Military training exercise tempo and scale monitoring
High-revisit optical and SAR imagery can quantify how often, how large, and where military forces train. Separating a rehearsal from a real deployment requires honest reading of observable indicators, not inference of intent.
Sensors
- Planet Dove constellation: 3–5 m resolution, daily global revisit at nadir. Ideal for tracking apron occupancy, vehicle track formation and range utilisation across time-series stacks. The high cadence is the primary asset; the resolution is sufficient to count parked aircraft and identify track patterns but not to identify individual vehicle types.
- Maxar WorldView-2 / WorldView-3: 30–46 cm panchromatic resolution on tasking. Used for spot confirmation: verifying vehicle type, counting individual aircraft on aprons, and reading impact-crater density on range surfaces. Revisit is 1–4.5 days depending on latitude and collection angle, but tasking competes with other customers.
- Sentinel-1 SAR (C-band): Interferometric Wide mode at 10 m resolution, 6-day repeat at mid-latitudes (12-day with a single satellite). Cloud-independent, which matters for coastal or tropical training areas. Coherence-change detection reveals disturbed ground from vehicle movement even when optical imagery is obscured. Does not resolve individual vehicles but detects area-scale disturbance reliably.
- Airbus Pléiades / Pléiades Neo: Pléiades Neo reaches 30 cm resolution with same-day or next-day tasking possible in many regions. Useful for confirming exercise scale at a specific moment, particularly for counting logistics vehicles or assessing field-camp footprints. Cost per task is higher than Dove, so it functions as a confirmation layer rather than a monitoring backbone.
What a training range actually records
Military training exercises are, physically, a problem of ground disturbance and asset concentration. Armoured vehicles leave track signatures in soil that persist for days to weeks depending on rainfall and vegetation. Artillery and air-delivered munitions create impact craters whose density and pattern are measurable from sub-metre imagery. Aircraft, helicopters and support vehicles concentrate on aprons and dispersal areas in ways that differ from peacetime garrison posture. Each of these is a signal that high-revisit optical imagery can track systematically.
The key analytical unit is change over a baseline. A single image of an airbase tells you little. A time-series of 90 daily Planet Dove acquisitions over the same site tells you the normal apron occupancy, the normal vehicle presence on adjacent ranges, and how sharply the current scene departs from that norm. Departure magnitude and duration are the two numbers that matter most.
The exercise-versus-deployment problem
This is the central difficulty, and any honest account of the method must say so plainly. Exercises and genuine operational deployments share many observable signatures: increased aircraft presence, logistics vehicle concentration, range activity, and communications infrastructure deployment. Imagery alone cannot read intent. What it can do is characterise observable indicators that, taken together, raise or lower the probability of each interpretation.
Published open-source intelligence work on Chinese People's Liberation Army exercises around Taiwan, notably the large-scale drills of August 2022, demonstrated the approach in practice. Analysts using commercial Planet imagery tracked amphibious vessel movements, airfield sortie rates and ground-force concentration in real time, and compared the observable footprint against prior exercises to assess whether the pattern was consistent with a rehearsal or a preparatory deployment posture. The conclusion was probabilistic, not definitive. That is the honest limit of the method.
Indicators that tend to favour an exercise interpretation include: pre-announced political context, return of forces to garrison within days, use of known permanent training ranges, and absence of forward logistics stockpiling. Indicators that complicate that reading include: exercises that begin without announcement, forces that do not return on the expected schedule, and concurrent activity at ports or airfields not associated with the named exercise area. Imagery provides the raw observation; the analyst provides the interpretation, and should say so.
Building a tempo baseline
Tempo analysis requires archive depth. Planet's Dove archive extends to 2016 in many regions, giving roughly eight years of daily imagery from which to construct a seasonal baseline of exercise frequency. Sentinel-1's archive runs from 2014. For a given training area, the analyst can compute monthly mean apron occupancy, mean track-disturbed area on ranges, and mean coherence-change extent, then express current observations as a deviation from that multi-year mean.
Frequency matters as much as scale. A force that trains at a particular range four times a year and then shifts to twelve times a year has changed its tempo even if individual exercises look similar in size. That shift is detectable from a Planet time-series. Scale, measured as geographic extent of disturbed ground or number of aircraft present, is a separate dimension. An increase in both simultaneously is a stronger signal than either alone.
Resolution floors and what they hide
Planet Dove at 3–5 m resolves parked aircraft as distinct objects and identifies vehicle track networks, but cannot reliably distinguish a main battle tank from a wheeled armoured personnel carrier. That distinction matters for force composition analysis. Maxar WorldView-3 at 30–46 cm can make that call in good lighting conditions, but tasking is not guaranteed and cloud cover remains a constraint over many exercise areas in East and South-East Asia.
