The Mystery of the Shrinking Barrows

Lidar investigations

Over twenty years, the round barrows on land returned to grass around Stonehenge, much of it National Trust land, appear to have lost height. The average loss is 5 cm, about two inches, and the worst is more than a hand's width. Barrows in neighbouring fields whose use never changed have not moved. Either something is wearing down scheduled monuments in one of the most closely protected landscapes in Britain, or the lidar is being fooled. This post is the investigation.

The LiDAR differences and features can be explored at

https://lidar-lens-public.tim-daw.workers.dev/stonehenge-change/

How it started

Since 2000 most of the arable land in the Stonehenge part of the World Heritage Site has gone back to grass. The National Trust stopped arable farming on its holding that year, and by 2019 more than 526 ha was in permanent pasture (National Trust; National Trust 2019). Map 3 of the 2009 WHS Management Plan dates the reversions field by field from 2000 to 2011. Map 4 of the 2015 plan adds further fields reverted by 2015, without dates (WHS Management Plan 2015).

Two Environment Agency lidar surveys bracket the change. The first was flown on 8 December 2001 (survey P_2053, 1 m last-return DTM), while much of the land was still under the plough. The second is the National LIDAR Programme survey of 4 April 2021. The aim was modest: to see whether faint earthworks had begun to reappear once ploughing stopped. They had not. A search of the reverted ground found five patches of new relief, and all of them are modern, the largest being the turning circle for the visitor buses at SU 1207 4236. The barrows, however, did something else.

The oddity

Each mound was measured the same way in both surveys: a plane fitted to the ground beyond the mound was subtracted, leaving the mound's height above its surroundings. Seven mounds in fields reverted between 2003 and 2007 were on average 0.06 m lower in 2021. Eight mounds in grass since before 2000 had not changed, and nor had 27 that were still arable in 2008. The worst cases were a group of three bowl barrows (NHLE 1009620, −0.15 m), the 2 m bowl barrow north of Springbottom Farm (1009625, −0.11 m) and the long barrow north of Normanton Gorse (1008953, −0.06 m).

Figure 1. Normanton Down, reverted to grass 2003–2007. Left: December 2001. Centre: April 2021. Right: change in local relief, 2021 minus 2001; red = relatively lower now. 1 m grid, σ = 1 m smoothing, hillshade from the NW at 35°.

Figure 2. The control: Old King Barrows and the Avenue, grassland by 2000. Same processing and scale as Figure 1. No systematic change.

A few centimetres is well inside the ±15 cm absolute accuracy quoted for the older Environment Agency surveys, and close to the 0.10 m scatter of a single featureless test point. Before asking what happened to the barrows, the first job was to establish that anything had happened at all.

Ruling out the instrument

Levels. The whole 2021 surface sits higher than the 2001 surface: the median difference on tarmac roads in tile SU1040 is +0.083 m (n = 8,008). Each tile was levelled to its own road offset. Horizontal misregistration is under 0.5 m.

Stripes. The levelled difference map (Figure 3) shows north–south bands with a dominant east–west period of 417 m, the fingerprint of mismatched 2001 flight strips, plus 5–10 cm steps at field boundaries. The 2001 files contain no strip outlines, so the bands were removed empirically. Within each 1 km block of northing, the median difference was taken down each column of stable ground, smoothed over 51 m and subtracted. The correction spans −0.17 to +0.15 m (1st–99th percentile). Stable ground was defined in two ways, long-term grass only and all open fields, to check whether the choice drives the result.

Figure 3. Co-registered difference, 2021 minus 2001, tiles SU1040 and SU1035, 5 m smoothing, ±0.10 m; red = higher in 2021, blue = lower. North–south flight-strip banding and field-shaped steps dominate; woodland and built areas saturate.

Neighbours. Each mound top was compared with a ring of ground 10–30 m beyond its foot, in the same field and within 25 m of the mound in easting, so that any leftover striping affects mound and ring alike.

Decoys. Random points away from monuments, measured the same way, show no change in the reverted fields: −0.002 m (n = 58) and −0.004 m (n = 26).

