55,769 British addresses sit on fast-sinking ground. 67% of that ground is peat.

Where the addresses are: the coalfield

The patches with the most addresses on them, every one in the Yorkshire–Nottinghamshire coalfield. The evidence column is the honest three-state result: a documented shallow-workings zone is strong evidence; “in the coalfield” is a consultation-area overlap — consistent with mining legacy, not proof.

The residue: 36 patches with no mapped cause

After every layer, 36 patches carry no mapped cause — and they are not scattered randomly. They clump. The Fen-edge group is the most telling: ground that was peat, wasted so far it is no longer mapped as peat, and still sinking — the skirt of the lost fens. The Bovey Basin group sits on Devon's ball-clay country, where quarrying appears in no national layer. Coordinates included so anyone can check.

The full verdict split

Of 3,633 patches: 2,740 explained by a strong layer (mapped peat for most of the area; documented coal workings; historic landfill; clay with a measured seasonal cycle), 699 with a candidate present but only weak, susceptibility-grade evidence — a suspect named is not a case closed — 158 Scottish patches that passed everything except the landfill test Scotland has no open register for, and 36 with no mapped cause. By addresses: 6,516 explained, 27,853 candidate-present, 55 on no-mapped-cause ground. Download everything: all 3,633 patches with verdicts and evidence (CSV).

Method — so the result can be checked

1. Measure: EGMS Ortho vertical velocities (2020-2024 release), 100 m grid, GNSS-anchored; 6,621,790 points on Great Britain after clipping to the coastline.
2. Threshold and cluster: cells at -5 mm/yr or below (225,760), merged with neighbours; clusters under 10 cells dropped as speckle. Sensitivity: at −3 the hit count is 578,932, at −8 it is 47,261.
3. Count addresses at the signal: an OS Open UPRN address counts only if its own 100 m cell is a hit cell — never because it is near one.
4. Subtract, layer by layer, three outcomes each — hit, pass, or not-tested: Coal Authority development high-risk areas and coalfield consultation areas; historic landfill (England and Wales; Scotland has no open register, so Scottish clusters are marked not-tested, never silently passed); mapped peat (Natural England, NatureScot carbon-and-peatland classes, Wales's deep-peat map); and six BGS GeoSure susceptibilities. Susceptibility is weak evidence by definition; documented deposits and workings are strong. Shrink–swell clay upgrades to strong only when the patch itself shows a measured seasonal cycle.
5. Controls before verdicts: the coalfield towns must surface with coal attribution; the Pennine and Fen peat must classify as peat; central London must not produce false clusters (its mean is -1.2 mm/yr); the Highlands must rise — Inverness shows +1.44 mm/yr of post-glacial uplift; and Ripon's gypsum belt taught us a limit worth stating: collapse there is episodic sinkholes, which a five-year velocity cannot see, so karst absence here is not karst safety.

Ground motion © European Union, Copernicus Land Monitoring Service (EGMS), 2020-2024. Addresses: OS Open UPRN. Hazard layers: Coal Authority, Environment Agency, Natural England, NatureScot, Welsh Government, BGS. We did not produce the measurements; we produced the join. Analysis dated 2026-08-06.

Is a sinking address a damaged house?

No. These are ground-surface velocities measured by satellite radar at 5 mm a year and faster — slow, broad movement of the ground itself, not a verdict on any building. Uniform settlement can leave a structure untouched; what damages buildings is differential movement across a single footprint, which a 100 m grid cannot see. This is screening-grade analysis from public data, not a structural survey.

Why is so much of it peat?

Because drained peat genuinely wastes away — it oxidises and the surface lowers, and on upland bogs the surface also rises and falls seasonally with water ("bog breathing"), which the radar sees as a seasonal cycle on top of the trend. 67% of Britain's fast-sinking ground sits on mapped peat: the Pennines, the Forest of Bowland, the Flow Country, and the lowland Fens. Hardly anyone lives on it — which is why the address map and the area map tell such different stories.

Are the coalfield towns being damaged by old mines?

We can say less than you might expect, and the table says exactly how much. The towns with the most addresses on fast-sinking ground — Doncaster, Wakefield, Mansfield, Nottingham's northern wards — are all in the coalfield, and sinking there is consistent with mining legacy. But our strong evidence layer (the Coal Authority's development high-risk areas) maps shallow workings, and most of these clusters sit outside it, so they carry the weaker "in the coalfield" attribution, openly. Deep-mining aftermath — collapsing old workings, minewater changes — is the plausible candidate the open data cannot confirm site by site.

Why "no mapped cause" instead of "unexplained"?

Because our tested set is eight mapped layers — coal (two layers), historic landfill, peat (three national maps), and six BGS GeoSure ground-hazard susceptibilities — and a cause can be real without being mapped. Groundwater change has no open GB-wide layer and is untested here, stated openly. Scotland has no open historic-landfill register, so 158 Scottish patches that passed everything else are counted separately as "untested remainder", never as mystery. What survives all of that is small — 36 patches — and it clumps in a handful of specific places, which is what makes it interesting.