Short answer

Synthetic Aperture Radar (SAR) is an active microwave sensor that illuminates the ground with its own signal and records the echo, so it works in darkness and sees through cloud — exactly when optical sensors fail during floods, eruptions and storms. For terrain and hazards you exploit three products: backscatter (surface roughness and moisture), coherence (change between two passes), and interferometry / InSAR (millimetre-scale ground deformation from the radar phase). The catch is geometry: SAR is side-looking, so steep terrain produces layover, foreshortening and shadow that must be respected before any measurement is trusted.

Why SAR, and why it is different

Optical remote sensing is passive — it records reflected sunlight and is blind at night and under cloud. SAR transmits microwave pulses (Sentinel-1 uses C-band, ~5.6 cm wavelength) and measures both the amplitude and the phase of the return. Three consequences matter for hazard work:

  1. All-weather, day-night. Microwaves penetrate cloud, smoke and rain, so SAR captures the scene during the disaster, not days later when the sky clears.
  2. Phase is information. Because SAR records phase, two acquisitions can be differenced to measure ground motion far below the pixel size — the basis of InSAR.
  3. Geometry, not illumination, dominates. Backscatter depends on surface roughness relative to wavelength, dielectric properties (water content), and the local incidence angle. Smooth water looks dark (specular reflection away from the sensor); rough or built-up surfaces look bright (double-bounce).

The workhorse free dataset is Sentinel-1 (Copernicus): C-band, dual-polarisation (VV+VH common over land), with the Interferometric Wide (IW) swath at ~5×20 m resolution and a 6–12 day repeat depending on coverage.

The three SAR products for hazards

Backscatter

Calibrated backscatter (σ0, in dB) maps surface character. Its strongest hazard use is flood mapping: open water becomes specular and returns almost nothing, appearing as dark patches against the rougher dry land. A change-detection between a pre-event and a during-event image isolates the inundated area. Backscatter also tracks soil moisture and vegetation structure, both relevant to landslide and drought context.

Coherence

Interferometric coherence measures how correlated two acquisitions are, pixel by pixel, from 0 (totally changed) to 1 (unchanged). A sudden coherence loss between two dates flags physical disturbance — a landslide scar, a lava flow, building collapse, or a freshly flooded field. Coherence change maps are a fast, robust way to delineate the footprint of an event even where backscatter alone is ambiguous.

InSAR — ground deformation

This is SAR's signature capability. The phase difference between two acquisitions over the same area encodes any change in the sensor-to-ground distance along the line of sight (LOS). After removing the topographic phase (using a DEM) and the flat-Earth phase, the residual interferogram maps deformation in fractions of the wavelength. With C-band, one full interferometric fringe corresponds to ~2.8 cm of LOS displacement (half the wavelength, because the path is two-way).

Single-pair (DInSAR) interferograms show co-seismic or co-eruptive deformation. For slow processes — subsidence, slope creep, mining-induced settlement — time-series InSAR methods, principally Persistent Scatterer Interferometry (PSI) and SBAS, stack dozens of acquisitions to reach millimetre-per-year precision over stable, coherent targets such as rock outcrops and buildings.

Geometric distortions you cannot ignore

SAR looks sideways, so relief distorts the image in ways that have no optical equivalent. In mountainous terrain — exactly where landslide and slope hazards live — three effects dominate:

  • Foreshortening: slopes facing the radar are compressed into fewer pixels, appearing bright and short.
  • Layover: when a slope is steeper than the incidence angle, the top of a feature is recorded before its base, so it "lays over" the terrain in front of it. The geometry is irreversibly scrambled there.
  • Radar shadow: slopes facing away from the radar steeper than the look angle receive no illumination and return nothing.

Because Sentinel-1 acquires in both ascending and descending orbits with opposite look directions, a slope in layover on one geometry may be visible on the other. Mapping deformation on steep terrain therefore usually combines both orbits, and a slope's true 3D motion can only be partly resolved from a single LOS — InSAR measures the LOS projection, not the full vector.

Worked example: a Sentinel-1 deformation interferogram in SNAP

Using ESA's free SNAP toolbox:

  1. Download two Sentinel-1 IW SLC products over the area, same orbit (both ascending or both descending), same relative orbit number, from the Copernicus Data Space Ecosystem.
  2. Coregister the pair with the S-1 TOPS Coregistration operator, which uses orbit files and a DEM (SRTM 1-arcsec by default) to align them to sub-pixel accuracy.
  3. Form the interferogram (Interferogram Formation) and inspect coherence; low coherence over vegetation means unreliable phase there.
  4. TOPS Deburst to merge the SLC bursts into a continuous scene.
  5. Remove topographic phase with Topographic Phase Removal using the DEM.
  6. Goldstein phase filtering to reduce noise, then Phase Unwrapping (via SNAPHU) to convert wrapped fringes into continuous displacement.
  7. Convert phase to displacement and Range-Doppler Terrain Correction to geocode the result into a map CRS (e.g. EPSG:4326), which is also when layover/shadow masks are applied.
  8. Validate against GNSS station velocities or levelling where available.

The output is a LOS displacement map. Interpreting it as "subsidence" or "uplift" requires knowing the look geometry and, ideally, combining ascending and descending tracks.

Pitfalls and why they happen

  • Decorrelation over vegetation and water. C-band coherence collapses over dense canopy and changing water between passes, so InSAR is unreliable there. It happens because the scatterers physically move; use coherence as a mask, not a result.
  • Atmospheric phase screen. Tropospheric water vapour adds phase that masquerades as deformation, sometimes centimetres in a single pair. Time-series methods average it out; single interferograms can be badly biased.
  • Reading LOS as vertical. A single interferogram measures the line-of-sight component only. Horizontal motion (e.g. a translational landslide) projects partially into LOS and is easily misread.
  • Ignoring geometric distortion. Measuring deformation in a layover zone is meaningless — the phase there is a mixture of multiple ground points.
  • Wrong polarisation for the task. VV is generally better for flood/open-water contrast; VH adds sensitivity to vegetation volume scattering. Choosing blindly weakens the signal.

Quality control

  • Check the coherence map first; trust phase only where coherence is high (typically > 0.3–0.4 for C-band).
  • Apply and inspect the layover/shadow mask after terrain correction; exclude those pixels from interpretation.
  • Cross-validate deformation against independent ground measurements (GNSS, levelling) or against the opposite orbit.
  • Confirm geocoding by overlaying the terrain-corrected product on a reference layer and checking alignment in flat areas.
  • For flood backscatter, verify the dark-water classification against known water bodies and the flood's plausible extent.

Bathyl perspective

We use SAR where optical data cannot answer the question — through cloud, in darkness, and for ground motion measured in millimetres. Every SAR product we deliver states the orbit geometry, polarisation, coherence threshold and distortion masks, so a reviewer knows where the measurement is solid and where the radar geometry forbids a confident read.

Related reading

Sources