More barrows. Adding OpenStreetMap tumuli to the scheduled monuments gives 157 measurable mounds in classed fields. After destriping and pairing:

Field history	n	Mound minus ring, mean change (m)	Lower / higher	p
Grass by 2000	58	+0.007	23 / 35	0.15
Reverted 2000–2011 (dated)	47	−0.055	38 / 9	0.00003
Reverted 2008–2015 (undated)	10	−0.007	7 / 3	0.34
Arable throughout	40	0.000	18 / 22	0.64

Two-sided binomial sign test. Long-term grass used as stable ground; using all open fields gives the same −0.055 m (38 / 9) and shifts no other class by more than 0.003 m.

The loss survives every step almost unchanged, and the control classes stay flat. Something real distinguishes the barrows in the early reversions. The later, undated reversions show nothing.

A second site. The best independent coverage is around Avebury, with a 0.5 m survey of May 2006 and National LIDAR Programme cover from winter 2018–19. Only three measurable mounds there lie in fields mapped as reverted, and those reversions are undated. All three are lower, by 0.024 m on average (p = 0.25), which is consistent with Stonehenge but not evidence. Cranborne Chase, the South Downs and the RSPB's Winterbourne Downs offered no field-located reversion, dated after an early survey, with enough mounds to test. Winterbourne Downs began reverting in 2006, after an October 2005 survey (Farm Wildlife 2020), but holds only four barrow listings.

The suspects

Sheep, rabbits and weather. Soil worn from the mounds should end up in the ring ditches. It has not: ring ditches in the 2000–2011 reversions are, if anything, 0.016 m deeper (n = 13, p = 0.58).

Settling fill. On the analogy of a Victorian well, loose ditch fill might compact and sink under grass. There is no sign of this at 1 m resolution, in ring ditches or in 188 transects across linear ditches (+0.006 m, p = 0.94).

Slumping plough soil. Soil heaped at barrow margins by ploughing would drop at the edge, not the top. The loss is greatest at the centre: −0.075 m within half the radius, against −0.024 m at 10–30 m.

Visible vegetation in 2001. The 2001 last-return surface sits about 6 cm above the DTM on mound tops, but equally on grassland mounds that did not shrink and on reverted ones that did. In the reverted classes it does not predict the size of the loss (r = −0.10, p = 0.51), and nor does 2001 ground roughness (r = −0.06, p = 0.68).

The 2001 processing. The striping shows that the 2001 processing left systematic errors, and the metadata name neither the sensor nor the ground filter. The neighbour comparison cancels anything that varies over tens of metres, but not an error that depends on surface type.

Grazing. In December 2001 many of these barrows were uncultivated islands in arable fields. Since reversion the National Trust has run "a tightly managed grazing regime", with many barrows mown or strimmed (National Trust). Island mounds that went to grass lost 0.044 m relative to their surroundings (34 down / 12 up, p = 0.002). Island mounds that stayed in arable did not (+0.005 m). No per-field grazing dates have been found.

The theory

The measurement compares a mound with the ground around it, and only in the early reversions did that contrast change between the flights. In December 2001 the mound carried rough, ungrazed grass and the ground around it was cultivated soil. By April 2021 both were short sward. A mat of dead grass and litter on an ungrazed island would hold the lidar ground surface a few centimetres above the soil on the mound but not on the field around it. The 2001 DSM-minus-DTM test cannot detect such a mat, because the last return and the filtered ground both land on top of it. Grassland mounds (grass all round at both dates) and arable mounds (the same island contrast at both dates) cancel out. Only mounds whose surroundings changed character should change, and that is what the table shows. On this reading the barrows have not shrunk at all. What has gone is the thatch.

Two pieces of evidence point the same way. On Normanton Down (Figure 1) the loss forms halos that enclose groups of mounds along the line of the former uncultivated strips, rather than circles round each mound. In 2021 the point cloud classes fewer returns as vegetation on these mound tops (3.3%) than on long-term grassland mounds (10.5%).

A witness

The barrows in Fargo Plantation, just south of the Cursus, are an independent test outside the reversions. In April 2009 the trees there still prevented a full earthwork survey (Bishop & Amadio 2010). Between the flights the plantation was thinned but left standing, and the vegetation was cleared from the barrows alone, a clearance I witnessed. Every one of those barrows is lower in 2021. Vegetation was removed from the mounds and not from their surroundings, and the lidar records it as lost height.

