Sensors
- Maxar WorldView-3: 31 cm panchromatic, 1.24 m multispectral, 3.7 m SWIR across 8 bands (1195–2365 nm). Sub-daily revisit at mid-latitudes under tasking. The SWIR bands are the primary discriminator: real steel and aluminium produce spectrally flat or slightly rising SWIR returns, while rubber and canvas drop sharply in the 1600 nm and 2200 nm windows.
- Sentinel-2 MSI: 10 m visible and NIR, 20 m SWIR (bands 11 and 12 at 1610 nm and 2190 nm). Free, 5-day revisit at the equator (2–3 days at higher latitudes with both satellites). Spatial resolution is too coarse to resolve individual vehicles reliably, but useful for area-wide spectral anomaly screening before tasking commercial assets.
- Landsat 8/9 TIRS: 100 m thermal infrared (resampled to 30 m in products), two bands at 10.6 and 12.5 µm. 16-day revisit per satellite, 8-day combined. Thermal inertia analysis across pre-dawn and afternoon acquisitions is the classical method for separating high-mass metal objects from low-mass rubber or wood. Resolution limits mean it works at formation level, not individual vehicle level.
- Planet SuperDove: 3–5 m resolution across 8 spectral bands including red-edge and NIR. Near-daily revisit globally. Best used for change detection and shadow-geometry consistency checks rather than spectral material discrimination, given limited SWIR coverage. Useful for flagging new objects that warrant thermal or SWIR follow-up.
Why physics outperforms camouflage netting
Military deception is ancient. What changed in the satellite era is that concealment now has to fool not just the visible spectrum but thermal infrared, short-wave infrared and, separately, synthetic aperture radar. No single material does all of that convincingly at low cost. Inflatable rubber decoys, wooden mock-ups and radar reflector arrays are each optimised against one threat: the optical reconnaissance pass. They are not optimised against a diurnal thermal cycle observed from orbit.
The physical principle is thermal inertia. A real armoured vehicle, whether a T-72 or a self-propelled howitzer, is a mass of steel averaging several tonnes. Steel has a volumetric heat capacity roughly 3,000 times that of air. It absorbs solar energy through the day and releases it slowly overnight. At pre-dawn, a real vehicle is measurably warmer than the surrounding soil. A rubber inflatable, by contrast, has negligible thermal mass. It equilibrates to ambient temperature within an hour or two of sunset. That gap, typically 2–6 K depending on vehicle type, ambient conditions and soil moisture, is detectable in Landsat TIRS data and more sharply in airborne or commercial thermal sensors.
What a SWIR band gives away that the eye cannot
Thermal analysis is powerful but coarse in resolution. The complementary method is short-wave infrared spectral discrimination. Steel, aluminium alloy and military-grade paint each have characteristic SWIR reflectance profiles that differ from rubber, canvas and painted wood. WorldView-3's eight SWIR bands, covering 1195 nm to 2365 nm at 3.7 m resolution, are well-suited to this. Published analysis of the sensor's SWIR capability notes that material classes separable in these bands include metal oxides, hydrated minerals and synthetic polymers. Military rubber compounds fall into a spectrally distinct class from painted steel.
The practical limit is that SWIR discrimination requires good solar illumination and low atmospheric water vapour. Haze, thin cloud or high humidity degrades the signal. At 3.7 m, individual vehicles are resolved as a handful of pixels, which is enough for a spectral sample but not for shape analysis. The method therefore works best as a cross-check: optical shape analysis identifies candidate objects, SWIR sampling tests material class, thermal inertia tests mass. Agreement across all three raises confidence. Disagreement on any one warrants caution.
The shadow test and other geometric tells
Real vehicles have consistent three-dimensional geometry. An inflatable decoy, particularly one that is partially deflated or unevenly weighted, casts a shadow that does not match the expected silhouette of the object it represents. At WorldView-3's sub-metre panchromatic resolution, shadow length and direction can be compared against known sun-elevation angles at the acquisition time to compute an implied object height. A T-72 hull is approximately 2.2 m tall to the turret roof. An inflatable replica that has sagged by 30 cm produces a measurably shorter shadow.
This is not foolproof. A well-maintained inflatable under active management can hold correct geometry. The shadow test is most useful when decoy fields are large, hastily deployed or subject to wind. It also catches wooden mock-ups that lack the correct proportions. The method requires a single high-resolution optical collect and basic trigonometry; no specialised processing chain is needed beyond accurate acquisition metadata.
