Mass grave and conflict burial-site surface disturbance detection
Freshly disturbed soil has a distinct spectral and dielectric signature detectable from orbit. The window is narrow: vegetation regrowth begins masking evidence within weeks, making rapid multispectral and SAR revisit the difference between documentation and impunity.
Sensors
- Maxar WorldView-3: 0.31 m panchromatic, 1.24 m 8-band multispectral including SWIR bands 1580–2365 nm. SWIR is particularly sensitive to soil moisture contrast between disturbed and compacted undisturbed ground. Revisit approximately 1–4.5 days depending on latitude and tasking priority. Commercial tasking required; archive depth from 2014.
- Planet SuperDove: 3 m resolution, 8 spectral bands including red-edge and NIR, daily global revisit. Spatial resolution is insufficient to resolve individual burial pits smaller than roughly 9–10 m across, but the daily cadence is the most operationally useful attribute for catching the short disturbance window before vegetation recovery begins.
- Sentinel-2 MSI: 10 m (visible/NIR) and 20 m (SWIR, red-edge) resolution, 5-day revisit at the equator with both satellites. Free and open archive from 2015. SWIR bands 11 and 12 (1610 nm and 2190 nm) respond to soil moisture and texture changes. Minimum detectable disturbed patch is roughly 30–50 m² under ideal conditions, larger in practice.
- Sentinel-1 SAR (C-band, 5.6 cm): Interferometric coherence at 5–20 m resolution (IW mode). Surface disturbance randomises the phase relationship between repeat passes, producing coherence loss over disturbed ground even under cloud or at night. Six-day repeat cycle in IW mode over most conflict-affected regions. Coherence loss is non-specific: it flags any ground change, so optical confirmation is always required.
What disturbed soil actually looks like from orbit
When soil is excavated and replaced, several physical properties change simultaneously. Bulk density drops, surface roughness increases, and residual moisture from deeper layers is exposed. In multispectral imagery this produces a characteristic darkening in SWIR bands relative to surrounding undisturbed ground, combined with a depression in the Normalised Difference Vegetation Index because bare soil replaces any surface cover. The Normalised Difference Soil Index (NDSI, using SWIR and NIR) and the Bare Soil Index (BSI, combining SWIR, NIR and visible red) both amplify this contrast against background.
In SAR, the mechanism is different. C-band radar (Sentinel-1 at 5.6 cm wavelength) penetrates dry vegetation lightly but scatters strongly off rough or moist disturbed soil. More usefully, repeat-pass interferometry measures phase coherence between two acquisitions. Undisturbed ground maintains coherence; any surface movement or texture change decorrelates the phase signal. A freshly dug and backfilled area appears as a coherence-loss patch in the interferogram. This works through cloud, at night, and even under light canopy, which matters in conflict zones where access is denied and weather is unreliable.
The clock starts at first disturbance
Published work by AAAS Science and Human Rights Coalition, Human Rights Watch, and UNOSAT has documented this methodology in detail, applied to sites including Srebrenica (where secondary mass graves were identified by comparing pre- and post-disturbance imagery) and, more recently, Mariupol and other Ukrainian sites following the 2022 invasion. The consistent finding is that the spectral signal degrades rapidly. In temperate and semi-arid climates, pioneer vegetation can establish within two to four weeks of disturbance, progressively filling in the NDVI depression. Soil moisture equilibrates with the surrounding matrix over a similar timescale, weakening the SWIR contrast.
SAR coherence loss persists somewhat longer because surface roughness changes are more durable than moisture anomalies, but even that signal can be confused by subsequent rainfall, wind erosion or deliberate surface treatment. The practical implication is that a detection programme cannot rely on monthly revisit. It needs near-daily optical tasking (Planet) for the first two weeks, backed by Sentinel-1 coherence pairs on their six-day cycle, with WorldView-3 tasked immediately on any positive flag for sub-metre confirmation.
Geometry as evidence: why shape matters as much as spectral response
Individual burial pits are typically irregular in outline, elongated, and clustered in groups rather than following agricultural field geometry. This geometric signature distinguishes them from ploughing, construction trenching or drainage works, all of which produce linear or rectilinear patterns. Automated change-detection algorithms flag all disturbed soil; human analysts then apply geometric filters and contextual reasoning (proximity to conflict activity, absence of agricultural calendar justification, vehicle tracks without corresponding infrastructure) to separate candidate sites from false positives.
At WorldView-3 resolution, individual pits of roughly 2 m × 5 m are resolvable. At Sentinel-2 resolution, only clusters or mass graves of significant extent are detectable as a single anomaly. Planet SuperDove at 3 m sits between these: it can resolve larger individual features and is the practical workhorse for rapid initial screening given its daily revisit. The honest limit is that none of these sensors can confirm what is buried. They detect surface disturbance. Ground truth, forensic investigation and legal attribution remain human responsibilities.
False positives are the professional hazard
Agricultural ploughing, construction, pipeline burial, shell-crater backfilling and deliberate landscape modification by forces aware of satellite monitoring all produce spectral and coherence signatures that overlap with burial-site disturbance. A credible detection workflow requires change-detection output to be cross-referenced against agricultural calendars, land-use baselines, known infrastructure projects and conflict event databases before any site is flagged with confidence.
