Yard tractor density as container terminal throughput proxy
Internal terminal tractors are the connective tissue of any container port, yet they are invisible to AIS and crane sensors. Sub-50 cm optical imagery makes them countable, and multi-temporal stacks make their patterns readable.
Sensors
- Planet SkySat: 50 cm native resolution, up to 12 tasked collects per day over a single site; the sub-daily revisit is what enables shift-pattern analysis rather than just a snapshot count
- Airbus Pléiades Neo: 30 cm panchromatic resolution, 4-band multispectral at 1.2 m; constellation of four satellites allows same-day stereo or repeat tasking; colour separation helps distinguish tractor liveries and moving dust plumes from stationary units
- Maxar WorldView-3: 31 cm panchromatic, 16 SWIR bands; the SWIR bands add little for tractor detection but aid yard-surface classification; archive depth back to 2014 supports long-run utilisation benchmarking
- ICEYE SAR (X-band): Spotlight mode at approximately 25 cm azimuth resolution; cloud-independent and capable of night imaging; moving tractors produce Doppler-induced position offsets in SAR geometry, which is a nuisance for counting but a signal for detecting active operations during dark shifts
Why tractors, not cranes
Ship-to-shore cranes are easy to monitor from space. They are large, fixed, and their boom angle is legible in medium-resolution imagery. What they cannot tell you is whether the horizontal transport layer between quayside and stack is keeping pace. A terminal can have all cranes operational and still throttle throughput if its internal tractor fleet is under-deployed, in maintenance, or simply mis-routed across the yard.
Internal terminal tractors (ITTs) are typically 4 to 6 metres long and 2 to 3 metres wide. At 50 cm resolution they occupy roughly 8 to 12 pixels in a panchromatic image, which is sufficient for detection but not trivial. At 30 cm they are unambiguous objects with visible chassis geometry. The practical floor for reliable automated detection is around 40 cm GSD; below that, shadow length and contrast against concrete become the limiting factors rather than pixel count.
The moving-versus-stationary problem
A single optical pass captures one instant. A tractor parked for maintenance and a tractor momentarily stationary between jobs look identical. This is not a solvable problem within a single image. It is, however, manageable with multi-temporal stacks. If a unit appears in the same pixel location across three collects separated by 30 to 90 minutes, it is almost certainly not in service. If it has moved, it was active.
Planet SkySat's tasking flexibility, collecting up to 12 passes over a single site within a day, is specifically what makes this feasible at operational cadence rather than as a research exercise. Pléiades Neo's same-day stereo tasking offers a different angle: the stereo pair, collected minutes apart, can reveal motion blur or positional offset in moving units. Neither approach gives you real-time telemetry, but together they can characterise the active fraction of a fleet across a shift boundary.
SAR adds a different dimension at night. ICEYE X-band Spotlight imagery of a terminal during a night shift will show moving tractors displaced from their true positions in the along-track direction due to Doppler shift, typically by tens of metres at tractor speeds of 20 to 30 km/h. That displacement is analytically awkward for counting but it is a reliable indicator that the night shift is running, which optical imagery cannot confirm.
Reading shift patterns from a temporal stack
A week of daily SkySat tasking, even at one or two passes per day, begins to reveal structure. Tractor density at 06:00 local versus 14:00 local versus a weekend morning encodes shift schedules, peak-window management, and whether a terminal is running two or three shifts. That information is commercially sensitive precisely because it is not published anywhere.
The analytical method is straightforward in principle: object detection on each image, positional clustering across the temporal stack, and density mapping normalised to the active yard area. The honest caveat is that cloud cover over tropical or monsoon-affected ports can break the stack unpredictably. A port like Singapore operates under frequent convective cloud; a Baltic terminal in winter faces low sun angles that lengthen shadows and degrade contrast. Neither problem is fatal, but both require contingency tasking or gap-filling with SAR.
What the density map actually tells a buyer
Tractor density is a proxy, not a direct throughput measurement. The relationship between ITT count and TEU moves per hour is not fixed; it depends on terminal layout, automation level, and the mix of 20-foot and 40-foot boxes being handled. A fully automated terminal like Rotterdam's Maasvlakte II uses automated guided vehicles rather than manned ITTs, so this method does not apply there. The technique is most informative at semi-automated and conventional terminals, which still represent the majority of global port capacity.
What a density map reliably indicates: whether a terminal is in a high-activity window or a low-activity window at the moment of imaging; whether the spatial distribution of tractors is concentrated near one berth (suggesting a single large vessel discharge) or spread across the yard (suggesting multiple operations); and, over time, whether utilisation is trending up or down relative to a baseline period. For a port authority benchmarking a concession holder, or a shipper deciding between two competing terminals, that directional signal has real decision value.
