Stereo satellite DSM time series for tell volume loss quantification
Sub-metre stereo satellite imagery converts archaeological tell damage from anecdote into auditable volume. DSM differencing across epochs measures what was removed, redeposited, or bulldozed across the entire mound, not just the pits a human analyst can count.
Sensors
- Pleiades 1A / 1B (Airbus): 50 cm panchromatic GSD in nanosatellite stereo mode; tri-stereo tasking available. Along-track stereo pairs acquired in a single pass, minimising shadow and illumination inconsistency between the two images. Archive extends to 2012.
- WorldView-1 / WorldView-2 (Maxar): WorldView-1 delivers 50 cm panchromatic stereo; WorldView-2 adds eight multispectral bands at 46 cm pan. Both support in-track stereo collection. WorldView-2 archive reaches back to 2009, useful for pre-conflict baseline construction where CORONA gaps exist.
- CORONA KH-4 / KH-4A / KH-4B (declassified USGS): Film-return reconnaissance imagery from 1960 to 1972, scanned at approximately 1.8 m effective GSD for KH-4B. Stereo coverage is inherent in the fore-and-aft camera geometry. Provides the only systematic sub-5 m baseline for tells in Syria, Iraq and Iran predating large-scale looting and conflict.
- Pleiades Neo 3 / 4 (Airbus): 30 cm GSD panchromatic, the current commercial floor for routine stereo tasking. Resolves individual spoil piles as small as roughly 1 m across. Revisit at mid-latitudes is approximately daily when both satellites are tasked together.
- SPOT 6 / 7 (Airbus): 1.5 m pan GSD; useful as a lower-cost intermediate epoch between CORONA and modern sub-metre stereo. Archive from 2012. Insufficient alone to resolve micro-topography on small tells but adequate for mounds larger than roughly 5 ha.
Why volume, and why the whole mound
Counting looting pits from a single high-resolution image is the conventional approach, and it is systematically wrong in two directions. It undercounts because recent backfill, shadow and vegetation cover pits from nadir view. It also misses the category of damage that is often largest in absolute mass terms: mechanised bulldozing that shears off entire stratigraphic layers without leaving the circular signature of a clandestine shaft.
A DSM time series sidesteps both problems. By subtracting a surface model from one epoch against a co-registered model from another, the analyst obtains a signed difference raster. Every cell that has lost elevation contributes a negative volume; every cell that has gained, from spoil redistribution or aeolian deposition, contributes positive. The net integral across the mound gives a defensible cubic-metre figure that can be cited in legal proceedings, UNESCO reports or insurance valuations. The method does not require the analyst to interpret what caused the change, only to measure that it occurred.
What a floating roof gives away
Stereo photogrammetry from satellite imagery produces a digital surface model, not a bare-earth digital terrain model. On a tell, that distinction matters less than it does in forest: the mound surface is largely the archaeological surface. The main confound is vegetation, which can be sparse scrub on Mesopotamian tells or denser growth on Anatolian höyüks. A normalised difference vegetation index layer derived from the multispectral bands of WorldView-2 or Pleiades Neo allows vegetated cells to be masked or corrected before differencing.
The practical vertical accuracy of a well-controlled Pleiades stereo DSM, with ground control points derived from corner reflectors or identifiable stable features, is typically cited in the literature at 0.3 to 0.5 m RMSE under favourable conditions. That means volume changes of a few hundred cubic metres on a mound of moderate size are detectable; changes of tens of cubic metres are not. Honest reporting requires stating this floor explicitly. A 2 m deep clandestine shaft of 1 m diameter displaces roughly 1.6 cubic metres, which is below detection. A machine-excavated trench 20 m long, 3 m wide and 1.5 m deep displaces around 90 cubic metres, which is detectable if the surrounding surface is stable.
CORONA as the zero epoch
The USGS EROS archive holds declassified CORONA imagery covering much of the Middle East and Central Asia from 1960 onwards. The KH-4B fore and aft cameras, separated by roughly 30 degrees of convergence angle, produce stereo pairs from which DSMs can be extracted using standard photogrammetric software. Effective GSD after scanning is approximately 1.8 m for KH-4B, coarser than modern commercial sensors but sufficient to define the gross morphology of a tell: its height, basal extent and summit plateau.
The value of CORONA is not precision. It is chronology. A CORONA-derived DSM dated to 1967 over a Syrian tell that has since been bulldozed provides the only plausible pre-disturbance volume baseline. Differencing that against a 2015 Pleiades DSM captures five decades of net change. The co-registration challenge is real: CORONA film distortion and the absence of rational polynomial coefficients require careful bundle adjustment, but published workflows exist and the results are usable for mound-scale volume accounting even if cell-level accuracy is modest.
Building the time series: epochs, co-registration and change attribution
A useful time series typically anchors on three to five epochs: a CORONA baseline, one or two intermediate WorldView or SPOT acquisitions, and a current sub-metre stereo pair. Each DSM is produced independently, then co-registered to a common datum using stable off-mound reference surfaces, roads or field boundaries that have not changed between epochs. The differencing is done in sequence: CORONA to WorldView-2, WorldView-2 to Pleiades Neo, and so on. This reveals not just total loss but its rate and timing, which is often the analytically important question.
