Armoured vehicle dispersal and hide-site detection
Sub-metre optical imagery and X-band SAR backscatter anomalies can reveal armoured vehicle dispersal from garrison to hide positions, as demonstrated publicly during the 2022 Russian pre-invasion build-up. Each sensor method has hard limits that any honest analyst must state.
Sensors
- Planet SkySat: 50 cm pan-sharpened optical; tasked on demand; revisit depends on constellation geometry but can achieve same-day collection over a named area. Sufficient for vehicle-class detection (tank versus wheeled APC) in open ground; ambiguous under netting or dense canopy.
- Maxar WorldView-3: 31 cm panchromatic, 1.24 m eight-band multispectral. At 31 cm, individual vehicle type discrimination becomes feasible on open ground, though camouflage netting reduces confidence sharply. Off-nadir agility gives revisit of roughly one day over a priority target.
- Capella Space X-band SAR: Spotlight modes down to approximately 50 cm resolution; X-band (9.65 GHz) backscatter is strongly elevated by metal-rich objects such as armoured hulls and gun barrels, particularly against low-backscatter backgrounds like fields or shallow snow. Day/night, all-weather collection.
- ICEYE X-band SAR: Spotlight resolution approximately 50 cm; revisit over a given latitude band can be sub-daily across the constellation. SAR coherence between repeat passes degrades where vegetation or soil has been disturbed, flagging vehicle movement even when the vehicles themselves have departed.
- Sentinel-1 C-band SAR (ESA): Free, 6-day repeat at 10 m IW mode. Too coarse for individual vehicle detection but useful for wide-area screening: persistent bright point targets or coherence-loss patches over large forest edges can cue higher-resolution tasking. Archive extends to 2014.
What a metal hull does to radar
X-band SAR works because armoured vehicles are, in electromagnetic terms, extremely loud. Steel hulls, gun barrels, and track links produce corner-reflector-like returns that can be 10 to 20 dB brighter than surrounding bare soil at X-band frequencies. A T-72 parked in a harvested field is not hiding from a SAR satellite; it is advertising itself.
The physics changes under dense canopy. At X-band, the signal attenuates through a closed forest canopy, reducing the contrast between a vehicle return and background clutter. C-band penetrates slightly better but at coarser resolution. This is the fundamental trade-off: the same wavelength that makes X-band sensitive to metal on open ground makes it less reliable in heavy woodland. Analysts who do not state this limit are selling something other than analysis.
The 2022 pre-invasion build-up: what the public record shows
Between October 2021 and February 2022, Planet, Maxar, and several open-source analyst groups published imagery documenting Russian armoured and artillery units moving from permanent garrisons to field positions around Ukraine's borders. The methodology was straightforward: compare archived baseline imagery of known garrison sites against fresh tasked collections, identify new vehicle concentrations, and track the progression of those concentrations toward the border.
SkySat and WorldView-3 collections showed battalion tactical group (BTG) formations, fuel and ammunition logistics trains, and medical support units that together indicated offensive rather than exercise posture. The key analytical contribution was not any single image but the change signal across dozens of sites over weeks. Volume, direction of movement, and the presence of support elements that have no training-exercise rationale were the real indicators.
This public episode established a benchmark for what unclassified commercial imagery can achieve: detection and rough unit-type attribution at the formation level, honest uncertainty at the individual vehicle level, and near-real-time dissemination to policymakers and the press. It did not, and cannot, replace signals intelligence or human reporting. It is a complement, not a substitute.
Optical change detection: the method and its floor
The core optical method is pixel-level or object-level change detection between a clean baseline image and a current collection over the same area. At 50 cm (SkySat), a main battle tank occupies roughly 10 by 30 pixels, enough to detect presence and general orientation. At 31 cm (WorldView-3), turret shape and track width become distinguishable, allowing experienced analysts to separate, say, a T-80 from a BMP-2 on open ground.
Camouflage netting is a genuine problem. Thermal-IR bands on WorldView-3 (3.7 m resolution) can sometimes detect residual heat from engines through netting, but this requires recent engine activity and favourable thermal contrast, typically early morning in cool weather. Multispectral analysis using the near-infrared band can flag netting that uses non-photosynthetic material, since real vegetation reflects strongly in NIR while painted fabric does not. Neither technique is reliable enough to be stated as a firm detection capability; both are probabilistic indicators.
SAR coherence as a movement indicator
When a SAR satellite passes over the same ground twice, the phase relationship between the two passes (coherence) is high over stable surfaces like roads or buildings and low over areas that have changed, whether from vegetation growth, soil disturbance, or vehicle movement. A column of vehicles that has driven across a field and departed will leave a coherence-loss signature even after the vehicles themselves are gone.