Combining with adjacent layers
Tractor density is most informative when read alongside crane activity and yard occupancy, which are covered in separate pages in this library. The three layers address quayside productivity, horizontal transport, and stack utilisation respectively. A terminal showing high crane activity but low tractor density has a bottleneck in horizontal transport. High tractor density with low yard occupancy suggests containers are moving but not accumulating, which could mean rapid truck gate throughput or, less favourably, poor yard organisation causing unnecessary moves.
Satellize runs this kind of multi-layer analysis as part of its satellite-data analytics work, combining open and commercial imagery sources depending on the revisit and resolution requirements a client specifies. The crop-estimation programme in Tonga uses a comparable multi-temporal stacking approach over a very different target, which illustrates that the underlying method transfers across domains even when the objects being counted are quite different in character.
Honest limits before you commission tasking
Detection accuracy in published object-detection studies on very-high-resolution imagery typically reaches 85 to 95 percent for vehicles of this size under good conditions, with false positives concentrated near shadow boundaries and at the edges of container stacks where tractor-sized gaps appear. Expect lower performance in terminals with heavy overhead canopies, in imagery collected at sun elevation angles below 30 degrees, or where terminal operators use white or light-grey tractor liveries that blend with concrete.
Archive depth is a practical constraint. SkySat's commercial archive goes back to around 2016 but tasked collects at sub-daily frequency over a specific terminal are unlikely to exist unless someone commissioned them. Pléiades Neo launched in 2021. Maxar WorldView-3 has the deepest archive among very-high-resolution commercial sensors, from 2014, but historic collect frequency over any given terminal is unpredictable. For a retrospective study of a specific port, the first step is always an archive search before assuming usable data exists.
Typical figures
| Spatial resolution (optical) | 30 cm (Pléiades Neo panchromatic), 31 cm (WorldView-3 panchromatic), 50 cm (SkySat) |
| Spatial resolution (SAR night) | ~25 cm azimuth in ICEYE Spotlight mode; moving targets displaced by Doppler offset |
| Revisit (tasked optical) | Up to 12 SkySat passes per day over a single site; Pléiades Neo same-day stereo feasible |
| Minimum detectable target | Vehicles approximately 4 m × 2.5 m reliably detectable at ≤50 cm GSD; marginal at 50–80 cm |
| Spectral bands | Panchromatic + 4-band multispectral (blue, green, red, NIR) for optical; X-band (9.6 GHz) for ICEYE SAR |
| Temporal stack for shift analysis | Minimum 5–7 same-day or consecutive-day collects; 3–4 weeks for shift-pattern characterisation |
| Cloud sensitivity | Optical unusable under cloud cover; SAR unaffected; tropical and monsoon ports require SAR contingency |
| Archive depth | WorldView-3 from 2014; SkySat from ~2016; Pléiades Neo from 2021; historic density at any specific terminal unpredictable |
| Delivery latency | Tasked imagery typically available 2–6 hours after collect; analytics pipeline adds 4–24 hours depending on stack size |
| Deliverable formats | GeoTIFF density rasters, GeoJSON object detections, CSV time-series of tractor counts by zone, PDF shift-pattern report |
Analytics Satellize can run
| Point-in-time tractor count and spatial density map | Supervised object detection on very-high-resolution panchromatic imagery; convolutional neural network or template-matching approaches documented in remote sensing literature | GeoJSON layer of detected ITT positions with confidence scores, plus GeoTIFF density raster by terminal zone |
| Active versus stationary fleet fraction | Cross-temporal position differencing across multi-pass SkySat stack within a single day; units with zero displacement flagged as stationary | CSV table per collect showing total detected, stationary count, and active fraction; updated per tasking window |
| Shift-pattern utilisation index | Time-series clustering of tractor density across a 3–4 week stack; peak and trough windows identified relative to local time | Weekly PDF report with annotated density time-series and shift-window characterisation |
| Night-shift activity flag (SAR) | ICEYE Spotlight SAR Doppler offset detection; presence of displaced moving targets in terminal apron area indicates active night operations | Binary alert per SAR collect: night shift active or inactive, with annotated SAR image |
| Throughput trend index (multi-month) | Rolling 30-day average tractor density normalised to active yard area; trend decomposition separating seasonal from structural change | Monthly GIS layer and trend chart suitable for concession benchmarking or competitive intelligence |
| Horizontal transport bottleneck indicator | Spatial correlation of tractor density map with crane-activity zones; low tractor density adjacent to high-activity cranes flags a transport constraint | Annotated overhead image with zone-level bottleneck flags, delivered as GeoPackage and PDF |
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.