Railway embankment and trackbed deformation from InSAR
Persistent Scatterer InSAR detects sub-centimetre settlement along railway embankments, correlating deformation with load cycles and groundwater. Published UK and Italian programmes show it works at scale.
Sensors
- Sentinel-1 A/B (C-band SAR, ESA): 5.6 cm wavelength, 5 x 20 m ground resolution in Interferometric Wide Swath mode, 6-day repeat at mid-latitudes with both satellites active. Archive from 2014 gives decade-long displacement time-series. Free and open.
- TerraSAR-X / TanDEM-X (X-band SAR, DLR/Airbus): 3.1 cm wavelength, Stripmap mode at 3 m resolution, 11-day exact repeat. Higher spatial density of persistent scatterers on narrow embankment features; better sensitivity to short-wavelength deformation. Commercial tasking required.
- COSMO-SkyMed (X-band SAR, ASI/Italian MoD): 3.1 cm wavelength, 3 m Stripmap resolution, constellation of four satellites enabling revisit of 1-4 days. Used operationally by Italian RFI for the national railway monitoring programme. Commercial access.
- Deployed corner reflectors (trihedral, ground-based): Passive aluminium reflectors bolted to sleepers or embankment shoulders create guaranteed coherent point targets in otherwise vegetated or ballasted corridors. ESA has published design specifications for Sentinel-1-compatible reflectors. Not a sensor, but a coherence-engineering tool.
Why embankments settle and why that matters to InSAR
Railway embankments built on soft alluvial or peat ground consolidate over decades. The mechanism is straightforward: repeated axle loads drive pore-water pressure cycles in the fill and the underlying substrate, and seasonal groundwater fluctuation adds a reversible component on top of the irreversible trend. On the UK network, many embankments were built in the Victorian era from whatever material was locally available, which means clay-rich fill with variable compaction. Rates of a few millimetres per year are common; localised problem sections can exceed 20 mm per year.
InSAR is sensitive to exactly this scale of motion. A single Sentinel-1 interferogram formed from two passes 6 days apart can detect line-of-sight displacement to roughly 3-5 mm under good coherence conditions. Stack-based PS-InSAR, which fits a displacement time-series to hundreds of interferograms, pushes the detection floor for trend estimation to around 1 mm per year on stable scatterers, though that figure depends heavily on atmospheric correction quality and scatterer density.
The geometry problem: a narrow linear target in a wide beam
A standard railway formation is 6-10 m wide at the trackbed. Sentinel-1 IW pixels are 5 m in azimuth and 20 m in range. The embankment therefore occupies at most one or two pixels in range, which limits the density of naturally occurring persistent scatterers. Metallic infrastructure helps: signal gantries, rail clips, drainage culverts and fencing posts all return strong, stable radar echoes. Studies using Sentinel-1 over UK Network Rail corridors have found usable PS densities of 10-50 scatterers per kilometre on electrified main lines, dropping sharply on rural unelectrified routes through vegetated cuttings.
Ascending and descending orbit geometries see the same embankment from different look angles, roughly 30-45 degrees from vertical in Sentinel-1 IW mode. Combining both geometries allows decomposition of the line-of-sight signal into vertical and horizontal components, which matters because embankment settlement is primarily vertical while lateral spreading of embankment shoulders produces a horizontal component. Without at least two geometries, the two contributions are ambiguous. This is not a limitation unique to railways, but it is particularly consequential here because asset managers need to distinguish consolidation settlement from slope instability.
Corner reflectors: engineering coherence into the corridor
Where natural scatterer density is too low, trihedral corner reflectors can be installed directly on the infrastructure. A correctly oriented 1.5 m trihedral has a radar cross-section of roughly 30-40 dBm² at C-band, dominating the surrounding clutter and producing a stable phase centre regardless of vegetation growth or ballast disturbance. ESA published design guidance for Sentinel-1-compatible reflectors as part of its GMES ground segment documentation.
The practical installation question is spacing. For settlement monitoring, reflectors every 50-100 m along a problem section give sufficient spatial resolution to distinguish differential settlement between spans. That density is achievable and has been demonstrated in Italian RFI pilot deployments. The cost is modest relative to the cost of a speed restriction or an emergency embankment repair, but it does require trackside access and maintenance agreements with the infrastructure manager. Reflectors are also visible in optical imagery, which can complicate any claim that the monitoring is covert.
What published programmes have actually found
Network Rail and the British Geological Survey have published results from Sentinel-1 PS-InSAR monitoring covering sections of the UK national rail network. The BGS national ground motion service, which uses Sentinel-1 data processed by the British Geological Survey in collaboration with Terrafirma, maps displacement across Great Britain at roughly 20-50 m posting. Railway corridors appear clearly in these maps, with identifiable settlement anomalies correlating with known problem embankments on soft ground in the Fens, the Thames estuary corridor and parts of the West Coast Main Line.
In Italy, the national rail infrastructure manager RFI has used COSMO-SkyMed data, processed with PS-InSAR methods, to monitor embankments on the high-speed network. Published results from sections of the Milan-Naples corridor show seasonal displacement amplitudes of 3-8 mm correlated with groundwater depth measurements, and multi-year trends of 2-5 mm per year on embankments crossing the Po Valley alluvial plain. These figures are consistent with independent geotechnical monitoring at instrumented sections.
