Rail network military freight and rolling-stock movement monitoring

Concentrations of military rolling stock at rail yards are visible days before a force moves. High-revisit optical and SAR imagery, cross-referenced against track-usage signatures, turns that signal into structured intelligence.

Sensors

Planet Dove constellation: 3–5 m resolution optical, daily global revisit. Sufficient to count and classify rolling-stock types (flatcar, boxcar, tank transporter) and detect yard density changes overnight. Cloud cover remains a persistent gap.
Maxar WorldView-3: 0.31 m panchromatic, 1.24 m multispectral. Resolves individual vehicle silhouettes on flatcars and distinguishes covered wagons from open ones. Tasked on demand; revisit at a given point is typically 1–4 days depending on orbital geometry and tasking priority.
Sentinel-1 SAR (C-band, ESA): Interferometric Wide Swath mode at 10 m resolution, 6-day repeat at mid-latitudes (12-day single-satellite). Cloud- and darkness-independent. Coherence differencing detects ballast disturbance and track-surface change caused by heavy repeated loading, even when rolling stock has moved on.
Airbus Pléiades: 0.5 m optical, tasked commercially. Useful for confirmation passes when Planet flags a density spike. Stereo acquisition available, which allows rudimentary height estimation of stacked or covered loads.

What a marshalling yard reveals before the order is given

Rail is the dominant mode for moving heavy armour over distances beyond a few hundred kilometres. A main battle tank weighs 45–65 tonnes; road transport wears out both the vehicle and the road, and is slow. Rail concentrates that mass efficiently, which is precisely why marshalling yards become the earliest legible signal of a large-scale military movement.

The methodology was applied publicly and in near-real time during late 2021, when open-source analysts and commercial imagery providers documented the accumulation of Russian military rolling stock at yards near Yelnya, Valuyki and other points along rail lines leading toward Ukraine. Planet Dove imagery, published openly, showed flatcar density rising over successive daily passes. The physical logic is simple: military flatcars have a distinctive aspect ratio and load profile, and when dozens appear at a yard that normally handles grain or timber, the anomaly is statistically obvious even before any vehicle on them is individually identified.

The physical indicators and how sensors read them

Three distinct signals compound each other. First, rolling-stock density: the count of flatcars, covered wagons and specialised tank transporters per unit yard area, measured from optical imagery. Planet's daily revisit makes it possible to track this count day-on-day; WorldView-3's sub-metre resolution makes it possible to confirm vehicle type on the flatcar deck.

Second, track-usage shadows. Repeated heavy axle loads compress and laterally displace ballast, creating a slightly darker, compacted appearance in high-resolution optical imagery. This is subtle but consistent, and it persists after the rolling stock departs, providing a retrospective record of which sidings carried heavy traffic.

Third, SAR coherence loss. Sentinel-1 interferometric coherence drops when the surface between two passes has changed. Fresh ballast disturbance, tamped track, or the simple presence and then absence of large metallic objects (wagons) all reduce coherence. A coherence-difference map between two Sentinel-1 passes, six days apart, highlights active sidings even through cloud cover and at night. This is the method's primary advantage in winter or overcast conditions, when optical imagery may be unavailable for days at a stretch.

Classification: what can and cannot be resolved

At Planet's 3–5 m resolution, individual vehicle type on a flatcar is generally not resolvable. What is resolvable is the flatcar itself, its length class, and whether it is loaded or empty. Loaded flatcars carrying armoured vehicles show a characteristic centred rectangular mass; tank transporters have a distinctive well-deck profile visible at 0.5–1 m resolution from Pléiades or WorldView-3.

Covered wagons, used to conceal wheeled vehicles or ammunition, are a genuine ambiguity. Their presence in unusual numbers at a military-adjacent yard is itself a signal, but the contents cannot be determined from optical imagery alone. Thermal sensors add little here because covered wagons are not meaningfully warmer than their surroundings at rest.

The hard limit is covered or tunnelled infrastructure. Rail lines that pass through tunnels, or yards where rolling stock is moved into hardened shelters, are opaque to all optical sensors and to SAR surface-coherence methods. Underground rail networks, such as those documented at certain hardened Soviet-era facilities, are entirely blind to this methodology. Analysts must note the entry and exit points of tunnels and treat the interval as a gap.

Cadence, latency and the intelligence window

The practical intelligence value of this method depends on how quickly imagery can be processed and delivered after collection. Planet's daily global mosaic is typically available within hours of acquisition. A change-detection alert on a specific yard, comparing yesterday's count against a rolling baseline, can be generated automatically and delivered the same day.

SAR coherence products from Sentinel-1 require two passes separated by six or twelve days, which means they are better suited to confirming a pattern than to detecting a sudden overnight change. The combination of daily optical alerts with periodic SAR coherence confirmation is therefore more useful than either sensor alone.

