Flood defence embankment condition and settlement monitoring
Persistent-scatterer InSAR detects millimetre-scale settlement on flood embankments years before visible failure, but vegetated earthworks are sparse in coherent scatterers and need augmentation. This page explains the physics, the workarounds and the honest limits.
Sensors
- Sentinel-1 (C-band SAR, ESA): 5.6 cm wavelength, 5 x 20 m IW mode ground resolution, 6-day repeat at mid-latitudes with both satellites. The workhorse for levee PS-InSAR: free, dense archive from 2014, and the wavelength penetrates light vegetation canopy enough to reach hard targets on the embankment crest. Coherence on pure grass is still poor; useful signal comes from structures, fencing and installed reflectors.
- COSMO-SkyMed (X-band SAR, ASI): 3.1 cm wavelength, Spotlight mode down to roughly 1 m resolution. Shorter wavelength improves sensitivity to small displacements on hard targets and resolves individual structures on narrow embankment crests. Revisit can be as short as 1 day with constellation tasking, which matters during flood-season monitoring campaigns.
- TerraSAR-X / TanDEM-X (X-band SAR, DLR/Airbus): StripMap at 3 m, Spotlight at 1 m. Similar wavelength advantages to COSMO-SkyMed. The TanDEM-X pairing enables high-quality baseline DEMs for deformation reference. Commercial tasking allows consistent geometry repeat, which is essential for long PS time-series on narrow linear infrastructure.
- ICEYE (X-band SAR, commercial constellation): Sub-metre Spot mode, revisit potentially daily over a specific site by tasking multiple satellites. Useful for rapid-response monitoring after a flood event or when a suspected weak zone needs intensive observation. Archive depth is shorter than Sentinel-1, limiting historical baseline construction.
Why millimetres matter on an earthen bank
Flood embankments and levees fail through processes that are largely invisible at the surface until they are not: internal erosion (piping), foundation consolidation, slope instability, and differential settlement driven by variable subsoil conditions. The visible crack or slump that prompts an emergency inspection is typically the end of a process that began months or years earlier with displacements measured in single-digit millimetres per year.
Persistent-scatterer InSAR (PS-InSAR) exploits the phase difference between dozens or hundreds of SAR acquisitions over the same target to resolve line-of-sight displacement at the 1 to 3 mm per year level on stable coherent scatterers. Applied to a levee network, it can map which sections are settling uniformly (often acceptable), which are settling differentially across short distances (a warning sign of variable foundation conditions), and which show acceleration over time (an operational alert). The Dutch Rijkswaterstaat and regional water authorities have used Sentinel-1 PS-InSAR for levee monitoring in the Netherlands since the mid-2010s, and published studies from the Po Valley in Italy have demonstrated detection of centimetre-scale seasonal deformation on river embankments using the same approach.
The coherence problem: grass is the enemy
Earthen embankments are, almost by design, vegetated. Grass cover stabilises the slope against surface erosion. It also destroys interferometric coherence. Vegetation scatters radar energy in ways that change with wind, moisture and growth stage, producing phase noise that overwhelms the deformation signal. On a well-maintained grass levee with no hard infrastructure, a C-band PS analysis may return fewer than one coherent scatterer per hectare. That is not enough to characterise deformation along a narrow linear structure.
The practical solutions are two. First, exploit whatever hard targets exist: inspection access roads, concrete aprons, drainage outfall structures, marker posts, fencing and any riprap or stone revetment. X-band sensors (TerraSAR-X, COSMO-SkyMed, ICEYE) are more sensitive to these small hard targets than C-band because the shorter wavelength produces stronger backscatter from centimetre-scale features. Second, install trihedral corner reflectors at planned intervals. A trihedral reflector, typically a folded metal sheet forming three mutually perpendicular faces, returns almost all incident radar energy directly back to the sensor. It appears as a point of very high signal-to-noise in every acquisition, giving a guaranteed coherent scatterer at a known physical location. Reflectors can be surveyed in with differential GPS to provide an absolute displacement reference.
Corner reflectors: what the public record shows
The use of installed corner reflectors to augment PS networks on flood infrastructure is documented in peer-reviewed literature and in operational programmes. ESA-funded work has demonstrated that trihedral reflectors with face dimensions of around 50 cm produce radar cross-sections detectable well above the noise floor in both Sentinel-1 and TerraSAR-X data. Spacing of reflectors at 200 to 500 m intervals along an embankment crest is a practical starting point, adjusted for the density of natural hard targets already present.
Reflector deployment is a physical site operation, not a remote-sensing task, and it carries costs and maintenance obligations. Reflectors can be knocked over by maintenance machinery, stolen, or shifted by frost heave, each of which introduces a false displacement signal. A quality-controlled monitoring programme needs periodic field verification and a protocol for flagging anomalous phase jumps that are inconsistent with neighbouring scatterers.
What the Netherlands and Po Valley programmes actually showed
Published analyses of Dutch levees using Sentinel-1 have identified sections showing line-of-sight displacement rates of several millimetres per year, with spatial patterns consistent with differential peat consolidation in the foundation. The value is not in any single displacement number but in the spatial pattern: a levee settling uniformly at 5 mm per year is far less concerning than one settling at 2 mm per year in most places and 12 mm per year over a 50-metre section.
