Urban sinkhole and karst subsidence detection
Sinkholes over dissolving limestone, gypsum and salt can swallow roads in moments, yet PS-InSAR reveals the slow bowl-shaped precursor deformation that precedes collapse. Satellite radar is a risk-screening tool, not a real-time alarm.
Sensors
- Sentinel-1 (C-band SAR, ESA): 6-day repeat at mid-latitudes with two satellites; ground range resolution approximately 5 × 20 m in IW mode. Free and global, making it the standard choice for city-wide PS-InSAR time-series. Phase coherence can degrade over vegetated or sandy surfaces, but dense urban reflector networks perform well.
- TerraSAR-X / TanDEM-X (X-band SAR, DLR/Airbus): Spotlight mode delivers approximately 1 m resolution; 11-day repeat, commercial tasking. X-band is more sensitive to millimetre-scale displacement and maintains coherence over shorter time intervals, which matters when precursor signals are subtle or the area of interest is small.
- COSMO-SkyMed (X-band SAR, ASI): Constellation of four satellites; revisit as short as 1 day under tasking. Stripmap mode at roughly 3 m resolution. Useful for rapid re-acquisition after a suspected collapse event or for high-priority urban corridors requiring frequent updates.
- RADARSAT-2 (C-band SAR, MDA): Ultra-fine mode reaches approximately 3 m resolution; 24-day repeat. C-band archive extends to 2007, enabling retrospective analysis to check whether precursor deformation predated known collapse events.
What dissolving rock looks like from orbit
Sinkholes form when soluble bedrock, limestone, gypsum or halite (rock salt), is gradually removed by groundwater, leaving a subsurface cavity that eventually loses its roof. The surface expression before catastrophic collapse is typically a shallow, roughly circular bowl of differential settlement, often only a few millimetres to a few centimetres deep and tens of metres across. That geometry is exactly what Persistent Scatterer InSAR (PS-InSAR) is built to detect.
PS-InSAR works by tracking the radar phase of individual bright, phase-stable reflectors, corners of buildings, road furniture, exposed rock, across a stack of SAR images acquired over months or years. Where the ground is subsiding, the two-way travel distance to a scatterer increases, shifting the phase. A time-series of phase measurements can resolve line-of-sight displacement to roughly 1 mm per year over a dense PS network. In a karst city, the pattern to look for is a localised cluster of scatterers all moving downward relative to their stable neighbours, forming the characteristic bowl shape that published studies associate with active dissolution.
What the published record actually shows
The Dead Sea shoreline is the most extensively documented karst-sinkhole environment in the satellite record. Dropping lake levels have exposed evaporite deposits to freshwater infiltration, producing hundreds of sinkholes since the 1980s. Studies using ERS, Envisat and Sentinel-1 data have detected precursory subsidence bowls of 5 to 30 mm cumulative displacement developing over periods of weeks to months before collapse. The spatial footprint of the precursor signal was typically 20 to 100 metres in diameter, which is resolvable with Sentinel-1 IW mode but only just, and is resolved more cleanly with X-band systems.
Spanish gypsum karst, particularly in the Ebro valley around Zaragoza, has been studied with Envisat and TerraSAR-X. Published work identified subsidence rates of several millimetres per year over known dissolution features, with localised acceleration preceding documented collapses. These studies also showed that not every dissolution feature produces a detectable surface signal before failure: some cavities propagate upward with no measurable precursor at all, a fundamental limit that any honest assessment must acknowledge.
Urban settings add a complication. Buildings themselves settle for reasons unrelated to karst: clay consolidation, groundwater abstraction, construction loading. Separating karst-driven deformation from background urban settlement requires geological context, a good geological map of soluble bedrock extent, and ideally borehole or geophysical data to constrain the subsurface.
The hard limit: InSAR cannot watch fast enough
Catastrophic sinkhole collapse, the kind that swallows a car or a building, can happen in seconds to minutes. Sentinel-1 revisits a given location every 6 days. TerraSAR-X, with commercial tasking, can achieve 1-day revisit, but that is still not a warning system in any operational sense. PS-InSAR is a precursor-screening tool: it identifies locations where the ground is already moving and flags them for ground investigation, geophysical surveys (microgravity, ground-penetrating radar, seismic refraction) that can characterise the subsurface cavity directly.
There is also a detection floor. Isolated sinkholes in areas with few PS candidates, parks, unpaved ground, sandy soils, will have sparse or absent scatterers. The method works best in dense built environments with hard surfaces. Where scatterer density drops below roughly 5 to 10 PS per km², time-series quality degrades significantly and small deformation signals become unreliable.
