Spoil-heap volume estimation at illegally excavated sites using stereo DSMs
Differencing pre- and post-disturbance digital surface models from commercial stereo satellite imagery quantifies the volume of material removed by illegal excavation, producing metric evidence usable in criminal and civil proceedings.
Sensors
- WorldView-1/2/3 (Maxar): Panchromatic stereo at 0.31–0.5 m GSD; along-track stereo pairs collected in a single pass minimise temporal decorrelation. Vertical accuracy in open terrain typically 0.3–0.5 m RMSE after rational-polynomial correction, sufficient to resolve individual spoil heaps of 0.5 m height or more.
- Pleiades-1A/1B (Airbus): 0.5 m panchromatic stereo with tri-stereo option; agile pointing allows same-day stereo acquisition. Published vertical accuracy of 0.3–0.5 m RMSE in flat to gently rolling terrain without ground control, improving to ~0.1 m with GCPs.
- Pleiades Neo 3/4 (Airbus): 0.3 m panchromatic stereo, the finest commercially available. Increases point-cloud density and improves detection of small or shallow pits; revisit of 1–2 days at mid-latitudes.
- SPOT-6/7 (Airbus): 1.5 m panchromatic stereo; lower cost per km² than Pleiades. Adequate for site-scale volume budgets where individual pits exceed roughly 3 m diameter, but insufficient for resolving small spoil heaps at dense looting fields.
What a spoil heap reveals that a pit alone cannot
An excavation pit tells you where looters dug. The spoil heap tells you how much they removed. Volume is the legally actionable quantity: it determines the scale of the offence, informs restitution calculations, and can be cross-checked against the number and size of objects that subsequently appear on the antiquities market. A pit visible in optical imagery might be 2 metres across; the displaced earth piled beside it, measurable in a differenced DSM, converts that observation into cubic metres of disturbed stratigraphy.
The principle is straightforward. A digital surface model captures the elevation of every point on the ground at a given date. Subtract a pre-disturbance DSM from a post-disturbance one and the signed difference volume gives you both the material removed (negative cells inside the pit) and the material deposited (positive cells over the spoil heap). The two numbers should balance within the accuracy envelope of the sensors. When they do not balance, it often means some spoil has been removed from the site entirely, which is itself evidentially significant.
How stereo photogrammetry turns imagery into elevation
Commercial stereo satellites collect two images of the same ground from different angles, typically with a base-to-height ratio of 0.4 to 0.6 for WorldView and Pleiades. Dense image matching algorithms, the same family of methods used in close-range photogrammetry, propagate correspondences across the image pair to generate a dense point cloud, which is then gridded into a DSM at a post-spacing matching the sensor GSD. For WorldView-3 stereo, that means a 0.3–0.5 m grid; for SPOT-7, around 1.5 m.
Vertical accuracy depends on terrain roughness, vegetation cover, and whether ground control points are available. In open, sparsely vegetated terrain, the published RMSE figures of 0.3–0.5 m for WorldView and Pleiades are achievable without GCPs, using rational polynomial coefficients alone. Add even a handful of GPS-surveyed control points and accuracy can approach 0.1–0.15 m. That matters because a typical looting spoil heap at a Bronze Age tumulus or a pre-Columbian platform mound may be only 0.4–0.8 m high; resolving it confidently requires vertical accuracy at least half that height.
Temporal baseline is critical. The pre-disturbance DSM should ideally predate the suspected looting episode; archive stereo imagery from Maxar and Airbus extends back to 2008–2010 for many regions. If no pre-event stereo exists, a bare-earth model derived from topographic databases or earlier photogrammetric surveys can serve as the baseline, though with higher uncertainty.
Published evidence from the Balkans and the Andes
The method is not theoretical. Studies published in the remote-sensing literature have applied stereo DSM differencing to looted Bronze Age cemeteries in Bulgaria and Serbia, where dense tumulus fields have been systematically pillaged since the 1990s. Researchers differenced WorldView and Pleiades DSMs against pre-looting baselines derived from archival imagery, recovering volume estimates per mound and aggregating them to site-level loss figures. The results were used to prioritise police investigation and to support UNESCO reporting.
In the Andes, similar workflows have been applied to pre-Columbian adobe platform mounds and cemetery complexes in coastal Peru, where looting, locally called huaqueo, has been documented by satellite since at least 2010. The flat, arid terrain is close to ideal for stereo photogrammetry: minimal vegetation, stable surface reflectance, and negligible seasonal change in surface texture. Volume estimates derived from Pleiades tri-stereo data have been cross-validated against field surveys with sub-metre agreement.
These studies share an important methodological note: the volume difference captures displaced sediment, not the number or nature of artefacts removed. Converting cubic metres of disturbed earth into an artefact count requires independent assumptions about deposit density that are site-specific and uncertain. Courts and prosecutors should be briefed on this distinction.
