Looting-pit detection and monitoring at conflict-zone heritage sites
Illegal excavation pits produce bright soil disturbances visible in sub-metre optical imagery. Change detection between pre- and post-conflict acquisitions has quantified thousands of pits at Apamea, Abusir el-Malek, and sites across southern Iraq.
Sensors
- WorldView-3 (Maxar): 0.31 m panchromatic, 1.24 m multispectral (8 bands including SWIR); revisit approximately 1 day at mid-latitudes. Resolves pits as small as 0.5–1 m diameter depending on soil contrast. SWIR bands help distinguish fresh disturbed soil from weathered surface.
- WorldView-2 (Maxar): 0.46 m panchromatic, 1.84 m multispectral (8 bands). Slightly coarser than WV-3 but the archive extends to 2009, making it the primary source for pre-conflict baselines at Syrian and Iraqi sites.
- Pléiades-1A/1B (Airbus): 0.5 m panchromatic, 2 m multispectral; daily revisit in stereo or tri-stereo mode. Stereo pairs support pit-depth estimation via DSM differencing, though depth accuracy is typically ±0.3–0.5 m under good contrast conditions.
- Planet SkySat: 0.5 m panchromatic, 1 m multispectral; tasked revisit within hours on request. Useful for rapid confirmation after a suspected looting event, though the archive depth is shallower than WorldView.
- Sentinel-2 (ESA Copernicus): 10 m multispectral (visible and NIR); free, 5-day revisit. Too coarse to resolve individual pits but useful for monitoring large-scale soil disturbance extent and for scheduling costly commercial tasking at the right moment.
What a looting pit looks like from orbit
A freshly dug pit exposes subsurface sediment that is typically lighter in colour and higher in reflectance than the surrounding compacted or crusted surface. In very-high-resolution (VHR) optical imagery, this produces a bright circular or irregular patch, often with a darker shadow on the downslope rim where spoil has been heaped. The contrast is sharpest in the red and near-infrared bands and fades over months as the disturbed surface weathers and re-crusts.
Pit diameters at known looted sites range from roughly 1 m (a single-person probe) to 10 m or more where mechanical equipment has been used. At Apamea in Syria, analysts using WorldView imagery published in 2013 counted more than 4,000 pits in a single site sector, many in the 2–5 m diameter range. That scale of disturbance is legible even at 0.5 m resolution, but the smallest probes sit right at the detection threshold: a 1 m pit occupies only four pixels in a WorldView-3 panchromatic image and is easily confused with natural surface heterogeneity or animal burrows.
Change detection: before, during, and after
The methodological backbone is bitemporal or multitemporal change detection. A pre-conflict or pre-crisis image establishes the undisturbed baseline. Subsequent acquisitions are co-registered to sub-pixel accuracy, then differenced spectrally or classified independently and compared. Pixels or objects that shift from dark to bright, or that acquire the spectral signature of loose sediment, are candidates for new disturbance.
The EAMENA project (Endangered Archaeology in the Middle East and North Africa), hosted at Oxford and Leicester, has applied this workflow systematically across Syrian, Iraqi, and Libyan sites using WorldView and Pléiades imagery. ASOR Cultural Heritage Initiatives produced quarterly monitoring reports during the Syrian conflict, documenting the acceleration of looting at Apamea and Dura-Europos in near-real time. Both programmes found that looting intensity correlated with periods of reduced government control rather than with active fighting, which has implications for monitoring cadence: the critical window is the months after security collapse, not the months of peak bombardment.
Object-based image analysis (OBIA) outperforms pixel-based differencing for pit detection because pits are spatially coherent objects with shape as well as spectral attributes. Published workflows segment the image into objects, then classify by brightness, circularity, and texture. Reported precision and recall figures in peer-reviewed studies vary considerably with site type and image quality, but false-positive rates of 20–40 per cent are common without manual validation, which means automated counts should always be treated as estimates with stated uncertainty.
Resolution floors and honest ambiguity
The 0.3–0.5 m panchromatic floor of current commercial VHR satellites is not sufficient to resolve every pit. A probe hole dug by one person with a shovel may be 60–80 cm across. It will appear as one or two pixels in WorldView-3 panchromatic data and is statistically indistinguishable from surface noise at that scale. Published studies acknowledge this by reporting counts only for pits above a stated minimum diameter, typically 1–2 m.
Shadow geometry matters too. Low sun-elevation angles at dawn or dusk, or in winter at high latitudes, produce longer shadows that make small pits more visible. This is well established in earthwork detection literature. Conversely, near-nadir imagery acquired at high solar elevation flattens shadows and reduces detection sensitivity for shallow features. Tasking requests for looting surveys should specify a sun elevation of 30–50 degrees where operationally possible.
Multispectral band ratios can help separate fresh disturbed soil from background, but performance degrades when the surrounding surface is already sandy or pale. At sites in the Syrian steppe or southern Iraq, where the natural surface is light-coloured alluvium, spectral contrast between pit spoil and background can be very low. In those conditions, texture and shape cues in the panchromatic band carry most of the discriminative weight.
