Military encampment growth and force-buildup detection
Optical time-series analysis can quantify tent lines, vehicle parks, and logistics depots as they grow, giving analysts days of warning before a force is ready to move. Revisit cadence is the binding constraint.
Sensors
- Planet Dove (PlanetScope): 3-metre native resolution, near-daily global revisit. The practical baseline for encampment change detection: frequent enough to catch overnight tent-line expansion, though individual vehicle identification requires confirmation from higher-resolution assets.
- Planet SkySat: 0.5-metre resolution, taskable on demand with same-day or next-day delivery. Resolves individual vehicle types, tent configurations, and field-hospital layout. Revisit is on-demand rather than systematic, so it supplements rather than replaces daily Dove coverage.
- Maxar WorldView-3: 0.31-metre panchromatic, 1.24-metre multispectral. Best available commercial resolution for structure classification and shadow-based height estimation. Revisit is tasking-dependent, typically one to a few days at mid-latitudes, cloud permitting.
- Landsat 8/9 TIRS: Thermal infrared bands 10 and 11 at 100-metre native resolution (resampled to 30 metres in products). Detects heat signatures from occupied camps at night: vehicle engines, generator exhausts, and concentrations of human activity produce measurable radiance differences against background terrain. Free, 16-day revisit limits tactical timeliness but supports pattern-of-life analysis.
What the pixel record actually captured in 2021-2022
The Russian force buildup on Ukraine's borders between late 2021 and February 2022 became the most publicly documented application of commercial satellite change detection in history. Analysts at the Institute for the Study of War, Bellingcat, and several commercial providers published annotated Planet and Maxar imagery showing battalion tactical group assembly areas, field hospitals, fuel depots, and bridging equipment appearing and expanding across sites in Yelnya, Voronezh, and Crimea over a period of weeks. The imagery was not classified. It was available to any subscriber.
That episode established a practical benchmark. A force buildup of brigade scale or larger, conducted over days rather than hours, is detectable from commercial optical imagery if daily revisit is available and the analyst knows what to look for. The question for any defence or intelligence buyer is not whether detection is possible but how quickly it can be converted into a structured, repeatable assessment.
What a shadow gives away
Shadow-based height estimation is the workhorse method for structure classification when imagery resolution is insufficient to identify an object directly. The principle is straightforward: if you know the solar elevation angle at the time of acquisition (recorded in every image's metadata) and you can measure the shadow length in pixels, you can calculate the object's height to within a metre or two at sub-metre resolution. A standard military general-purpose tent is roughly 2 to 3 metres at the ridge. A field hospital or command post is typically 4 to 6 metres. A prefabricated logistics shelter can exceed 8 metres. These ranges overlap, but combined with footprint area and the geometric regularity of military layouts, they allow probabilistic classification.
The method degrades at low solar elevation angles, common in high-latitude winter operations, where shadows become very long and measurement errors compound. It also fails when structures are under camouflage netting that distorts the shadow profile. Analysts should treat shadow-derived heights as supporting evidence rather than definitive identification.
Thermal infrared: occupied or abandoned?
An encampment that looks identical in two daytime optical images may have been vacated between acquisitions. Thermal infrared at night resolves the ambiguity in many cases. Landsat 8/9 TIRS captures emitted thermal radiance in two bands centred near 10.9 and 12 micrometres. A camp with active generators, idling vehicles, and hundreds of personnel produces a measurable thermal excess above background terrain, typically detectable even at TIRS's 100-metre native resolution when the concentration of heat sources is sufficient.
The honest limits are significant. TIRS revisit is 16 days, which is too slow for tactical warning. Atmospheric correction is required and introduces uncertainty. Smaller encampments, or those with strict emissions discipline, may not produce a signal above the noise floor. For higher-confidence thermal discrimination at shorter revisit, airborne or space-based mid-wave infrared sensors are more capable, but their data is generally not available on open commercial terms. TIRS is the publicly accessible floor, not the ceiling.
