Urban land subsidence mapping for infrastructure risk
InSAR time-series analysis detects millimetre-scale vertical ground motion across entire cities, pinpointing differential settlement before it cracks foundations, buckles pipelines or derails trains.
Sensors
- Sentinel-1 IW SLC (C-band, 5.6 cm wavelength): 6-day repeat at mid-latitudes when both satellites are operational, 5 x 20 m ground range resolution in Interferometric Wide swath mode. Free and open archive from 2014. C-band is sensitive to surface and shallow subsurface motion but loses coherence quickly over vegetation and bare soil; urban areas with hard surfaces retain coherence well across multi-year stacks.
- COSMO-SkyMed Second Generation (X-band, 3.1 cm wavelength): Spotlight modes reach 1 x 1 m resolution; Stripmap delivers 3 x 3 m. Revisit as short as 4 days with the four-satellite constellation. Shorter wavelength improves deformation sensitivity but also increases decorrelation risk in wet or vegetated scenes. Suited to dense urban cores where high PS density is assured.
- TerraSAR-X / TanDEM-X (X-band): High-resolution Spotlight mode at approximately 1 m resolution, with StripMap at 3 m. Repeat cycle 11 days for a single satellite. Long commercial archive from 2007 enables historical subsidence trend reconstruction. Often used alongside COSMO-SkyMed to increase acquisition density for PS-InSAR stacks.
- ALOS-2 PALSAR-2 (L-band, 23.6 cm wavelength): 14-day repeat, 3 x 6 m in Fine Beam Single mode. L-band penetrates vegetation canopy and maintains coherence in areas where C- and X-band fail, making it useful for peri-urban fringes and parks within the study extent. Deformation sensitivity per interferogram is lower than X-band, but multi-temporal stacking recovers millimetre-level precision.
Why a city sinks, and why it matters now
Urban land subsidence is rarely dramatic. A building does not disappear overnight; it tilts by half a degree over five years while its drainage connections slowly shear. The damage accumulates invisibly until a road cracks, a sewer collapses or a flood-defence embankment no longer reaches design height. By then, the remediation cost typically dwarfs any plausible monitoring budget.
Three mechanisms dominate. Groundwater extraction compresses aquifer systems, sometimes irreversibly. Underground construction, metro tunnels especially, displaces soil and induces settlement in a radius that can extend 30 metres or more from the excavation face. Compressible fill, the reclaimed land and informal embankments common in coastal and delta cities, consolidates under its own weight for decades. These processes interact, and they are spatially heterogeneous: two adjacent buildings on different foundations can move at rates differing by an order of magnitude. That heterogeneity is precisely what satellite InSAR captures and conventional levelling surveys miss.
What a persistent scatterer gives away
Interferometric SAR works by comparing the phase of radar echoes from the same point on the ground across multiple satellite passes. If the ground has moved toward or away from the satellite between passes, the phase shifts by an amount proportional to displacement. At C-band (5.6 cm wavelength), a half-wavelength shift corresponds to 2.8 cm of range change. With time-series methods, sub-millimetre annual rates become detectable.
Persistent Scatterer InSAR (PS-InSAR) identifies pixels that remain coherent across the entire acquisition stack, typically corner reflectors, metal rooftops, exposed concrete and lamp posts. Each PS acts as a natural benchmark. The Small Baseline Subset (SBAS) method relaxes the coherence requirement by pairing acquisitions with short temporal and spatial baselines, which recovers more pixels in transitional urban areas but introduces slightly greater phase ambiguity. Both methods require a minimum stack depth: published practice consistently cites at least 20 acquisitions as the threshold below which velocity estimates become unreliable. With Sentinel-1's 6-day repeat, a 20-image stack can be assembled in roughly four months, though a year or more of data produces substantially better separation of subsidence signal from atmospheric noise.
Atmospheric phase delay is the chief artefact. Water vapour gradients across a city can mimic several centimetres of apparent deformation in a single interferogram. Time-series averaging suppresses this, and external weather model data (ERA5 reanalysis, for instance) can be used to correct residual tropospheric signal. The honest caveat: even after correction, atmospheric artefacts set a practical detection floor of roughly 1 to 2 mm per year for annual velocity maps, and point estimates in any single interferogram carry much larger uncertainty.
Where the method breaks down
Decorrelation is the fundamental limit. Parks, golf courses, tree-lined boulevards and any surface that changes between passes lose phase coherence and produce no PS or SBAS measurements. In cities with extensive green cover or frequent construction disturbance, the spatial coverage of the deformation map will have gaps. L-band PALSAR-2 partially fills these gaps because its longer wavelength penetrates vegetation and maintains coherence over rougher surfaces, but it does not eliminate the problem.
Layover and shadow are geometric artefacts specific to SAR. Tall buildings lean toward the sensor in the image, and their bases may be obscured by shadow. This means that the base of a skyscraper, often the most structurally critical measurement point, can be harder to retrieve than the roof. Ascending and descending orbit combinations help decompose line-of-sight displacement into vertical and horizontal components, but the decomposition assumes negligible east-west motion, which is not always valid near active faults or large excavations.
Finally, InSAR measures relative displacement, not absolute elevation. Results are anchored to a reference point assumed to be stable. If that reference is itself subsiding, all velocities shift. Ground-truth levelling at a small number of stable benchmarks remains necessary to convert relative rates into absolute subsidence figures.