Archive depth matters too. Planet's optical archive extends back to 2016 for many areas; Sentinel-1 SAR data is available from 2014. Establishing a baseline of normal yard activity, seasonal variation in commercial freight, and typical rolling-stock counts before an anomaly is assessed is not optional. Without it, a busy harvest season at a grain-handling yard looks indistinguishable from military pre-positioning.

Integrating the picture: from yard counts to movement estimates

A single yard count is a data point. A sequence of counts across a network of yards, linked by known rail lines, becomes a movement estimate. If flatcar density rises at Yard A, falls three days later, and then rises at Yard B along the same line, the inference is a transit, not a static build-up. This network-level reading requires systematic monitoring of multiple nodes simultaneously, which is where automated change detection across a constellation like Planet's earns its keep.

Cross-referencing with open-source rail timetable data, known line capacities, and published military logistics doctrine (Russian military rail doctrine, for instance, is extensively documented in open academic and defence literature) allows an analyst to estimate throughput: how many vehicles per day a given line can move, and therefore how long a build-up of observed scale would take to deploy.

The analytical framework is not proprietary; the discipline is in applying it consistently, at scale, and without gaps in the monitoring calendar.

Honest limits and what they mean for a buyer

Cloud cover is the most operationally significant constraint. Persistent cloud over a yard for five or more consecutive days can blind the optical layer entirely during a critical window. SAR partially compensates, but coherence methods detect surface change, not rolling-stock presence in real time. A buyer relying on this methodology in a high-cloud environment, such as the Russian Far East in autumn or Central Europe in winter, should plan for optical gaps and weight the SAR layer accordingly.

Deception is a real adversarial risk. Dummy flatcars, decoy loads, or deliberate use of commercial freight to mask military movements are all documented historical practices. No remote-sensing methodology can reliably distinguish a genuine tank transporter from a well-constructed decoy at Planet resolution; WorldView-3 reduces but does not eliminate that ambiguity. The methodology is strongest as an early-warning and pattern-of-life tool, not as a definitive order-of-battle count.

Finally, rail network coverage in imagery archives is uneven. Yards in densely monitored regions such as Europe are well-covered historically; yards in less-tasked regions may have sparse baseline data, making anomaly detection less reliable. A programme design should account for this by establishing baseline collection before a monitoring requirement becomes urgent.

Typical figures

Best optical resolution (commercial)	0.31 m panchromatic (WorldView-3); 0.5 m (Pléiades)
Routine optical resolution (daily revisit)	3–5 m (Planet Dove)
SAR resolution	10 m (Sentinel-1 IW mode)
Optical revisit at a fixed point	Daily (Planet); 1–4 days tasked (WorldView-3, Pléiades)
SAR revisit at a fixed point	6 days (Sentinel-1 two-satellite pair, mid-latitudes); 12 days single satellite
Minimum detectable rolling-stock unit	Individual flatcar at 3–5 m; vehicle type on flatcar at 0.5–1 m
Coherence-change detection sensitivity	Ballast disturbance and large metallic object presence/absence detectable at 10 m SAR
Archive depth	Planet optical from 2016; Sentinel-1 SAR from 2014
Latency (optical alert)	Same-day to 24 hours after acquisition (Planet daily mosaic)
Cloud penetration	SAR only; optical entirely cloud-blocked during persistent overcast

Analytics Satellize can run

Yard density change alert	Automated object-count change detection on daily Planet mosaic, compared against rolling 30-day baseline	Daily alert feed with flatcar-count delta per monitored yard, delivered as GeoJSON or structured report
Rolling-stock classification map	Visual and semi-automated classification of flatcar, tank transporter and covered-wagon types from sub-metre tasked imagery (WorldView-3 or Pléiades)	Annotated imagery tile with vehicle-type labels and confidence rating, delivered as GIS layer
SAR coherence-change product	Sentinel-1 interferometric coherence differencing between successive 6- or 12-day passes over monitored yard network	Coherence-difference raster highlighting active sidings and ballast disturbance, delivered as GeoTIFF with analyst annotation
Network-level movement estimate	Multi-node time-series analysis linking yard-count sequences along known rail lines, cross-referenced with published line-capacity data	Weekly intelligence summary estimating transit direction, volume and probable destination, delivered as structured PDF or secure data feed
Baseline pattern-of-life report	Statistical characterisation of normal yard activity from archive imagery, establishing seasonal and commercial freight norms before anomaly thresholds are set	One-time baseline report per yard or yard network, with recommended alert thresholds
Covered-wagon anomaly flag	Detection of unusual covered-wagon concentrations at military-adjacent yards using optical count and yard-type context	Flagged imagery with analyst note on ambiguity and recommended confirmation tasking

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.