High-rise building tilt and differential settlement from SAR
Multi-temporal SAR interferometry can resolve differential displacement between adjacent points on a tall building facade to sub-millimetre precision, revealing tilt or foundation settlement invisible to conventional survey. This page explains the geometry, the sensors capable of doing it, and where the method breaks down.
Sensors
- TerraSAR-X Staring Spotlight: X-band (9.65 GHz), ground resolution approximately 0.25 m in azimuth by 0.6 m in range in Staring Spotlight mode. Revisit at mid-latitudes roughly 11 days, reducible to 2–3 days with competing ascending and descending acquisitions. The finest commercially available spaceborne SAR resolution for individual building analysis.
- COSMO-SkyMed Second Generation (CSG): X-band, Spotlight-2 mode delivers approximately 0.35 m azimuth resolution. Constellation of two satellites gives a revisit of 12 hours to a few days depending on latitude and tasking priority. Polarimetric capability added over the first generation aids coherence assessment.
- ICEYE X-band SAR: Spotlight mode resolution approximately 0.5 m. Growing constellation of more than 30 satellites offers revisit intervals of hours for high-priority targets, useful for rapid-onset settlement events. Archive depth is shorter than heritage systems.
- Sentinel-1 (A/B/C): C-band (5.405 GHz), Interferometric Wide Swath mode at 5 × 20 m resolution. Insufficient for isolating individual buildings in a dense urban block, but useful for neighbourhood-scale subsidence context and for establishing long baselines (archive from 2014). Free and open data.
Why a single displacement number is not enough
Standard persistent-scatterer InSAR gives one line-of-sight displacement value per coherent scatterer. For a tall building, that is adequate for detecting that something is moving. It is not adequate for detecting that a building is tilting, because tilt is a differential quantity: the top of the structure moves relative to the base, and the difference between those two numbers is the signal you need.
A 100-metre tower tilting at 1 milliradian, roughly 0.06 degrees, will show approximately 100 mm of horizontal displacement at the roof relative to the foundation. That is detectable. But a uniform ground subsidence of 50 mm beneath the whole building will also move the roof scatterers by 50 mm in the line-of-sight direction, with no tilt at all. Without comparing scatterers at multiple heights on the same facade, the two signatures are indistinguishable. Multi-looking SAR interferometry, applied to a stack of images from the same orbital geometry, extracts a displacement time series for each persistent scatterer independently. The vertical profile of those time series, from ground-floor corners to roof-edge returns, is what reveals tilt.
The geometry problem: line-of-sight is not vertical
SAR satellites observe from the side, not from directly overhead. TerraSAR-X in Staring Spotlight typically operates at an incidence angle between 20 and 55 degrees, with 30 to 45 degrees common for urban tasking. The measured displacement is the projection of the true 3-D displacement vector onto the satellite's line-of-sight (LOS) direction. A purely vertical settlement of 10 mm produces a LOS signal of roughly 8–9 mm at a 30-degree incidence angle. A purely horizontal displacement of the same magnitude, oriented along the satellite's range direction, produces a comparable LOS signal. The two are geometrically confused in a single-pass dataset.
The standard solution is to combine ascending and descending pass acquisitions. Because the two geometries view the building from opposite sides in the east-west plane, their LOS vectors are linearly independent. Solving the two-equation system yields separate estimates of vertical displacement and east-west horizontal displacement. North-south displacement remains largely invisible to both geometries because SAR satellites fly near-polar orbits and have almost no sensitivity in the along-track direction. For a building settling or tilting primarily in the north-south plane, this is a genuine blind spot that no amount of processing corrects.
What coherence loss tells you, and when it ruins the analysis
Persistent-scatterer analysis depends on finding scatterers whose radar return is stable across dozens of acquisitions spanning months or years. On a glass curtain-wall tower, stable returns come from window frames, balcony railings, air-conditioning units and corner reflectors formed where facades meet at right angles. These are reliable. What destroys them is physical change: scaffolding erected for renovation, new cladding panels replacing old ones, or major facade cleaning that alters surface geometry at the centimetre scale.
When coherence drops below roughly 0.3 on a persistent-scatterer candidate, that scatterer is lost from the time series. A building undergoing a two-year facade refurbishment may lose most of its upper-floor scatterers precisely during the period when differential settlement from construction loading on adjacent sites is most likely. The analyst must flag coherence loss explicitly rather than treat it as missing data. It is information: it tells you the surface changed, but it cannot tell you whether the structure moved.
X-band sensors (TerraSAR-X, CSG, ICEYE) are more sensitive to small surface changes than C-band Sentinel-1, because the shorter wavelength means a smaller physical change produces a larger phase shift. That sensitivity is a double-edged property. Higher resolution, more scatterers per building, finer displacement sensitivity, but also faster decorrelation when anything on the facade is disturbed.