From deformation maps to infrastructure risk scores
A deformation velocity map is not, by itself, an actionable product for an infrastructure authority. The analytical step that matters is overlaying subsidence rates and their spatial gradients onto asset registers: pipeline networks, road centrelines, building footprints, rail alignments and flood-defence embankments. Differential settlement, the rate difference between two adjacent points, is more damaging to rigid infrastructure than uniform subsidence. A pipeline crossing a zone where one bank subsides at 15 mm per year and the other at 2 mm per year faces cumulative angular distortion that rigid joints cannot absorb indefinitely.
Combining InSAR-derived velocity fields with soil type, aquifer data and infrastructure age produces a tiered risk classification. Zones of high velocity, high gradient and vulnerable assets warrant physical inspection; zones of moderate velocity over new infrastructure with flexible joints can be monitored at longer intervals. This prioritisation is where satellite data earns its cost advantage over blanket ground-survey programmes. Satellize applies this layered analysis approach in its urban analytics work, drawing on the same open-constellation data and overlay methods described here.
Time-series outputs also support early-warning functions. If a zone that was stable for three years begins accelerating, that change in rate is often detectable within one to two Sentinel-1 repeat cycles, provided the PS density is adequate. Automated change-point detection on PS velocity time series can flag anomalies for human review without requiring an analyst to inspect every pixel manually.
Choosing the right sensor stack for your city
For most national mapping agencies and municipal authorities, Sentinel-1 is the sensible starting point. The archive is free, the 6-day revisit builds stacks quickly, and the 250 km swath covers an entire metropolitan area in a single pass. The trade-off is spatial resolution: at 5 x 20 m, individual buildings are not resolved, and PS density in low-rise informal areas can be sparse.
Where the client needs building-level attribution, or where a specific corridor (a new metro line, a waterfront reclamation zone) demands higher precision, COSMO-SkyMed or TerraSAR-X tasking adds a focused high-resolution layer. The cost is real: commercial X-band acquisitions are priced per scene and per tasking request, and a dense time series over a large city accumulates quickly. A practical architecture pairs Sentinel-1 for city-wide monitoring with commercial X-band for targeted high-risk zones identified in the first Sentinel analysis pass.
ALOS-2 PALSAR-2 deserves consideration in cities with substantial vegetated cover or in tropical climates where wet-season coherence loss is severe. JAXA's archives extend back to 2014 and the data is available through research agreements, though not freely in the way Sentinel-1 is.
Typical figures
| Spatial resolution (Sentinel-1 IW) | 5 x 20 m (ground range x azimuth); PS point spacing in dense urban areas typically 20-100 m |
| Spatial resolution (COSMO-SkyMed / TerraSAR-X Spotlight) | 1-3 m; PS spacing can reach 5-15 m in high-density urban cores |
| Revisit period | 6 days (Sentinel-1 dual satellite, mid-latitudes); 4 days (COSMO-SkyMed 4-satellite); 11 days (TerraSAR-X single); 14 days (ALOS-2) |
| Minimum detectable annual velocity | 1-2 mm/year for annual mean velocity after atmospheric correction; single-interferogram noise floor 5-20 mm |
| Minimum acquisition stack for reliable PS-InSAR | 20 images (published consensus); 30-50 preferred for robust velocity and seasonal separation |
| Radar frequency / wavelength | C-band 5.6 cm (Sentinel-1); X-band 3.1 cm (COSMO-SkyMed, TerraSAR-X); L-band 23.6 cm (ALOS-2 PALSAR-2) |
| Swath width | 250 km (Sentinel-1 IW); 40 km (COSMO-SkyMed Stripmap); 30 km (TerraSAR-X Stripmap); 70 km (ALOS-2 Fine Beam) |
| Archive depth | Sentinel-1: from April 2014. TerraSAR-X: from 2007. COSMO-SkyMed: from 2007. ALOS-2: from 2014 |
| Delivery formats | GeoTIFF velocity rasters, Shapefile or GeoPackage PS point clouds, CSV time-series per asset, PDF risk-zone report |
| Atmospheric correction | ERA5 reanalysis tropospheric correction applicable; residual error typically 1-3 mm in annual velocity |
Analytics Satellize can run
| City-wide mean velocity map | PS-InSAR or SBAS time-series on Sentinel-1 IW SLC stack (minimum 20 acquisitions) | GeoTIFF raster and PS point-cloud shapefile showing annual vertical displacement rate in mm/year, colour-classified by severity threshold |
| Differential settlement index along linear infrastructure | Spatial gradient computation on PS velocity field, overlaid on pipeline and road-network GIS layers | GIS line layer with per-segment differential rate and risk tier; priority inspection list as CSV |
| Subsidence acceleration alert | Change-point detection on per-PS velocity time series using sequential statistical testing | Automated alert report flagging PS clusters where rate has changed by more than a defined threshold within the most recent rolling window |
| High-resolution corridor deformation profile | PS-InSAR on commercial X-band (COSMO-SkyMed or TerraSAR-X) for targeted zones identified in Sentinel-1 pass | Building-attributed PS point cloud with time series plots per structure; PDF summary for engineering review |
| Ascending and descending orbit decomposition | Vector decomposition of line-of-sight velocities from two orbit geometries into vertical and east-west components | Separate GeoTIFF layers for vertical and horizontal displacement rates; uncertainty bounds per pixel |
| Historical subsidence trend reconstruction | SBAS processing of archived Sentinel-1 or TerraSAR-X SLC scenes back to 2014 or 2007 respectively | Multi-year time-series chart per zone of interest; decadal cumulative displacement map |
| Infrastructure risk score layer | Weighted overlay of subsidence velocity, gradient, soil compressibility proxy and asset age from client GIS register | Ranked asset list and zonal risk map suitable for maintenance budget prioritisation; GeoPackage plus PDF |
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.