Highway and road-cut slope instability monitoring by InSAR
Time-series InSAR detects millimetre-scale displacement on road-cut slopes and embankments months before failure. Sentinel-1, COSMO-SkyMed and ALOS-2 PALSAR-2 cover different trade-offs of resolution, revisit and vegetation penetration.
Sensors
- Sentinel-1 A/B (C-band SAR): 6-day repeat at mid-latitudes using combined ascending and descending passes; ground range resolution approximately 5 x 20 m in Interferometric Wide Swath mode. Sufficient for corridor-scale mapping of embankments and large cut faces. Free archive from 2014 onwards via Copernicus Data Space.
- COSMO-SkyMed Second Generation (X-band SAR, spotlight mode): Spotlight mode delivers approximately 1 x 1 m resolution on individual slope faces, enabling detection of small block movements and tension crack widening on a single cut. Revisit is taskable to 1-3 days over a target. Commercial tasking required.
- ALOS-2 PALSAR-2 (L-band SAR): L-band (23.6 cm wavelength) penetrates vegetation canopy better than C- or X-band, recovering coherence on forested or scrub-covered cut slopes where Sentinel-1 loses phase. Fine-beam mode offers 6 m resolution; revisit is 14 days. Critical for tropical or temperate vegetated corridors.
- Copernicus DEM GLO-10: 10 m global DEM derived from TanDEM-X, used to remove topographic phase from interferograms and to compute slope aspect and gradient for stability back-analysis. Vertical accuracy approximately 4 m LE90 globally; locally better in well-mapped regions.
What a cut slope gives away before it fails
When a highway engineer blasts and excavates a slope face, the remaining rock or soil mass begins adjusting to its new stress state immediately. That adjustment is rarely catastrophic from day one. More commonly, discontinuities open slowly, pore pressures redistribute over seasons, and the slope creeps along a forming shear surface at rates measured in millimetres per month. InSAR measures exactly that signal.
Synthetic aperture radar interferometry compares the phase of radar returns from two passes over the same ground. Any displacement of the surface along the satellite's line of sight between passes shifts the phase by a fraction of the radar wavelength. Sentinel-1 operates at C-band (5.6 cm wavelength), so a full phase cycle corresponds to 2.8 cm of line-of-sight displacement. With time-series methods such as Persistent Scatterer InSAR or Small Baseline Subset (SBAS), displacement histories can be resolved to roughly 1-2 mm per year for stable, coherent targets. On an active slope, accelerating trends of 5-20 mm over a few months are detectable well before a visible failure surface develops.
Ascending and descending geometry: why you need both
InSAR measures displacement only along the satellite's line of sight, which is a mixture of vertical and horizontal motion weighted by the look angle (typically 30-46 degrees from vertical for Sentinel-1). A slope moving directly away from the satellite looks identical in phase to a slope subsiding vertically. On a mountain highway, the slope face may be oriented at any azimuth relative to the satellite track, and the true displacement vector is slope-parallel, not vertical.
Combining ascending passes (satellite moving northward, looking right, roughly east-facing illumination) with descending passes (moving southward, looking right, roughly west-facing illumination) provides two independent line-of-sight measurements. With a DEM-derived slope normal to constrain the third dimension, the two measurements can be decomposed into east-west and vertical components, and from those into slope-parallel displacement. This decomposition is not algebraically perfect when the slope faces north or south, because both geometries then have similar sensitivity to east-west motion and poor sensitivity to north-south motion. Honest reporting of the decomposition uncertainty is part of any credible deliverable.
The coherence problem on vegetated cuts
Interferometric coherence is the correlation of radar phase between two passes. Bare rock and concrete structures maintain high coherence over months. Vegetation does not: leaves and branches move between passes, randomising the phase and destroying the displacement signal. On a newly cut slope with exposed rock, C-band Sentinel-1 typically achieves good coherence. Once pioneer vegetation colonises the bench faces, coherence drops sharply, often within one to two growing seasons.
L-band radar (ALOS-2 PALSAR-2, wavelength 23.6 cm) penetrates light-to-moderate vegetation and scatters from the soil and rock beneath, recovering coherence where C-band has failed. The trade-off is coarser resolution in standard modes and a 14-day revisit that may miss rapid acceleration events. X-band (COSMO-SkyMed, wavelength 3.1 cm) is more sensitive to surface motion per unit displacement but loses coherence faster than C-band on any vegetation. The practical answer for vegetated tropical or temperate corridors is to run L-band as the primary time-series sensor and task X-band spotlight only on specific bare-face targets where fine spatial detail matters.
Temporal decorrelation is not the only coherence killer. Wet snow, heavy rain and soil moisture changes all affect phase stability. Atmospheric water vapour introduces apparent displacement signals of several centimetres that must be corrected using auxiliary weather data or empirical models. In mountain terrain, where orographic rainfall is spatially variable at scales of a few kilometres, atmospheric correction is harder than on flat ground and residual errors of 5-10 mm per epoch are realistic.
Feeding back-analysis, not replacing it
InSAR displacement maps are not stability assessments. A slope showing 8 mm of movement over six months may be settling elastically into a new equilibrium, or it may be approaching the residual friction angle on a clay-filled discontinuity. Distinguishing the two requires a geotechnical engineer with a slope model, shear strength parameters and pore-pressure data. What InSAR provides is the spatial pattern and temporal evolution of displacement: which bench is moving, at what rate, and whether that rate is accelerating.
