Radar interferometry detection of subsurface tomb cavities and hypogea
Repeat-pass InSAR detects millimetre-scale surface strain caused by differential settling above shallow rock-cut tombs, catacombs and hypogea. Sentinel-1 and COSMO-SkyMed archives supply the temporal density needed to separate cavity signals from seasonal moisture noise.
Sensors
- Sentinel-1 (C-band SAR, ESA): 5.6 cm wavelength; Interferometric Wide Swath mode at 5 x 20 m resolution, 6-day repeat at mid-latitudes with both satellites active. Free archive from 2014 gives the long time series essential for distinguishing a persistent deformation signal from seasonal soil-moisture phase shifts.
- COSMO-SkyMed (X-band SAR, ASI): 3.1 cm wavelength; Spotlight mode reaches 1 m resolution, Stripmap at 3-15 m. Shorter wavelength improves sensitivity to sub-centimetre displacement but also increases decorrelation over vegetated or disturbed surfaces. Revisit can be tasked to 1-4 days for targeted monitoring campaigns.
- TanDEM-X (X-band SAR, DLR): Bistatic single-pass mode generates a global 12 m DEM (0.2 m relative vertical accuracy in flat terrain) used as a baseline topographic reference. Repeat-pass interferograms from the TDX archive can extend X-band coherence analysis back to 2010.
- Sentinel-1 Persistent Scatterer / SBAS processing: Not a separate sensor but a processing mode applied to the Sentinel-1 archive. PS-InSAR and SBAS algorithms extract per-pixel displacement time series at millimetre precision from stacks of 30 or more interferograms, which is the practical minimum for reliable cavity-signal isolation.
Why a hollow in the rock shows up in a radar phase map
A rock-cut tomb or catacomb removes material that would otherwise support the overburden. Even in limestone, the thin cap above a shallow hypogeum flexes under its own weight, thermal cycling and groundwater fluctuation. That flexure is tiny: fractions of a millimetre per season in competent rock, a few millimetres per year where the cap is thinner or the fill above is unconsolidated soil. Radar interferometry measures the round-trip phase difference between two passes of the same satellite. A displacement of half the radar wavelength (2.8 cm for C-band Sentinel-1, 1.55 cm for X-band COSMO-SkyMed) shifts the phase by a full cycle. Motions far smaller than that are detectable statistically when dozens of interferograms are stacked.
The signal of interest is a localised, persistent phase anomaly: a small bowl of subsidence, typically 5-30 m in diameter, that reappears consistently across seasons rather than tracking the broad spatial pattern of rainfall-driven soil swelling. That persistence is the discriminator. Seasonal moisture effects are spatially coherent across hundreds of metres; cavity-related strain is not.
What the physics rules out
This method has a hard geological ceiling. For surface strain to reach a detectable threshold, the overburden above the cavity must be compliant enough to deform. In massive, unjointed limestone or granite with more than roughly 8-12 metres of intact rock above the void, the strain at the surface falls below the noise floor of any current spaceborne SAR system. Published studies of known catacomb sites in Rome and Malta have detected signals where roof thickness was under approximately 5 metres and the fill above included soil or fractured rock. Deeper or harder-capped sites are effectively invisible to this technique.
Vegetation adds another constraint. Dense canopy decorrelates the interferogram entirely. C-band Sentinel-1 loses coherence over forest in 6-12 days; X-band decorrelates even faster. The method works best on bare soil, sparse scrub, low crops and paved urban surfaces. Sites in the Levant, North Africa, the Maltese archipelago and the Italian peninsula sit in this sweet spot geographically. Sites buried under tropical forest do not.
Urban thermal noise, construction vibration and traffic loading all produce millimetre-scale surface motion that can mimic or mask a cavity signal. Separating them requires spatial context: a known-void signal should be stationary in plan position across years, not migrating or correlated with construction activity nearby.
Building the time series: from raw interferograms to a deformation map
A single interferogram is rarely enough. Atmospheric water vapour introduces path-delay errors of up to several centimetres, dwarfing the cavity signal. The standard mitigation is to process a stack of at least 30 co-registered interferograms using either Persistent Scatterer InSAR (PS-InSAR) or Small Baseline Subset (SBAS) algorithms. PS-InSAR identifies pixels that maintain phase coherence across the full stack, typically hard surfaces, exposed rock or building corners, and models a displacement time series for each. SBAS uses short-baseline interferogram pairs to preserve coherence over softer ground. Both approaches reduce atmospheric noise by an order of magnitude relative to a single interferogram.
For a site with six years of Sentinel-1 coverage, a stack of 200 or more acquisitions is routine. That density allows detection of displacement rates as low as 1-2 mm per year in favourable conditions, which is within the expected range for shallow hypogea in semi-arid limestone terrain. Coherence maps derived from the same stack highlight areas where the ground surface has changed between passes, flagging recent disturbance or looter activity as a secondary product.
Calibrating against known voids before trusting the anomalies
Any credible InSAR prospection campaign begins with a calibration step: processing the archive over a site where the subsurface geometry is independently known from ground-penetrating radar, borehole logs or published excavation records. This establishes the local noise floor and confirms that the processing chain is sensitive to the expected displacement magnitude. Without this step, the rate of false positives from differential compaction, buried utilities and geological heterogeneity is too high to be operationally useful.
