Archaeological site looting pit detection and progression monitoring
Illicit excavation leaves crater fields and spoil heaps visible in very-high-resolution optical imagery. Mapping pit density, disturbed area and progression over time gives heritage authorities and enforcement agencies the evidence base to act.
Sensors
- Maxar WorldView-3: 0.31 m panchromatic resolution; 8-band VNIR plus SWIR. At this resolution, individual pits of roughly 1 m diameter cast measurable shadows, enabling shadow-based depth estimation. Revisit at mid-latitudes is approximately 1 to 4.5 days depending on off-nadir tasking.
- Airbus Pléiades Neo: 0.30 m native panchromatic, 4-band multispectral at 1.2 m. Stereo and tri-stereo tasking supports surface-model differencing for volume estimation. Revisit is up to twice daily per satellite, with two satellites in the constellation.
- Planet SkySat: 0.50 m panchromatic, 1 m multispectral. Lower cost per revisit than Pléiades Neo or WorldView-3, making it practical for frequent monitoring of large site clusters. Video mode can confirm active digging activity.
- CORONA declassified archive (KH-4, KH-9): Film-return imagery from 1960 to 1972 at roughly 1.8 to 7.5 m ground resolution, now held by the USGS EROS archive. Provides pre-conflict and pre-looting baselines for sites in Syria, Iraq and Egypt that have no other pre-disturbance record.
What a looting pit looks like from orbit
A freshly dug pit in dry alluvial soil produces three detectable signals in sub-metre imagery: a dark circular or oval void, a pale spoil heap immediately adjacent (excavated sediment is brighter than the compacted surface), and a shadow whose length encodes depth when the sun angle is known. At WorldView-3 resolution, pits as small as 1 to 2 m in diameter are individually countable. At Pléiades Neo, stereo pairs allow surface-model differencing that approximates disturbed volume, though uncertainty grows quickly below roughly 0.5 m of relief.
The signal degrades in several conditions worth stating plainly. Vegetation cover, even sparse scrub, obscures pit edges. Sand infill after rain or wind can erase a pit's spectral contrast within weeks. Cloud is rare over Syria and Iraq but not over Mali, where the wet season runs from June to September and can interrupt monitoring for weeks at a time. Night digging, which is common in active conflict zones, leaves no thermal trace detectable by optical sensors and would require SAR coherence methods that carry their own interpretation ambiguity at this scale.
Pre-conflict baselines and the CORONA advantage
Quantifying looting requires knowing what was there before. For sites in Syria and Iraq, the CORONA programme's declassified imagery, shot between 1960 and 1972, is the only consistent pre-disturbance record predating both the 2003 Iraq War and the 2011 Syrian conflict. The USGS EROS archive holds this material and it is publicly accessible. EAMENA and ASOR Cultural Heritage Initiatives have both used CORONA systematically to establish pit-count baselines, then compared them against modern Maxar and Pléiades acquisitions to attribute new disturbance to specific conflict periods.
The practical limitation is registration accuracy. CORONA imagery was acquired from a moving film-return capsule, and geometric correction to match modern orthorectified imagery introduces positional error of several metres. Change detection therefore works best at the site level rather than the individual pit level when spanning CORONA to present-day imagery. For sites where Maxar or Pléiades imagery exists from before 2011, the baseline problem is much simpler.
Shadow-based depth estimation: what it can and cannot tell you
If the solar elevation angle at image acquisition time is known (it is recorded in metadata), the length of a pit's shadow in the image gives a geometric estimate of pit depth. The method is straightforward in principle. In practice, pit walls are rarely vertical, shadow edges are blurred by the point-spread function of the sensor, and the pit floor may be partially illuminated. Published work by researchers associated with EAMENA suggests depth estimates carry uncertainty of roughly plus or minus 30 to 50 percent at sub-metre resolution. That is enough to distinguish a shallow probe trench from a deep shaft, but not enough to produce reliable volume figures for legal proceedings without ground-truth confirmation.
Volume estimates derived from stereo digital surface models are more defensible, but stereo collection must be planned in advance and adds cost. For monitoring programmes covering dozens of sites, the practical approach is to use shadow depth as a screening indicator and reserve stereo tasking for sites where legal documentation is the goal.
Mapping progression: how a site dies over time
Single-date pit counts describe a state. Time-series analysis describes a process, and the process matters for enforcement prioritisation. A site with 200 pits that has not changed in three years is a different problem from one with 20 pits that doubled last month. Change detection between sequential acquisitions can be done manually by trained analysts or semi-automatically using image differencing on the panchromatic band, with new bright spoil heaps and new dark voids flagged as candidate detections for human review.
ASOR's published monitoring of sites in Syria during the 2013 to 2017 period demonstrated that looting intensity correlated with shifts in territorial control, with some sites showing near-complete cessation of new pit formation when armed groups with an interest in revenue extraction consolidated control. That pattern matters analytically: it means progression monitoring can sometimes distinguish opportunistic looting from organised extraction, which has implications for sanctions and cultural-property crime investigations.