Attribution of change to cause, whether illicit digging, agricultural encroachment, erosion or conflict damage, requires ancillary evidence. The DSM difference alone cannot distinguish a bulldozed layer from a severe erosion event. Optical texture analysis of the contemporaneous panchromatic imagery, combined with knowledge of conflict chronology or reported site events, provides the attribution layer. The volume figure is objective; the cause label is interpretive and should be presented as such.
Limits the method cannot escape
Cloud cover is the most immediate operational constraint. Optical stereo requires clear skies at acquisition time, and parts of northern Syria and southern Turkey experience persistent winter cloud that can delay tasking by weeks. This is not a solvable problem with the sensors described here; it is a reason to task during the dry season and to accept that some epochs will have gaps.
The 0.3 to 0.5 m vertical RMSE figure assumes adequate ground control. In active conflict zones, deploying ground control targets is impossible, and the analyst must rely on tie-points to stable off-site features or on published geoid models. Accuracy degrades accordingly, sometimes to 1 m or worse. Volume estimates derived from poorly controlled DSMs should carry explicit uncertainty bounds, not point estimates. Finally, the method measures surface change only. Tunnelling from the side of a mound, a known technique at some sites, leaves the top surface intact and is invisible to this approach entirely.
From raster difference to evidential record
The output of a DSM differencing workflow is a GeoTIFF of signed elevation change, but that is not what a heritage ministry or international court needs. The deliverable is a report that states: the mound had an estimated volume of X cubic metres at epoch one; at epoch two it had Y; the net loss is Z, with an uncertainty of plus or minus W based on the co-registration residuals and the sensor vertical accuracy. Colour-coded change maps, cross-sections through the mound axis and tabulated epoch-by-epoch figures make the evidence legible to a non-specialist audience.
Satellize structures this workflow as a repeatable analytics pipeline, running stereo DSM generation and differencing on client-licensed commercial imagery and delivering georeferenced change layers alongside a written assessment. The Tonga crop-estimation programme demonstrated the same underlying principle of multi-epoch raster differencing for quantitative change accounting, applied to a very different domain. For heritage clients, the pipeline is adapted to mound morphology and the specific co-registration demands of long-baseline CORONA-to-modern comparisons. If you are managing a portfolio of at-risk tells and need a defensible volume-loss record before the next reporting cycle, the practical next step is to define the site list and the target epochs so that tasking windows can be planned around seasonal cloud climatology.
Typical figures
| Best available panchromatic GSD | 30 cm (Pleiades Neo); 46–50 cm (WorldView-1/2, Pleiades 1A/1B) |
| CORONA KH-4B effective GSD (scanned) | Approximately 1.8 m after film scanning; stereo inherent in fore/aft camera geometry |
| Typical DSM vertical accuracy (well-controlled) | 0.3–0.5 m RMSE for Pleiades/WorldView stereo; 1–3 m for CORONA-derived DSMs |
| Minimum detectable volume change | Approximately 50–200 m³ net change on a stable mound surface; smaller changes below noise floor |
| Revisit for new stereo tasking | Pleiades Neo: approximately daily at mid-latitudes (two satellites). WorldView constellation: 1–4 days depending on latitude and tasking priority |
| Archive depth (commercial stereo) | WorldView-2 from 2009; Pleiades 1A/1B from 2012; CORONA from 1960 (USGS EROS) |
| Cloud constraint | Optical stereo requires clear skies; persistent winter cloud in northern Levant and Anatolia can delay acquisition by weeks |
| Primary delivery formats | GeoTIFF DSM and difference raster; GeoPackage or Shapefile change polygons; PDF/georeferenced report with epoch-by-epoch volume table |
| Spectral bands used | Panchromatic for DSM generation; NDVI from multispectral bands (WorldView-2: 8 bands; Pleiades: 4 bands) for vegetation masking |
| Stereo convergence angle (typical) | In-track pairs typically 20–30° convergence; CORONA fore/aft approximately 30° |
Analytics Satellize can run
| Epoch-pair DSM difference raster | Dense image matching (semi-global matching or similar photogrammetric pipeline) applied to stereo pairs; co-registered by stable off-mound tie-points | Signed GeoTIFF elevation-change layer with per-cell uncertainty estimate |
| Mound volume-loss table | Spatial integration of negative-change cells within a mound boundary polygon; positive cells reported separately as redeposition | CSV and PDF table of cubic-metre loss and gain per epoch interval, with uncertainty bounds |
| CORONA-to-modern baseline comparison | Bundle-adjusted CORONA KH-4 stereo DSM co-registered to current sub-metre DSM; long-baseline differencing with documented co-registration residuals | Georeferenced change map and written assessment of decadal volume loss suitable for UNESCO or legal reporting |
| Vegetation-masked change analysis | NDVI thresholding on multispectral imagery to exclude vegetated cells from volumetric calculation; flagged separately in output | GeoTIFF with vegetation-masked change layer and coverage-quality metadata |
| Damage-rate time series | Sequential epoch differencing across three or more DSMs; annualised loss rate calculated per interval | Chart and data table of annualised cubic-metre loss rate; supports prioritisation of monitoring resources across a site portfolio |
| Morphological cross-section profiles | Extracted elevation profiles along user-defined transects through mound axis at each epoch | Vector profile lines and stacked elevation plots in PDF and DXF formats |
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.