Pre-failure landslide precursor motion detection
Slow creep on an unstable slope can precede catastrophic failure by months or years. InSAR time-series methods detect that creep at millimetre-per-year sensitivity, giving engineers and civil authorities a measurable warning window.
Sensors
- Sentinel-1 A/B (C-band SAR): 6- to 12-day repeat (single satellite since 2021 Sentinel-1B loss; dual-satellite revisit expected to return with Sentinel-1C). Ground range resolution 5 x 20 m in IW mode. C-band is sensitive to vegetation moisture, which degrades coherence on densely forested slopes. Free and open archive from 2014.
- ALOS-2 PALSAR-2 (L-band SAR): L-band (1.27 GHz) penetrates forest canopy significantly better than C-band, preserving coherence on vegetated slopes. Stripmap mode delivers 3 m resolution; standard repeat is 14 days. Critical for slopes in the Three Gorges area and tropical or densely wooded terrain where Sentinel-1 coherence collapses.
- COSMO-SkyMed (X-band SAR): X-band at 1–3 m resolution in Spotlight mode enables Persistent Scatterer analysis on individual rock faces, retaining walls and built structures on slopes. Constellation of four satellites gives revisit of less than 24 hours on tasked targets, useful for accelerating-phase monitoring. Commercial tasking required.
- TerraSAR-X / TanDEM-X (X-band SAR): Spotlight mode reaches 1 m resolution. High phase stability makes it well suited to PS-InSAR on sparse, rocky terrain. Staring Spotlight mode (0.25 m) can resolve motion on individual boulders. Commercial archive and tasking. Revisit approximately 11 days without tasking, shorter with.
Why a slope telegraphs failure before it falls
Catastrophic landslides rarely begin without precursory deformation. The physical reason is straightforward: shear strength along a failure plane is mobilised progressively, and as the safety factor approaches unity, the slope accelerates. This acceleration phase can last weeks to years, depending on lithology, pore-water conditions and slope geometry. On the Vajont reservoir slope in Italy, retrospective analysis of survey data showed measurable acceleration in the months before the 1963 failure that killed nearly two thousand people. The signal was there; the monitoring density was not.
InSAR time-series methods detect the line-of-sight displacement component of that creep at sensitivities that ground-based survey networks rarely match at scale. Published PS-InSAR studies on the Mud Creek landslide in California and on multiple slopes above the Three Gorges reservoir have retrospectively identified acceleration signals of 10–30 mm in the months before failure events, using Sentinel-1 archives. The caveat is important: retrospective identification is easier than prospective detection, because analysts know where to look. Operational early warning requires systematic monitoring of many slopes simultaneously, most of which will never fail.
What PS-InSAR and SBAS actually measure, and where they disagree
Persistent Scatterer InSAR (PS-InSAR) identifies pixels that maintain stable radar reflectivity across a long image stack, typically man-made structures, bare rock outcrops or coarse debris. It fits a deformation model to the phase history of each scatterer and can resolve velocities to roughly 1–2 mm/yr in C-band under good conditions. Small Baseline Subset (SBAS) processing uses distributed scatterers by averaging phase across spatially coherent patches, which works better on agricultural terraces and less-urbanised slopes but at coarser spatial resolution.
The two methods are complementary rather than redundant. On a single slope, PS-InSAR may anchor the velocity field to a road cutting or a farmhouse, while SBAS fills in the broader deformation pattern across the vegetated hillside. Published studies on the Ivancich landslide in Assisi and on slopes in the Apennines have used both in combination. Neither method measures absolute 3-D displacement directly. Each measures only the projection of motion onto the satellite line of sight, which means a slope moving purely horizontally and perpendicular to the satellite track contributes almost nothing to the measured phase. This is not a minor caveat: it is a fundamental geometric constraint that can cause a real landslide to appear stationary in a single-geometry dataset.
Ascending and descending orbits: the geometry you cannot skip
Sentinel-1 acquires the same area from both ascending (roughly west-looking) and descending (roughly east-looking) passes. Combining the two line-of-sight measurements allows decomposition into vertical and east-west displacement components. North-south motion remains poorly constrained by C-band InSAR because the satellite orbit is nearly polar. For slopes with a dominant downslope direction, knowing whether motion is primarily vertical (settling) or horizontal (sliding) is essential to distinguishing deep-seated rotational failure from shallow translational movement.
Published work on Three Gorges reservoir slopes, particularly the Xintan and Qianjiangping landslides, has demonstrated that ascending-only or descending-only analysis can underestimate total displacement by a factor of two or more, depending on slope aspect. East-facing slopes are better observed from descending geometry; west-facing slopes from ascending. A monitoring programme that uses only one geometry for a slope with unfavourable aspect may miss most of the signal. This is a design choice that must be made before data acquisition begins, not after.
The coherence problem on vegetated slopes
Vegetation is the principal enemy of InSAR coherence. C-band wavelength (approximately 5.6 cm for Sentinel-1) is scattered by leaves and small branches, and the scattering geometry changes with wind, growth and moisture. On densely forested slopes, interferometric coherence can drop below usable thresholds (typically coherence below 0.3) across entire scenes, leaving no measurable phase. This is not a processing artefact; it is a physical property of the wavelength.
L-band (ALOS-2 PALSAR-2, wavelength approximately 23 cm) penetrates canopy to the ground surface and to shallow subsurface scatterers, maintaining coherence in conditions where C-band fails entirely. The trade-off is coarser resolution in standard modes and a commercial or restricted-access data policy compared with Sentinel-1's open archive. For slopes in the Alps above the treeline, C-band is often adequate. For the Apennines, where slopes are heavily forested, or for tropical analogues, L-band is frequently the only viable option. No single wavelength is universally correct.
