Methodology

How we measure, and where the limits are.

Radar interferometry (InSAR)

Radar satellites image the Earth on a fixed repeat cycle and record the phase of the returned signal. Comparing the phase between passes over the same ground resolves changes in distance at a fraction of the radar wavelength. Stacking hundreds of scenes over years turns this into a time series of ground motion with millimeter scale sensitivity, for millions of measurement points at once.

The public radar archive over the Gulf is continuous since 2015. This is the core of what we do: the history of ground behavior already exists, for every parcel, before any instrument is installed on site.

Why the Gulf is a strong environment for this technique

Radar interferometry performs best on dry, stable, built surfaces. Concrete, asphalt, and rock are excellent reflectors, and arid climates minimize the vegetation change that degrades the signal elsewhere. Dense urban districts in the Gulf yield measurement point densities well above global averages.

The supervised AI pipeline

Our agent system operates the full analysis chain: archive screening and scene selection, interferometric processing in the cloud, atmospheric correction, time series inversion, and quality control. Each step produces an auditable record. Work that traditionally requires specialist teams for weeks completes in hours.

Quality gates

No result reaches a client unless it passes explicit thresholds, including temporal coherence, network coherence, velocity plausibility, minimum time series length, and reference point stability. Results near a threshold are escalated for expert review rather than passed automatically. Every reported velocity carries a confidence interval.

Expert review and signature

Automation provides scale; it does not provide assurance on its own. Every client report is reviewed and signed by our geodesy scientific advisor, and alarm thresholds for monitoring engagements are calibrated together with the client's own geotechnical team during onboarding.

Published precedent

The technique is not experimental. Peer reviewed studies have applied the same satellite interferometry to reclaimed land and arid terrain, validated against ground truth:

  • Remah, UAE. Sentinel-1 persistent scatterer analysis produced dense measurements over sandy terrain and resolved a groundwater driven subsidence bowl, validated against well records (Science of the Total Environment, 2021).
  • Mokpo, South Korea. 79 Sentinel-1 scenes over a city built largely on reclaimed land; 12,000+ measurement points, all above the 0.70 coherence bar we also apply (2025).
  • Shenzhen and Hong Kong. Interferometric results cross validated against 18 GNSS stations; subsidence concentrated in reclamation zones and shallow foundation buildings (2019).
  • Beijing. InSAR verified against precise leveling at a mean error of 1.4 mm/yr across 26 benchmarks (ENVISAT era study, 2016).
  • Busan New Port, South Korea. Long term satellite monitoring of reclaimed port land settling for more than 15 years after placement (2022).

Honest limits

  • Open water and bare, loose sand cannot be measured reliably. Sparse areas are marked as low confidence zones in every report.
  • Coastal humidity gradients add atmospheric noise. We correct with weather model data and network filtering, and the residual uncertainty is reflected in the reported confidence intervals.
  • Satellite measurement complements ground investigation; it does not replace it. Our reports are decision support data, not a geotechnical design service. They tell you where to look, and where to drill.

Want the full technical picture for your portfolio? Request pilot access.

Model transparency

The layers added on top of the measurement.

Damage cost exposure model

Projected settlement is mapped to published building damage categories (Burland and Wroth 1974; Burland, Broms and de Mello 1977) through the angular distortion tiers described above. Each category carries a conservative damage ratio band expressed as a fraction of reinstatement value. TerraWatch never states absolute repair costs. Any currency figure shown is derived from a client supplied indicative asset value and is labeled indicative pending quantity surveyor calibration.

As a qualitative anchor, the NIBS Natural Hazard Mitigation Saves study (2019) reports mitigation spend returning on the order of six to one across hazards. It is presented as an order of magnitude framing only, not a settlement specific figure.

Construction screening inputs

Pre acquisition screening translates observed settlement rates into time to published foundation serviceability references (approximately 25 mm for isolated footings on sand, approximately 50 mm for rafts; Eurocode 7 Annex H, informative, hogging values halved). These are decision support data, not a geotechnical design service. Foundation design parameters must be confirmed by the client's licensed geotechnical engineer, and every issued screening report carries an expert signature.

AI interpretive layer

An expert system generates draft interpretive text per measurement point under a strict rule set: probabilistic language only, uncertainty carried on every figure, line of sight geometry stated, anonymous point codes, and a forbidden vocabulary filter. Generation happens at build time, is validated automatically, and remains a labeled draft until reviewed by the named human expert. No AI text reaches an issued report without that review.

Topographic context

Terrain panels use the Copernicus DEM GLO-30 surface model delivered with the InSAR processing chain, normalized to local water level for readability. Topography is shown as context only; it is not a ground motion product and carries no risk meaning by itself.