Historical precedent: Soviet and Iranian programmes
The analytical challenge is not hypothetical. During the 1991 Gulf War, post-conflict surveys found that coalition air strikes had destroyed a much smaller fraction of Iraqi armour than battle-damage assessments suggested, partly because decoys had been struck in preference to real vehicles. The Iraqi programme drew on Soviet doctrine that had been developing since the 1960s, including inflatable T-72 replicas and radar corner reflectors designed to mimic tank radar cross-sections.
More recently, open-source analysts examining Iranian military exercises have documented the use of inflatable aircraft and vehicle decoys at airfields and staging areas. These are visible in commercial optical imagery but require thermal or SWIR follow-up to discriminate from real assets. The published literature on this, including work by researchers using Landsat thermal archives, demonstrates that the diurnal temperature differential method is reproducible and does not require classified sensors. The honest caveat is that a sophisticated adversary aware of the method can deploy active thermal emitters inside decoys to spoof the signature. That countermeasure adds cost and logistics burden, which is itself an intelligence indicator.
Honest limits of the method
Cloud cover is the most immediate operational constraint. Optical and SWIR methods are entirely blocked by cloud. Thermal infrared penetrates thin cloud partially but degrades. SAR is the complement for denied-weather conditions, though SAR discrimination of decoys from real vehicles is a separate and harder problem: a radar corner reflector can be tuned to mimic the radar cross-section of a tank with reasonable fidelity.
Revisit rate matters enormously for thermal inertia analysis. Landsat's 8-day combined revisit means that in a fast-moving operational situation, the pre-dawn and afternoon thermal pair needed for inertia calculation may not be available on demand. Commercial thermal tasking from sensors such as those on some Planet or Satellogic platforms offers faster revisit but at lower radiometric precision than TIRS. WorldView-3 SWIR tasking is available on client licence but is a finite resource competed across many users. Analysts should treat the multi-sensor cross-referencing approach as a probabilistic confidence score, not a binary verdict. A single-sensor result, however suggestive, warrants corroboration before any consequential decision rests on it.
Satellize runs this multi-layer discrimination workflow on open and commercial data under client licence.
Typical figures
| Best optical resolution (WorldView-3 pan) | 0.31 m |
| SWIR resolution (WorldView-3) | 3.7 m across 8 bands, 1195–2365 nm |
| Thermal resolution (Landsat 8/9 TIRS) | 100 m native (30 m resampled); two bands at 10.6 µm and 12.5 µm |
| Revisit (Sentinel-2 combined) | 2–5 days depending on latitude |
| Revisit (Landsat 8+9 combined) | 8 days |
| Minimum detectable thermal differential | Approximately 0.1 K (TIRS NEdT); operationally meaningful vehicle-to-ground differentials typically 2–6 K at pre-dawn |
| Minimum resolvable object (WorldView-3 pan) | Individual vehicles resolved; shadow geometry measurable for objects taller than ~0.5 m |
| Archive depth | Landsat from 1972; Sentinel-2 from 2015; WorldView-3 from 2014 (tasked archive) |
| Cloud limitation | Optical and SWIR fully blocked; TIRS degraded under thin cloud; SAR required as complement |
| Delivery formats | GeoTIFF classification layers, GIS-ready shapefiles, confidence-scored object reports |
Analytics Satellize can run
| Diurnal thermal inertia classification | Paired pre-dawn and afternoon Landsat TIRS acquisition differencing; thermal inertia proxy computed per object cluster | GeoTIFF layer flagging high-inertia (probable real metal) vs low-inertia (probable decoy or empty) object zones, with confidence band |
| SWIR material-class sampling | Per-pixel spectral angle mapping against reference spectra for steel, rubber and painted wood using WorldView-3 SWIR bands | Object-level material-class report with spectral distance scores; delivered as attributed shapefile |
| Shadow-geometry height estimation | Sun-elevation trigonometry applied to WorldView-3 or Planet panchromatic imagery; implied height compared against known vehicle-type profiles | Tabular report of object implied heights vs expected heights for nominated vehicle types |
| Multi-sensor confidence scoring | Bayesian combination of thermal, SWIR and geometric test results into a single discrimination confidence score per object | Scored object list in GIS format; objects above threshold flagged for analyst review |
| Change detection for decoy emplacement events | Planet SuperDove daily difference imagery to detect new object appearance; triggers tasking of WorldView-3 or thermal follow-up | Automated alert with coordinates and first-appearance date; follow-up collect recommendation |
| Area-wide spectral anomaly screening | Sentinel-2 band 11/12 SWIR ratio analysis over broad areas to identify spectral outliers warranting high-resolution tasking | Priority tasking list with anomaly coordinates and SWIR ratio values |
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.