One honest caveat: PS-InSAR gives displacement at the scatterer location, which is usually the rail infrastructure surface or nearby metalwork. It does not directly measure what is happening inside the embankment body or at the fill-substrate interface. Correlation with geotechnical models is required to convert surface displacement to engineering parameters such as void ratio change or factor of safety.
Limits the asset manager should know before commissioning a survey
Cloud cover does not affect SAR. That is one of its genuine advantages over optical monitoring. However, C-band coherence degrades over dense vegetation, which means cuttings through woodland can produce very few usable scatterers. X-band (TerraSAR-X, COSMO-SkyMed) is more sensitive to vegetation decorrelation, not less, so switching to higher-resolution commercial SAR does not always solve the coherence problem in vegetated corridors.
Atmospheric artefacts, particularly tropospheric water vapour gradients, introduce apparent displacement signals of 5-20 mm per interferogram at C-band. PS-InSAR stacking suppresses this by treating the atmosphere as temporally uncorrelated noise, but short stacks (fewer than 20-30 interferograms) are poorly corrected. This means a monitoring programme needs at least 6-12 months of data before trend estimates are reliable. For emergency assessment after a flood or a known embankment failure, InSAR is not the right primary tool. It is a trend monitor, not an event detector.
Satellize processes Sentinel-1 stacks for linear infrastructure corridors and can incorporate commercial X-band tasking where scatterer density requires it. The workflow for a new railway corridor typically begins with a scatterer density audit on the existing archive before any corner reflectors are specified.
From displacement map to maintenance decision
The engineering output that asset managers actually need is not a displacement map but a ranked list of sections exceeding a threshold rate, with confidence intervals and a seasonal model separated from the trend. A section showing 8 mm per year of irreversible settlement is a different maintenance priority from one showing 8 mm of seasonal oscillation that recovers each spring.
Published practice from UK and Italian programmes uses a three-tier classification: sections below 2 mm per year are treated as stable, 2-10 mm per year triggers increased inspection frequency, and above 10 mm per year triggers geotechnical investigation. These thresholds are not universal and depend on track geometry tolerances, which vary by line speed and axle load. The InSAR product should be delivered as a GIS layer with per-scatterer velocity and uncertainty, not as a static report, so that the asset manager's own engineers can apply their own thresholds and integrate the data with existing track geometry records.
Typical figures
| Spatial resolution (Sentinel-1 IW) | 5 m azimuth × 20 m range per pixel; PS density 10-50 per km on electrified lines |
| Spatial resolution (TerraSAR-X / COSMO-SkyMed Stripmap) | ~3 m; higher PS density on narrow embankment features |
| Revisit period | 6 days (Sentinel-1 A+B at mid-latitudes); 1-4 days (COSMO-SkyMed constellation); 11 days (TerraSAR-X) |
| Line-of-sight displacement precision (single interferogram) | 3-5 mm under good coherence at C-band |
| Annual velocity detection floor (PS-InSAR stack) | ~1 mm per year on stable scatterers with adequate stack depth and atmospheric correction |
| Minimum stack depth for reliable trend | 20-30 interferograms; typically 6-12 months of acquisition |
| Archive depth (Sentinel-1) | 2014 to present (Sentinel-1A); 2016 to present (Sentinel-1B, with gap from August 2021 to December 2023 following satellite anomaly) |
| Corner reflector radar cross-section (1.5 m trihedral, C-band) | ~30-40 dBm²; dominates surrounding clutter, stable phase centre |
| Coherence limitation | Vegetated cuttings and woodland corridors significantly reduce PS density; X-band more affected than C-band in dense canopy |
| Delivery format | GIS vector layer (per-scatterer velocity, uncertainty, seasonal amplitude); optionally CSV or GeoJSON for integration with track geometry management systems |
Analytics Satellize can run
| Corridor PS-InSAR velocity map | Persistent Scatterer InSAR (StaMPS or similar published algorithm) applied to Sentinel-1 IW stack; atmospheric correction via spatial-temporal filtering | GeoJSON or Shapefile of scatterer points with mean line-of-sight velocity (mm/yr), standard deviation, and time-series for each point |
| Vertical/horizontal displacement decomposition | Ascending and descending orbit combination using published geometric decomposition; assumes negligible along-track displacement | Separate GIS layers for vertical settlement rate and horizontal (cross-track) displacement rate per scatterer |
| Seasonal groundwater correlation model | Harmonic regression on PS time-series to separate annual sinusoidal component from linear trend; comparison with published groundwater level records where available | Per-section report tabulating irreversible trend versus seasonal amplitude, with confidence intervals |
| Exceedance threshold alert layer | Threshold classification applied to velocity estimates; thresholds configurable by client against their track geometry tolerance standards | Prioritised section list (GIS and PDF) flagging segments exceeding 2, 5 and 10 mm/yr with scatterer density and uncertainty noted |
| Corner reflector placement recommendation | Scatterer density audit on existing archive; identification of gaps exceeding 200 m with fewer than 5 usable PS; site suitability scoring against trackside access constraints | Installation plan report with proposed reflector coordinates, orientation angles for Sentinel-1 geometry, and expected radar cross-section |
| Multi-epoch change detection (X-band uplift to Sentinel-1 baseline) | Cross-sensor displacement comparison using published co-registration methods; accounts for wavelength difference in phase-to-displacement conversion | Time-series chart per problem section showing displacement from 2014 Sentinel-1 baseline through to current commercial X-band acquisition |
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.