Post-strike battle damage assessment from optical imagery
Pre- and post-event optical imagery can quantify structural collapse, cratering, and fire scarring at military and dual-use sites. Resolution determines what you can actually measure: 0.31 m resolves a vehicle-sized crater; 10 m resolves a city block.
Sensors
- Maxar WorldView-3: 0.31 m panchromatic, 1.24 m multispectral (8 bands including SWIR). The only commercially available sensor that resolves individual vehicle-sized craters and distinguishes partial from total structural collapse. Revisit roughly 1 day at mid-latitudes with tasking priority, though cloud and competing demand affect actual delivery. Archive extends to 2014.
- Airbus Pléiades Neo: 0.30 m panchromatic, 0.75 m multispectral. Comparable resolution to WorldView-3; four satellites give a nominal revisit of less than 24 hours over most targets. Stereo and tri-stereo tasking enables crater-depth estimation via photogrammetric DSM differencing.
- Planet SkySat: 0.50 m panchromatic, approximately 0.72 m multispectral. Revisit can be tasked to multiple times per day over a fixed point, which is useful for monitoring active strikes or post-strike clearance activity. Swath width is narrow (roughly 6.6 km), so area coverage requires careful tasking.
- Sentinel-2 MSI: 10 m resolution at visible and near-infrared bands; 20 m for red-edge and SWIR. Free and openly archived with 5-day revisit (2-3 days with both satellites). Sufficient for block-level destruction mapping and burn-scar delineation using the normalised burn ratio (NBR), but cannot resolve individual craters or confirm single-building collapse. Useful for screening large areas before commercial tasking.
What the imagery actually measures, and what it does not
Battle damage assessment from optical imagery is a change-detection problem. A pre-event scene establishes a baseline: roof integrity, surface texture, shadow geometry, and spectral signature. The post-event scene is differenced against that baseline. What changes are rooftop collapse (visible as a shift from uniform surface to rubble texture), cratering (circular depressions with raised ejecta rings), fire scarring (darkened ground and altered SWIR reflectance), and debris scatter. Analysts at UNOSAT, AAAS, and Bellingcat have applied these workflows systematically in Syria, Yemen, and Ukraine, publishing methodology alongside findings.
The method has clear limits. Optical sensors measure surface reflectance. They cannot see through roofs that remain standing over destroyed interiors, cannot distinguish a building that was evacuated before a strike from one that was not, and cannot assess subsurface damage. A hardened bunker with an intact concrete lid looks undamaged from orbit even if the interior is destroyed. Honest assessment requires pairing optical findings with other intelligence, not treating them as complete.
The resolution floor problem
Resolution is not a marketing variable here; it determines what analytic conclusions are legally and operationally defensible. At 0.31 m (WorldView-3) or 0.30 m (Pléiades Neo), an analyst can measure crater diameter, estimate depth from shadow geometry, distinguish a direct hit on a hardstand from a near-miss, and count collapsed bays in a multi-bay warehouse. At 0.50 m (SkySat), single-vehicle craters are marginal; at 10 m (Sentinel-2), a crater must be roughly 15 m across before it registers reliably as a discrete feature.
The practical consequence: Sentinel-2 is appropriate for screening large areas, identifying which districts or installations warrant closer attention, and mapping fire scars across tens of thousands of hectares. It is not appropriate for assessing whether a specific hardened aircraft shelter was penetrated or whether a fuel depot was fully destroyed versus partially damaged. For those questions, sub-metre commercial tasking is required, and the tasking must happen before reconstruction or debris clearance obscures the evidence.
Cloud cover over active conflict zones
Cloud is the most persistent operational constraint. The Donbas in winter, the Syrian coastal range, and highland Yemen all experience extended cloud periods. A strike that occurs under overcast conditions may not be imageable at optical wavelengths for days or weeks. By then, debris clearance, reconstruction, or secondary strikes will have altered the scene.
There is no optical solution to this. The honest answer is that cloud-affected periods require synthetic aperture radar (SAR), specifically coherence-loss methods using Sentinel-1 or commercial X-band systems, which are covered in the sibling page on SAR coherence-loss damage detection. Optical and SAR are complementary, not competitive. A workflow that uses Sentinel-2 for rapid screening, SAR for cloud-affected periods, and sub-metre optical for confirmed-damage characterisation is more reliable than any single-sensor approach. Clients who plan for cloud from the outset, rather than treating it as an exception, get better results.
Published workflows: how Bellingcat, UNOSAT, and AAAS do it
The methodological record is public. UNOSAT's damage assessments in Syria and Ukraine use a combination of visual interpretation by trained analysts and semi-automated change detection, with results classified into four damage categories: destroyed, severely damaged, moderately damaged, and possibly damaged. The AAAS programme applied similar four-tier classification to sites in North Korea and Syria, using WorldView imagery and photogrammetric shadow analysis to estimate building height loss. Bellingcat's Ukraine work has demonstrated that dated satellite imagery can geolocate and timestamp events to rebut false claims, a function that depends entirely on archive depth and acquisition metadata integrity.
