Protected site compliance monitoring for military hospitals and cultural heritage
Satellite change detection tracks whether facilities protected under international humanitarian law show evidence of military use or deliberate targeting, producing dated visual evidence admissible in accountability processes.
Sensors
- Maxar WorldView Legion: 30 cm panchromatic resolution; up to 15 revisits per day over a single target in mid-latitudes. The primary sensor for detecting individual vehicles, fortification sandbags, weapon emplacements and structural damage at the level of detail courts and investigators require.
- Planet SkySat: 50 cm native resolution, tasked on demand. Used operationally by UNOSAT and open-source investigators for before-and-after comparison at named protected sites. Video mode can resolve vehicle movement within a compound over a 90-second collect.
- Airbus Pléiades Neo: 30 cm panchromatic, 1.2 m multispectral; same-day stereo pairs possible. Stereo collection allows 3-D reconstruction of new earthworks or fortification walls, distinguishing deliberate construction from battle damage.
- Sentinel-1 SAR (C-band): Free, open-archive SAR at 5 x 20 m (IW mode) or 1.5 x 3.5 m (Stripmap). Penetrates cloud and smoke. Coherence-change analysis flags structural damage or ground disturbance even when optical tasking is denied. Six-day repeat at the equator; shorter at higher latitudes.
What international humanitarian law actually requires of imagery
Articles 18 and 19 of the Fourth Geneva Convention protect civilian hospitals from attack, and Protocol I extends that protection to cultural property, schools and places of worship. The protection is conditional: it lapses if a facility is used to commit acts harmful to the enemy. That conditionality is the crux of the legal and evidentiary problem. Both sides in almost every modern conflict assert the other has forfeited protection. Satellite imagery does not resolve intent, but it documents observable physical state at a known time and location, which is precisely what investigators, prosecutors and UN commissions of inquiry need.
UNOSAT has published damage assessments for sites in Syria, Gaza, Ukraine and Yemen using this principle. The Syrian Archive (now Mnemonic) cross-referenced Maxar and Planet imagery with geolocated social-media footage to build multi-source records of hospital strikes. Neither organisation claims imagery proves a strike was unlawful on its own. They claim, correctly, that imagery establishes what was physically present and what changed, leaving the legal inference to the tribunal.
What a vehicle concentration inside a protected perimeter gives away
The observable signatures of military use at a nominally protected site fall into a small number of categories. Vehicle concentrations are the most common: military-pattern trucks, armoured personnel carriers or artillery pieces parked inside a hospital courtyard or mosque compound are visible at 30 to 50 cm resolution. At that resolution, vehicle type is often identifiable by shadow length, wheelbase and roof profile. A cluster of such vehicles that was absent in a baseline image and present in a subsequent collect is a documented change, regardless of what either party claims.
New earthworks are the second major signature. Sandbag revetments, firing positions, trenches dug across a compound perimeter or berms pushed up around a building all appear as new linear or irregular features in optical imagery and as coherence loss in repeat-pass SAR. Pléiades Neo stereo pairs can measure the height of a new berm to within roughly one metre, which distinguishes a defensive fighting position from a flood-protection embankment.
Structural damage is the third category, and here the method is most straightforward. A building that is intact in a WorldView collect on day zero and shows roof collapse, blast scarring or rubble scatter on day three has been damaged in that interval. Sentinel-1 coherence change, which measures how much the radar backscatter from a surface has changed between two passes, can detect building collapse at the block level even through cloud cover or smoke.
The archive problem: why baseline matters as much as the strike image
A single post-event image proves very little. The evidentiary value comes from the comparison: what was the site's condition before, and what changed? Commercial archives now hold imagery of most conflict-affected cities going back ten to fifteen years. Maxar's archive predates the Syrian civil war for most major Syrian cities. Planet's daily collect, which began at global scale around 2017, provides a continuous record for the period since. Sentinel-1, operating since 2014, adds a free SAR baseline.
The practical challenge is retrieval and registration. Images from different sensors, collected at different angles and times, must be co-registered to sub-pixel accuracy before change is measured rather than assumed. Standard image-to-image co-registration using feature matching can achieve 0.3 to 0.5 pixel residual error on well-textured urban scenes. That is sufficient to distinguish a new vehicle from a shadow artefact, but analysts must document the registration quality in any evidentiary product. Courts are increasingly familiar with these methods; the International Criminal Court's Office of the Prosecutor has accepted satellite imagery as evidence in multiple cases.
Honest limits: what imagery cannot establish
Resolution has a floor. At 30 cm, you can identify a tank. You cannot read a unit marking or confirm whether weapons are loaded. You cannot see inside a building. A hospital that has been converted to a command post with all activity conducted indoors will not show the diagnostic external signatures described above, and imagery will not detect it.
Revisit is not continuous. Even WorldView Legion's 15 daily revisits leave gaps. A vehicle convoy that arrives, conducts an operation and departs within a two-hour window may not be captured at all. This is a genuine limit, not a caveat to be buried. Investigators who rely solely on imagery will miss events that occurred between collects.
