Construction vibration impact proxy from SAR coherence
SAR interferometric coherence drops when millimetre-scale surface change disrupts phase relationships between repeat passes. That decorrelation, mapped spatially, acts as an indirect proxy for vibration disturbance zones around active piling, blasting or heavy plant operations.
Sensors
- Sentinel-1 (C-band, 5.6 cm wavelength): Free, global, 6-day repeat at mid-latitudes with two satellites active. Interferometric Wide Swath mode gives 5 x 20 m resolution. C-band coherence is sensitive to vegetation and loose soil but still resolves coherence loss on hard urban surfaces. Archive from 2014.
- TerraSAR-X / TanDEM-X (X-band, 3.1 cm wavelength): Commercial tasking at 1 m Spotlight resolution with 11-day repeat (shorter with off-nadir tasking). Shorter wavelength increases sensitivity to small surface displacements, which is advantageous for detecting subtle vibration-induced change on rigid structures.
- COSMO-SkyMed Second Generation (X-band): Italian Space Agency constellation offering 1 m Spotlight resolution and flexible revisit down to roughly 4 days. Dual-polarisation modes support coherence analysis on both VV and VH channels, useful for separating building-face returns from ground returns.
- ICEYE (X-band): Commercial small-SAR constellation with sub-1 m Spotlight capability and same-day tasking in some orbits. Useful for rapid-response coherence assessment when a specific piling event is scheduled, though short temporal baselines (hours to days) can conflate weather-driven change with vibration effects.
Why coherence falls when the ground shakes
Repeat-pass SAR interferometry works by comparing the phase of radar echoes from the same patch of ground acquired days or weeks apart. When the surface is stable at the sub-wavelength scale, the two echoes correlate tightly and the coherence value approaches 1. When anything moves or changes between passes, phase becomes random and coherence drops toward 0.
Construction vibration introduces exactly that kind of sub-wavelength disturbance. Piling operations transmit energy through the ground at frequencies typically between 1 Hz and 100 Hz, producing particle velocities that, even at distances of 50 to 200 metres, can reach several millimetres per second. At those amplitudes, loose render, roof tiles, window frames and soil surfaces shift by fractions of a millimetre. That is enough to randomise the phase of an X-band (3.1 cm wavelength) or C-band (5.6 cm wavelength) radar echo. The result is a coherence map where buildings close to the piling rig show anomalously low values compared with the same structures in a pre-construction baseline period.
Reading the spatial pattern, not just the number
The diagnostic signal is not a single low-coherence pixel. It is a spatial gradient: coherence is lowest immediately adjacent to the works and recovers with distance, broadly following the geometric spreading and soil attenuation of ground-borne vibration. In practice, published construction-vibration standards such as BS 5228-2 and DIN 4150-3 define distance-dependent thresholds for particle velocity. The coherence pattern does not directly measure particle velocity, but the zone of statistically significant coherence loss can be compared with those distance thresholds to flag which structures warrant instrument-based follow-up.
Urban geometry complicates the picture. Buildings facing the construction site on their gable ends behave differently from those presenting only rooftops to the satellite. Multi-look processing and careful selection of stable reference pixels within the scene help separate genuine vibration-driven decorrelation from look-angle artefacts. Ascending and descending passes, processed independently, can partially resolve this ambiguity because each geometry illuminates different building faces.
What coherence loss cannot tell you, and why that matters
Coherence loss is non-specific. Rainfall wets surfaces and decorrelates them. Wind moves vegetation. A scaffolding erection or a parked lorry on a rooftop changes the radar cross-section. All of these produce low coherence that is indistinguishable, pixel by pixel, from vibration-induced change. This is not a minor caveat; it is the central interpretive challenge.
The method becomes defensible only when coherence maps are read alongside a construction activity log. If coherence drops sharply in a ring around the site on the same days that piling records show active driving, and recovers during plant shutdown periods, the circumstantial case strengthens considerably. Without that temporal alignment, the coherence map is suggestive at best. Clients should treat the output as a screening tool that directs geotechnical attention, not as a substitute for vibration monitoring instruments such as geophones or seismometers placed at the affected structures.
Temporal baseline choice and the coherence window
The temporal baseline between the two SAR acquisitions used to form an interferogram controls what the coherence value represents. A 6-day Sentinel-1 pair captures change accumulated over that interval. A 12-day pair integrates more change and will show lower coherence in disturbed areas, but also more background decorrelation from weather. For vibration monitoring, short baselines of 6 to 12 days are generally preferred because they reduce the chance that unrelated environmental change dominates the signal.
