CORONA and KH-4 declassified imagery baseline for site-loss quantification
Declassified CORONA and KH-4 reconnaissance imagery, freely archived by the USGS, gives archaeologists a pre-disturbance baseline stretching back to 1960. Co-registered against modern sensors, it makes site loss measurable rather than anecdotal.
Sensors
- CORONA KH-4 / KH-4A / KH-4B (USGS archive): Panchromatic film returned from orbit 1960–1972. Ground resolution approximately 1.8 m for KH-4B at nadir after scanning; actual usable detail varies with film grain, scan quality and obliquity. Stereo pairs available for many frames. Declassified 1995 and freely downloadable via USGS Earth Explorer.
- WorldView-2 / WorldView-3 (Maxar): Panchromatic resolution 0.31 m (WV-3) and 0.46 m (WV-2) at nadir; 8-band multispectral at 1.24 m and 1.85 m respectively. The primary modern comparator sensor for sub-metre change detection at individual sites. Tasked commercially; archive coverage of heritage regions variable.
- PlanetScope: Daily 3 m four-band (RGBNIR) global coverage. Revisit of roughly once per day in most regions. Useful for monitoring active disturbance at lower cost than tasked commercial imagery; resolution insufficient for small pit detection but adequate for field-boundary encroachment and mound footprint change.
- Sentinel-2 MSI (ESA Copernicus): 10 m visible and NIR bands, 20 m red-edge and SWIR, 5-day revisit at the equator (2–3 days with both satellites). Free and open. Appropriate for landscape-scale site-density change across large study areas; not suited to resolving individual features smaller than roughly 20–30 m.
What the spy satellites actually captured
The CORONA programme was a United States strategic reconnaissance effort that ran from August 1960 to May 1972. Its primary purpose was Soviet missile-site mapping. Its accidental gift to archaeology was systematic photographic coverage of the Near East, Central Asia, North Africa and parts of South Asia at a moment before the agricultural intensification, urban expansion and conflict of the late twentieth century had reshaped the landscape.
The KH-4 series, and its successors KH-4A and KH-4B, returned film canisters by parachute for mid-air recovery. KH-4B achieved a nominal ground resolution of approximately 1.8 m. That is coarser than a modern commercial sensor, but it is fine enough to resolve tell mounds, enclosure ditches, field boundaries, road networks and the absence of modern disturbance. When the imagery was declassified in 1995 under Executive Order 12951, roughly 860,000 frames became available. The USGS Earth Explorer portal now serves them freely. No subscription required.
Why geometric correction is the whole problem
A CORONA frame is not a map. It is a scanned strip of film exposed through a rotating panoramic camera on a moving, slightly unstable platform. Radial distortion, Earth curvature, terrain relief and imprecise orbital parameters all warp the geometry. Before any pixel-level comparison with a modern image is meaningful, that distortion must be removed to sub-pixel accuracy.
The standard workflow involves identifying ground control points, GCPs, visible in both the historical and modern image. Stable, sharp-cornered features work best: road intersections, ancient stone structures, irrigation canal junctions. Rubber-sheet polynomial warping or, better, rigorous rational polynomial coefficient correction is applied, and the result is resampled to a common grid. Published studies in the Near East, notably work across Syrian and Iraqi sites documented before and after conflict, have achieved co-registration residuals of under one pixel at the CORONA native resolution, roughly 1.5–2 m. Below that threshold, genuine change can be separated from geometric artefact. Above it, you are measuring noise.
Quantifying loss: what the differencing actually shows
Once two epochs are co-registered, change is computed by image differencing, normalised difference of a suitable index, or supervised classification of surface type in each epoch followed by transition-matrix analysis. The choice depends on what is being measured.
For mound volume loss caused by agricultural ploughing, the most informative approach combines image differencing with a digital surface model from the modern epoch. The CORONA image establishes the original mound footprint and, where stereo pairs exist, a rough historical surface. Modern stereo DSMs from WorldView or Pleiades give the current surface. The volume deficit is the integral of the elevation difference across the mound area. For field-boundary encroachment, binary classification of cultivated versus uncultivated land in each epoch, followed by a simple transition count, gives hectares lost per decade. This method has been applied systematically across the Fertile Crescent. Studies published in the journal Remote Sensing and in the Journal of Archaeological Science have documented losses of 30–70% of mound area at sites in Iraq and Syria between the CORONA era and the 2000s, with the pace accelerating after 2011.
Looting pits are a harder target. CORONA resolution is generally insufficient to resolve individual pits smaller than a few metres. That detection problem belongs to sub-metre modern sensors and is covered separately in the looting-pit monitoring page. What CORONA does resolve is the pre-disturbance state of the site surface, confirming whether pits visible in modern imagery are genuinely new or were already present decades ago.
Honest limits of the archive
CORONA coverage is not uniform. Frame density varies by the programme's strategic priorities, which were Soviet, not archaeological. Some regions of high heritage significance have sparse coverage; others were photographed multiple times in a single year. Cloud cover at the moment of acquisition is permanent: if the frame is cloudy, there is no second chance.
Scan quality varies. Early digitisation runs produced lower-dynamic-range scans than later efforts. The USGS has progressively rescanned frames at higher radiometric quality, but not every frame in the archive has been rescanned. Users should check scan generation before assuming radiometric comparability across frames.
The 1.8 m nominal resolution of KH-4B is a best-case figure. Oblique frames, poor film development, or atmospheric haze during acquisition can degrade effective resolution to 3–5 m or worse. And the archive ends in 1972. For sites in regions that experienced major agricultural transformation in the 1950s, the CORONA baseline may already post-date significant loss. Earlier coverage from KH-1 through KH-3 exists but at coarser resolution and sparser coverage.
