InSAR basics for geotechnical engineers

Who this page is for: civil and geotechnical engineers who have been handed an InSAR result — a velocity map, a colleague's CSV, a consultant's figure — and need to understand what the measurement actually is before interpreting or defending it. No remote-sensing background is assumed; the radar physics is kept to the minimum that interpretation requires. What it covers: what interferometry measures, what the line-of-sight geometry means for deformation interpretation, what coherence is and why it decides where measurements exist at all, and how the three main processing families differ. What it does not cover: how to run the processors themselves — the workflow manual and the software documentation handle that.

What InSAR measures

Synthetic Aperture Radar satellites illuminate the ground with microwave pulses and record the echoed signal's phase as well as its strength. Compare the phase of two images of the same area taken from nearly the same position at different times, and the phase difference contains the ground's motion along the direction the antenna was looking during those seconds — the line of sight, or LOS. Unwrap the phase differences across an image stack and you can convert them into a displacement history per pixel, then fit a rate per point. That is the essence of InSAR: millimetre-level displacement sensing over wide areas, from space, without touching the ground.

The "synthetic aperture" part is what makes the millimetres possible: as the satellite flies, its antenna synthesizes an aperture hundreds of times larger than the physical one, achieving metre-level ground resolution from 700 kilometres away. Sentinel-1, the workhorse mission for this kind of work, repeats every orbit with the same geometry so that successive images are directly comparable — a twelve-day revisit at the equator, six days where both orbit units overlap.

The line of sight — and why it is the first thing to fix

The satellite does not measure "subsidence". It measures the projection of the full displacement vector onto the LOS direction, which for Sentinel-1 is tilted roughly 30–46° from vertical depending on position in the image swath. A purely vertical movement appears in the LOS at only cos θ of its true magnitude; the rest of the vector is invisible to that geometry. Consequently the most common interpretation error in engineering practice — reading LOS velocity directly as settlement — silently understates vertical rates by twenty to thirty percent. The correction is simple division by cos θ, but it must be done with each point's own incidence angle and with the sign convention of the export understood; the LOS Converter exists to make that step systematic. The deeper limitation follows from the same geometry: a single viewing direction cannot separate vertical from horizontal motion at all. Where slopes may be moving along their own dip, the LOS signal mixes both, and only a second geometry — an ascending and descending track pair — allows the components to be separated.

Coherence — where measurements can exist

Phase comparison only works where the ground surface returns a stable radar signature between the two epochs. This stability is coherence. Urban fabric, rock faces and infrastructure hold coherent scatterers for years; vegetation decorrelates C-band signal within weeks as leaves move and grow; water decorrelates instantly; snow buries the surface seasonally. Coherence therefore decides where InSAR can measure anything at all, long before accuracy enters the discussion. It also explains the central mission choice: C-band (Sentinel-1, ~5.6 cm wavelength) is free and systematic but penetrates vegetation poorly, while L-band (ALOS-2/4, ~24 cm) penetrates canopy substantially better at the cost of access and resolution. When a site pre-check flags forest under C-band as a red flag, this is the physics it is describing.

PS, SBAS and DS — three strategies for the same signal

Raw interferograms are corrupted by atmosphere, orbit error and decorrelation. The processing families are strategies for extracting stable motion anyway. PS-InSAR (persistent scatterer) tracks individual phase-stable targets — buildings, rocks, infrastructure — through long image stacks, accepting sparse spatial coverage in exchange for very stable point behaviour. It suits urban and rocky sites and slow, steady signals. SBAS (small baseline subset) builds many interferograms from short baseline pairs, averages down the noise, and retrieves smoother, denser fields — well suited to subsidence bowls and medium-coherence land cover. DS-InSAR (distributed scatterer) relaxes the point-target requirement by statistically combining neighbouring pixels, recovering measurements over moderate vegetation and mixed terrain where PS fails — at the price of heavier computation. Rapid events (landslides, mining collapses) additionally use DInSAR — single interferogram pairs — where the deformation happens between two epochs and magnitude, not long-term rate, is the question. The site pre-checker recommends among these based on exactly the land-cover and coherence expectations this page describes.

What corrupts the measurement

Four corruption families matter to the interpreter. Atmospheric phase delay — humidity and stratified troposphere imprint centimetre-scale phase patterns unrelated to ground motion; correction models and spatial averaging suppress it, and it is usually the largest single term in an error budget. Geometric uncertainty — imprecise orbits and DEM errors leave phase residuals and shift pixels; precise orbits and modern global DEMs reduce it. Decorrelation noise — the random phase loss described above, worst over vegetation and long time intervals. Unwrapping failure — the phase-unwrapping step can integrate incorrectly in noisy or rapidly deforming areas, producing isolated gross errors that no amount of averaging repairs; these surface as steps and outliers, which is why the time-series tool watches for them. None of these can be fully eliminated; they are bounded, stated and lived with — which is why the error budget exists as a deliverable section rather than a footnote.

What this means for engineering use

Three dispositions follow. First, InSAR is a screening instrument: wide-area, history-rich, cheap per point — ideal for finding where to look — but each point carries geometric and atmospheric uncertainty that field methods do not. Second, every InSAR number needs its geometry corrected and its uncertainty stated before it enters an engineering document; the workbench's conversion and budget tools exist for precisely those two steps. Third, credibility comes from ground truth: GNSS and levelling comparisons are what turn a plausible map into a defensible result. The workflow page chains these dispositions into a pipeline, and the sample dataset guide lets you rehearse the full chain on data with a known answer.

Mapping the vocabulary: what processors call it versus what engineers call it

A surprising share of cross-discipline friction comes from the two professions describing the same thing with different words, so a short glossary earns its place on a basics page. What a processor report calls a persistent scatterer (PS), an engineer can safely read as "a point on something phase-stable — usually a structure or rock — whose motion history is trustworthy". Coherence is the radar's own signal-to-noise statement about a location: high coherence means the phase comparison was valid, low means the measurement there is weaker or absent. LOS velocity is the raw projection described above; deformation rate or displacement without a geometry qualifier is exactly the term to distrust, because it hides whether the geometry correction happened. A reference point in an InSAR product plays the role of a benchmark in levelling: everything else is measured relative to it, so its own stability is inherited by the whole map. Atmospheric phase screen (APS) names the correction layer for the atmosphere family above. Ascending/descending mean the two viewing geometries from opposite flight directions, which together unlock the vertical/horizontal separation. And a time series in the processing sense is precisely what the workbench's time-series tool expects: displacement per epoch per point, from which rates, acceleration and steps are derived. When a consultant's report uses one of these words loosely, asking which precise sense is meant is not pedantry — it is usually where the interpretation error is hiding.

Where to go next

Practical next steps: the CSV format specification explains what your processor's exports look like and how to get them into the workbench cleanly; the sample dataset guide walks a complete worked example; the workflow page maps every stage to its tool; and the LOS Converter is where most engineers should start.