Honest limits of the method
Cloud cover is the most immediate operational constraint. Stereo acquisition requires two clear-sky images collected within minutes of each other on the same pass, or within a day or two for multi-pass stereo. In humid tropical or Mediterranean winter conditions, tasking windows can be weeks apart. Archive searches sometimes reveal no usable stereo for a given site and season.
Vegetation is the second limit. Dense shrub or tree cover prevents the matching algorithm from seeing the ground surface; the DSM represents the canopy, not the terrain. Spoil heaps hidden beneath scrub regrowth may be invisible or underestimated. Partial canopy can be handled by filtering point clouds to ground returns, but this degrades point density and increases uncertainty.
Small or shallow disturbances fall below the detection floor. A pit less than roughly 1–1.5 m in diameter, or a spoil heap less than 0.3–0.4 m high, sits within the vertical noise of even the best commercial stereo DSMs. Systematic looting of shallow Bronze Age graves, where disturbed depth may be only 20–30 cm, will be missed or underestimated. The method is most reliable for sites with substantial relief change: tell mounds, platform mounds, tumuli, and shaft tombs.
Finally, the legal chain of custody for satellite imagery must be established before evidence is submitted. Imagery providers can supply certificates of authenticity and metadata attestation; this should be arranged at the time of tasking, not retrospectively.
From DSM difference to court-ready volume report
A defensible volume estimate requires documented processing steps: image metadata, sensor calibration records, the matching algorithm and its parameters, co-registration method, and uncertainty propagation from pixel-level matching error through to final volume figure. Uncertainty should be reported as a confidence interval, not a single number. A statement such as '4,200 ± 600 cubic metres removed between March 2019 and November 2021' is more credible to a court than a bare figure.
Satellize structures volume-estimation deliverables to meet evidentiary standards: a georeferenced difference raster, a pit-and-heap polygon layer with per-feature volume and uncertainty, a processing log, and a plain-language summary written for non-specialist readers including legal counsel. The workflow draws on the same photogrammetric methods validated in the published Balkan and Andean studies. For clients requiring independent validation, co-registration accuracy can be verified against GPS-surveyed check points supplied by the client or a third-party surveyor.
For heritage agencies running systematic monitoring programmes across large looting-prone regions, the same stereo archive can be processed as a time series, tracking the rate of volume loss per site per year. That cumulative record is often more persuasive in policy and funding arguments than any single incident report.
Typical figures
| Best available spatial resolution (stereo DSM post-spacing) | 0.3 m (Pleiades Neo); 0.3–0.5 m (WorldView-1/2/3); 0.5 m (Pleiades-1A/1B); 1.5 m (SPOT-6/7) |
| Vertical accuracy, open terrain, no GCPs | 0.3–0.5 m RMSE (WorldView, Pleiades); improves to ~0.1–0.15 m with GPS ground control |
| Minimum detectable spoil heap height | ~0.4–0.6 m (practical floor given vertical noise; shallower features require GCPs) |
| Minimum detectable pit diameter | ~1–1.5 m at 0.3–0.5 m GSD; larger at SPOT-6/7 resolution |
| Revisit / tasking latency | 1–4 days for Pleiades Neo and WorldView-3 at mid-latitudes; subject to cloud and tasking priority |
| Archive depth | Maxar archive from ~2008; Airbus Pleiades from 2012; SPOT-6/7 from 2012 |
| Temporal baseline for differencing | Any two cloud-free stereo acquisitions; minimum meaningful interval depends on looting rate |
| Delivery formats | GeoTIFF difference raster, GeoPackage / Shapefile polygon layer, PDF evidence report, processing log |
| Cloud cover constraint | Both stereo images must be cloud-free over the site; cloud fraction >10% over the area of interest typically unusable |
Analytics Satellize can run
| Pre/post DSM pair and signed difference raster | Dense image matching (SGM or similar) applied to stereo pairs; co-registration by least-squares surface matching | GeoTIFF difference raster with per-pixel elevation change and uncertainty band |
| Pit and spoil-heap polygon inventory with per-feature volumes | Threshold segmentation of difference raster; volume integrated by trapezoidal rule over negative (pit) and positive (heap) cells | GeoPackage polygon layer with volume, uncertainty, and date attributes; CSV summary table |
| Site-level volume budget and mass-balance check | Summation of pit volumes versus spoil-heap volumes; imbalance flagged as potential off-site removal | Section of evidence report with balance table and interpretation note |
| Time-series volume-loss rate per site | Multi-epoch DSM differencing across archive stereo pairs; linear or piecewise regression on cumulative volume loss | Chart and tabular time series; GIS layer with epoch-by-epoch change polygons |
| Uncertainty-quantified volume confidence intervals | Propagation of co-registration error and matching noise through volume integration; Monte Carlo or analytical error budget | Uncertainty table included in evidence report; suitable for citation in legal submissions |
| Court-ready evidence package | Documented processing chain with image metadata, calibration records, algorithm parameters, and chain-of-custody attestation | PDF report with plain-language summary, georeferenced figures, and processing appendix |
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.