Underground facility detection from surface signatures
Subsurface military construction leaves measurable marks above ground. Combining InSAR deformation mapping with high-resolution optical analysis of spoil heaps, shaft arrays and access infrastructure lets analysts infer underground activity before a facility becomes operational.
Sensors
- Sentinel-1 SAR (IW mode): C-band SAR at 5 cm wavelength; IW mode delivers 5 x 20 m ground resolution across a 250 km swath. Repeat-pass InSAR with a 6-day revisit (12-day for a single satellite) can resolve ground deformation to sub-centimetre precision, detecting the subtle surface settling that overlies underground excavation. Free and open archive from 2014.
- Maxar WorldView-3: Panchromatic resolution of 0.31 m and 8-band multispectral at 1.24 m. Sufficient to resolve individual ventilation shaft collars, spoil-heap texture and construction vehicle tracks. Tasking latency is typically 1–3 days; archive extends to 2014.
- Planet SkySat: 0.5 m panchromatic, up to 15 frames per day over a target. Useful for monitoring construction tempo: counting spoil-removal truck movements and tracking day-on-day changes in heap volume. Less archival depth than WorldView-3 but higher revisit for active monitoring.
- Airbus Pléiades Neo: 0.30 m native resolution, tri-stereo collection available. Stereo pairs allow digital surface model generation, enabling volumetric estimation of spoil heaps, which correlates loosely with excavated tunnel volume. Revisit of up to twice daily at mid-latitudes.
What the surface cannot conceal
Underground military construction is designed to be invisible. The facility itself is. The construction process is not. Boring or blasting through rock produces spoil: excavated material that must go somewhere. A facility large enough to shelter command infrastructure, missile storage or weapons production generates tens of thousands of cubic metres of spoil. That material piles up on the surface, and its geometry, colour and texture are readable from orbit.
Beyond spoil, underground facilities require air. Ventilation shaft arrays are a reliable indicator: they appear as small circular or rectangular concrete collars, often in linear or grid arrangements on hillsides above the main entrance. Power is a second indicator. Underground facilities of any scale draw significant electrical load, so new high-tension line extensions or transformer stations appearing near otherwise uninhabited terrain are worth examining. Access roads are a third. Hardened underground sites typically require roads built to carry heavy loads, with wider turning radii and reinforced surfaces visible in high-resolution optical imagery.
InSAR and the weight of excavation
Interferometric SAR measures the phase difference between two radar passes over the same ground. When the ground moves, even by millimetres, the phase shifts. Underground excavation removes material and reduces the mechanical support beneath the surface, causing slow subsidence above the cavity. Published studies of known facilities, including documented analysis of North Korean sites at Punggye-ri and construction at Iranian hardened sites, have demonstrated that InSAR time-series can detect deformation patterns consistent with subsurface voids forming over weeks to months.
Sentinel-1's 6-day revisit makes it practical to build dense time-series stacks. Persistent Scatterer InSAR (PSInSAR) or Small Baseline Subset (SBAS) processing of those stacks can isolate deformation signals at the centimetre-to-millimetre level, separating genuine ground movement from atmospheric noise. The spatial footprint of the deformation, its rate of change, and its correlation with surface construction activity collectively inform probability judgements. This is not a depth sounder. InSAR tells you the ground moved; it does not tell you how far down the cavity is or what is inside it.
Spoil heap geometry as an excavation proxy
Stereo optical imagery from Pléiades Neo or WorldView-3 allows analysts to generate digital surface models accurate to roughly 0.5 m vertically under good contrast conditions. Differencing two such models taken weeks apart gives a volumetric spoil-accumulation rate. That rate, combined with published empirical relationships between tunnel cross-section and spoil volume, gives a rough estimate of the excavation rate and, over time, total tunnel length.
The word 'rough' deserves emphasis. Rock density varies. Spoil is sometimes removed from site by truck, which means the visible heap understates total excavation. Trucks can be counted in high-revisit SkySat imagery, but only during daylight and only when cloud cover permits. The method produces a probability range, not a blueprint. Open-source analysts at groups including the James Martin Center for Nonproliferation Studies and 38 North have published exactly this kind of multi-indicator assessment for North Korean facilities, and their methodology is the documented baseline from which satellite-data analysis in this domain starts.
Honest limits of the surface-signature method
The method is inferential throughout. No combination of optical resolution and InSAR precision can determine facility function, depth, or the nature of what is being stored or produced underground. An analyst can say, with reasonable confidence, that significant underground construction is occurring at a given location. Saying what the construction is for requires additional intelligence that satellite data alone cannot supply.
