Dam and embankment deformation detection with SAR
Satellite SAR can detect sub-centimetre surface displacement on dam faces and embankments weeks before visible distress appears. This page explains which sensors work, where coherence fails, and how satellite data integrates with piezometer readings.
Sensors
- Sentinel-1 (C-band, 5.6 cm wavelength): IW mode gives 5 x 20 m ground resolution with 6-day revisit at mid-latitudes using both satellites. Free and open archive from 2014. C-band coherence degrades rapidly over vegetated embankments; bare concrete or rip-rap faces perform far better. Displacement sensitivity via DInSAR is typically 3-10 mm in the line-of-sight direction under good coherence conditions.
- ALOS-2 PALSAR-2 (L-band, 23.6 cm wavelength): L-band penetrates vegetation canopy and maintains coherence over grass-covered embankments where C-band loses phase entirely. Stripmap mode delivers 3 m resolution; ScanSAR offers 60-350 m at wider swath. Revisit is 14 days for a single satellite. Particularly valuable for vegetated earth-fill dams in tropical or temperate climates.
- ICEYE X-band constellation: X-band (3.1 cm wavelength) at 0.5 m Spotlight resolution enables very high spatial detail on concrete dam faces, spillway walls and instrumented sections. The ICEYE constellation can achieve sub-daily revisit through tasking coordination. Short wavelength means high phase sensitivity but also faster coherence loss if any vegetation or moisture change occurs between passes.
- COSMO-SkyMed (X-band): Italian Space Agency constellation of four satellites. Spotlight mode reaches 1 m resolution with revisit intervals of a few days depending on latitude and tasking priority. Established archive from 2007 provides historical baselines useful for detecting slow, multi-year deformation trends on ageing structures.
What a millimetre of movement actually means
Most dam failures are not sudden. Seepage piping, slope creep and foundation settlement all produce surface displacement that accumulates over months or years before any visible crack appears. The engineering question is not whether a dam moves, because all large structures settle to some degree, but whether the rate or pattern of movement is changing. That is precisely what differential SAR interferometry (DInSAR) is designed to detect.
DInSAR works by comparing the phase of radar returns from two passes over the same area. A change in the distance between satellite and surface of half a wavelength produces a full phase cycle. For Sentinel-1 at C-band, that half-wavelength is roughly 2.8 cm; for ALOS-2 at L-band it is about 11.8 cm. Persistent Scatterer Interferometry (PSI) and Small Baseline Subset (SBAS) methods stack many interferograms to push displacement sensitivity below 1 mm per year on stable, coherent targets. A concrete dam face, a rip-rap apron or an exposed instrument housing can all serve as persistent scatterers. The embankment slope covered in grass is a different problem entirely.
Where coherence goes and what to do about it
Coherence is the correlation between phase values across two SAR acquisitions. It drops toward zero when surface scatterers change between passes, which happens with vegetation growth, soil moisture variation, freeze-thaw cycles and surface disturbance. A grass-covered earth-fill embankment in a temperate climate can produce coherence values below 0.3 at C-band over a 12-day interval, rendering standard DInSAR unreliable. L-band mitigates this because longer wavelengths interact with the soil beneath the canopy rather than the canopy itself, preserving coherence over moderate vegetation cover. This is the primary operational reason to use ALOS-2 on vegetated earth-fill dams despite its higher data cost and 14-day revisit.
When coherence is lost entirely, offset tracking (also called intensity cross-correlation or pixel-offset tracking) provides a fallback. Rather than comparing phase, it measures the shift of amplitude patterns between two images by cross-correlating small sub-windows. Detection limits are coarser, typically around one-tenth to one-twentieth of the pixel spacing, so roughly 0.5-2 m for Sentinel-1 IW mode. That is too coarse for routine monitoring but adequate for detecting rapid, large-scale slope failures or embankment slumping during a flood event. High-resolution X-band data from ICEYE or COSMO-SkyMed narrows the detection floor significantly when offset tracking is the only viable method.