Cloud cover is the other hard constraint. Optical sensors are blind through it. A sustained overcast period of two to three weeks in a temperate climate can allow vegetation recovery to erase the optical signature entirely before the first clear acquisition. SAR coherence partially compensates, but the combination of cloud and deliberate surface treatment (gravel spreading, re-seeding) can defeat even a well-resourced monitoring programme. Analysts should document detection confidence explicitly, distinguishing between 'confirmed surface disturbance of unknown cause' and 'probable burial activity', and never conflate the two in reporting.
Evidentiary chain and the role of open data
For this use case, the evidentiary chain matters as much as the detection itself. Imagery used in international criminal proceedings must be sourced from providers with documented chain of custody, known sensor calibration records and verifiable acquisition timestamps. Sentinel-2 and Sentinel-1 data, distributed under the Copernicus Open Access policy with full metadata, have been used in published human rights documentation and carry well-established provenance. Commercial imagery from Maxar carries its own licensing and metadata standards that have withstood scrutiny in legal contexts.
Satellize runs change-detection analytics on open Sentinel constellations and can add commercial tasking on client licence, applying the same soil-index and coherence-loss methods described here. The Kingdom of Tonga crop-estimation programme is a different domain, but the underlying soil-index methodology is directly transferable. For defence and accountability clients, the output format matters: GIS layers with acquisition metadata, uncertainty scores and analyst notes carry more weight in documentation workflows than image exports alone.
What a monitoring programme actually requires
A credible burial-site monitoring programme for an active conflict zone needs four components. A baseline: pre-conflict or pre-event imagery of the area of interest, ideally from multiple sensors and seasons, to establish what normal disturbance looks like. A trigger layer: a near-daily optical feed (Planet) or SAR coherence pair stack that flags anomalies automatically. A confirmation asset: commercial tasking of WorldView-3 or equivalent within 24–48 hours of a flag, while the signal is still strong. And an analytical record: timestamped, metadata-rich outputs archived in a format suitable for legal or investigative use.
The archive depth of Sentinel-2 (from 2015) and the commercial archives of Maxar (WorldView-3 from 2014) mean that retrospective analysis is often possible even when real-time monitoring was not in place. This has been central to post-hoc accountability investigations. It does not replace timely detection, but it provides a fallback that has proven legally significant in past proceedings.
Typical figures
| Best spatial resolution (optical) | 0.31 m pan / 1.24 m multispectral (WorldView-3); 3 m (Planet SuperDove); 10–20 m (Sentinel-2) |
| Best spatial resolution (SAR coherence) | 5–20 m (Sentinel-1 IW mode) |
| Revisit cadence | Daily (Planet); 5–6 days (Sentinel-2 and Sentinel-1 at mid-latitudes); 1–4.5 days on tasking (WorldView-3) |
| Spectral bands used | SWIR (1610 nm, 2190 nm), NIR (842 nm), Red-edge (705–783 nm), Red (665 nm); C-band SAR (5.6 cm) for coherence |
| Minimum detectable disturbed patch (optical) | Approximately 30–50 m² under ideal conditions (Sentinel-2); approximately 9–25 m² (Planet); approximately 4 m² (WorldView-3 multispectral) |
| Detection window before vegetation masking | Typically 2–4 weeks in temperate/semi-arid climates; shorter in high-rainfall tropical settings |
| Cloud penetration | SAR coherence only; optical sensors fully blocked by cloud |
| Archive depth | Sentinel-2 from 2015; WorldView-3 from 2014; Planet from approximately 2016 at global scale |
| Delivery formats | GeoTIFF change-detection layers, GeoPackage / Shapefile vector outputs with acquisition metadata, timestamped PDF analyst reports |
| Latency (flag to confirmed delivery) | 24–72 hours from satellite pass to analyst-reviewed output, depending on cloud and tasking queue |
Analytics Satellize can run
| Soil-index change map | Bitemporal NDSI and BSI differencing on Sentinel-2 or Planet SuperDove imagery, thresholded against pre-event baseline | GeoTIFF layer with per-pixel change magnitude and flagged anomaly polygons, delivered per acquisition |
| SAR coherence-loss layer | Sentinel-1 repeat-pass interferometric coherence differencing (IW mode, 6-day pairs), masked to areas of interest | GeoPackage coherence-loss polygons with acquisition date pair metadata, updated on each new Sentinel-1 pass |
| Candidate site confirmation report | WorldView-3 8-band multispectral analysis of flagged polygons, including geometric characterisation and contextual assessment against agricultural calendar and conflict event data | Timestamped PDF analyst report with annotated imagery, confidence classification and chain-of-custody metadata |
| Temporal signature archive | Multi-date stack of Sentinel-2 and Sentinel-1 acquisitions over sites of interest, enabling retrospective analysis of disturbance onset and vegetation recovery rate | Archived GeoTIFF time series with full Copernicus metadata, formatted for submission to investigative or legal workflows |
| False-positive screening layer | Cross-reference of flagged disturbance polygons against land-use baseline, agricultural parcel boundaries and known construction activity, with geometric shape analysis to filter linear or rectilinear features | Filtered candidate list with false-positive probability score per site, delivered as GIS layer with analyst notes |
| Monitoring alert feed | Automated near-daily Planet SuperDove ingestion with NDVI and BSI anomaly detection, triggering alert on threshold exceedance within defined area of interest | Email or API alert within 24 hours of anomaly detection, with thumbnail and coordinates, followed by full layer delivery |
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.