Reference
How the instruments work, and what each one cannot see
Every page here answers one question and carries one number we own or one paper traced to its DOI. Written to be checked.
Lidar
What is lidar, and what can it not see?
Lidar measures ground height by timing laser pulses from an aircraft. Filter out the returns that hit vegetation and you get a model of the bare earth under the trees. It sees shape, not age, not material and not anything flush with the surrounding ground. It also cannot tell you what a shape is.
Environment Agency 1 m composite, Open Government Licence v3
Why is one sun angle barely better than noise?
A hillshade lights the terrain from an imaginary sun. Anything running parallel to that light casts no shadow and disappears. We measured it: across 4 square kilometres, eight sun angles found 1,195 distinct features between them, and the average single angle found 17.2% of them. The floor you would get from pure noise is 12.5%.
17.2% mean single-azimuth recall against a 12.5% noise floor
What is a local relief model, and what does it hide?
A local relief model subtracts a smoothed copy of the terrain from the terrain itself, leaving only what sits on top. A half-metre barrow on a steep hillside becomes as obvious as one on a flat field. It also deletes anything wider than its smoothing kernel, and it fails near the edges of your tiles.
A 45 metre kernel needs 22 metres of real ground on every side
Why a survey that only finds bumps will never find a pond barrow
A pond barrow is a deliberate hollow, not a mound. Our detector accepts positive relief only, so it cannot find one at any setting. Over Salisbury Plain it recovered 0 of 5, and no tuning would have changed that. A survey can be blind to a whole monument class and still report a respectable overall figure.
0 of 5 pond barrows, 0 of 6 disc barrows, 0 of 2 henges, 0 of 1 cursus
The tile boundary that hid a scheduled monument
We missed a scheduled round barrow twice over, for two unrelated reasons. Its footprint was 145 square metres against a 150 square metre minimum. It also sat 12 metres from a tile edge, inside the border where the relief model is computed from padding. That border is 8.6% of every tile.
Rejected by 5 square metres, and 12 metres from an edge. 0.52 km² unreliable across one survey
Scoring
Recall and precision, explained on real archaeology
Recall is the share of the real monuments your survey found. Precision is the share of your candidates that are real. They pull against each other, and almost every published lidar survey reports neither. Over 16 square kilometres of Salisbury Plain we scored 33.1% recall and 23.8% precision. Over Bodmin Moor, 28.7% and 5.7%.
Salisbury 33.1% recall, 23.8% precision. Bodmin 28.7% and 5.7%
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.
The tolerance problem: one setting, and the recall figure moves by a factor of five
To score a survey you must decide how close a candidate has to be to count as a hit. That distance is the tolerance, it is almost never published, and it decides the answer. On Bodmin Moor our recall runs from 9.6% at 25 metres to 46.8% at 100 metres. Same detections, same monuments, one setting.
Bodmin recall 9.6% at 25 m, 46.8% at 100 m. Salisbury 21.5% to 48.8%
How to score your own survey against a national monument record
Choose the block by a written rule before you fetch any data. Run the detector unchanged. Pull the monument records for the block. Match candidates to monuments at a stated tolerance, sweep it, and break the result down by monument type. The type breakdown is where the useful information is, because that is where the zeros show up.
Block chosen by rule at 18:20:42, lidar fetched at 18:21:24. Both timestamps in the run file
What would a random candidate list have scored?
Recall rises when you add candidates, whether or not they detect anything. So the number that says what a recall figure means is the one a scatter of random points scores on the same ground. On Bodmin Moor, 470 random points score 28.2%. Our detector scored 28.7% with 470 candidates.
Bodmin: detector 28.7%, random 28.2%. Salisbury: detector 33.1%, random 12.0%
Nobody searches for this. Measured demand across the whole subject is roughly 250 a month, and "machine learning archaeology" is zero. These pages exist because a method question answered with a number gets cited, and because the arguments elsewhere on this site are worth nothing without them.
The surveys behind most of these numbers are written up in full in the books, and advance readers can read them before they go on sale.