informative. Given the solar elevation angle at acquisition time, a shadow cast by a vertical tail fin encodes the fin height, which differs measurably between aircraft families. A 737 fin stands roughly 12.6 m; a 777-300ER fin reaches 18.5 m. The ratio between fuselage-length measurement and shadow-derived fin height provides a second discriminating feature independent of image resolution noise.

Converting stand occupancy into an economic signal

An airport stand occupied by a wide-body aircraft for three hours represents, at minimum, several hundred departing passengers and a corresponding cargo belly load. Aggregate stand occupancy across a terminal, measured across a time series of images, correlates with published passenger throughput figures in a way that can be calibrated against public airport statistics and then used to estimate activity in periods when official data is delayed or absent.

The proxy is not precise. A stand occupied by a parked-out-of-service aircraft contributes nothing to throughput. Remote cargo stands turn over differently from passenger gates. Seasonal charter patterns create spikes that look like structural growth. These ambiguities are real, and any occupancy-based throughput model needs ground-truth calibration against at least one period of known traffic before it can be trusted for inference. The signal is most reliable when used comparatively: the same airport across time, or a set of airports against each other under a consistent methodology.

The timing problem nobody can fully solve

Morning banking waves at hub airports, typically between 06:00 and 09:00 local time, are when stand utilisation peaks. Afternoon banking waves follow a similar pattern. A satellite image captured at 11:30 local time, when stands are partially cleared, will systematically undercount activity relative to peak. No current commercial tasking system, including Pléiades Neo's daily revisit or SkySat's multi-pass capability, can guarantee overpass timing within a one-hour window on a specified date. Cloud cover compounds the problem: a 30 percent cloud probability per pass means that capturing a specific banking wave on a specific day is genuinely uncertain.

The practical response is to build a statistical model from many acquisitions spread across different times of day and days of week, then weight observations by their acquisition time relative to published schedule data. This smooths the timing bias over a large sample. For a single-event assessment, such as measuring activity at an airport on a specific date for regulatory or competitive intelligence purposes, the timing constraint should be stated explicitly in any deliverable. Honest analysis acknowledges this; analysis that does not is misleading.

Automated detection pipelines and their published limits

Object detection on airport imagery has been a benchmark task in the remote-sensing machine-learning community for over a decade. The DOTA dataset (Dataset for Object deTection in Aerial images), published by Wuhan University and widely cited, includes aircraft as a target class in very-high-resolution imagery and has been used to benchmark dozens of detection architectures. Reported mean average precision figures for aircraft detection on DOTA reach above 0.80 for modern oriented bounding-box detectors, but these figures are measured on imagery collected under favourable conditions and do not transfer directly to operational tasking scenarios with variable sun angle, partial cloud shadow, or aircraft under jetway bridges.

Minimum detectable target size is a function of both resolution and contrast. At 0.5 m, a regional turboprop with a 27 m fuselage is detectable but classification confidence drops. At 0.31 m, the same aircraft is classifiable with reasonable confidence. Aircraft parked in shade, partially occluded by terminal structures, or towed without engines are systematically missed or misclassified by automated pipelines. Human-in-the-loop review remains standard practice for any high-stakes count.

Building a time-series occupancy index

A single image is an anecdote. A time series is evidence. For airports with published monthly passenger statistics, a multi-year archive of WorldView or Pléiades imagery can be used to build a regression between observed stand occupancy at known acquisition times and reported throughput. Once calibrated, the model produces occupancy-based throughput estimates for months where official data is not yet published, providing a lead indicator of roughly four to eight weeks depending on the airport's reporting lag.

Satellize applies this class of time-series occupancy analysis across several transport and logistics contexts. The underlying methodology, pairing high-cadence optical archive with published ground-truth statistics for calibration, is the same approach used in the Tonga crop-estimation programme, where satellite-derived canopy indices were calibrated against ground survey data before being used for inference in unsurveyed periods.

The archive depth available from WorldView satellites extends to 2007 for some airports, and Pléiades archive coverage begins around 2012. This gives a genuine multi-cycle baseline for airports in commercially active tasking zones, though archive density varies significantly by location and historical tasking demand.

What this analysis cannot replace

Stand utilisation from satellite imagery is a supply-side observation: it tells you what was physically present on the apron at the moment of acquisition. It does not tell you flight origin, passenger load factor, revenue per available seat kilometre, or whether an aircraft was operating commercially or in maintenance. Those data come from ADS-B feeds, airline schedules and ground handling records, none of which are captured in the image. The satellite layer is most powerful when fused with schedule data and ADS-B history, not when used in isolation.

Cargo freighters present a specific ambiguity. A 747-400F parked for six hours may be loading, unloading, awaiting a slot, or undergoing a line check. Dwell time from a time series of images distinguishes these cases only coarsely. For detailed cargo throughput analysis, the optical stand-occupancy method is a first filter, not a final answer.

Typical figures

Best available spatial resolution	0.30 m panchromatic (Pléiades Neo, WorldView-3)
Minimum classifiable aircraft length	Approximately 27–30 m fuselage at 0.31 m GSD (regional turboprops marginal; narrow-bodies reliable)
Revisit under tasking	Daily (Pléiades Neo 2-satellite constellation); up to 12 passes/day (SkySat); ~1 day (WorldView-3 at mid-latitudes)
Spectral bands	Panchromatic + 4-band multispectral standard; 8-band VNIR and 8-band SWIR available on WorldView-3
Cloud risk per pass	Varies by location and season; 20–50% cloud probability per individual pass is common at many hub airports
Tasking latency (image to delivery)	Typically 24–48 hours from acquisition for standard commercial tasking; priority pipelines can reach 4–6 hours
Archive depth	WorldView from 2007; Pléiades from 2012; coverage density varies strongly by commercial tasking history of location
Timing guarantee against banking waves	None. Overpass time is constrained by orbital geometry and cloud; peak-hour capture is probabilistic, not assured
Typical area per image chip	WorldView-3 single strip: up to 13.1 km swath width; most major airports fit within a single collect

Analytics Satellize can run

Aircraft count per stand category	Oriented bounding-box object detection (CNN-based, trained on published benchmark datasets including DOTA); human QA review on ambiguous detections	GIS layer (GeoJSON or shapefile) with per-stand occupancy flag and aircraft bounding box, delivered per image acquisition
Aircraft type classification	Geometric feature extraction: fuselage length, wingspan, shadow-derived tail height, nacelle count; rule-based classifier mapped to ICAO type categories	Tabular report with aircraft type, confidence band and stand ID per acquisition; flagged where classification is ambiguous
Stand occupancy rate time series	Aggregation of per-image counts across a tasked archive; occupancy rate expressed as fraction of total stands occupied at acquisition time, normalised by acquisition hour	Monthly occupancy index chart and underlying CSV, with acquisition-time metadata for each observation
Throughput proxy index	Regression of time-normalised occupancy against published IATA or airport authority passenger statistics for calibration period; out-of-sample inference for subsequent periods	Lead-indicator throughput estimate with stated confidence interval, updated on each new acquisition; methodology note included
Anomaly alert: unusual stand configuration	Statistical process control on rolling occupancy baseline; alert triggered when observed count deviates from expected range by more than two standard deviations	Email or API alert with annotated image chip and deviation metric; suitable for operational monitoring of competitor or partner hub airports
Historical baseline report	Archive mining of available WorldView and Pléiades collections over a defined airport and date range; manual and automated count reconciliation	PDF report with annual occupancy trends, type-mix evolution and timing-bias caveats; suitable for regulatory submission or investment due diligence

Sources

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.