Clandestine grave and mass burial site detection
Freshly disturbed soil and decomposition-driven vegetation die-off leave spectral and topographic signatures detectable from orbit. Satellite remote sensing, documented in forensic literature, can cue ground investigators to clandestine burials linked to organised crime or atrocity events.
Sensors
- Maxar WorldView-3: 31 cm panchromatic, 1.24 m multispectral (8 VNIR bands plus 8 SWIR bands). Stereo collection enables 0.3–1 m relative vertical accuracy DEMs for detecting micro-topographic mounds as shallow as 20–30 cm above surrounding grade. Tasked on demand; archive extends to 2014.
- Airbus Pléiades Neo: 30 cm panchromatic, 1.2 m multispectral (4 bands). Tri-stereo mode produces dense point clouds useful for subtle surface deformation mapping. Revisit of 1–2 days at mid-latitudes when the full constellation is tasked.
- Sentinel-2 MSI: 10 m visible and NIR bands, 20 m red-edge and SWIR bands. Free, 5-day revisit (10 m bands). Too coarse for individual graves but capable of detecting vegetation anomalies over mass burial pits of tens of square metres or larger, and of tracking spectral change over months.
- TanDEM-X DEM products: Global DEM at 12 m posting (TanDEM-X 12 m), with relative vertical accuracy around 2 m for the global product. Custom high-resolution stripmap products can approach sub-metre vertical precision over small areas. Useful for detecting landscape-scale topographic anomalies and comparing against archive baselines.
What disturbed soil actually signals from orbit
When soil is excavated and returned to a pit, its physical character changes in ways that persist for months to years. Particle size distribution is altered, compaction is reduced, moisture retention differs from undisturbed surroundings, and organic material from depth is mixed to the surface. Each of these changes modifies spectral reflectance across visible, near-infrared and short-wave infrared wavelengths. Published forensic remote-sensing studies, including work by researchers at Cranfield University and others published in journals such as Remote Sensing and Forensic Science International, have documented that disturbed soil produces measurable NDVI suppression and shifts in SWIR reflectance relative to background.
The physical mechanism is straightforward. Freshly turned soil is typically darker in the visible bands immediately after disturbance, then lightens as it dries. Moisture anomalies persist longer in clay-rich soils. In SWIR bands around 1.6 µm and 2.2 µm, soil mineral composition and moisture content both influence reflectance, giving WorldView-3's eight SWIR bands particular diagnostic value at spatial scales relevant to individual or small-cluster graves.
Vegetation die-off as a delayed but persistent indicator
Decomposition produces ammonia, hydrogen sulphide and volatile fatty acids that leach into surrounding soil and suppress root function. The resulting vegetation stress is detectable in near-infrared reflectance: healthy green vegetation has high NIR reflectance driven by leaf cell structure, and stressed or dead vegetation loses this signature. On Sentinel-2, this appears as a depression in the NIR band (Band 8, 842 nm) and in red-edge bands (705 nm and 740 nm), which are sensitive to early-stage chlorophyll loss before visible yellowing occurs.
The timing matters. Spectral anomalies from soil disturbance alone may fade within weeks in humid climates. Vegetation die-off typically peaks weeks to months after burial, depending on depth, climate and season. This means a detection programme needs multi-temporal coverage at consistent intervals rather than a single acquisition. In arid environments, where natural vegetation is sparse, the soil spectral signal itself is more persistent and the vegetation indicator less useful. Analysts must calibrate expectations to local ecology.
Micro-topography: the mound that cannot be flattened
Refilled soil rarely returns to the exact original surface level. Disturbed material has lower bulk density, and decomposition of organic matter causes further subsidence over time, creating a characteristic sequence: initial slight mounding followed by a depression as decomposition progresses. At 30 cm resolution, stereo-derived DEMs from WorldView-3 or Pléiades Neo can detect relative height anomalies of 20–40 cm against a stable surrounding surface, provided the surrounding terrain is reasonably flat and the DEM is differenced against a pre-event baseline.
Differencing two DEMs acquired months apart is the standard method. The challenge is co-registration error: sub-pixel misalignment between acquisitions introduces artefacts that can mimic or mask real signals. Rigorous ground control and careful co-registration are prerequisites. TanDEM-X global products are too coarse for individual grave detection but can establish regional topographic baselines and flag landscape-scale anomalies worth tasking with optical stereo.
Where the method works and where it fails
The method has documented application in contexts including investigations into cartel activity in Mexico, where researchers have used multispectral imagery to identify candidate sites later confirmed by ground teams. It is not a forensic instrument in itself. Satellite analysis produces a probability-ranked list of candidate locations, not confirmed identifications. Ground truth remains essential, and false positives from agricultural disturbance, construction, animal activity or natural erosion are common.
