Tailings dam surface deformation monitoring with InSAR
Interferometric SAR detects millimetre-scale surface displacement on tailings storage facility embankments, providing early warning of precursor deformation weeks or months before visible distress. Sentinel-1 supplies the long archive; ICEYE and Capella add resolution and revisit where it matters most.
Sensors
- Sentinel-1 (C-band SAR, ESA): 5.6 cm wavelength, 5 × 20 m resolution in Interferometric Wide Swath mode, 6-day repeat at mid-latitudes with two satellites. Free archive from 2014 enables multi-year velocity trend analysis via PS-InSAR and SBAS.
- ICEYE X-band SAR constellation: ~1 m resolution spotlight mode, sub-daily revisit capability on tasked targets. X-band (3.1 cm wavelength) improves coherence on dry, compacted embankment surfaces relative to C-band, and resolves finer structural features.
- Capella Space X-band SAR: Spotlight imagery to ~0.5 m resolution. Useful for isolating deformation on narrow crest sections or specific instrumented zones. On-demand tasking with turnaround measured in hours.
- COSMO-SkyMed (X-band SAR, ASI): Stripmap and spotlight modes to 1 m, with a four-satellite constellation offering revisit of 1–4 days. Long operational history since 2007 provides a secondary archive for sites where Sentinel-1 geometry is unfavourable.
Why embankments telegraph failure before it happens
Tailings storage facility failures are rarely instantaneous. The 2019 Brumadinho collapse and the 2015 Fundão failure were both preceded by measurable precursor signals that existing ground instrumentation either missed or reported too late for evacuation. Surface deformation, specifically the slow outward creep or settlement of an embankment face, is one of the most reliable of those precursors. Rates as low as a few millimetres per month, sustained over weeks, indicate internal pore pressure changes, piping, or foundation softening that will eventually overwhelm the structure.
InSAR measures line-of-sight displacement between a radar satellite and the ground by comparing the phase of two SAR acquisitions taken at different times. Where the surface has moved toward or away from the satellite, the phase difference encodes that displacement to sub-centimetre precision. On a tailings dam, that translates to a spatial map of which sections of the embankment are stable and which are accelerating, updated with every satellite pass.
What the processing chain actually does
Raw interferograms are noisy. Atmospheric water vapour alone can introduce apparent displacement signals of several centimetres, easily masking real structural movement. Two processing frameworks address this. Persistent Scatterer InSAR (PS-InSAR) identifies individual pixels, typically rock outcrops, concrete structures or coarse gravel on the crest, that remain phase-coherent across a long time series. By fitting a displacement model to dozens of acquisitions simultaneously, it separates real ground motion from atmospheric artefacts. Small Baseline Subset (SBAS) processing works across distributed scatterers rather than point targets, trading some precision for better spatial coverage on smoother embankment faces.
For a well-maintained, dry embankment surface, PS-InSAR on Sentinel-1 can resolve displacement rates down to roughly 1–2 mm per year in the mean velocity field, with individual epoch uncertainties of 3–5 mm. Those figures degrade on wet or vegetated surfaces, where phase coherence collapses. A freshly deposited tailings beach, a dam face covered in grass for rehabilitation, or a crest saturated after heavy rainfall will all produce sparse or unreliable measurements. That is not a flaw to hide; it is a design constraint that determines where ground-based instruments must fill the gap.
The archive advantage and where it runs out
Sentinel-1 has been acquiring systematically over most of the world since 2014, and ESA's data policy makes that archive freely accessible. For a tailings facility that has never been instrumented with InSAR before, it is often possible to reconstruct a decade of surface displacement history from existing data within days of commissioning an analysis. That retrospective view is genuinely useful: it can reveal whether a section of embankment that looks stable today was actually creeping slowly three years ago, before a raise.
The archive has gaps. Sentinel-1B failed in August 2021, halving the constellation's revisit rate until Sentinel-1C launched in late 2023. Over some regions, particularly parts of the Americas and Africa, systematic acquisition coverage was inconsistent in the early years. For facilities in those gaps, or for any site requiring revisit shorter than six days, commercial X-band satellites are the practical answer. ICEYE's constellation can, in principle, revisit a single point multiple times per day, though the cost of continuous tasking at that cadence is significant and should be weighed against the actual velocity of movement being tracked.
Reading the displacement map: what the numbers mean operationally
A velocity map showing 2 mm per year across the entire downstream face of an embankment is almost certainly normal consolidation. A cluster of pixels accelerating from 5 mm per year to 15 mm per year over three months, concentrated on the upper third of one abutment, is a different matter entirely. The operational value of InSAR is not a single displacement number but the spatial pattern and the time derivative of that pattern.
