Prisoner of war and detention facility detection from imagery
Sub-metre commercial imagery can reveal newly built detention compounds through perimeter geometry, guard-tower shadows and barrack density. Timeline reconstruction from Planet and Maxar archives supports legal documentation and accountability investigations.
Sensors
- Planet SkySat: 50 cm native resolution (resampled to 50 cm product), tasked revisit of under 24 hours to a specific point. Sufficient to resolve individual fence posts, guard-tower platforms and vehicle types at a compound entrance.
- Maxar WorldView-3: 31 cm panchromatic, 1.24 m multispectral. The sharpest routinely available commercial optical sensor. Resolves double-fence spacing, shadow geometry of watchtowers and construction equipment. Archive extends to 2014.
- Airbus Pléiades Neo: 30 cm panchromatic, 1.2 m multispectral, stereo and tri-stereo tasking available. Stereo pairs allow height estimation of perimeter walls and towers to roughly 0.5 m vertical accuracy, useful for distinguishing a 3 m agricultural fence from a 5 m security perimeter.
- Sentinel-2 MSI: 10 m resolution, 5-day revisit at mid-latitudes. Too coarse to resolve fence lines, but useful for detecting the thermal and spectral signature of rapid ground clearing, new road cuts and large-scale construction activity that precedes facility construction. Free and openly archived from 2015.
What the fence line tells an analyst before anything else
Detention compounds have a distinctive spatial grammar. A single perimeter fence is standard for a factory or a school. A double fence with a cleared buffer strip of 3 to 10 metres between the two runs is a security feature with a specific purpose: it prevents physical contact with the outer fence from inside. At sub-metre resolution, that buffer strip is unambiguous. WorldView-3 at 31 cm resolves individual fence posts. Pléiades Neo stereo pairs allow the fence height to be estimated from shadow length and solar elevation angle, a published photogrammetric method requiring no ground truth.
Corner and mid-span guard towers cast diagnostic shadows. A watchtower platform elevated 4 to 6 metres above grade, typical of facilities documented in UNOSAT analyses of Xinjiang, produces a shadow of calculable length given the acquisition time and solar geometry. That shadow length is independent of image radiometry and survives JPEG compression. It is one of the most reliable geometric signatures in detention-facility analysis, and it cannot be spoofed by painting the roof a different colour.
Rapid construction as the first detectable signal
Most facilities of concern are not discovered fully built. The intelligence value lies in detecting construction early, before the compound is operational. Sentinel-2, despite its 10 m pixels, reliably detects large-scale ground clearing: bare soil has a distinct spectral signature in the shortwave infrared bands (SWIR, bands 11 and 12), and a sudden rectangular clearing in previously vegetated or undisturbed ground is anomalous. The 5-day revisit means that a clearing event can be dated to within a week. That timestamp anchors the subsequent sub-metre timeline.
Once a Sentinel-2 anomaly is flagged, commercial tasking on SkySat or WorldView-3 provides the structural detail. The sequence is economical: open data for triage, commercial data for confirmation. In the Xinjiang case, journalists and researchers at the Australian Strategic Policy Institute used exactly this layered approach, cross-referencing medium-resolution change signals with high-resolution commercial imagery to document facility expansion over time. The method is reproducible and the underlying imagery is independently verifiable.
Shelter density and the arithmetic of occupancy
High-density temporary shelter clusters are a documented indicator. Prefabricated barracks or modular units packed at spacings of 3 to 5 metres, with no vehicular access between rows, suggest occupant density inconsistent with voluntary accommodation. Published UNOSAT work on conflict-zone detention sites uses structure count and footprint area to estimate minimum occupancy ranges, acknowledging that actual density depends on unknown internal layout.
The honest limit here is significant. Imagery shows structures, not people. A compound that looks like a detention facility might be a quarantine site, a military barracks or a labour camp. The geometry is necessary evidence, not sufficient evidence. What imagery can establish is construction chronology, structural typology and the presence or absence of features, such as exercise yards, isolation blocks and vehicle inspection pits, that are consistent with published descriptions of detention infrastructure. The legal and evidentiary interpretation requires human analysts with context.
Timeline reconstruction for legal documentation
Archive depth matters as much as current resolution. WorldView-3 archive imagery is available from 2014, SkySat from approximately 2017, and Sentinel-2 from 2015. For a facility alleged to have been constructed during a specific period, analysts can construct a dated sequence of images showing bare ground, ground clearing, foundation work, perimeter erection and completed compound. Each image carries embedded metadata including acquisition timestamp, satellite position and solar geometry, all of which are independently verifiable.
This reconstruction capability has direct application in international humanitarian law investigations. The International Criminal Court and UN commissions of inquiry have accepted satellite imagery as documentary evidence in multiple proceedings. The standard required is not photographic certainty but corroborated consistency: does the imagery record match the alleged timeline? A well-constructed image sequence, with honest annotation of what is and is not visible, meets that standard. Gaps in cloud-free coverage must be acknowledged explicitly; a gap is not evidence of absence.