Cloud cover is a persistent constraint in certain regions. Sentinel-1 is unaffected by cloud, but optical confirmation of surface features requires clear conditions. At high latitudes or in monsoon climates, optical revisit may be functionally much lower than the nominal figure. Spoil can be removed at night or dispersed to reduce its signature. Ventilation shafts can be camouflaged with netting or vegetation. Sophisticated state actors are aware of satellite collection schedules and can time surface activity accordingly, though this requires operational discipline that is itself costly to sustain. The method is most reliable when multiple independent indicators converge: deformation, spoil, ventilation infrastructure and power-line extension all pointing to the same location.
Building an indicator framework
Practical analysis of underground facility signatures works through a tiered indicator framework. Tier one indicators are detectable at moderate resolution and low cost: new access roads, power-line extensions, and changes in surface reflectance consistent with excavated material. These trigger collection of tier two data: high-resolution optical tasking and InSAR baseline establishment. Tier three is time-series analysis: volumetric spoil accumulation, deformation rate mapping, and construction tempo assessment from repeated optical passes.
Satellize structures analytics for clients in this domain around exactly this kind of layered tasking logic, drawing on open Sentinel-1 archive for deformation baselines and adding commercial optical tasking when indicators warrant it. The approach is similar in principle to the crop-estimation programme Satellize runs for the Kingdom of Tonga, where open-data baselines are combined with targeted commercial collection to answer a specific question with appropriate precision. The question here is harder, and the honest answer is always probabilistic.
What a finished assessment looks like
A completed underground facility surface-signature assessment is a geospatial report that maps each observable indicator, assigns a confidence level to each, and synthesises them into an overall probability judgement about whether significant underground construction is occurring. It includes an InSAR deformation map with displacement time-series plots, annotated optical imagery identifying shaft arrays, spoil extent and access infrastructure, a spoil volume estimate with stated uncertainty bounds, and a construction tempo assessment based on change detection across the archive.
What it does not include is a facility schematic, a depth estimate, or a function attribution. Those claims would require data the method cannot provide. A credible assessment states its limits as clearly as its findings. Buyers who want certainty about what is underground should understand that no unclassified remote-sensing method delivers it. What the method does deliver is a well-evidenced, reproducible basis for deciding whether a location warrants closer attention through other means.
Typical figures
| Optical resolution (commercial) | 0.30 m (Pléiades Neo, WorldView-3 pan); 0.50 m (SkySat) |
| SAR resolution (Sentinel-1 IW) | 5 m range x 20 m azimuth; 250 km swath |
| InSAR deformation sensitivity | Sub-centimetre (millimetre-scale with PSInSAR stacking over multiple passes) |
| Revisit (SAR deformation monitoring) | 6 days (two-satellite Sentinel-1 constellation); 12 days single satellite |
| Revisit (optical, commercial tasking) | 1–2 days (SkySat, Pléiades Neo); 1–3 days tasking latency (WorldView-3) |
| Spoil volume estimation accuracy | Roughly ±15–30% depending on stereo geometry, contrast and removal activity |
| Minimum detectable surface feature | Ventilation shaft collar approx. 1 m diameter detectable at 0.30 m resolution; spoil heaps from approx. 500 m² area |
| SAR archive depth | Sentinel-1A from April 2014; commercial SAR (Capella, ICEYE) from 2019 |
| Optical archive depth | WorldView-3 from 2014; Pléiades Neo from 2021; SkySat from 2016 |
| Deliverable formats | GeoTIFF deformation maps, annotated JPEG/PNG optical composites, GeoJSON indicator layers, PDF assessment report |
Analytics Satellize can run
| InSAR deformation time-series | PSInSAR or SBAS processing of Sentinel-1 IW stack; atmospheric correction via ERA5 or GACOS | GeoTIFF displacement maps with per-pixel time-series plots; anomaly alert when deformation rate exceeds defined threshold |
| Spoil heap volumetric change | Stereo DSM differencing from Pléiades Neo or WorldView-3 stereo pairs; change detection against baseline DSM | Volume estimate with uncertainty range in PDF report; GeoJSON polygon of heap extent per epoch |
| Surface indicator mapping | Object-based image analysis (OBIA) on high-resolution optical imagery; manual expert annotation for ambiguous features | Annotated optical composite with georeferenced indicator points (shafts, roads, power lines); GIS layer |
| Construction tempo assessment | Multi-date change detection on SkySat or Planet daily imagery; vehicle count and track-density analysis | Activity timeline chart; alert feed triggered by new surface disturbance events |
| Multi-indicator probability assessment | Structured analytic technique combining deformation, spoil volume, surface infrastructure and power-line indicators; confidence-weighted scoring | Formal written assessment with stated confidence levels and indicator provenance; suitable for intelligence briefing format |
| Baseline establishment for new sites | Archive trawl of Sentinel-1 and open optical (Sentinel-2, Landsat 8/9) to establish pre-activity surface state; change detection relative to baseline | Baseline report with pre-construction DSM and SAR coherence map; archive access to 2014 |
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.