Acceleration thresholds for alert escalation are site-specific and should be set in consultation with the facility's geotechnical engineer of record. InSAR does not replace that expertise. What it does is give the engineer a spatially continuous picture of the embankment that no practical network of surface extensometers or total stations can replicate. A typical monitoring grid of prism targets might cover twenty to thirty points on a large embankment; a Sentinel-1 PS-InSAR result over the same structure might return several hundred coherent measurement points, and an X-band spotlight acquisition can push that into the thousands.
Satellize integrates InSAR time-series processing with commercial tasking on client licence, and applies the same analytical discipline it brings to its crop-estimation work in Tonga: specific outputs, honest uncertainty bounds, no decorrelated pixels dressed up as data.
Honest limits and how to work around them
Vegetation is the primary enemy of embankment InSAR. Grass or shrub cover on a downstream slope scatters the radar signal differently with every acquisition, destroying the phase coherence that the technique depends on. Facilities that have deliberately vegetated their embankments for erosion control or rehabilitation will see measurement density fall sharply. Short-wavelength X-band is more susceptible to this than C-band, not less, because the signal interacts more strongly with small leaves and stems.
Geometry matters too. InSAR measures displacement in the satellite's line of sight, not in three-dimensional space. A slope moving directly toward or away from the satellite is well-measured; a slope moving horizontally perpendicular to the orbit track is nearly invisible. Most analyses combine ascending and descending orbit passes to decompose the signal into vertical and east-west components, but north-south movement remains poorly constrained. For embankments oriented north-south with predominantly lateral movement, InSAR results should be treated with additional caution.
Finally, very rapid displacement, more than roughly 10–15 cm between acquisitions at C-band, causes phase wrapping ambiguity that standard processing cannot resolve without additional assumptions. By the time a tailings dam is moving at that rate, it is almost certainly already in emergency response. InSAR is a tool for the slow, pre-failure phase, not for real-time crisis tracking.
Typical figures
| Spatial resolution (Sentinel-1 IW mode) | 5 × 20 m ground range; PS measurement spacing depends on coherent target density, typically 20–200 m on embankments |
| Spatial resolution (ICEYE / Capella spotlight) | 0.5–1 m, enabling crest-scale feature discrimination |
| Revisit interval | 6 days (Sentinel-1 two-satellite, mid-latitude); sub-daily possible with ICEYE tasking |
| Minimum detectable displacement rate (PS-InSAR, Sentinel-1) | 1–2 mm per year in mean velocity; single-epoch precision 3–5 mm line-of-sight |
| Radar frequency | C-band 5.405 GHz (Sentinel-1); X-band ~9.6 GHz (ICEYE, Capella, COSMO-SkyMed) |
| Archive depth | Sentinel-1: 2014 to present; COSMO-SkyMed: 2007 to present (tasked); commercial X-band: 2019 to present |
| Processing latency (operational monitoring) | 12–48 hours after acquisition for incremental interferogram; full time-series reprocessing 2–5 days |
| Displacement components resolved | Vertical and east-west via ascending/descending combination; north-south component poorly constrained |
| Delivery formats | GeoTIFF velocity and displacement rasters, GIS point layers (PS/SBAS), time-series CSV per measurement point, PDF engineering summary |
| Surface conditions limiting coherence | Vegetation, saturated tailings beach, fresh loose material; dry compacted rock or concrete surfaces perform best |
Analytics Satellize can run
| Baseline velocity map | PS-InSAR or SBAS time-series inversion on Sentinel-1 archive (minimum 20 acquisitions recommended) | GeoTIFF and GIS point layer showing mean line-of-sight displacement rate across the embankment, with uncertainty per point |
| Incremental displacement alert | Differential interferogram between consecutive acquisitions, thresholded against site-specific baseline | Automated alert (email or API) when any embankment zone exceeds the agreed displacement increment between passes |
| Acceleration detection report | Time-series velocity change detection using moving-window regression on PS/SBAS displacement history | Monthly PDF identifying zones where displacement rate has increased by more than a configurable threshold, with annotated time-series plots |
| Ascending/descending decomposition | Two-orbit geometry inversion to separate vertical and east-west displacement components | Paired GeoTIFF layers (vertical, horizontal) with vector displacement arrows overlaid on optical basemap |
| Retrospective archive analysis | Full PS-InSAR reprocessing of available Sentinel-1 stack from 2014 or earliest available date | Historical displacement timeline for the facility, identifying any prior episodes of anomalous movement before the current monitoring period |
| High-resolution crest monitoring (active watch) | SBAS processing of commercial X-band time series (ICEYE or Capella) on tasked acquisition schedule | Weekly GIS layer at 1–2 m posting on crest and upper embankment zones, integrated with Sentinel-1 background monitoring |
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.