InSAR measurement of urban ground subsidence
Interferometric SAR detects millimetre-scale urban subsidence from groundwater extraction, tunnelling and fill compaction. This page explains PS-InSAR and SBAS processing, their honest limits, and the satellite archives that make city-scale monitoring practical.
Sensors
- Sentinel-1 (C-band, 5.6 cm wavelength): Free archive from 2014. Interferometric Wide Swath mode covers 250 km at 5 × 20 m ground range resolution. Six-day repeat at mid-latitudes with both satellites active, giving dense temporal stacks essential for PS-InSAR. Phase coherence degrades over vegetated surfaces; urban hard targets perform well.
- COSMO-SkyMed (X-band, 3.1 cm wavelength): Italian Space Agency constellation of four satellites. Stripmap mode delivers 3 m resolution; Spotlight mode reaches 1 m. Shorter wavelength improves sensitivity to small displacements but loses coherence faster over soil or grass. Revisit can be as short as one day with tasking. Commercial archive from 2007.
- TerraSAR-X / TanDEM-X (X-band, 3.1 cm wavelength): German Aerospace Centre commercial system. Spotlight mode at 1–2 m resolution makes it the preferred choice for monitoring individual structures or narrow corridors. Staring Spotlight mode achieves sub-metre resolution. Archive from 2007. Tasking is licensed per acquisition.
- ALOS-2 PALSAR-2 (L-band, 23.6 cm wavelength): Longer wavelength maintains coherence over vegetated or loosely built areas where C- and X-band fail. Useful for peri-urban fringe monitoring or cities with significant tree cover. Stripmap resolution is 3–10 m; revisit is 14 days. Archive from 2014 via JAXA.
Why a city sinks, and why it matters
Urban ground subsidence has three dominant causes: groundwater extraction that removes pore pressure from aquifer systems, causing sediment compaction; tunnelling and underground construction that disturbs the stress field in surrounding soil; and the long-term consolidation of engineered fill placed on soft alluvial ground during rapid urban expansion. All three produce vertical displacement that is slow enough to be invisible to the naked eye and fast enough to crack foundations, misalign drainage gradients and invalidate engineering assumptions made at the time of construction.
The practical problem for infrastructure owners and city planners is that conventional monitoring, whether precise levelling surveys or point-based GPS networks, is expensive to maintain at city scale and produces sparse spatial coverage. A levelling campaign might measure hundreds of benchmarks across a metropolitan area; a Sentinel-1 PS-InSAR result over the same area can return hundreds of thousands of measurement points. The density difference is not cosmetic. It reveals spatial patterns, gradients between adjacent blocks and the footprint of individual extraction wells that point-based methods simply cannot resolve.
How interferometry turns phase into displacement
A SAR satellite transmits microwave pulses and records the two-way travel time and phase of the return signal. When two acquisitions over the same area are combined into an interferogram, the phase difference encodes the change in range between the sensor and the ground. One full phase cycle (2π) corresponds to half the radar wavelength of displacement along the line of sight. For Sentinel-1 C-band that is approximately 2.8 mm per fringe; for X-band systems it is roughly 1.55 mm. The theoretical sensitivity is sub-millimetre, though practical noise floors in urban PS analysis are typically cited at 1–2 mm per year for well-processed stacks of 30 or more scenes.
The atmosphere is the principal source of error. Tropospheric water vapour introduces path-length delays that can mimic or mask centimetre-scale displacement signals. Ionospheric effects are more pronounced at L-band. Both are mitigated by temporal averaging across large scene stacks, by using auxiliary weather model data (ERA5 reanalysis is commonly applied), or by exploiting the spatial correlation structure of atmospheric noise, which differs from the localised patterns of true ground motion.
PS-InSAR and SBAS: two approaches, different trade-offs
Persistent Scatterer InSAR identifies pixels whose radar backscatter is stable across the entire archive: metal structures, concrete corners, exposed rock faces. Each persistent scatterer is treated as a phase-stable reference point, and a time series of displacement is extracted for it. Scatterer density in a dense urban core can exceed 500 per square kilometre; in a suburban area with detached housing and gardens it may fall below 50, making the method unreliable. PS-InSAR is the standard choice where high spatial precision over hard infrastructure is the objective.
Small Baseline Subset (SBAS) processing takes a different approach. Rather than selecting individual stable pixels, it forms interferograms only between acquisitions with short spatial and temporal baselines, then inverts the network of phase measurements to produce a displacement time series for distributed scatterers. SBAS trades point-level precision for broader spatial coverage and performs better in peri-urban or partially vegetated zones. In practice, analysts often run both chains and compare results: agreement between them strengthens confidence; divergence flags either a processing artefact or a genuine localised signal worth investigating.
Spatial resolution in both methods is ultimately governed by scatterer density, not sensor pixel size. A 5 m SAR pixel does not automatically yield 5 m displacement maps. In areas of sparse scatterers, the effective spatial resolution of the displacement field degrades to 20 m or worse. This is the limit that matters for engineering decisions, and it should be stated clearly in any deliverable.