Loose ends

The later reversions show no loss, possibly because their mostly low mounds were ploughed over rather than kept as islands, or because some of that ground was still cultivated close to 2021: one field mapped as reverted by 2015 carries tramlines 20–25 m apart in 2021. The 13 named barrows of the Normanton Down group lose 0.034 m on average but split only 7 down / 6 up. In a small subsample, mounds reverted in 2007 lost more (−0.106 m, n = 3) than those reverted in 2003 (−0.025 m, n = 3).

Something else in the difference map

The same map holds a second, unrelated signal. West of Stonehenge, the field between the former A344 and the A303 is lower overall in 2021, but narrow strips within it have not changed (Figure 4). They run westward from the stones for a considerable distance, opening into a V, along the lines of former tracks to the monument. Ground compacted by long use as a trackway would settle less than the loosened plough soil around it. The old routes to the stones are recorded in how the ground has settled.

Figure 4. Co-registered difference, 2021 minus 2001, west of Stonehenge; red = higher in 2021, blue = lower, ±0.10 m. Pale east–west strips within the lowered field between the former A344 and the A303 mark ground that has not changed.

The verdict, for now

The signal is real and repeatable. Barrows in fields reverted between 2000 and 2011 measure about 5 cm lower relative to their surroundings in 2021 than in 2001, while barrows in long-term grass and long-term arable do not. The most economical explanation is that the barrows are intact. On the uncultivated islands, the surface the laser took for ground in 2001 lay above the soil, and grazing and clearance have since removed it. This is not yet proven, and an error in the 2001 processing that depends on surface type has not been excluded. Four checks would close the case:

1. Sensor and ground-filter details for the December 2001 flight from the Environment Agency, and from Bewley, Crutchley & Shell (2005), who describe the early lidar survey of the WHS.
2. Aerial photographs from 2000–2003, to map the uncultivated islands and test whether the loss stops at their edges.
3. Per-field grazing and mowing histories from the National Trust, the RSPB and Natural England.
4. A probe measurement of thatch depth on an ungrazed barrow and a grazed one nearby, the cheapest test of the missing 5 cm.

Methods

Environment Agency DTM, 8 December 2001 (P_2053, 1 m), and National LIDAR Programme DTM, 4 April 2021 (1 m), on a common grid with Gaussian smoothing (σ = 1 m). Mound centres were refined on the 2021 survey. A plane fitted beyond radius + 8 m was removed, and values were taken from azimuthally averaged radial profiles. Field histories come from the hand-georeferenced plan maps (15–50 m accuracy). Outside the WHS, Priority Habitat Inventory and CROME 2020–21 were used. Monuments are NHLE scheduled monuments plus OpenStreetMap tumuli.

References and data

• Bewley, R.H., Crutchley, S.P. & Shell, C.A. 2005. New light on an ancient landscape: lidar survey in the Stonehenge World Heritage Site. Antiquity 79(305), 636–647. doi:10.1017/S0003598X00114577
• Bishop, S. & Amadio, L. 2010. Stonehenge World Heritage Site Landscape Project: The Cursus Barrows and Surrounding Area. English Heritage Research Department Report Series 85-2010. Historic England
• Stonehenge and Avebury WHS Steering Committees 2015. Stonehenge, Avebury and Associated Sites World Heritage Site Management Plan 2015. PDF
• National Trust. Our work at the Stonehenge Landscape. nationaltrust.org.uk
• National Trust 2019. Written Representation, A303 Amesbury to Berwick Down (TR010025). PDF
• Farm Wildlife 2020. Case study: reversion of arable land to lowland chalk grassland. farmwildlife.info
• Environment Agency / Defra lidar. Defra Data Services Platform; LIDAR DTM Time Stamped Extents
• Historic England, NHLE scheduled monuments. Data downloads
• Natural England Open Data (Priority Habitat Inventory); RPA Crop Map of England. naturalengland-defra.opendata.arcgis.com
• OpenStreetMap contributors (ODbL), via the Overpass API.