Building a monitoring programme that is actually useful
A credible urban karst-monitoring programme has three components. First, a geological screening layer: map the extent of soluble bedrock from existing geological surveys and overlay it with infrastructure criticality (hospitals, tunnels, water mains, heritage structures). This defines the priority zone for InSAR analysis and avoids wasting effort on areas underlain by stable granite.
Second, a PS-InSAR baseline: process the full Sentinel-1 archive for the city, which now extends to late 2014, to establish background deformation rates and identify any anomalies already in the record. Anomalies that have been accelerating for years are higher priority than stable areas. Third, an ongoing monitoring cadence: 6-day Sentinel-1 updates flagging any new bowl-shaped deformation clusters exceeding a defined threshold, with commercial X-band tasking triggered for high-priority anomalies to improve spatial resolution and confirm the signal.
Satellize structures this kind of layered programme for government clients, combining open-constellation processing with commercial tasking on client licence. The Tonga crop-estimation programme is a different domain, but the operational logic, open data as the backbone, commercial sensors for critical detail, applies equally here.
Frequency, archive depth and what they cost you
The Sentinel-1 archive is free and open, which makes retrospective analysis genuinely affordable. Processing a city-scale PS-InSAR time-series over a decade of acquisitions is computationally intensive but not prohibitive with cloud infrastructure. The main cost is analyst time: interpreting deformation anomalies in geological context, filtering out non-karst signals, and writing defensible risk assessments that a city engineer can act on.
Commercial X-band data (TerraSAR-X, COSMO-SkyMed) adds cost but improves spatial resolution by roughly a factor of five compared with Sentinel-1 IW mode. For a city with a known high-risk district, the economics usually favour X-band tasking over that district alongside Sentinel-1 coverage of the broader area. Archive depth for commercial sensors is shorter (TerraSAR-X data from 2007, COSMO-SkyMed from 2008) but still long enough to establish multi-year deformation trends.
Typical figures
| Sentinel-1 IW spatial resolution | ~5 m (range) × 20 m (azimuth); PS-InSAR localises scatterers to sub-pixel accuracy |
| TerraSAR-X Spotlight resolution | ~1 m; enables detection of smaller deformation bowls |
| Sentinel-1 revisit (two-satellite) | 6 days at mid-latitudes; 12 days with single satellite |
| COSMO-SkyMed minimum revisit | ~1 day under tasking (four-satellite constellation) |
| Minimum detectable displacement rate (PS-InSAR) | ~1 mm/year line-of-sight over dense PS networks; single-epoch precision ~5–10 mm |
| Sentinel-1 archive depth | Late 2014 to present (Europe/Middle East); varies by region |
| Radar frequency (Sentinel-1) | C-band, 5.405 GHz; TerraSAR-X / COSMO-SkyMed: X-band, ~9.6 GHz |
| Typical PS density required for reliable time-series | ≥5–10 PS/km²; dense urban cores typically 50–500 PS/km² |
| Deformation bowl diameter detectable | ~20–100 m from published Dead Sea studies; smaller features require X-band |
| Delivery formats | GeoTIFF displacement maps, GeoPackage PS point clouds, PDF risk-zone reports, GIS-ready shapefiles |
Analytics Satellize can run
| City-wide PS-InSAR deformation baseline | Persistent Scatterer InSAR time-series (StaMPS or similar open algorithm) applied to full Sentinel-1 archive | GeoPackage of PS points with velocity and time-series; PDF summary of anomalous zones ranked by rate and geological context |
| Karst-risk screening layer | Overlay of PS velocity anomalies with soluble-bedrock geology and infrastructure criticality index | GIS polygon layer classifying risk zones; tabular priority list for ground-investigation scheduling |
| Precursor bowl detection alert | Automated spatial clustering of PS points showing coherent downward velocity exceeding user-defined threshold, updated each Sentinel-1 cycle | 6-day alert feed (email or API) with coordinates, magnitude and time-series plot for each new anomaly |
| High-resolution confirmation acquisition | Commercial TerraSAR-X or COSMO-SkyMed tasking over flagged anomaly; short-baseline InSAR or offset tracking for rapid deformation check | Single-pair interferogram and displacement estimate within 48 hours of tasking; analyst interpretation note |
| Retrospective collapse attribution | PS-InSAR time-series analysis of archive data preceding a known collapse event to quantify precursor signal duration and magnitude | Technical report with displacement time-series, velocity maps and interpretation of precursor detectability |
| Ongoing quarterly deformation report | Rolling PS-InSAR update with change detection relative to baseline; flagging of accelerating scatterers | Quarterly PDF report and updated GIS layers; trend plots for each monitored priority zone |
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.