Revisit frequency is the binding constraint for progression monitoring. WorldView-3 and Pléiades Neo can in principle revisit a site daily, but tasking costs make weekly or fortnightly cadence more realistic for large site clusters. Planet SkySat's lower per-image cost makes it a practical option for frequent screening across a wide area, with higher-resolution assets tasked only when SkySat detects new activity.
What the data cannot settle on its own
Satellite imagery establishes that disturbance occurred, approximately when it occurred, and at roughly what scale. It does not establish who dug the pits, what was removed, or where objects went. Attribution to specific actors requires intelligence fusion that imagery alone cannot supply. Courts and cultural-property prosecutors have used satellite evidence from ASOR and EAMENA as corroborating documentation, not as primary proof of individual criminal acts.
There is also an interpretive ambiguity that honest analysts acknowledge: not every pit in an archaeological landscape is a looting pit. Agricultural planting holes, irrigation features, animal burrows and old archaeological excavation trenches can all produce similar spectral signatures at sub-metre resolution. Ground-truth validation, or at minimum comparison against known site records from national antiquities authorities, is necessary before pit counts are treated as looting counts. Satellize's analytics workflows for this use case incorporate that validation step explicitly, drawing on open site databases where available, as the team does in its crop-estimation work for the Kingdom of Tonga.
Building a monitoring programme that holds up to scrutiny
A defensible monitoring programme combines three things: a documented baseline, a consistent acquisition protocol, and an analysis methodology that can be described and replicated. The baseline should use the best available pre-disturbance imagery, whether CORONA, early Maxar, or national archive material. The acquisition protocol should fix sensor, sun-angle constraints and off-nadir limits so that images are geometrically comparable across time. The analysis methodology should specify whether pit detection is manual, semi-automatic or fully automated, and should report inter-analyst agreement rates.
For organisations building the legal record needed to support cultural-property crime prosecutions or UNESCO reporting obligations, the chain of custody for imagery metadata matters as much as the imagery itself. Acquisition certificates from Maxar and Airbus are available and should be retained alongside the imagery. That is not a technical point; it is an evidentiary one.
Typical figures
| Best optical resolution (commercial) | 0.30 m (Pléiades Neo), 0.31 m (WorldView-3) |
| Minimum detectable pit diameter | Approximately 1 to 2 m at 0.31 m resolution; larger pits more reliably mapped |
| Revisit cadence | Up to daily (Pléiades Neo, two satellites); 1 to 4.5 days (WorldView-3 off-nadir); near-daily (SkySat) |
| Archive depth | CORONA from 1960; Maxar commercial archive from approximately 2001; Pléiades from 2012 |
| Spectral bands | Panchromatic plus 4 to 8 VNIR bands (WorldView-3 adds SWIR); no thermal on these sensors |
| Shadow depth estimation uncertainty | Approximately ±30 to 50% of estimated depth at sub-metre resolution |
| Cloud limitation | Optical only; cloud blocks acquisition. Wet-season gaps of weeks possible in Sahel (Mali). SAR not covered on this page. |
| Stereo surface-model accuracy | Pléiades Neo tri-stereo: vertical accuracy approximately 0.3 to 0.5 m in open terrain |
| Delivery formats | Orthorectified GeoTIFF, pit-count GeoJSON, change-detection raster, PDF site report |
| Latency (tasked acquisition to delivery) | Typically 24 to 72 hours after image acquisition for commercial providers |
Analytics Satellize can run
| Pit density map | Manual or semi-automated pit delineation on panchromatic imagery; density calculated per hectare across site polygon | GeoJSON layer of individual pit centroids and polygons, plus density raster; site-level summary table in PDF report |
| Disturbed-area estimate | Binary change mask (disturbed vs undisturbed) derived from image differencing or supervised classification against baseline imagery | Raster mask and total disturbed-area figure in hectares, with confidence interval based on analyst validation sample |
| Shadow-based depth screening | Geometric shadow-length measurement using recorded solar elevation angle from image metadata; applied per pit centroid | Attribute table appended to pit GeoJSON with estimated depth range and stated uncertainty band |
| Progression timeline | Multi-date image differencing at fixed intervals; new pit detections flagged and dated to the acquisition window in which they first appear | Time-series chart of cumulative pit count and new-pit rate per period; animated GIF or video for briefing use |
| Stereo volume model | Digital surface model differencing between pre- and post-disturbance stereo acquisitions (Pléiades Neo tri-stereo); volume computed from depth raster | Difference DSM raster, per-pit volume estimates, site-total disturbed volume in cubic metres with stated vertical-accuracy caveats |
| Priority site alert | Automated screening of new SkySat acquisitions for spectral change exceeding threshold within registered site polygons; human review before alert dispatch | Email or API alert with thumbnail, coordinates and change-magnitude score; links to full imagery for analyst review |
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.