Reading the acceleration curve: from creep rate to alert threshold
A slope moving at a steady 5 mm/yr is a chronic hazard but not an imminent one. The same slope accelerating from 5 mm/yr to 50 mm/yr over six months is a different problem. The Fukuzono model, published in 1985 and widely used in operational landslide monitoring, predicts failure time from the inverse velocity trend: as failure approaches, the inverse of velocity decreases linearly toward zero. InSAR time series can, in principle, feed this model if the temporal sampling is dense enough and the coherence is maintained through the acceleration phase.
In practice, the acceleration phase often coincides with wet seasons, when vegetation moisture peaks and coherence degrades. This is the worst possible timing. Published studies on the Maoxian landslide in Sichuan (which failed in 2017 killing over 100 people) used ALOS-2 data to reconstruct a pre-failure velocity increase, but the Sentinel-1 coherence in that forested terrain was insufficient for the same analysis. The lesson is that monitoring programmes must be designed with the expected failure season in mind, and sensor choice must match the coherence conditions of that season specifically.
Satellize's InSAR analytics workflows apply SBAS and PS processing to open and commercial SAR archives, with velocity maps and time-series outputs delivered as GIS layers. The same analytical infrastructure that supports the Tonga crop-estimation programme handles multi-temporal SAR stacks; the domain expertise differs but the processing pipeline is shared.
Honest limits: what InSAR cannot tell you
InSAR measures surface displacement, not subsurface failure-plane geometry. A slope can show zero surface deformation while a deep failure plane is slowly weakening, if the deforming mass is too rigid to express differential surface motion at detectable scales. Conversely, a slope can show large InSAR velocities from shallow, non-catastrophic creep that will never produce a major failure. Distinguishing these cases requires integration with geological mapping, borehole data and pore-pressure records. InSAR is a screening and monitoring tool, not a standalone failure predictor.
The minimum detectable velocity in C-band is approximately 1–2 mm/yr under ideal conditions: good coherence, long archive, stable atmospheric corrections. In practice, atmospheric phase delays (tropospheric water vapour gradients over mountainous terrain) can introduce apparent deformation signals of 10–20 mm per acquisition, which must be corrected using ERA5 reanalysis or GACOS models before velocity fields are reliable. In steep terrain, geometric distortions including layover and shadow further reduce coverage. A south-facing cliff illuminated from the north may simply not appear in any usable SAR geometry. These are not problems that better processing solves; they are constraints imposed by physics and orbital geometry.
Typical figures
| Spatial resolution (Sentinel-1 IW mode) | 5 m range x 20 m azimuth; PS/SBAS products typically gridded at 20–100 m |
| Spatial resolution (COSMO-SkyMed / TerraSAR-X Spotlight) | 1–3 m; PS products can resolve individual structures on slopes |
| Revisit interval | 6–12 days (Sentinel-1 single satellite); 14 days (ALOS-2); less than 24 hours possible with COSMO-SkyMed tasking |
| Minimum detectable velocity (C-band, good coherence) | Approximately 1–2 mm/yr over a long stack; 5–10 mm/yr in practice on vegetated terrain |
| Frequency / wavelength | C-band 5.6 cm (Sentinel-1); L-band 23 cm (ALOS-2); X-band 3.1 cm (COSMO-SkyMed, TerraSAR-X) |
| SAR archive depth | Sentinel-1 from 2014; ALOS-2 from 2014; ERS/Envisat C-band archive from 1992 available for retrospective studies |
| Atmospheric correction | ERA5 or GACOS tropospheric models; residual errors 5–20 mm per scene in mountainous terrain |
| Geometric limitations | Layover and shadow on slopes steeper than incidence angle; north-south motion poorly constrained; single geometry underestimates total displacement on unfavourably oriented slopes |
| Coherence floor | C-band unusable (coherence below ~0.3) on densely forested slopes; L-band maintains coherence in most vegetated terrain |
| Deliverable formats | GeoTIFF velocity maps, CSV time-series per PS/SBAS point, GeoPackage or Shapefile, alert reports in PDF |
Analytics Satellize can run
| Mean annual velocity map | PS-InSAR or SBAS time-series inversion over multi-year Sentinel-1 stack | GeoTIFF raster and vector point layer with velocity in mm/yr, line-of-sight direction specified |
| Displacement time series per scatterer | PS-InSAR phase history extraction with atmospheric correction (GACOS or ERA5) | CSV time-series per persistent scatterer, suitable for Fukuzono inverse-velocity analysis |
| Ascending/descending decomposition | Geometric decomposition of dual-geometry line-of-sight velocities into vertical and east-west components | Two-band GeoTIFF (vertical, horizontal) with uncertainty estimates; PDF methodology note |
| Acceleration detection and alert | Change-point detection on rolling inverse-velocity time series; threshold set per slope baseline rate | Automated alert report flagging slopes where inverse velocity trend crosses defined threshold, with supporting time-series plots |
| Coherence loss mapping | Spatial coherence estimation across interferogram stack; comparison of C-band and L-band coverage | GeoTIFF coherence map identifying monitoring gaps; sensor recommendation report for vegetated zones |
| Retrospective precursor analysis | SBAS inversion over historical archive (Sentinel-1 from 2014, ERS/Envisat where available) to reconstruct pre-failure velocity history | Time-series report showing velocity evolution in the months to years before a documented failure event |
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.