Revisit cadence is the binding constraint
A force can erect a battalion-scale assembly area in 24 to 48 hours. At Planet's Dove constellation revisit of approximately one day, the window between 'nothing there' and 'fully established camp' may be captured in a single image pair. At Landsat's 16-day revisit, the same buildup is invisible until it is long complete. This is why Planet's daily coverage is the practical baseline for encampment monitoring, not because 3-metre resolution is ideal but because frequency of observation is what produces warning.
Cloud cover is the secondary constraint and is not trivial. Persistent cloud over a region of interest can blank optical coverage for days at a time. Synthetic aperture radar can penetrate cloud and operates day and night, but SAR change detection for encampment growth is a different analytical problem covered separately in the SAR coherence-loss page in this library. Optical and SAR should be treated as complementary rather than interchangeable.
Building a repeatable monitoring cadence
Ad hoc analysis of a single image pair is how the 2021-2022 buildup was first reported publicly. It is not how a government intelligence function should operate. A repeatable encampment-monitoring programme requires a defined area of interest, a consistent image archive going back far enough to establish baseline land cover, a change-detection algorithm calibrated to the local terrain and vegetation phenology, and an alert threshold that distinguishes genuine new construction from agricultural activity, seasonal grazing, or civil construction.
Satellize structures this kind of programme for government clients, running change detection across open and commercial imagery and delivering structured alerts with confidence scores rather than raw imagery. The Tonga crop-estimation programme uses a similar time-series architecture, applied to a very different problem, which illustrates that the underlying analytical pipeline is domain-agnostic even when the application is not.
The output a defence client actually needs is not a folder of flagged images. It is a dated, georeferenced record of site extent, structure count, and vehicle-park area, updated on each cloud-free acquisition, with a trend line that shows rate of buildup. That record supports both current intelligence and historical analysis if a dispute arises later about when a force was assembled.
Typical figures
| Best available optical resolution | 0.31 m panchromatic (Maxar WorldView-3); 0.5 m (Planet SkySat); 3 m (Planet Dove) |
| Systematic daily revisit | ~1 day globally (Planet Dove); 16 days (Landsat 8/9); WorldView-3 and SkySat on-demand |
| Thermal infrared revisit | 16 days (Landsat 8/9 TIRS, bands 10-11, 100 m native resolution) |
| Shadow-height estimation accuracy | ±1-2 m at sub-metre resolution under favourable solar geometry; degrades at solar elevation below ~20° |
| Minimum detectable new construction | Approximately 10 m × 10 m footprint at 3 m resolution with multi-date differencing; individual vehicles at 0.5 m |
| Spectral bands used | Visible/NIR (RGB+NIR, Dove and SkySat); SWIR and panchromatic (WorldView-3); TIR 10.9 µm and 12 µm (TIRS) |
| Cloud limitation | Optical coverage blanked by persistent cloud; no mitigation without SAR integration |
| Archive depth | Planet Dove from ~2016; Landsat from 1972; WorldView-3 from 2014 |
| Typical alert latency | Same day to 24 hours from image acquisition, depending on tasking and processing pipeline |
| Delivery formats | GeoTIFF change layers, GeoJSON site polygons, PDF intelligence summaries, API alert feed |
Analytics Satellize can run
| Encampment footprint time-series | Multi-date image differencing and object-based image analysis on Planet Dove archive | GeoJSON polygon layer per acquisition date, with area in hectares and rate-of-change trend |
| Structure-type classification | Shadow-based height estimation combined with footprint geometry at SkySat or WorldView-3 resolution | Annotated GeoTIFF with structure class labels (tent, hard shelter, vehicle, logistics) and confidence scores |
| Vehicle-park count and area | Object detection on 0.5 m or better imagery using publicly documented template-matching and deep-learning methods | Tabular count by vehicle-park zone, updated per cloud-free acquisition |
| Thermal occupancy indicator | Landsat 8/9 TIRS nighttime radiance anomaly against multi-year baseline for site location | Binary occupied/unoccupied flag per 16-day acquisition with radiance excess value |
| Buildup rate alert | Statistical threshold on rate-of-change in footprint area across rolling 7-day window | Automated alert with site ID, coordinates, area delta, and source image date |
| Baseline land-cover reference | Pre-event Landsat and Sentinel-2 composite to distinguish new military construction from pre-existing features | Classified baseline raster with change mask overlaid on current acquisition |
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.