What the archive depth buys you
Sentinel-1A was launched in April 2014, Sentinel-1B followed in 2016 (though it failed in 2021, reducing revisit to 12 days at mid-latitudes). That archive now spans a decade. For subsidence monitoring, depth matters more than near-real-time latency: a ten-year velocity map separates genuine long-term trends from seasonal signals driven by rainfall recharge and temperature-induced thermal expansion of structures. Seasonal elastic deformation of aquifer systems can reach several centimetres annually in cities with large seasonal groundwater fluctuations; without multi-year stacks, it is easy to misinterpret a recharge recovery as subsidence arrest.
COSMO-SkyMed and TerraSAR-X archives extend to 2007 and 2008 respectively, offering pre-Sentinel baselines for cities where historical context is legally or contractually significant. Combining X-band historical data with a Sentinel-1 long-term stack requires careful co-registration and cross-calibration, but the technique is established in the published literature and extends the usable time series by six to seven years.
Honest limits: what InSAR cannot tell you
InSAR measures displacement along the satellite's line of sight, not purely vertical. Decomposing line-of-sight measurements into vertical and horizontal components requires acquisitions from both ascending and descending orbits, and even then the east-west component is recoverable only approximately; north-south motion is nearly invisible to polar-orbiting SAR. For subsidence monitoring this is usually acceptable, because vertical motion dominates, but in areas of lateral ground movement near tunnels or slip zones the interpretation requires care.
Coherence loss is the other hard limit. Fresh construction spoil, disturbed soil, water bodies and dense vegetation all depolarise the radar return, producing decorrelation that makes phase measurement impossible. A newly excavated construction site may be precisely the area of greatest interest and precisely the area where InSAR returns no usable signal. Analysts should flag decorrelated zones explicitly rather than interpolating across them. Persistent scatterers on the perimeter of a site can constrain the displacement field at the boundary, but the interior remains a gap.
Applying the analysis at city scale
A practical city-scale subsidence programme typically begins with a Sentinel-1 PS-InSAR baseline velocity map covering the full urban footprint, processed over the available archive. This identifies subsidence bowls, stable zones and any anomalous accelerations. High-velocity zones (commonly defined as greater than 10 mm per year in alluvial megacities) are then prioritised for higher-resolution COSMO-SkyMed or TerraSAR-X tasking to resolve individual structures. The two-tier approach balances cost against spatial detail.
Satellize runs this kind of tiered processing on open Sentinel-1 archives and adds commercial tasking where clients hold the appropriate licences. The analytical workflow is the same class of method used in the published literature on cities including Jakarta, Mexico City and Tehran, where subsidence rates exceeding 25 cm per year have been documented using C-band stacks. Deliverables are georeferenced GIS layers and structured reports, not raw interferograms. The next step for a prospective client is a scoping call to define the city boundary, the archive period of interest and the infrastructure assets that need per-structure resolution.
Typical figures
| Typical PS scatterer spacing (urban core) | 5–20 m effective spatial resolution; depends on building density and surface material |
| Sentinel-1 revisit (single satellite) | 12 days at mid-latitudes; 6 days when both satellites are operational |
| COSMO-SkyMed revisit | 1–16 days depending on tasking priority and orbit geometry |
| Radar frequency / wavelength | C-band 5.405 GHz / 5.6 cm (Sentinel-1); X-band ~9.6 GHz / 3.1 cm (COSMO-SkyMed, TerraSAR-X); L-band 1.27 GHz / 23.6 cm (ALOS-2) |
| Minimum detectable displacement rate (PS-InSAR, 30+ scene stack) | Approximately 1–2 mm per year under good coherence conditions |
| Sentinel-1 archive depth | From April 2014; COSMO-SkyMed and TerraSAR-X from 2007–2008 |
| Processing latency (archive analysis) | Weeks for full-archive PS run; near-real-time update possible with 12-day Sentinel-1 cadence once baseline stack is established |
| Delivery formats | GeoTIFF displacement velocity rasters, shapefiles of PS point time series, PDF engineering reports, GIS-ready layers (GeoPackage or PostGIS) |
| Coverage per Sentinel-1 pass | 250 km swath width in Interferometric Wide Swath mode; city-scale coverage typically achieved in one to three frames |
Analytics Satellize can run
| City-scale subsidence velocity map | PS-InSAR over multi-year Sentinel-1 archive; standard persistent scatterer time-series inversion | GeoTIFF raster and PS point shapefile showing mean annual displacement rate in mm/year, with uncertainty estimates per point |
| Seasonal deformation decomposition | SBAS time-series harmonic analysis separating linear trend from annual and semi-annual cycles | Time-series plots per zone and raster layers of trend, amplitude and phase of seasonal signal |
| High-resolution structural settlement profile | TerraSAR-X or COSMO-SkyMed Spotlight PS-InSAR over targeted asset footprint | Per-building displacement time series in GIS layer; flagged exceedance alerts when rate exceeds client-defined threshold |
| Ascending / descending orbit decomposition | Two-geometry line-of-sight decomposition to isolate vertical and east-west displacement components | Separate vertical and horizontal velocity rasters with documented decomposition assumptions and residual uncertainty |
| Subsidence risk classification layer | Velocity thresholding combined with building footprint and infrastructure corridor overlays | Risk-scored GIS polygon layer (low / moderate / high / critical) for use in urban planning or project-finance due diligence |
| Historical baseline extension | Cross-calibrated fusion of pre-Sentinel COSMO-SkyMed or TerraSAR-X archive with Sentinel-1 long-term stack | Extended time series from 2007 to present; PDF report documenting calibration method and confidence intervals |
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.