Acceleration is the critical diagnostic. A slope moving at a constant 2 mm per month for two years is very different from one that moved 2 mm per month for eighteen months and then 8 mm per month in the last six weeks. The Fukuzono inverse-velocity method, published in the 1980s and widely used in geotechnical practice, projects time to failure by fitting a power-law curve to the inverse of displacement velocity. InSAR time-series provides the input data for that method, applied remotely across an entire road corridor rather than at isolated extensometer pins. The method has known failure modes: it requires a clearly accelerating trend, and it performs poorly on slopes that accelerate then stabilise before failing later.
The output that is operationally useful is not a single failure-probability number but a ranked list of slope segments showing displacement rate, acceleration trend, spatial coherence of the moving area, and flags for segments where the pattern matches known pre-failure signatures. That list then directs field inspection resources to the highest-priority sites.
Resolution floors and what they mean for road corridors
A standard Sentinel-1 IW pixel covers roughly 5 x 20 m on the ground after multi-looking. A road cut face may be only 15-30 m high and 50-100 m wide. At Sentinel-1 resolution, a single unstable bench may occupy only two or three pixels, making it difficult to distinguish internal deformation patterns from slope-wide motion. COSMO-SkyMed spotlight at 1 m resolution resolves individual benches, tension cracks and retaining wall panels, but at commercial tasking cost and with a narrower swath that must be specifically pointed.
For a national highway authority managing several hundred kilometres of mountain road, the practical workflow is to run Sentinel-1 time-series across the full corridor to identify anomalous segments, then task COSMO-SkyMed or ALOS-2 spotlight over the ten or twenty sites that show the most concerning signals. This tiered approach matches sensor cost to information need. Satellize structures corridor monitoring programmes on exactly this logic, with the Copernicus open archive covering the baseline survey and commercial tasking added only where the data justify it.
Archive depth and what history buys you
Sentinel-1A has been acquiring data since April 2014. For any highway corridor in a region with consistent Sentinel-1 coverage, a decade of displacement history is available without new tasking. That archive is valuable for two reasons. First, it reveals whether a slope that looks stable today has been creeping slowly for years, a pattern invisible to a single-epoch survey. Second, it allows correlation of displacement episodes with rainfall records, snowmelt timing and seismic events, which directly informs the triggering thresholds used in early-warning systems.
Archive processing is not free of cost or complication. Atmospheric correction over long time series in mountain terrain requires careful attention to seasonal signals that can mimic real displacement trends. Orbit corrections and DEM errors introduce long-wavelength ramps that must be removed before the displacement signal is credible. These are solvable problems with established methods, but they take time and expertise. A ten-year Sentinel-1 time-series over a 200 km corridor is a multi-week processing task, not an overnight product.
Typical figures
| Spatial resolution (Sentinel-1 IW) | ~5 x 20 m (range x azimuth) after standard multi-looking |
| Spatial resolution (COSMO-SkyMed spotlight) | ~1 x 1 m; swath approximately 10 x 10 km |
| Spatial resolution (ALOS-2 PALSAR-2 fine beam) | ~6 m; swath 70 km |
| Revisit period | Sentinel-1: 6 days (mid-latitudes, combined A+B); ALOS-2: 14 days; COSMO-SkyMed: 1-3 days (taskable) |
| Minimum detectable displacement rate | ~1-2 mm/year for stable PS targets (Sentinel-1 time-series); single-interferogram precision ~5-10 mm line-of-sight |
| Radar frequency / wavelength | C-band 5.6 cm (Sentinel-1); X-band 3.1 cm (COSMO-SkyMed); L-band 23.6 cm (ALOS-2) |
| DEM vertical accuracy (GLO-10) | ~4 m LE90 global; used for topographic phase removal and slope geometry |
| Sentinel-1 archive depth | April 2014 to present (Sentinel-1A); consistent coverage varies by region |
| Processing latency (operational monitoring) | Typically 2-5 days after scene acquisition for alert-grade products; full time-series reprocessing 2-4 weeks |
| Delivery formats | GeoTIFF displacement maps, GeoPackage/shapefile point clouds, CSV velocity time-series, PDF slope-segment reports |
Analytics Satellize can run
| Corridor displacement velocity map | SBAS or Persistent Scatterer InSAR time-series (Sentinel-1 IW, ascending + descending stacks) | GeoTIFF raster and point shapefile showing mean annual line-of-sight velocity across the full road corridor, flagged anomaly polygons |
| Slope-parallel displacement decomposition | Two-geometry line-of-sight decomposition constrained by GLO-10 slope aspect and gradient | Vector GIS layer of slope-parallel velocity per identified slope segment, with decomposition uncertainty estimate |
| Acceleration alert | Rolling inverse-velocity trend fitting (Fukuzono method) applied to PS/SBAS time-series at flagged segments | Automated alert report listing segments with statistically significant acceleration, ranked by rate-of-change |
| High-resolution single-slope displacement map | Differential InSAR or offset tracking on COSMO-SkyMed spotlight pairs over priority sites | 1 m resolution displacement map of individual cut face, delivered as GeoTIFF with annotated PDF interpretation |
| Coherence loss map (vegetation encroachment indicator) | Multi-temporal coherence stack analysis on Sentinel-1 and ALOS-2 pairs | GIS layer showing coherence trend per slope segment, flagging areas where vegetation has degraded monitoring confidence |
| Historical displacement archive report | Full Sentinel-1 archive SBAS processing from 2014, with rainfall and seismic event overlay | PDF report and GeoPackage correlating multi-year displacement episodes with triggering records, for use in back-analysis |
| Geotechnical input dataset | Displacement time-series extraction at user-specified slope-segment centroids, formatted for limit-equilibrium or numerical stability models | CSV time-series per slope segment, slope geometry from GLO-10, ready for import into geotechnical modelling software |
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.