An open ontology for land register integrity, tested against all 33 Scottish registration counties

Published 21 August 2026. Data retrieved 21 August 2026 under the Open Government Licence v3.0.

The Land Reform (Scotland) Act 2025 regulates land by area. It attaches duties to landholdings of 1,000 hectares or more. Scotland’s national cadastral dataset publishes no areas.

The INSPIRE Cadastral Parcels dataset, which Registers of Scotland publishes to satisfy the INSPIRE Directive, defines an areavalue attribute as the “registered area value giving quantification of the area projected on the horizontal plane of the cadastral parcel”. We downloaded all 33 registration counties on 21 August 2026, 1,564,345 parcels in total, and checked what that field contains. It contains the literal string UNPOPULATED on every parcel, and that is the only value the field takes anywhere in Scotland.

Five further attributes behave the same way: referencepoint, beginlifespanversion, endlifespanversion, validfrom and validto. That is six mandated attributes across 1,564,345 parcels with one distinct value between them.

Why a magic word is worse than an empty field

INSPIRE provides a mechanism for an attribute that has no value. It is called a nil reason, and it carries why the value is absent: unknown, inapplicable, withheld. Writing the word UNPOPULATED into the field instead does two things. It destroys the declared type, so a quantity becomes text, which is why the field arrives as a string in every downstream republication we examined. And it collapses every possible reason for the absence into one uninformative token, so a consumer cannot tell whether the area is unknown to the Keeper, inapplicable to this parcel, or simply not carried in this particular product.

The area is computable, and that is the problem

The polygons are published, so an area can be computed. There are two obvious ways to do it, both defensible, and they do not agree.

Summing double counts, and it does so by design. Registers of Scotland states that overlapping and stacked polygons are a valid part of its data model, that holes such as removals and exceptions are indicated with an overlapping polygon, and that polygons representing holes are not removed from ownership polygons before they are included in the dataset. Add the parcels up and you count some ground twice, and some excluded ground as though it were included.

There is no correction factor

If the distortion were roughly constant, a practitioner could apply a haircut and move on. It is not. Across the 33 registration counties it ranges from 1.33 per cent to 95.40 per cent, a spread of about seventy to one.

Fourteen of the 33 counties shown. The full table is in the repository.

The threshold flips on real land

This is not a theoretical concern. We grouped the published parcels into connected clusters of touching or overlapping polygons, and measured each cluster both ways. Of the 270 clusters that reach 1,000 hectares by summation, 20 fall below 1,000 hectares by union. The largest is a cluster of 341 parcels in Angus which measures 2,950.2 hectares if you add its parcels up and 172.2 hectares if you take its union, an inflation of about seventeenfold. A cluster of 1,131 parcels in Midlothian measures 1,983.4 hectares by summation and 61.2 by union.

These clusters are spatial groupings, not ownership holdings, and we make no claim that anyone owns them. They are here to show that the aggregation arithmetic the Act requires is unsafe when performed on this dataset, which is a statement about the data and not about any landowner.

What this means for the threshold

A landholding under the Act is an aggregate, assembled across a single owner and connected persons, and contiguity is not required. Assessing it against 1,000 hectares means adding parcels up. That is precisely the operation the register’s own data model makes unsafe. The consequence is that a holding’s statutory status can be an artefact of the arithmetic chosen rather than a fact about the land, and the ontology in this repository is built to make that explicit: an area is modelled as a dated claim by a named source computed by a stated method, never as a property of a parcel.

The policy backdrop, from the government’s own figures

Scottish Government statistics, published as RDF and queryable without a key, record the area of land in community ownership rising from 51,971.5 hectares in 2000 to 213,803.22 hectares in 2024. The trajectory has flattened hard. Between 2023 and 2024 the total grew by 8.46 hectares.

What we did not find

We tested every parcel against the identifier rules Registers of Scotland states in its own specification, namely SCT followed by a title id padded with zeros to ten digits, and the namespaced INSPIRE form of the same value. There were zero violations of either rule and zero duplicate labels across all 1,564,345 parcels. That hypothesis died, and it is reported here because a null result is a result. On this evidence Registers of Scotland conforms to its own declared identifier rules.

Where the register is good, measured

We also tested three guarantees Registers of Scotland states in its own specification, across all 1,564,345 parcels. It passes all three.

Its specification says “there should be no duplicated polygons within the dataset”, and there are none. Geometry validity is two failures in 1.56 million. This also corrected a gap in our own first pass, which repaired invalid geometry with a zero-width buffer without counting how often it had to, and so would have let a source with poor geometry hygiene pass as clean.

Carbon, and why this matters beyond conveyancing

A land management plan under the Act is expected to address carbon, which in Scotland mostly means peat. Sizing peat on a holding means overlaying it on the national habitat map, and that calculation inherits every problem on this page before it meets problems of its own. We measured that separately in measuring peat on a Scottish landholding, which compares all 8.1 billion pixel pairs of the two national habitat maps. Scotland’s raised and blanket bog covers 1,330,616 hectares, and the peat class is clean in both maps.

What we got wrong, and corrected

Two errors of ours are worth stating, because a study that audits other people’s data should hold its own to the same standard.

We reported that our second validation engine could not match focus nodes for a common SHACL targeting construct. That was wrong, and it was wrong because we had invoked the tool incorrectly. Run correctly, the two engines agree exactly on the structural layer and on the cross-source layer that carries the argument. The single genuine disagreement isolates to one unimplemented constraint, which has since been fixed in that engine along with two related defects, all of which caused it to report a pass over data it had not checked.

Running the two engines against each other also exposed a real error in our own shapes, an unconstrained cross product that inflated one layer’s results, and a later audit found two modelling errors and a self-contradiction in our own ontology: a property declared on too narrow a class, and an identifier scheme whose declared length disagreed with its own pattern. All are fixed, and both audits now run in continuous integration so they cannot return. The corrections are recorded in the build report rather than applied silently.

Scope, stated plainly

These findings are about what the register says and does not say. They are not estimates of who owns Scotland, and no ownership data is used anywhere. Resolving an INSPIRE identifier to a title number requires a paid Registers of Scotland service. Registers of Scotland states that approximately 30 per cent of Land Register titles are absent from this dataset, that leases and properties without a determinable single ground level owner are excluded by design, and that the Land Register is still replacing the General Register of Sasines, with county operational dates staggered between 1981 and 2003. Every area figure above is an area of published cadastral parcels, never an area of Scotland.

How it was verified

Every headline is computed twice, once set-based in Python over the measurement output and once by SPARQL over the emitted assertion graph, by code paths that share nothing. The governance script exits non-zero if the two disagree. The ontology, the scheme registry and all three SHACL layers are validated and linted by two independent engines. The harvest is resumable, the data is regenerable rather than committed, and the build report records what could not be obtained, including that the Registers of Scotland web host returns HTTP 403 to this client on every path we tried, which blocks the official identifier cross-reference file.

Repository

The ontology, the SHACL layers, the pipeline and the build report are public.

github.com/fabio-rovai/scotland-land-register-ontology (opens in new tab)

Contains public sector information licensed under the Open Government Licence v3.0. © Crown copyright. Reproduced with the permission of Registers of Scotland. Contains OS data. © Crown copyright and database right 2026.