Cloud cover and smoke can deny optical collection entirely. In heavily contested areas, deliberate burning of tyres or agricultural fields has been used to obscure sites during critical periods. Sentinel-1 SAR is the fallback, but at IW-mode resolution it cannot resolve individual vehicles, only building-scale damage.
Finally, imagery establishes physical state, not legal responsibility. Identifying who fired a missile, who ordered a strike or whether a commander knew a site was protected requires additional evidence. Imagery is one input to an accountability process, not the whole of it.
Building a monitoring list: coordinates, baselines and alert thresholds
Effective compliance monitoring starts with a curated coordinate list. UNESCO publishes the World Heritage List with GPS coordinates. ICRC field teams and national health ministries often publish hospital locations. These coordinates become the seed for a tasking and alert pipeline.
For each protected site, a baseline composite is assembled from the best available cloud-free imagery before the period of interest. Change is then measured against that baseline at each new collect, using normalised difference in pixel values, object detection for vehicles and coherence-change metrics from SAR. Alert thresholds are set by analysts, not automated classifiers alone, because false positives in this domain carry serious consequences. A construction crane misclassified as a military vehicle in a report submitted to a UN commission is not a minor error.
Satellize builds and maintains this kind of coordinate-anchored monitoring pipeline for government clients. The same architecture that supports the Tonga crop-estimation programme, adapted for change detection rather than spectral classification, underpins the protected-site workflow. Analysts review flagged changes before any product leaves the system.
From pixel change to evidentiary product
The output of a compliance-monitoring workflow is not a raw image. It is a structured report that documents the site's coordinates, the baseline image date and source, the comparison image date and source, the co-registration quality metric, the specific pixels or objects that changed, and the analyst's interpretation of what the change represents. That structure mirrors the chain-of-custody requirements that legal processes impose on digital evidence.
UNOSAT's published damage assessment methodology, which is openly documented, follows this structure. The Syrian Archive added a layer of cross-referencing with open-source video and social-media metadata to corroborate the imagery interpretation. The combination of satellite evidence and ground-sourced corroboration is now the standard that serious accountability organisations apply. Imagery alone is admissible; imagery corroborated by independent ground evidence is persuasive.
Typical figures
| Best optical resolution (commercial) | 30 cm (Maxar WorldView Legion, Airbus Pléiades Neo) |
| Best optical resolution (open) | 3 m (Planet Dove, daily global); 10 m (Sentinel-2, 5-day revisit) |
| SAR resolution | 1.5 x 3.5 m (Sentinel-1 Stripmap); 5 x 20 m (IW mode) |
| Tasked revisit (commercial) | Up to 15 passes/day over a single target (WorldView Legion at mid-latitudes) |
| SAR revisit | 6 days (Sentinel-1, equatorial); shorter at higher latitudes |
| Minimum detectable vehicle | ~2 m length at 30 cm optical resolution; not reliably detectable in Sentinel-1 IW mode |
| Archive depth | Maxar: 20+ years for major cities; Planet daily global: from ~2017; Sentinel-1: from 2014 |
| Spectral bands (optical) | Panchromatic + 4-band multispectral (RGB + NIR) standard; 8-band superspectral on WorldView-3 |
| Latency (tasked collect to delivery) | 2–24 hours for commercial priority tasking; Sentinel open data typically 1–3 days |
| Deliverable formats | Georeferenced GeoTIFF, annotated PDF report, GeoJSON change polygons, KMZ for court submission |
Analytics Satellize can run
| Protected-site baseline composite | Multi-image median compositing from archive optical and SAR data, co-registered to sub-pixel accuracy | Georeferenced baseline image set per coordinate, with acquisition metadata and co-registration quality report |
| Vehicle concentration alert | Object detection (convolutional neural network trained on labelled military vehicle imagery) applied to new collects; analyst review before dispatch | Alert report with annotated image chip, vehicle count estimate and confidence level |
| Structural damage change map | Pixel-difference and coherence-change analysis between baseline and current collect; damage graded by UNOSAT-aligned severity classes (destroyed, severely damaged, moderately damaged) | GeoJSON polygon layer with damage grade per building footprint; companion PDF for legal submission |
| New earthworks and fortification detection | Normalised surface-model differencing from stereo optical pairs (Pléiades Neo); texture and spectral change detection for SAR-only periods | Annotated image report with height estimates for new features; flagged coordinates for follow-up tasking |
| Continuous monitoring feed for a named coordinate list | Automated ingestion of new Sentinel-1 and Planet Dove passes; change score computed per site; threshold-based queue for analyst review | Weekly summary report; immediate alert on threshold exceedance; GIS layer updated per collect cycle |
| Evidentiary image package | Structured documentation following UNOSAT chain-of-custody conventions: source, date, co-registration metric, analyst interpretation, limitations statement | Court-ready PDF with metadata annex; source imagery in original format with unmodified EXIF/NITF headers |
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.