Building a site-loss time series across multiple epochs
CORONA provides the 1960–1972 baseline. Landsat's archive, beginning with Landsat 1 in 1972 and continuous through Landsat 8 and 9 today, extends the time series at 30 m resolution. That is too coarse for individual site analysis but useful for landscape-scale disturbance trends. Sentinel-2, from 2015, adds 10 m multispectral coverage with a consistent radiometric calibration. Commercial sub-metre imagery from WorldView, Pleiades or PlanetScope fills the gap for priority sites requiring fine spatial detail in recent years.
A properly constructed time series therefore has at least three nodes: the CORONA baseline, a mid-period Landsat assessment, and a current high-resolution image. Change rates can be computed between any pair of nodes, and the acceleration or deceleration of loss between periods is often as informative as the total loss figure. A site losing 2% of its area per year in the 1980s and 8% per year since 2010 tells a different policy story from one with a steady 3% throughout.
Satellize has applied multitemporal change-detection pipelines operationally, including for the Kingdom of Tonga crop-estimation programme where seasonal agricultural change is tracked across small island landmasses. The same co-registration and differencing logic transfers directly to archaeological site monitoring, with the CORONA frame substituting for the earliest Landsat epoch.
From pixel counts to evidence a heritage authority can use
A change-detection raster is not, by itself, a policy instrument. Converting it into one requires three additional steps. First, the change polygons must be attributed: is a given area of mound loss attributable to ploughing, to urban construction, or to deliberate levelling? Spectral signatures, texture analysis and contextual reasoning from ancillary data all contribute to that attribution. Second, the loss must be expressed in units that non-specialist decision-makers understand: hectares, percentage of original site area, estimated cubic metres of archaeological deposit removed. Third, the analysis must be reproducible and documented to a standard that will survive legal or diplomatic scrutiny.
National heritage authorities, UNESCO advisory bodies and international legal frameworks increasingly require quantified evidence of site condition change. Satellite-derived metrics, when properly georeferenced and methodologically documented, meet that evidentiary bar in a way that field observation alone cannot, particularly for sites in conflict zones or jurisdictions where access is restricted. The CORONA archive, precisely because it predates almost all modern disturbance agents, is the closest thing archaeology has to a controlled pre-treatment observation.
Typical figures
| CORONA KH-4B nominal ground resolution | Approximately 1.8 m at nadir; effective resolution 2–5 m depending on frame obliquity, scan quality and atmospheric conditions at acquisition |
| CORONA archive temporal range | August 1960 to May 1972; KH-4 series operational 1962–1972 |
| Modern comparator resolution (WorldView-3) | 0.31 m panchromatic, 1.24 m 8-band multispectral at nadir |
| Modern comparator resolution (Sentinel-2) | 10 m visible/NIR bands; 20 m red-edge and SWIR; 5-day revisit (2–3 days with both satellites) |
| Co-registration accuracy achievable | Sub-pixel (under 1.5–2 m residual) with adequate GCPs and rigorous correction; degrades with sparse GCPs or oblique frames |
| Minimum detectable surface change | Approximately 4–9 m² at WorldView resolution post-registration; landscape features above 0.1 ha detectable in CORONA-to-Sentinel-2 comparisons |
| CORONA archive access | Free download via USGS Earth Explorer; no licence fee; scanned at up to 7 micron pixel pitch in recent rescanning runs |
| Spectral coverage | CORONA: panchromatic only. WorldView-3: panchromatic plus 8 VNIR and 8 SWIR bands. Sentinel-2: 13 bands 443–2190 nm |
| Stereo availability | CORONA stereo pairs available for many frames; base-to-height ratio approximately 0.3–0.4, giving vertical precision of roughly 5–15 m from historical stereo DSMs |
| Delivery formats | GeoTIFF change rasters, attributed vector polygons (GeoPackage / Shapefile), quantitative loss tables (CSV), PDF technical report with methodology appendix |
Analytics Satellize can run
| Co-registered CORONA / modern image pair | Rational polynomial coefficient orthorectification with GCP-based refinement; resampling to common UTM grid | GeoTIFF stack with CORONA and modern image in identical projection and pixel alignment, residual error report |
| Site footprint change polygon set | Binary land-cover classification (disturbed / undisturbed) in each epoch followed by transition-matrix differencing | Vector polygon layer attributed with area lost (ha), percentage of original extent, and disturbance category (agricultural, urban, other) |
| Mound surface-area loss time series | Multitemporal classification across CORONA, Landsat and Sentinel-2 epochs; area statistics extracted per site ID | CSV table of area by epoch per site; chart-ready output for inclusion in heritage authority reports |
| Estimated volume loss at priority tells | Differencing of historical CORONA stereo DSM against modern WorldView or Pleiades stereo DSM; integration of elevation deficit over mound footprint | Raster of elevation change (m) and scalar volume-loss estimate (m³) with confidence interval derived from co-registration residual |
| Disturbance attribution classification | Spectral and textural analysis of change polygons using WorldView 8-band or Sentinel-2 data; rule-based attribution to agricultural tillage, construction, or unclassified | Attributed GIS layer with disturbance type field; summary table by category for reporting to UNESCO or national authority |
| Multi-epoch loss rate report | Linear and piecewise regression of area loss against time across all available epochs; identification of acceleration periods | PDF technical report with methodology, epoch-by-epoch statistics, rate-change analysis and map figures; suitable for legal or diplomatic submission |
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.