The Maltese hypogea and the Roman catacomb network are two publicly documented test environments where InSAR has been evaluated against known geometry. Results from those contexts suggest that the method reliably flags cavities with roof spans above roughly 3 metres and overburden below 5 metres in soft limestone, while producing few detectable signals over deeper or harder-capped voids. Those figures should be treated as indicative rather than universal: local geology dominates the outcome.
Combining C-band and X-band to reduce ambiguity
C-band and X-band respond differently to the same surface. Where both wavelengths show a coincident, persistent phase anomaly of consistent spatial extent, confidence in a real subsurface feature rises substantially. Where only one band shows a signal, the candidate is more likely to be a moisture or vegetation artefact. Running Sentinel-1 and COSMO-SkyMed stacks in parallel over a priority site is therefore worth the additional processing cost for any site where the stakes of a false positive are high.
TanDEM-X contributes a third layer: a high-precision topographic baseline that removes orbital and terrain phase from both stacks and can reveal micro-topographic depressions at the surface that correlate spatially with the InSAR anomaly. A 10-20 cm surface bowl above a known tomb, invisible to the eye on flat terrain, is often detectable in a TanDEM-X DEM differenced against a historical topographic survey. Satellize incorporates this multi-frequency stack approach in its heritage analytics workflow, alongside the crop-estimation methodology it has applied for the Kingdom of Tonga.
No combination of radar frequencies replaces ground investigation. InSAR prospection narrows the search area and prioritises fieldwork; it does not confirm the presence, date or contents of a void.
What a delivered product actually looks like
The output of an InSAR cavity-prospection campaign is a georeferenced displacement-rate raster, typically at 5-20 m posting depending on the sensor and processing mode, overlaid with a ranked anomaly point layer. Each candidate anomaly carries a displacement rate in mm per year, a coherence score, a spatial extent estimate and a confidence tier based on multi-frequency agreement and calibration-site performance. The layer is delivered in GeoTIFF and GeoPackage formats compatible with standard GIS environments.
A written interpretation report accompanies the data, stating explicitly which geological and land-cover conditions at the site support or undermine the InSAR approach, which anomalies are recommended for ground-truth prioritisation, and which areas are simply outside the detection envelope of the method. Honest reporting of the envelope is not a disclaimer; it is the part of the analysis that saves a client from commissioning unnecessary fieldwork in areas where radar physics cannot help.
Typical figures
| Spatial resolution (Sentinel-1 IW mode) | 5 x 20 m (range x azimuth); resampled to ~14 m ground range |
| Spatial resolution (COSMO-SkyMed Spotlight) | 1 m; Stripmap 3-15 m depending on mode |
| Minimum detectable displacement rate (PS-InSAR, favourable conditions) | 1-2 mm per year over stacks of 30+ interferograms; ~5 mm per year for single interferogram |
| Revisit period (Sentinel-1, both satellites) | 6 days at mid-latitudes; 12 days with single satellite |
| Radar frequency / wavelength | C-band 5.405 GHz / 5.6 cm (Sentinel-1); X-band ~9.6 GHz / 3.1 cm (COSMO-SkyMed, TanDEM-X) |
| Archive depth | Sentinel-1 from April 2014; COSMO-SkyMed from 2007 (tasked archive); TanDEM-X from 2010 |
| Overburden depth detection limit (indicative) | Roughly 0-8 m in compliant limestone or soil fill; deeper or harder rock is below detection threshold |
| Minimum cavity span (indicative) | Approximately 3 m roof span for detectable surface strain in soft limestone |
| Swath width (Sentinel-1 IW) | 250 km; enables landscape-scale survey in a single pass |
| Delivery formats | GeoTIFF displacement-rate raster, GeoPackage anomaly point layer, PDF interpretation report |
Analytics Satellize can run
| Displacement-rate map | PS-InSAR or SBAS processing of Sentinel-1 or COSMO-SkyMed stack (30+ interferograms); atmospheric correction via ERA5 or GACOS tropospheric delay models | Georeferenced GeoTIFF raster of line-of-sight velocity in mm per year, with uncertainty layer |
| Ranked cavity-candidate anomaly layer | Spatial clustering of persistent negative displacement pixels; multi-frequency agreement scoring between C-band and X-band stacks | GeoPackage point layer with displacement rate, spatial extent, coherence score and confidence tier per candidate |
| Coherence-change map | Interferometric coherence differencing between baseline and monitoring epochs to detect surface disturbance consistent with illicit excavation or collapse | GeoTIFF change layer flagging coherence-loss events, with timestamps and area estimates |
| Micro-topographic depression analysis | TanDEM-X 12 m DEM differenced against SRTM or historical survey; depression morphology compared spatially with InSAR anomaly positions | Co-registered DEM difference raster and overlay shapefile of corroborating surface depressions |
| Seasonal noise separation report | Time-series decomposition of per-pixel displacement into trend, seasonal and residual components; cavity signal isolated in residual layer | PDF report with time-series plots for each priority anomaly, showing seasonal versus persistent displacement components |
| Fieldwork prioritisation matrix | Multi-criteria scoring combining displacement rate, multi-frequency agreement, micro-topographic correlation and geological suitability index | Ranked table of candidate sites with recommended investigation method (GPR transect, coring or visual inspection) and estimated survey effort |
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.