Cloud cover is a fundamental constraint in humid tropical regions, precisely where many clandestine burials occur in forested or jungle terrain. Sentinel-2 and optical commercial sensors are blind through cloud. SAR (synthetic aperture radar) can penetrate cloud and detect surface roughness changes, but published forensic literature on SAR for burial detection is limited compared with optical methods, and the interpretation is less mature. Dense forest canopy presents a separate problem: optical sensors see canopy, not ground, making detection in closed-canopy jungle essentially impossible with current civilian sensors. Detection is most reliable in open, semi-arid or agricultural terrain with low to moderate vegetation cover.
Individual graves of one to two bodies are at or below the detection threshold of any current civilian satellite. The method is most credible for mass burial sites covering tens of square metres or more, or for clusters of individual graves that collectively produce a detectable spectral or topographic anomaly.
Building a detection workflow
A practical workflow combines three data streams. First, a Sentinel-2 time series covering the area of interest at 10-day intervals establishes a spectral baseline and flags candidate change pixels using indices such as NDVI, BSI (Bare Soil Index) and a SWIR-based moisture index. Second, commercial tasking of WorldView-3 or Pléiades Neo over flagged candidates provides the spatial resolution needed to assess texture, shape and the morphological characteristics of disturbance. Third, stereo DEM differencing over the highest-priority candidates tests for micro-topographic anomaly.
Prioritisation matters because commercial tasking is expensive and cannot cover large territories at high revisit. An effective programme uses Sentinel-2 as the wide-area screening layer and reserves commercial tasking for sites that pass a spectral anomaly threshold. Satellize structures this kind of tiered tasking workflow for clients who need systematic, repeatable coverage rather than one-off acquisitions.
Analysts should document the confidence level of each candidate explicitly: spectral anomaly only, spectral plus topographic, or all three convergent lines of evidence. Courts and investigators require that chain of custody and analytic methodology be transparent. Satellite evidence has been admitted in international tribunal proceedings, but only when methodology is rigorously documented.
Typical figures
| Best spatial resolution (optical) | 30 cm (WorldView-3 panchromatic, Pléiades Neo panchromatic) |
| Stereo DEM vertical accuracy | 0.3–1 m relative (WorldView-3 stereo, well-controlled); TanDEM-X global product ~2 m |
| Minimum detectable burial area | Approximately 20–50 m² for spectral anomaly (mass graves or clusters); individual graves below current civilian detection threshold |
| Spectral bands used | VNIR 400–900 nm (NDVI, red-edge stress); SWIR 1.6 µm and 2.2 µm (soil moisture, mineralogy); WorldView-3 adds 8 SWIR bands |
| Revisit (screening layer) | 5 days at equator (Sentinel-2 twin satellites); 1–2 days (Pléiades Neo tasked) |
| Cloud penetration | None for optical sensors. SAR (e.g. Sentinel-1 C-band) penetrates cloud but forensic interpretation less established |
| Archive depth | Sentinel-2 from 2015; WorldView-3 from 2014; Landsat archive from 1972 (30 m, useful only for large sites) |
| Delivery formats | GeoTIFF change maps, GeoPackage candidate-site polygons, PDF ranked-site reports with confidence scoring |
| Latency (from tasking to delivery) | Sentinel-2 change alert: 24–72 hours after acquisition. Commercial stereo DEM: 5–15 days depending on cloud and tasking queue |
Analytics Satellize can run
| Spectral anomaly change map | Multi-temporal NDVI, BSI and SWIR moisture-index differencing on Sentinel-2 time series; z-score thresholding against local baseline | GeoTIFF raster and GeoPackage polygon layer of candidate anomaly pixels, updated on each clear-sky Sentinel-2 pass |
| Ranked candidate-site report | Multi-criteria scoring combining spectral anomaly magnitude, site morphology (shape regularity, area), proximity to access tracks, and temporal persistence | PDF report with georeferenced site cards, confidence tier (spectral only / spectral + morphological / all three lines), and recommended ground-priority order |
| Micro-topographic anomaly layer | DEM differencing between pre-event and post-event stereo acquisitions; co-registration via tie-point matching; anomaly extraction by height-difference threshold | GeoTIFF height-difference raster and polygon layer of mound or subsidence candidates, with co-registration error budget stated |
| Vegetation stress time series | Red-edge chlorophyll index (Sentinel-2 Bands 5 and 6 at 705 nm and 740 nm) tracked monthly; anomaly onset and duration logged per candidate polygon | CSV time-series table per candidate site, with annotated chart showing stress onset relative to suspected event window |
| High-resolution site orthoimage | Commercial tasking of WorldView-3 or Pléiades Neo over priority candidates; pan-sharpened to 30 cm; analyst annotation of textural and morphological features | GeoTIFF orthoimage with analyst markup layer (GeoPackage) identifying disturbed-soil boundaries, access tracks and contextual features |
| Methodology documentation package | Structured analytic narrative following published forensic remote-sensing standards, with sensor metadata, processing chain, uncertainty quantification and chain-of-custody log | PDF methodology report suitable for submission to investigative or judicial bodies, with all input data hashes and processing parameters recorded |
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.