Decomposing tilt from the scatterer height profile
Once a displacement time series exists for each persistent scatterer on a building, the next step is to assign each scatterer an approximate height. In Staring Spotlight mode, the SAR image is sharp enough that scatterers can often be matched to specific floors by comparing their range and azimuth positions against a building footprint from cadastral data or a photogrammetric model. The displacement rate is then plotted against height. A linear gradient, with the roof moving faster than the base in a consistent direction, is the signature of rigid-body tilt. A non-linear profile, where the middle floors show anomalous displacement, suggests internal structural deformation rather than foundation rotation.
The minimum detectable tilt depends on the displacement precision per scatterer and the height separation between the highest and lowest reliable scatterers. Published studies using TerraSAR-X Staring Spotlight over urban towers report LOS displacement precisions of 0.3 to 1.0 mm per epoch under good coherence conditions. For a 150-metre building with scatterers spanning 120 metres of height, that translates to a detectable tilt on the order of 0.01 to 0.05 milliradians per observation epoch, before temporal averaging. Averaging over a year of monthly acquisitions improves this further, but only if the building is not changing.
Practical limits buyers should understand before commissioning
Resolution is not the only constraint. Dense urban canyons produce layover and shadow: a building's facade facing away from the satellite is in shadow and contributes no scatterers. A very tall building will overlay its own base in the SAR image, compressing many floors into a few pixels. The decomposition into vertical and horizontal components requires simultaneous coherent coverage from both ascending and descending geometries, which means two separate tasking contracts and compatible acquisition schedules. If the building of interest sits at the edge of a standard scene, one geometry may clip it.
Archive depth matters for detecting slow processes. A settlement rate of 2 mm per year requires at least two to three years of data to separate from noise at typical precision levels. TerraSAR-X archive over some urban areas extends to 2008, CSG only to 2019. For a newly constructed tower with no prior SAR archive, a baseline acquisition campaign must begin before any concern arises, not after.
Satellize structures multi-temporal SAR analytics for infrastructure clients by combining open Sentinel-1 data for neighbourhood context with commercial X-band tasking for building-level precision, then delivers displacement time series as georeferenced GIS layers with per-scatterer confidence flags. The approach is the same one underpinning our Tonga crop-estimation work: open data for context, commercial data where resolution is the deciding factor.
Typical figures
| Best available spatial resolution (SAR) | ~0.25 m azimuth × ~0.6 m range (TerraSAR-X Staring Spotlight) |
| Typical displacement precision per scatterer | 0.3–1.0 mm LOS per epoch under good coherence (published TerraSAR-X Staring Spotlight studies) |
| Minimum detectable tilt (indicative) | ~0.01–0.05 mrad per epoch for a 150 m building; improves with temporal averaging |
| Revisit (TerraSAR-X, single geometry) | ~11 days; 2–3 days combining ascending and descending passes |
| Revisit (ICEYE Spotlight) | Hours to days depending on constellation tasking priority |
| Frequency bands used | X-band (9.65 GHz, TerraSAR-X / CSG / ICEYE) for building-level; C-band (5.405 GHz, Sentinel-1) for area context |
| Archive depth | TerraSAR-X from ~2008; Sentinel-1 from 2014; CSG from 2019; ICEYE from ~2019 |
| North-south displacement sensitivity | Near-zero for all near-polar SAR geometries; a known geometric blind spot |
| Coherence threshold for reliable PS | Typically >0.3; facade renovation or cladding change can drop coherence below this, invalidating affected scatterers |
| Delivery formats | GeoTIFF displacement rasters, GeoPackage or Shapefile PS point clouds with time-series attributes, CSV time-series tables |
Analytics Satellize can run
| Per-building tilt gradient map | Persistent-scatterer InSAR (PSInSAR) with scatterer height attribution from cadastral or photogrammetric data; linear regression of displacement rate against scatterer elevation | GIS point layer with tilt rate (mrad/yr) and uncertainty per building, updated each new acquisition |
| Ascending/descending LOS decomposition | Two-geometry InSAR combination to separate vertical and east-west horizontal displacement components | Paired raster layers (vertical and horizontal displacement rates) with coverage mask showing decomposition validity |
| Differential settlement alert | Threshold exceedance on displacement rate difference between base-level and roof-level scatterer clusters on the same structure | Automated alert (email or API push) when differential rate exceeds a client-defined threshold |
| Coherence loss flag and renovation detection | Temporal coherence monitoring across the PS candidate stack; drop below 0.3 flagged as surface-change event | Annotation layer marking affected scatterers and date of coherence loss, included in time-series report |
| Multi-year displacement time series | Small-baseline subset (SBAS) or PSInSAR on archive stack; Sentinel-1 for long baseline, X-band for precision epochs | Per-scatterer CSV time series with epoch dates, LOS displacement, and coherence values; visualised in interactive dashboard |
| Neighbourhood subsidence context layer | Sentinel-1 SBAS InSAR over a 2–5 km radius around the building of interest | GeoTIFF mean velocity map (mm/yr) as contextual background for the building-level analysis |
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.