Cloud cover, seasonality, and the humid-season gap
Most of the heavily looted sites in the public record sit in arid or semi-arid zones where cloud cover is rarely the binding constraint. Syria, Iraq, and Egypt have cloud-free seasons lasting six months or more, which is why optical monitoring has been practical there.
The picture changes at sites in the Sahel, sub-Saharan Africa, or South and Southeast Asia. During monsoon or wet seasons, persistent cloud cover can block optical acquisition for weeks at a time. SAR sensors such as Sentinel-1 are unaffected by cloud, but at 5–20 m resolution they cannot resolve individual pits. The practical consequence is a monitoring gap: SAR can flag large-scale surface change but cannot confirm pit morphology. Analysts working in humid environments must plan for this gap explicitly, scheduling optical acquisitions in the dry season and accepting that wet-season looting may go undetected until conditions improve.
Tasking latency is a separate issue. Commercial VHR satellites are in high demand, and conflict zones are sometimes subject to collection restrictions. Archive imagery may have gaps of weeks or months precisely in the periods of highest looting risk. Any monitoring programme should audit archive coverage before committing to a change-detection baseline.
From pixel counts to legal and policy use
Satellite-derived pit counts have entered legal and policy discussions. The UNESCO-commissioned assessments of Syrian sites and the ASOR reports were cited in UN Security Council deliberations and informed the 2015 UN Security Council Resolution 2347 on the protection of cultural heritage in conflict. That path from pixel to policy requires the analytic outputs to be reproducible, documented, and accompanied by honest uncertainty statements.
Deliverables for this use case therefore need to include not just a pit-count map but a confidence classification: high-confidence pits (clear circular morphology, strong spectral contrast, area above the detection threshold), medium-confidence candidates (shape or spectral cue present but not both), and rejected detections. The spatial precision of pit centroids should be reported alongside the positional accuracy of the imagery, which for orthorectified WorldView products is typically 3–5 m CE90 without ground control.
Satellize can structure this workflow for heritage agencies or legal teams that need defensible, reproducible outputs rather than a one-off map. The analytical approach mirrors the crop-estimation methodology we run for the Kingdom of Tonga: systematic, documented, and built to be re-run as new imagery arrives. Enquiries about site-specific monitoring scope are the right starting point.
Typical figures
| Best available spatial resolution (panchromatic) | 0.31 m (WorldView-3); 0.46 m (WorldView-2); 0.5 m (Pléiades-1, SkySat) |
| Minimum reliably detectable pit diameter | Approximately 1–2 m under good soil contrast; sub-1 m pits are ambiguous at current commercial resolution |
| Multispectral bands relevant to soil disturbance | Red, NIR, and SWIR (WorldView-3 SWIR at 3.7 m resolution); red-edge useful for separating vegetation from bare soil |
| Typical revisit (tasked commercial VHR) | 1–4 days (WorldView constellation); same-day possible with SkySat on priority tasking |
| Archive depth | WorldView-2 from 2009; WorldView-3 from 2014; Pléiades from 2012; SkySat from approximately 2014 |
| Positional accuracy (orthorectified, no GCP) | 3–5 m CE90 typical for WorldView and Pléiades standard products |
| Cloud-cover constraint | Optical only; arid sites (Syria, Iraq, Egypt) have 6+ month cloud-free seasons; humid-zone sites may have multi-week optical gaps |
| Stereo DSM vertical accuracy for pit-depth estimation | ±0.3–0.5 m under good contrast; degrades in shadow and low-contrast terrain |
| Sentinel-2 role | 10 m free baseline for large-area triage; cannot resolve individual pits |
Analytics Satellize can run
| Baseline pit inventory | Object-based image analysis (OBIA) on pre-crisis VHR image; segmentation by brightness, circularity, and texture; manual validation tier | GIS polygon layer of pre-crisis pit candidates with confidence classification (high / medium / uncertain) and attribute table of diameter and area |
| Change-detection pit count | Bitemporal spectral differencing and OBIA classification on co-registered image pairs; new bright objects flagged as candidate new pits | GIS layer of newly detected pits per acquisition epoch, with uncertainty bounds on total count and spatial density map (pits per hectare) |
| Looting-rate time series | Multitemporal stack of classified images; cumulative pit count and new-pit rate plotted per month or quarter | Time-series chart and tabular report suitable for legal or policy citation, with stated image-availability gaps noted explicitly |
| Spoil-heap volume estimate | Stereo DSM differencing (Pléiades tri-stereo or WorldView stereo) between pre- and post-looting acquisitions; volume computed from DSM difference raster | Per-pit volume estimate in cubic metres with stated vertical uncertainty; aggregated site-level disturbed volume figure |
| Spectral soil-disturbance index map | Band-ratio analysis (NIR/Red, SWIR contrast) to highlight fresh exposed sediment; calibrated against known pit locations | Raster layer of soil-disturbance probability, thresholded at analyst-defined confidence level; GeoTIFF and web-tile formats |
| Monitoring alert cadence plan | Archive gap analysis combined with cloud-climatology data to identify optimal tasking windows per season | Written tasking schedule with recommended acquisition dates, sun-elevation targets, and contingency for cloud or access denial |
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.