At ICEYE or Capella spotlight resolution, this technique can detect the disturbance footprint of a vehicle concentration with a spatial precision of roughly one to two metres. The limit is temporal: coherence decorrelates naturally over days in vegetated terrain even without vehicle activity, so the technique is most reliable with short repeat intervals of 24 to 72 hours. Sentinel-1 at 10 m and 6-day repeat is too coarse and too slow for this application at the individual-site level, though it remains useful for wide-area cueing.
What the method cannot do
No current commercial optical or SAR system can reliably count individual vehicles under a closed deciduous canopy in full leaf. Hyperspectral sensors can flag disturbed soil at forest edges, and SAR can detect the general backscatter elevation from a metal-rich area, but neither provides a confident vehicle count. Analysts who report precise numbers from forested hide sites are extrapolating, not measuring.
Revisit is the other constraint. Even with aggressive commercial tasking, a 12-hour gap between collections is operationally significant: a BTG can move 200 km in that window. Satellite imagery characterises positions, not movements in real time. It is most useful for establishing pattern-of-life baselines, detecting departures from those baselines, and providing post-hoc documentation. For near-real-time tracking, it is a data layer in a broader intelligence picture, not the whole picture.
Satellize integrates commercial tasking from constellations including SkySat and commercial SAR providers with open Sentinel-1 wide-area screening, delivering change-detection alerts and annotated GIS layers. The analytical methodology is the same class used publicly during the 2022 build-up monitoring work.
Designing a monitoring programme that holds up
A credible armoured vehicle dispersal programme needs three components working together. First, a baseline: clean, cloud-free imagery of all known garrison and marshalling sites at a resolution of 50 cm or better, dated and archived before any period of interest. Second, a triggered tasking protocol: Sentinel-1 wide-area screening flags anomalous backscatter or coherence-loss patches, which then cue sub-metre optical or SAR spotlight collection within hours. Third, an analytical layer that fuses optical change detection with SAR backscatter intensity and, where available, thermal-IR engine-heat signatures.
The output of such a programme is not a definitive vehicle count. It is a probabilistic assessment: this site shows signatures consistent with the presence of 20 to 40 metal-bodied vehicles, confidence high; vehicle type is consistent with tracked armour, confidence moderate; camouflage netting is present, type discrimination is low confidence. That kind of structured, hedged reporting is what separates analysis from speculation, and it is the only kind that remains credible when ground truth eventually becomes available.
Typical figures
| Best optical resolution (commercial) | 31 cm (Maxar WorldView-3 panchromatic) |
| Best SAR resolution (commercial spotlight) | ~50 cm (Capella Space, ICEYE X-band spotlight modes) |
| Wide-area SAR screening resolution | 10 m (Sentinel-1 IW mode, free, 6-day repeat) |
| Revisit for tasked commercial optical | Sub-daily to 1 day (WorldView-3 off-nadir); 1 day (SkySat constellation) |
| SAR frequency (commercial) | X-band, ~9.6 GHz (Capella, ICEYE); C-band, 5.4 GHz (Sentinel-1) |
| Minimum detectable target (open ground, optical) | Vehicle-class detection at 50 cm; type discrimination at 31 cm, open ground only |
| Detection confidence under camouflage netting | Low to moderate; NIR and thermal-IR provide probabilistic indicators only |
| Archive depth | Sentinel-1 from 2014; WorldView archive from ~2009; SkySat from ~2017 |
| Latency (tasked commercial collection to delivery) | Typically 2 to 6 hours post-pass for priority tasking |
| Delivery formats | GeoTIFF, annotated GIS vector layers (GeoJSON/Shapefile), PDF intelligence summary |
Analytics Satellize can run
| Garrison baseline inventory | Object-based image analysis (OBIA) on sub-metre optical; manual QA by trained imagery analyst | GIS layer with vehicle-concentration polygons, confidence scores, and baseline date stamps |
| Wide-area dispersal screening alert | Sentinel-1 SAR backscatter intensity change and coherence-loss detection across a defined area of interest | Automated alert feed with candidate anomaly coordinates, triggering commercial tasking queue |
| Hide-site confirmation report | Sub-metre optical change detection fused with X-band SAR backscatter intensity; NIR netting-discrimination check | Annotated image report with structured confidence assessment per site (presence, vehicle class, camouflage state) |
| Movement trajectory reconstruction | Multi-date optical and SAR coherence-loss time series; track and disturbed-soil feature extraction | Temporal GIS layer showing probable route and timing of dispersal movement |
| Formation-size estimation | Pixel-counting and shadow-length analysis on high-resolution optical; SAR bright-target density mapping | Probabilistic vehicle-count range per site with stated uncertainty bounds |
| Pattern-of-life baseline and deviation alert | Statistical change detection across rolling 30-day optical and SAR archive for named garrison sites | Weekly monitoring digest with flagged deviations from established activity norms |
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.