Railway track geometry subsidence from InSAR
Persistent-scatterer InSAR detects millimetre-scale differential settlement along railway alignments using rail infrastructure as coherent radar targets, giving asset managers early warning before geometry degradation triggers speed restrictions.
Sensors
- TerraSAR-X / TanDEM-X: X-band (9.65 GHz), stripmap mode at 3 m resolution, spotlight at ~1 m. Short wavelength maximises phase sensitivity to sub-centimetre displacement and resolves individual sleeper bays as discrete persistent scatterers. Revisit 11 days; commercial tasking can shorten this.
- COSMO-SkyMed Second Generation: X-band constellation of four satellites; spotlight mode reaches ~0.35 m resolution. Revisit as short as 12 hours with full constellation tasking. Particularly useful for monitoring active settlement events on high-speed lines where weekly cadence is insufficient.
- Sentinel-1 (A/B): C-band (5.405 GHz), IW mode at 5 × 20 m ground range resolution. Free and open archive from 2014. Revisit 6 days over Europe with two satellites. Lower spatial resolution limits PS density on narrow rail corridors, but long archive supports trend analysis and seasonal correction.
- ICEYE: X-band commercial SAR microsatellites; stripmap at ~3 m, spot mode at ~1 m. Flexible tasking and growing constellation allow revisit intervals below 24 hours over priority sites. Useful for rapid-response monitoring after a flood or earthworks event near the alignment.
What a sleeper and a rail fastener give a radar
Persistent-scatterer InSAR works by identifying pixels that return a stable, coherent radar signal across dozens of acquisitions. Railway infrastructure is unusually cooperative. Steel rails, cast-iron baseplates, concrete sleepers and metal fasteners all present corner-reflector or flat-plate geometries to an imaging radar. At X-band wavelengths (roughly 3 cm), a single sleeper bay can register as a distinct persistent scatterer, giving PS densities along a rail corridor that can exceed 500 points per kilometre on a well-maintained ballasted track.
C-band Sentinel-1 is less discriminating: at 5 × 20 m IW resolution, a single pixel typically integrates several sleeper bays, and PS density drops sharply on single-track rural lines. The trade-off is cost and archive depth. A Sentinel-1 stack going back to 2014 can reveal slow, chronic settlement trends that no track geometry car survey has captured, precisely because those surveys happen infrequently and record point-in-time geometry rather than cumulative displacement history.
From line-of-sight displacement to a geometry degradation signal
InSAR measures displacement in the satellite's line-of-sight direction, not vertically. Converting to vertical subsidence requires knowledge of the incidence angle (typically 20 to 45 degrees for the sensors used here) and an assumption that horizontal motion is negligible, which is reasonable for settlement-dominated processes but less so near embankment shoulders where lateral creep can occur. Decomposing ascending and descending pass data into vertical and east-west components removes most of that ambiguity, though north-south motion remains largely invisible to polar-orbiting SAR.
The operationally useful metric is not absolute displacement but differential displacement over short along-track baselines. A settlement of 10 mm spread uniformly over 500 m of track changes nothing mechanically. The same 10 mm concentrated over 20 m produces a twist or longitudinal level defect that, under EN 13848 or Network Rail standards, can mandate a speed restriction or emergency tamping. Differencing PS velocities between adjacent scatterers and normalising by baseline length produces a gradient map that maps directly onto the geometry parameters track engineers already use.
The honest limits of the method
InSAR does not measure the rail head. It measures the displacement of whatever structure returned the dominant radar signal in that pixel, which is most likely a sleeper, a fastener plate or a trackside structure such as a signal gantry or cable trough. On ballasted track the rail floats on the ballast bed; differential ballast settlement and rail displacement are closely coupled but not identical. Tamping redistributes ballast without changing the underlying ground, so a post-tamping SAR acquisition will show apparent uplift that reflects the sleeper rising, not the ground recovering.
Cloud cover is irrelevant for SAR, which is a genuine advantage over optical methods. However, dense vegetation overhanging the track, deep cuttings with steep side slopes, and electrification masts with complex multi-bounce geometries can all reduce PS density or introduce systematic phase errors. Temporal decorrelation during harvest or flooding can break the PS chain entirely for one or two acquisitions. Detection thresholds for Sentinel-1 PS processing are typically quoted at 1 to 2 mm per year for long stacks; X-band sensors can reach sub-millimetre annual velocity precision under good conditions, but that precision applies to the scatterer, not to a certified track geometry measurement.
The method is a screening and prioritisation tool. It identifies where to look, and at what urgency. Correlation with track geometry car records, which measure actual rail head position to sub-millimetre accuracy, is essential before any engineering decision is made. InSAR and geometry car data are complementary: one is continuous in time and space, the other is authoritative but sparse.