A pre-construction coherence stack, formed from 10 or more interferograms acquired before works begin, establishes the natural decorrelation level for each pixel in the scene. Pixels that drop below their historical mean by a statistically defined margin during the construction period are flagged. This change-detection framing is more rigorous than comparing a single construction-phase interferogram against an arbitrary threshold.
Practical workflow from tasking to output
For a major urban project, the workflow typically begins 3 to 6 months before piling starts, acquiring a baseline stack on a consistent geometry. Sentinel-1 is often used for this phase because the archive is free and the repeat is predictable. Once works begin, commercial X-band tasking from TerraSAR-X or COSMO-SkyMed is added for higher spatial resolution in the immediate vicinity of the site, where the gradient of coherence loss is steepest and the most sensitive structures are located.
Coherence maps are geocoded to a building footprint layer, so each structure receives a coherence anomaly score rather than a raw raster value. That score is compared with the pre-construction baseline and flagged if it exceeds a threshold agreed with the client's geotechnical team. The output is a GIS layer updated after each satellite pass, overlaid on the site's building condition survey records. Satellize applies this coherence-change workflow on open and commercial constellations, with the analytic configuration adapted to the site geometry and the specific construction method in use.
One honest constraint: processing latency. Sentinel-1 data is typically available within 1 to 3 hours of acquisition via the Copernicus Data Space, but interferogram generation, geocoding and quality-checking add hours to a day of processing time. For a 6-day revisit sensor, the result is still timely enough to inform weekly site meetings, but it is not a real-time vibration alarm.
Typical figures
| Spatial resolution (Sentinel-1 IW mode) | 5 x 20 m (range x azimuth); multi-looked to ~14 m for coherence estimation |
| Spatial resolution (TerraSAR-X / COSMO-SkyMed Spotlight) | ~1 m; coherence typically estimated at 3-5 m after multi-looking |
| Revisit (Sentinel-1, two satellites) | 6 days at mid-latitudes; 12 days with single satellite |
| Revisit (commercial X-band tasking) | 4-11 days standard; shorter with off-nadir scheduling |
| Radar frequency | C-band (5.405 GHz, Sentinel-1); X-band (~9.6 GHz, TerraSAR-X, COSMO-SkyMed, ICEYE) |
| Minimum detectable surface change (phase sensitivity) | Sub-millimetre at X-band; ~2-3 mm at C-band before phase wrapping |
| Coherence baseline stack depth | Sentinel-1 archive from April 2014; TerraSAR-X from 2008 |
| Processing latency after acquisition | Typically 4-24 hours for geocoded coherence layer, depending on pipeline configuration |
| Delivery format | GeoTIFF coherence raster; building-footprint anomaly score layer (GeoPackage or Shapefile); PDF report |
| Cloud sensitivity | None. SAR is all-weather; cloud and rain do not block the signal, though wet surfaces increase decorrelation noise |
Analytics Satellize can run
| Pre-construction coherence baseline | Multi-temporal coherence averaging over 10+ interferograms; per-pixel mean and standard deviation | GeoTIFF baseline coherence map with per-pixel variability layer; used as reference for all subsequent change detection |
| Construction-phase coherence anomaly map | Per-pass coherence differencing against pre-construction baseline; z-score thresholding per pixel | GeoTIFF anomaly raster updated after each satellite pass; flagged pixels exported as vector layer |
| Building-level coherence anomaly score | Zonal statistics over building footprint polygons; median coherence anomaly per structure | GeoPackage or Shapefile with anomaly score attribute; colour-coded for weekly site meeting review |
| Zone-of-influence boundary | Radial coherence profile analysis; distance at which anomaly score returns to baseline noise level | Polygon layer showing estimated disturbance radius; compared against BS 5228-2 or DIN 4150-3 distance thresholds |
| Temporal coherence time series per structure | Per-building coherence value extracted from each interferogram in the stack; plotted against construction activity log | CSV and chart report showing coherence vs. piling activity dates; supports legal and insurance documentation |
| Multi-geometry corroboration analysis | Independent coherence maps from ascending and descending passes compared to separate building-face decorrelation from look-angle artefacts | Dual-geometry anomaly overlay report; confidence classification (confirmed, probable, ambiguous) per structure |
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.