Cloud cover is the principal operational constraint in many conflict theatres. Optical sensors are blind through cloud. In persistently overcast regions, a 5-day Sentinel-2 revisit may yield only one or two usable acquisitions per month. SAR sensors such as Sentinel-1 can detect large-scale ground disturbance through cloud, but at 5 to 20 m resolution they cannot resolve fence geometry. The practical approach is to use SAR for continuity during cloud gaps and optical for structural characterisation when skies clear.
Road patterns, access control and what vehicles reveal
Controlled-access roads have a characteristic form: a single entry point, a vehicle inspection area with barriers visible as parallel lines across the road, and parking compounds separated from the main enclosure. At WorldView-3 resolution, vehicle types are distinguishable. Prison transport vans have a different aspect ratio and roof profile from civilian minibuses. Guard vehicles parked in regular patterns at shift-change intervals suggest an operational facility rather than a construction site.
Road surface condition is also informative. A newly constructed compound served by a heavily rutted unpaved road suggests high vehicle throughput inconsistent with a low-occupancy facility. Conversely, a large compound with minimal road wear may be newly built but not yet operational. Neither inference is conclusive, but both are documentable and datable.
How this analysis reaches decision-makers
Satellize structures this type of work as a monitored site programme: a defined set of coordinates under periodic tasking, with change alerts triggered by analyst-reviewed structural differences between acquisitions. The output is a dated image series with annotated overlays, a written assessment of observable indicators, and an explicit statement of what the imagery cannot determine. That last element is not a disclaimer; it is what distinguishes a credible intelligence product from advocacy.
For governments or legal teams requiring evidence packages, the deliverable includes acquisition metadata, sensor calibration references and a methodology note suitable for submission to an international body. The Satellize analytics team has experience structuring geospatial evidence for non-specialist audiences, drawing on the same methodological discipline applied to the Tonga crop-estimation programme, where the evidentiary standard for a government statistical release is comparably strict. Enquiries about a specific site or region of concern should include the coordinates, the alleged date range and the intended use of the output.
Typical figures
| Best available spatial resolution | 31 cm panchromatic (Maxar WorldView-3); 30 cm (Airbus Pléiades Neo); 50 cm (Planet SkySat) |
| Multispectral resolution | 1.24 m (WorldView-3); 1.2 m (Pléiades Neo); 10 m (Sentinel-2) |
| Tasked revisit cadence | Under 24 hours to a named point (SkySat); 1 to 3 days (WorldView-3, Pléiades Neo depending on off-nadir tasking) |
| Open-data revisit (Sentinel-2) | 5 days at mid-latitudes; cloud-free acquisition frequency varies by region and season |
| Archive depth | Sentinel-2 from 2015; WorldView-3 from 2014; SkySat from approximately 2017 |
| Minimum resolvable feature | Individual fence posts and guard-tower platforms at 31 to 50 cm resolution; double-fence buffer strip detectable at 50 cm if width exceeds approximately 2 m |
| Vertical accuracy (stereo) | Approximately 0.5 m RMS for wall and tower height from Pléiades Neo tri-stereo pairs |
| Cloud limitation | All optical sensors blind through cloud; SAR (Sentinel-1, 5 to 20 m) provides cloud-penetrating continuity for large-scale disturbance only |
| Delivery formats | Georeferenced GeoTIFF, annotated PDF report, KMZ/KML overlay, GIS-ready vector polygons, metadata package for legal submission |
Analytics Satellize can run
| Compound detection and structural typology classification | Object-based image analysis (OBIA) on sub-metre optical imagery; shadow-geometry photogrammetry for tower height estimation | Annotated site report with georeferenced perimeter polygon, feature inventory and structural typology assessment |
| Construction timeline reconstruction | Multi-date change detection across Sentinel-2 and commercial archive imagery; bare-soil SWIR spectral indexing for ground-clearing date estimation | Dated image series with annotated change overlays and a written chronology suitable for legal submission |
| Facility expansion monitoring | Periodic tasked acquisition on defined coordinates; automated footprint-area differencing between acquisitions; analyst review of structural changes | Monthly or on-demand change alert with before/after image pair and written assessment of new construction |
| Shelter density and occupancy estimation | Structure counting and footprint measurement from sub-metre imagery; density comparison against published UNOSAT facility typologies | Structure count table with estimated minimum occupancy range and explicit confidence bounds |
| Access road and vehicle-activity characterisation | Road morphology analysis; vehicle detection and classification at WorldView-3 resolution using published aspect-ratio signatures | Vehicle presence log with timestamps, road-condition assessment and access-control feature inventory |
| Evidentiary metadata package | Extraction and formatting of sensor acquisition metadata, solar geometry parameters and calibration references per international evidence standards | Structured metadata dossier with methodology note formatted for submission to UN bodies or international legal proceedings |
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.