What is ground truthing in remote sensing?

Checking a remotely sensed detection against independent evidence. That can mean excavation, a field visit, or comparison against an existing records database. The three are not equivalent and only the first can date anything.

Does comparing against a heritage list count as ground truthing?

It is the weakest form. It gives a genuine recall figure, because the list is independent of your survey, but it cannot validate a candidate that is not on the list. Absence from a record is not evidence of anything.

Have the candidates on this site been visited?

No. None of them. Every result here was produced from open lidar data on a screen and scored against a records database, and no site has been walked or excavated by us.

How much of the corpus? (Ground truthing: what it means, and who actually does it)

Every candidate in runs 004 to 006 was scored against the National Heritage List. None has been visited.

What was recovered? (Ground truthing: what it means, and who actually does it)

Recall figures against a genuinely independent record, and a precision figure that is a lower bound by construction.

What did we say about what we could not do? (Ground truthing: what it means, and who actually does it)

That no candidate on this site has been checked on the ground by anybody, and that the author has never done fieldwork.

Did anybody check it independently? (Ground truthing: what it means, and who actually does it)

The heritage list is external and was not produced by us, which is what makes the recall figure meaningful. Nobody external has checked our candidates.

Method · Scoring

Ground truthing: what it means, and who actually does it

Ground truthing means checking a remote detection against something independent. Walking the site is the strong form. Comparing against an existing heritage record is the weak form, and it is the one we use, because the author of these surveys has never been in the field. Most lidar coverage does neither and reports the candidate list as the finding.

Records-based checking, not fieldwork. The distinction matters and is stated on every run.

A detection algorithm produces coordinates. Ground truthing is everything you do afterwards to find out whether the thing at those coordinates is what you hoped. There are three grades of it and they are not equivalent, though the same phrase covers all three.

Grade	What it is	What it settles
Excavation	Someone digs.	What the feature is, and often when. The only method that can date anything.
Field visit	Someone walks to the coordinates and looks.	Whether there is anything there, and whether it is modern. Cannot date it.
Records check	Compare the coordinates against an existing heritage database.	Whether the feature is already known. Says nothing about anything unrecorded.

We use the third. It is worth saying plainly why: the author of these surveys has never been in the field, not once, and everything on this site came off a screen. Pretending otherwise would be the cheapest kind of lie available to this project and it would also be the easiest to catch.

What a records check can and cannot tell you

Scoring against the National Heritage List gives you a real recall figure, and that is not nothing. If 121 scheduled monuments sit inside your block and your detector recovers 40, you have learned something solid about what it misses, and you have learned it without leaving your desk.

What it cannot do is validate a candidate. A shape that is not on the list is not thereby refuted. It might be a genuine unrecorded monument, and it might be a slurry pit, and the records check has no opinion. Every precision figure we publish is therefore a lower bound, and every candidate stays a candidate.

Candidate

Most of what our detector produces is unverified and probably modern. We publish the counts anyway, because a candidate list with the unexplained majority quietly removed is a different document from the one the algorithm produced.

Runs 005 and 006, scored 6 August 2026. 128 of 168 Salisbury candidates and 443 of 470 Bodmin candidates fall outside tolerance of any scheduled monument.

The inversion worth watching for

In a lot of coverage, the fieldwork comes first and the lidar gets the credit. A team has been excavating a site for years, a survey is flown, the survey confirms and extends what the excavation already showed, and the headline says the machines found it.

That is not a complaint about the surveys, which are usually careful about this in the paper itself. It is a complaint about what happens between the paper and the press release, and it matters because it teaches readers that remote sensing settles things it does not settle.

The honest form of words

If you have run a records check and nothing else, the sentence is: this candidate is not currently on the heritage record for this area. Not: this is an undiscovered monument. The first is checkable and true. The second is a guess wearing the first one's clothes.

The four questions, applied

The same four we put to every result on this site, turned on this method.

How much of the corpus?: Every candidate in runs 004 to 006 was scored against the National Heritage List. None has been visited.
What was recovered?: Recall figures against a genuinely independent record, and a precision figure that is a lower bound by construction.
What did we say about what we could not do?: That no candidate on this site has been checked on the ground by anybody, and that the author has never done fieldwork.
Did anybody check it independently?: The heritage list is external and was not produced by us, which is what makes the recall figure meaningful. Nobody external has checked our candidates.

Sources

National Heritage List for England. Historic England, Open Government Licence v3, 2026.
LIDAR Composite Digital Terrain Model, 1 metre. Environment Agency, Open Government Licence v3, 2026.

Where this is written up in full

Lost Under the Canopy

Six surveys, 907 candidates, one confirmed. Then the same four questions turned on the most famous survey results in the world.

All four are written and none is on sale. Advance readers can read them first, in exchange for an honest review.