Sentinel-1 SAR removes the cloud problem but introduces its own ambiguity. C-band backscatter from a disturbed range surface can resemble backscatter from other rough or wet ground. Coherence-change detection is more reliable than single-image backscatter for ground-disturbance attribution, but it requires two acquisitions 6 or 12 days apart, which means a fast-moving exercise that begins and ends within that window may leave only a partial signature. Night exercises are invisible to optical sensors entirely; SAR captures them but at coarser resolution than daytime optical confirmation.
Structuring a monitoring programme
A practical monitoring programme for exercise tempo combines three layers. The first is a daily Planet Dove feed over a defined set of sites, processed for apron occupancy counts and range-disturbance area using change-detection algorithms. The second is triggered Sentinel-1 coherence analysis when the optical layer shows a significant departure from baseline, providing cloud-independent confirmation and ground-disturbance extent. The third is on-demand Pléiades or WorldView tasking for specific sites when the first two layers indicate an event of sufficient interest to justify the cost.
Satellize structures analytics programmes of this type for government clients, drawing on open-constellation data and adding commercial tasking on client licence. For a client requiring systematic, non-public monitoring of specific regions, the architecture is the same; the tasking cadence and reporting format are agreed at programme outset.
What the imagery cannot settle
Force readiness is not visible from orbit. An apron full of aircraft tells you aircraft are present; it does not tell you crew hours, maintenance status, or weapons load. Vehicle tracks on a range confirm movement; they do not confirm the tactical proficiency of the force that made them. Exercise imagery, even at sub-metre resolution, is evidence of activity, not capability.
Intent is similarly opaque. The August 2022 PLA exercises around Taiwan were read by different analysts as a rehearsal, a coercive signal, and a capability demonstration simultaneously, and all three interpretations were consistent with the observable imagery. The value of systematic monitoring is not that it resolves that ambiguity but that it provides a factual, timestamped record of what was observable, which is a necessary input to any serious assessment. Claiming more than that is the error to avoid.
Typical figures
| Best optical resolution (commercial tasking) | 30 cm (Maxar WorldView-3 panchromatic) |
| Daily monitoring resolution (Dove) | 3–5 m |
| SAR resolution (Sentinel-1 IW mode) | 10 m ground range |
| Optical revisit (Planet Dove) | Daily at nadir globally |
| SAR revisit (Sentinel-1, two-satellite) | 6 days at mid-latitudes |
| Minimum detectable aircraft on apron | Single aircraft at 30–46 cm; clusters of 2–3 at 3–5 m |
| Archive depth | Planet Dove from 2016; Sentinel-1 from 2014; Maxar archive from 2001 |
| Cloud limitation | Optical sensors fully obscured by cloud; SAR unaffected |
| Latency (Planet daily collect to analysis) | Typically 12–24 hours after acquisition |
| Delivery formats | GeoTIFF change layers, GeoJSON feature files, PDF/HTML site reports, time-series CSV |
Analytics Satellize can run
| Apron occupancy time-series | Object detection and counting on Planet Dove daily stack using supervised classification; validated against WorldView spot checks | Weekly CSV and GIS layer showing aircraft count per apron per day, with baseline deviation score |
| Range disturbance area mapping | Normalised Difference Vegetation Index change and bare-soil exposure index applied to Dove multispectral time-series; Sentinel-1 coherence-change as cloud-independent confirmation | Monthly GeoTIFF showing cumulative disturbed area per range polygon, with trend chart |
| Vehicle track network extraction | Line-feature extraction on sub-metre tasked imagery (Pléiades Neo or WorldView-3) using edge-detection and morphological filtering | GeoJSON track-network layer with date stamp and estimated vehicle-pass density |
| Exercise frequency and tempo index | Event segmentation on apron-occupancy and range-disturbance time-series; comparison against multi-year seasonal baseline to compute departure magnitude and frequency | Quarterly site report with tempo index, year-on-year comparison table and annotated imagery |
| Exercise-versus-deployment indicator scorecard | Multi-indicator fusion: apron occupancy, logistics vehicle presence, range activity, forward-site construction, port activity cross-reference; scored against published open-source indicator frameworks | Analytical assessment report with explicit confidence bounds and indicator-by-indicator evidence table |
| Impact-crater density mapping | Circular feature detection on high-resolution optical imagery; crater count and spatial clustering analysis to estimate munitions expenditure rate | GeoTIFF crater-density heatmap and summary table of estimated rounds per range area per period |
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.