Temporal decorrelation is not the only coherence enemy. Geometric decorrelation matters when the satellite viewing angle differs between passes, and volumetric decorrelation occurs when radar penetrates a volume rather than reflecting from a discrete surface. Operators should check perpendicular baseline values carefully; for Sentinel-1, baselines above roughly 150 m begin to degrade interferogram quality meaningfully.
Reading the geometry: what line-of-sight displacement actually measures
SAR measures displacement only in the radar line-of-sight (LOS) direction, which is a combination of vertical and horizontal movement projected onto the slant range vector. For a typical Sentinel-1 ascending pass at around 39 degrees incidence angle, the LOS is sensitive to roughly 78% of vertical displacement and around 60% of horizontal displacement in the range direction. Movement along the satellite flight track (azimuth) is almost invisible in standard interferometry.
For a dam embankment, this geometry has practical consequences. Horizontal crest movement toward the downstream face, which is a classic precursor to slope failure, has a significant range component and will appear in the LOS signal. Vertical settlement at the crest will also appear clearly. Lateral spreading parallel to the dam axis is the hardest displacement mode to detect from a single geometry. Combining ascending and descending passes, which Sentinel-1 can provide over most land areas, allows decomposition into approximate vertical and east-west horizontal components, though the north-south component remains poorly constrained.
Cross-validating satellite data with piezometers
Piezometers measure pore-water pressure within an embankment. Rising pore pressure reduces effective stress and is the proximate cause of most seepage-related failures. Satellite SAR cannot measure pore pressure directly. What it can do is detect the surface expression of the internal process: if rising pore pressure is causing the embankment to swell or deform, that deformation may appear in the LOS displacement field before it reaches the threshold that triggers a piezometer alarm.
The integration is therefore complementary rather than redundant. A piezometer network provides high-frequency, subsurface, point measurements. SAR provides spatially continuous surface displacement at the satellite revisit interval. Agreement between an anomalous piezometer reading and a co-located surface displacement signal substantially increases confidence in both. Disagreement is also informative: surface displacement without piezometer response may indicate shallow slope movement unrelated to seepage; piezometer response without surface displacement may indicate deep internal erosion not yet expressed at the surface. Neither sensor alone tells the full story.
Practically, integrating these data streams requires a common time reference and a georeferenced piezometer inventory. Satellize supports this integration by delivering displacement time-series as georeferenced GIS layers that can be overlaid directly on instrumentation maps, a workflow we apply across open-constellation and commercially tasked data.
Sensor selection by dam type and setting
Concrete gravity and arch dams with exposed faces are the easiest targets. X-band sensors such as ICEYE and COSMO-SkyMed provide millimetre-scale sensitivity on the dam face itself, and persistent scatterers are abundant on the concrete surface. Sentinel-1 C-band is adequate for routine monitoring of concrete structures and is cost-effective given its open-data status.
Vegetated earth-fill and rock-fill embankments in humid climates demand L-band. ALOS-2 PALSAR-2 is currently the only widely available spaceborne L-band SAR with a public tasking pathway, though NISAR (a NASA-ISRO mission) is expected to add free L-band data after its planned launch. Until then, operators monitoring grass-covered embankments should treat C-band DInSAR results with caution and use L-band as the primary coherent signal, supplemented by C-band offset tracking for event-scale rapid assessment.
Archive depth matters for ageing infrastructure. COSMO-SkyMed data extends to 2007, Sentinel-1 to 2014, and ALOS-2 to 2014 (with ALOS-1 PALSAR providing earlier L-band coverage back to 2006). A structure showing accelerating displacement over a decade of archive data presents a very different risk profile from one that has been stable throughout. That historical baseline is often the most valuable product a satellite analysis can deliver to an asset owner or regulator.