Automated change detection typically uses one or more of the following: image differencing on co-registered scenes, normalised difference indices (NBR for burn scars, NDVI for vegetation loss around blast zones), object-based image analysis (OBIA) to segment rooftop objects and flag texture changes, and deep-learning classifiers trained on labelled conflict-damage datasets. The COPERNICUS Emergency Management Service publishes grading maps using these methods. None of these automated approaches eliminates the need for human review before a finding is reported; false positives from seasonal vegetation change, construction activity, or sensor artefacts are common enough to be a standing concern.
Archive depth and the evidentiary chain
For legal, accountability, or treaty-verification purposes, the evidentiary value of a satellite image depends on its provenance: acquisition timestamp, sensor metadata, chain of custody, and whether the scene has been altered. Maxar's archive extends to 2014 for WorldView-3; Planet's SkySat archive is shallower but growing. Sentinel-2 data is freely available from 2015 and is archived by ESA with full metadata. These archives allow analysts to establish not just what changed, but when it changed, by stepping through a time series rather than comparing only two scenes.
Satellize applies these published change-detection workflows on behalf of government and enterprise clients, drawing on open Sentinel data and commercial tasking added on client licence. The analytic approach is the same one used in the Tonga crop-estimation programme, adapted from agricultural to security contexts: systematic, documented, and reproducible. Clients receive findings with explicit confidence levels and the underlying imagery, not just conclusions.
What a credible assessment report contains
A deliverable that is operationally useful specifies: the acquisition dates and sensors for both pre- and post-event scenes; the co-registration method and residual error; the damage classification scheme and the confidence threshold applied; a count and spatial coordinates of confirmed craters, collapsed structures, and burn scars; and an explicit statement of what could not be assessed due to cloud, shadow, or resolution limits.
Omitting the limits does not strengthen the assessment. It weakens it, because a reader who later identifies an uncaveated error will discount everything else. The most credible BDA products in the public record, including UNOSAT's Ukraine work, are explicit about what the imagery cannot confirm. That candour is the standard against which any commercial offering should be measured.
Typical figures
| Best available spatial resolution (commercial) | 0.30 m panchromatic (Pléiades Neo, WorldView-3) |
| Typical multispectral resolution | 0.75–1.24 m (Pléiades Neo / WorldView-3); 10 m (Sentinel-2) |
| Revisit (tasked commercial) | Less than 24 hours at most latitudes with priority tasking; SkySat can achieve multiple passes per day over a fixed point |
| Revisit (Sentinel-2, free) | 2–5 days cloud-free; cloud-affected periods may extend gap to weeks |
| Minimum detectable crater diameter | Approximately 1–2 m at 0.30 m resolution; approximately 15 m at 10 m resolution (Sentinel-2) |
| Spectral bands relevant to BDA | Panchromatic, visible RGB, NIR, SWIR (burn-scar detection via NBR); WorldView-3 adds 8 SWIR bands |
| Archive depth | WorldView-3 from 2014; Sentinel-2 from 2015 (open access); Planet SkySat from approximately 2017 |
| Latency from tasking to delivery | Typically 24–72 hours for commercial tasking; Sentinel-2 open data available within hours of acquisition |
| Damage classification standard | UNOSAT four-tier scheme (destroyed / severely damaged / moderately damaged / possibly damaged) is the de facto reference |
| Delivery formats | GeoTIFF orthoimage, vector damage polygon layer (GeoJSON / Shapefile), georeferenced PDF report, optional COG for web delivery |
Analytics Satellize can run
| Damage extent map | Object-based image analysis (OBIA) on co-registered pre/post scene pair, with four-tier UNOSAT classification scheme | Vector polygon layer with per-structure damage grade and confidence score; GeoJSON or Shapefile |
| Crater count and diameter measurement | Circular Hough transform or manual digitisation on sub-metre imagery; shadow-length photogrammetry for depth estimate | Point layer with crater coordinates, measured diameter, estimated depth range, and weapon-effects context notes |
| Burn-scar delineation | Normalised burn ratio (NBR = (NIR − SWIR) / (NIR + SWIR)) differenced between pre- and post-event scenes; applicable at Sentinel-2 scale for large fires | Raster burn-severity map and vector perimeter; area statistics in hectares |
| Time-series change chronology | Sequential scene differencing across archive acquisitions to bound the event date window; applicable where cloud-free archive exists | Annotated image strip with acquisition timestamps and change-onset date range |
| Reconstruction and clearance monitoring | Repeat tasking at weekly or monthly cadence; OBIA texture and roof-signature change detection to flag debris removal or new construction | Periodic update report with change flags and imagery; optional automated alert on threshold exceedance |
| Evidentiary image package | Archive retrieval with full acquisition metadata, co-registration documentation, and analyst annotation; methodology aligned with published AAAS and Bellingcat practice | Signed PDF report with provenance chain, annotated imagery, and explicit confidence and limitation statements |
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.