Seasonal signals and the problem of heave
Expansive clay subgrades common in the UK, parts of northern France and much of the US Gulf Coast produce seasonal vertical motion of 5 to 20 mm driven by soil moisture change. This is not settlement; it is reversible heave. A PS time series on such a formation shows a sinusoidal annual signal superimposed on any long-term subsidence trend. Separating the two requires either a long stack (three or more years to fit the seasonal model reliably) or an independent soil-moisture proxy such as Sentinel-1 backscatter or ESA's CCI Soil Moisture product.
Permafrost degradation presents a related problem on Arctic and sub-Arctic railways. Thaw settlement there can be rapid, irreversible and spatially abrupt. COSMO-SkyMed and ICEYE tasking at high latitudes is feasible, but the summer acquisition window is short and the absence of a long open archive makes trend separation harder. Sentinel-1 coverage extends to high latitudes and the archive is the best available starting point.
Integrating InSAR into a railway asset management workflow
The practical output of a PS-InSAR analysis for a railway operator is a georeferenced displacement-velocity layer aligned to the track centreline, with gradient statistics computed at configurable baseline lengths (typically 10 m, 20 m and 50 m). Thresholds derived from the operator's geometry standards trigger alerts at two levels: advisory (monitor more frequently) and action (correlate with geometry car data before next high-speed service).
Satellize runs this workflow on both open Sentinel-1 stacks and commercially tasked X-band acquisitions, depending on the required spatial resolution and revisit. The analytics pipeline produces GIS layers, time-series plots per track-kilometre post, and a prioritised maintenance segment list. One practical note: the first useful output from a new X-band stack typically requires 15 to 20 acquisitions to establish stable PS candidates, which means a commissioning period of three to five months before operational alerts are reliable. Sentinel-1 stacks from the open archive can be processed immediately and often cover six or more years of history.
For operators considering this approach, the most useful first step is a pilot corridor analysis on a section with known settlement history, so the InSAR output can be validated against existing geometry car records before the method is deployed network-wide.
Typical figures
| Spatial resolution (X-band spotlight) | ~1 m (TerraSAR-X, COSMO-SkyMed SG, ICEYE spot mode) |
| Spatial resolution (C-band IW) | 5 × 20 m ground range (Sentinel-1 IW) |
| Revisit interval | 6 days (Sentinel-1, Europe); 11 days (TerraSAR-X); 12 h minimum (COSMO-SkyMed SG full constellation); sub-24 h (ICEYE tasked) |
| Displacement detection threshold | 1–2 mm/yr velocity (Sentinel-1 long stack); sub-mm/yr (X-band long stack, good PS density) |
| Minimum PS density for reliable gradient | Typically >100 PS/km on ballasted track in X-band; 20–80 PS/km in C-band depending on infrastructure density |
| Radar frequency | X-band 9.65 GHz (TerraSAR-X, COSMO-SkyMed, ICEYE); C-band 5.405 GHz (Sentinel-1) |
| Archive depth | Sentinel-1: from April 2014 (open). TerraSAR-X: from 2007 (commercial licence). COSMO-SkyMed: from 2007 (commercial licence). |
| Commissioning period for new X-band stack | 15–20 acquisitions (~3–5 months) before PS candidates stabilise |
| Delivery formats | GeoTIFF displacement rasters, GeoPackage PS point layers, CSV time-series per track-km post, PDF segment-priority report |
| Weather sensitivity | None for cloud or rain; temporal decorrelation possible during flooding or dense vegetation growth |
Analytics Satellize can run
| PS displacement velocity map, track-aligned | Persistent-scatterer InSAR (PSInSAR / SqueeSAR class); PS candidate selection by amplitude dispersion index | GeoPackage point layer with line-of-sight and decomposed vertical velocity per scatterer, coloured by rate |
| Along-track differential settlement gradient | Pairwise differencing of adjacent PS velocities normalised by along-track baseline; thresholded against EN 13848 or operator-defined geometry limits | GIS polyline layer segmented by gradient severity; advisory and action threshold flags |
| Seasonal vs. long-term trend decomposition | Harmonic regression on PS time series to separate annual heave cycle from monotonic settlement trend | Per-scatterer trend and amplitude parameters; time-series plots for flagged segments |
| Historical subsidence baseline from open archive | Sentinel-1 PS processing on archive stack (2014–present); no new tasking cost | Retrospective velocity map and cumulative displacement plots; PDF corridor summary report |
| Rapid-response displacement change after incident | Short-stack coherent change detection on ICEYE or COSMO-SkyMed acquisitions bracketing a flood, earthworks or derailment event | Before/after displacement difference raster; prioritised inspection segment list delivered within 48 h of final acquisition |
| Maintenance prioritisation index | Composite scoring of displacement rate, gradient severity and trend acceleration; ranked by risk to line speed | Spreadsheet and GIS layer of track segments ranked by intervention urgency, ready for integration with asset management system |
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.