Our method · Parallax

Measured 100 m squares in the UK: 7.1 million
UK postcodes with a result: 99%
Radar passes per square: 301
Period covered: 2020–2024

1. Radar from orbit

The European Union’s Sentinel-1 satellites carry a C-band synthetic aperture radar with a wavelength of about 5.55 cm. Radar brings its own illumination, so it works at night and sees through cloud. Each satellite circles the Earth about 14.6 times a day, roughly 700 km up, on a sun-synchronous orbit that brings it over Britain around 6 am and 6 pm.

The satellites pass on both northbound (rising) and southbound (falling) tracks, so every spot is seen from two different angles. That is the idea behind our name: two viewpoints let you separate up-and-down movement from sideways movement.

2. From echoes to millimetres

Every radar echo records not just how strong the return was but its phase, the exact point in the wave’s cycle when it came back. If the ground moves towards or away from the satellite between two passes, the phase shifts. A shift of one full cycle corresponds to half a wavelength, about 2.8 cm, so small fractions of a cycle reveal movements of a millimetre or less.

Comparing hundreds of passes over the same spot, and keeping only points whose echoes stay stable over time (mostly buildings, walls and paved ground), gives a time series of height every six to twelve days.

3. The European Ground Motion Service

We build on the European Ground Motion Service (EGMS), part of the Copernicus Land Monitoring Service. EGMS processes the raw Sentinel-1 data for all of Europe and publishes it openly. We use its Level 3 “Ortho” product: vertical movement on a 100 m grid, combining rising and falling tracks and anchored to GNSS (GPS) stations. Our current data is release 2020–2024.

4. Comparing with the ground around you

Whole regions of Britain drift slowly up or down together, by about a millimetre a year in many places, partly as the land still adjusts after the last ice age. Shown raw, that would make almost every home look as though it were sinking.

So for every 100 m square we build a local reference: the median height history of the ground in a ring 1 to 3 km away, worked out in 1 km blocks. Subtracting it removes the regional drift and also noise that affects a whole area on the same day, such as moisture in the atmosphere. What remains is how your square moves relative to the ground around it, which is what matters for a building.

5. Fitting five years of measurements

For each square we fit a physical model to the 301 measurements:

height(t) = trend × t + ½ × acceleration × t² + yearly cycle + noise

Robust fitting. Occasional bad measurements (for example phase-unwrapping errors of about 28 mm) are down-weighted automatically rather than allowed to drag the trend.
Correlated noise. Radar errors on consecutive passes are related. We model that explicitly (generalised least squares with first-order autoregressive noise) so the uncertainty isn’t understated.
Only real acceleration. A speeding-up or slowing-down term is kept only when the data clearly supports it, using a statistical model-selection test.
The yearly cycle captures clay soils swelling in winter and shrinking in summer.

6. How sure we are

Every result comes with a 95% range. It combines the statistical uncertainty from the fit with a cautious allowance of 0.5 mm a year for errors shared across the whole dataset, which the fit alone can’t see. We test this on simulated data with known answers: our 95% ranges contain the true value 92–95% of the time, even with strongly correlated noise.

See accuracy and limits for what this means for an individual home.

7. Turning numbers into words

Result	When we use it
Steady	Movement under 1 mm a year, or not clearly different from zero once uncertainty is included
Slow movement	1 to 3 mm a year, and more than twice its uncertainty
Worth checking	More than 3 mm a year, and more than twice its uncertainty

Unusually large movements (over 10 mm a year) and irregular patterns carry an extra note recommending independent confirmation. We never describe a home as safe or unsafe.

8. How we check our work

Our fitting engine reproduces the official EGMS velocities to within 0.06 mm a year at test locations.
A nationwide sample of random real postcodes across all ten UK regions found a measurement for 99% of them.
The website and our analysis engine are tested against each other on thousands of cases, so they always give identical answers.
Every report lists the exact EGMS point, data release and processing version it came from.