Po Valley studies have documented seasonal displacement cycles on river embankments, with subsidence in summer linked to soil desiccation and partial recovery in winter. Separating this seasonal signal from a genuine long-term trend requires a time series of at least two to three years. Shorter campaigns risk misclassifying seasonal motion as structural deterioration, or missing a slow trend beneath seasonal noise. This is an honest limit of the method, and any credible monitoring programme should state it plainly.
Combining sensors and integrating with inspection programmes
No single sensor solves the problem. A practical embankment monitoring system layers Sentinel-1 for continuous wide-area coverage and historical baseline, commercial X-band for higher-resolution campaigns on priority sections, and installed reflectors where scatterer density is insufficient. The satellite data does not replace physical inspection. It directs inspectors to the sections that warrant attention, reducing the cost of blanket inspection programmes while improving the probability of finding incipient problems early.
Satellize structures this kind of layered analysis for infrastructure asset owners, running PS-InSAR workflows on open Sentinel-1 archives and adding commercial tasking where the geometry or resolution demands it.
Output is most useful when delivered as a GIS layer showing displacement rate per scatterer, overlaid on the embankment centreline, with time-series plots for flagged sections and a change-alert feed when any scatterer exceeds a defined velocity threshold. Integration with the asset owner's existing inspection database allows satellite-flagged sections to be prioritised in the next scheduled inspection cycle.
Honest limits and what satellite data cannot tell you
PS-InSAR measures displacement in the satellite line-of-sight direction, not vertical settlement directly. Converting to vertical requires geometric correction using the sensor incidence angle, and the result carries an assumption that horizontal motion is negligible. On an embankment that is both settling and moving laterally (slope creep), the decomposition is ambiguous without ascending and descending pass data combined.
Cloud cover does not affect SAR. Temporal decorrelation does. A gap in acquisitions caused by satellite maintenance, or a period of very wet weather that saturates the embankment surface and changes its dielectric properties, can degrade coherence for several acquisitions. The method also has a spatial resolution floor: a feature narrower than roughly one resolution cell will not be resolved, and a deforming zone smaller than the scatterer spacing will be missed entirely. Finally, PS-InSAR detects surface displacement. It cannot directly image the internal condition of an embankment. A piping failure developing deep in the core may produce no measurable surface signal until it is advanced. Satellite monitoring is an early-warning layer, not a substitute for geotechnical investigation.
Typical figures
| Spatial resolution (Sentinel-1 IW) | 5 x 20 m ground range; PS displacement mapped at individual scatterer locations, not as a continuous field |
| Spatial resolution (X-band Spotlight) | 1 to 3 m (TerraSAR-X, COSMO-SkyMed, ICEYE), enabling detection of scatterers on narrow embankment crests |
| Revisit period | 6 days (Sentinel-1 dual-satellite, mid-latitudes); 1 to 4 days with commercial X-band tasking |
| Minimum detectable displacement rate | 1 to 3 mm per year on stable coherent scatterers with multi-year time series; single-acquisition precision lower |
| Measurement direction | Line-of-sight; vertical component extracted with incidence-angle correction, full 3D decomposition requires ascending and descending passes |
| Archive depth | Sentinel-1: from April 2014 (S1-A) and April 2016 (S1-B); TerraSAR-X: from 2007; COSMO-SkyMed: from 2007 |
| Weather sensitivity | SAR is cloud- and rain-independent; surface dielectric changes during saturation events can degrade coherence |
| Minimum time series for reliable trend | 2 to 3 years to separate seasonal deformation cycles from genuine long-term settlement trends |
| Delivery formats | GeoTIFF displacement maps, GeoPackage or Shapefile PS point layers, CSV time-series per scatterer, change-alert JSON feed |
Analytics Satellize can run
| Baseline displacement-rate map | PS-InSAR (StaMPS or similar persistent-scatterer processor) on multi-year Sentinel-1 archive | GIS point layer with line-of-sight velocity (mm/yr) per scatterer, colour-coded by rate and flagged where differential settlement exceeds a defined threshold across adjacent scatterers |
| Seasonal deformation decomposition | Harmonic regression on PS time series to separate annual cycle from linear trend | Per-scatterer report showing amplitude of seasonal signal, linear trend rate and residual, enabling distinction between desiccation cycling and structural settlement |
| Priority section ranking | Spatial clustering of high-velocity or high-differential scatterers along the embankment centreline | Ranked list of embankment sections by deformation severity, formatted for integration with inspection scheduling systems |
| Corner reflector network design | Gap analysis of natural coherent scatterer density along the embankment, cross-referenced with incidence angle and sensor geometry | Recommended reflector spacing and orientation specification, with GPS coordinates for installation, based on Sentinel-1 and/or X-band geometry |
| Change-alert feed | Automated threshold monitoring on rolling PS velocity estimates after each new acquisition | JSON or email alert triggered when any monitored scatterer exceeds a velocity threshold or shows a step-change displacement inconsistent with adjacent scatterers |
| Historical event reconstruction | Retrospective PS-InSAR on archived Sentinel-1 data prior to a known embankment incident | Time-series report showing whether precursor displacement was detectable in the archive, with displacement-rate trajectory in the months before the 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.