Why do lidar surveys miss features near tile boundaries?

Filters such as a local relief model need real terrain data on every side of a point. Near the edge of a processing tile that data is missing and gets padded, so the filter output there is computed partly from invented values. On a 1,000 metre tile with a 45 pixel kernel, that unreliable border is 8.6% of the tile area.

How do you avoid the tile edge problem?

Fetch each tile with a halo at least as wide as the filter kernel radius, run the filter on the padded tile, then discard any detection that falls inside the halo. Alternatively mosaic the whole survey area before filtering at all.

How small a difference can cause a lidar survey to miss a monument?

Ours missed a scheduled round barrow whose footprint was 145 square metres against a minimum area of 150. Five square metres, or about 3%, on a threshold that had never been tested against anything.

How much of the corpus? (The tile boundary that hid a scheduled monument)

Six square kilometres of Dorset, four runs over the same ground, one monument tracked through all of them.

What was recovered? (The tile boundary that hid a scheduled monument)

Two independent causes of a single miss, and a measured area over which the second cause applies.

What did we say about what we could not do? (The tile boundary that hid a scheduled monument)

That three published runs contained an unreported 0.52 square kilometres where the relief model was computed from padding, and that we did not know until we asked why a specific known monument was absent.

Did anybody check it independently? (The tile boundary that hid a scheduled monument)

The barrow is on the national record, so its existence is independent of us. The distances were recomputed from the stored candidate lists rather than quoted from memory, which is a check on the writing rather than on the survey.

Method · Lidar

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

There is a round barrow north east of Hog Hill Barn in Dorset. It is on the National Heritage List, entry 1003237, at E365235 N94988. Three of our survey runs went straight over it and none of them saw it.

It is worth walking through why, because neither reason is a bug and both are the kind of thing that is invisible until you go looking for a specific absence.

Cause one: five square metres

The barrow passed every shape test comfortably. Elongation 1.33 against a limit of 1.8. Circularity 0.72 against a floor of 0.55. Crest ratio 1.99 against 1.35. On every measure of "is this the right shape" it was a clear yes.

Its footprint above the 0.20 metre relief threshold was 145 square metres. The minimum area was 150.

Refuted

Rejected by 5 square metres. The threshold was not chosen against anything. It was a round number that seemed sensible when the filter was written, and it drew a line 3% away from a real monument.

Finding 001, 5 August 2026. Round barrow north east of Hog Hill Barn, National Heritage List entry 1003237.

Lowering the floor to 100 finds it, when the barrow is centred in a tile. The cost, over the same 6 square kilometres, is candidates rising from 67 to 102 before deduplication and from 58 to 83 after. Roughly 50% more triage for one find, which is a real trade and not obviously the wrong one.

Cause two: the edge of a tile

Lowering the threshold was not enough, and the reason took longer to find.

The barrow sits 12 metres from the northern edge of tile 365000_94000. The local relief model uses a 45 pixel smoothing kernel, which needs 22 metres of real ground on every side of a point to compute a meaningful value. At 12 metres from the edge, that context does not exist and the software pads it.

So the relief model there was computed partly from invented data. At every minimum area we tested, 150, 100 and 60, the barrow was missed while it sat near that edge.

How much ground this affects

A 22 metre unreliable border around a 1,000 metre square tile is 8.6% of its area. Across the Dorset survey that came to 0.52 square kilometres where the method did not work, and there was nothing in the output to indicate it.

Which ground fell inside that border was decided by where we started the grid. We started at E364000 N93000 because those are round numbers. That choice, and nothing else, made this particular barrow invisible.

The distances, run by run

Run	Nearest candidate to the barrow
001	144.1 m
002	107.8 m
003	107.8 m
004, with a 40 m tile halo	1.4 m

Straight-line distance from the barrow centre to the nearest candidate in each stored run, recomputed from the candidate lists on 6 August 2026 rather than remembered.

Runs 002 and 003 are identical because lowering the area floor added candidates elsewhere and none of them near the barrow. Run 004 fetched every tile with a 40 metre halo, computed the relief model on the padded tile, then discarded any detection falling inside the halo. The barrow appears at 1.4 metres.

The fix is four lines. The point of this page is not the fix. It is that a survey can have a hole in it that is 8.6% of every tile, produce a clean-looking candidate list, and give you no way at all to notice.

The four questions, applied

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

How much of the corpus?: Six square kilometres of Dorset, four runs over the same ground, one monument tracked through all of them.
What was recovered?: Two independent causes of a single miss, and a measured area over which the second cause applies.
What did we say about what we could not do?: That three published runs contained an unreported 0.52 square kilometres where the relief model was computed from padding, and that we did not know until we asked why a specific known monument was absent.
Did anybody check it independently?: The barrow is on the national record, so its existence is independent of us. The distances were recomputed from the stored candidate lists rather than quoted from memory, which is a check on the writing rather than on the survey.

Sources

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