Honest limits and what satellite SAR cannot do
SAR displacement monitoring has genuine constraints that any responsible procurement should acknowledge. Cloud cover does not affect SAR, which is a real advantage over optical sensors, but heavy rainfall between passes changes soil moisture and can decorrelate C-band data over embankments even without vegetation. Atmospheric water vapour introduces apparent path-length changes that mimic displacement signals at the centimetre scale; tropospheric correction using weather model data or GNSS networks is standard practice but adds processing complexity and residual uncertainty.
The minimum detectable displacement for routine monitoring, after atmospheric correction and temporal averaging, is realistically 3-5 mm in LOS for PSI methods on coherent targets, and perhaps 10-20 mm for less coherent surfaces using SBAS. Offset tracking on standard-resolution data has a floor closer to 0.3-1 m. These figures assume adequate coherence; below a coherence threshold of roughly 0.3, phase-based methods should not be trusted regardless of what the displacement map shows.
Finally, satellite SAR cannot replace structural inspection or geotechnical investigation. It identifies where to look and when to look harder. A displacement anomaly detected from orbit should trigger ground investigation, not substitute for it.
Typical figures
| Spatial resolution (best available) | 0.5 m (ICEYE Spotlight); 1 m (COSMO-SkyMed Spotlight); 3 m (ALOS-2 Stripmap); 5 x 20 m (Sentinel-1 IW) |
| Revisit interval | Sub-daily (ICEYE tasked); 6 days (Sentinel-1, dual satellite); 14 days (ALOS-2); 3-5 days (COSMO-SkyMed tasked) |
| Radar frequency / wavelength | C-band 5.6 cm (Sentinel-1); L-band 23.6 cm (ALOS-2); X-band 3.1 cm (ICEYE, COSMO-SkyMed) |
| Displacement sensitivity (PSI/SBAS, coherent target) | 1-5 mm per year in line-of-sight under good coherence; ~10-20 mm for low-coherence SBAS |
| Offset-tracking detection floor | ~0.3-2 m depending on pixel spacing; suitable for large rapid movements only |
| Coherence suitability (vegetated embankment) | L-band preferred; C-band often decorrelates below 0.3 threshold over grass in 6-12 days |
| Archive depth | Sentinel-1 from 2014; ALOS-2 from 2014; COSMO-SkyMed from 2007; ALOS-1 L-band from 2006 |
| Delivery formats | GeoTIFF displacement maps, CSV time-series per point, GeoPackage / Shapefile alert layers, PDF monitoring reports |
| Atmospheric correction | ERA5 / GACOS tropospheric correction applicable; residual uncertainty typically 5-15 mm per acquisition |
Analytics Satellize can run
| Baseline displacement map | SBAS InSAR stacking over 12-24 month archive | GeoTIFF velocity map (mm/year in LOS) with coherence mask; PDF summary report |
| Persistent Scatterer time-series | PSI on coherent point targets (concrete, rip-rap, instrument housings) | GeoPackage of PS points with full displacement time-series; quarterly trend report |
| Ascending / descending decomposition | Geometric decomposition of dual-geometry LOS into vertical and east-west horizontal components | Paired GeoTIFF layers (vertical settlement, horizontal range displacement) for engineering review |
| Coherence loss alert | Automated coherence threshold monitoring per acquisition pair | Alert feed flagging embankment zones where coherence drops below 0.3, triggering L-band or offset-tracking fallback |
| Offset-tracking displacement (event response) | Intensity cross-correlation on pre- and post-event image pairs | GeoTIFF showing range and azimuth displacement fields; rapid-turnaround PDF within 48 hours of tasking |
| Piezometer cross-validation overlay | Spatial join of InSAR displacement time-series with georeferenced piezometer network | GIS layer correlating surface displacement anomalies with piezometer locations; annotated for engineering handover |
| Long-term trend change detection | Statistical breakpoint analysis on multi-year PS or SBAS time-series | Report identifying acceleration events with date, magnitude and spatial extent; suitable for regulatory submission |
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.