ORCID: 0000-0002-6377-2177
Cannings Cross Farm, Wiltshire SN10 3NP, UK
tim.daw@gmail.com • www.sarsen.org
July 2026
This work is licensed under a Creative Commons Attribution 4.0 International License (CC BY 4.0). https://creativecommons.org/licenses/by/4.0/
Abstract
The Stonehenge Altar Stone has been reattributed from Wales to the Orcadian Basin of northeast Scotland (Clarke et al. 2024), with the sampled localities of Mainland Orkney subsequently excluded as its specific source (Bevins et al. 2024). The Orcadian Basin itself, however, extends from the Moray Firth to Shetland and remains largely unsampled beyond a small number of hand specimens. We conducted an independent, open-data desk screen of the entire basin using BGS G-BASE stream-sediment barium and rubidium grids (500 m resolution) cross-referenced, pixel by pixel, against the BGS Geology 625k bedrock polygon dataset. Screening on the Altar Stone's published barium floor (Bevins et al. 2023; >1025 ppm) combined with a basin-relative Ba/Rb ratio threshold, we identify two priority candidate areas: a 42.5 km² area of the East Caithness coast, centred approximately 10 km southwest of Sarclet (98% confirmed genuine Middle Old Red Sandstone), and Shetland (Melby/Walls, 85% confirmed). The East Caithness result independently converges, within roughly 10 km, with a peer-reviewed detrital zircon geochronology study (Clarke et al. 2026) that identified Sarclet itself as its strongest statistical match (p = 0.96) using entirely unrelated data, assumptions, and methods. We report two negative or partially negative refinements as part of the method (a fault-proximity discriminant that did not survive testing; an unresolved glacial-till provenance question) and provide full data provenance for independent replication, including independent replication of the core grid-screening component by a separate AI system.
1. Introduction
The Altar Stone is the central sandstone megalith at Stonehenge, a six-tonne block long assumed, on stylistic grounds, to share the Anglo-Welsh Basin origin of the site's other non-sarsen ‘bluestones’. Clarke et al. (2024) overturned this using detrital zircon and apatite U–Pb geochronology, showing the Altar Stone's age spectrum matches the Orcadian Basin of northeast Scotland and is inconsistent with any Anglo-Welsh source. This relocated the search for the Altar Stone's origin to a basin roughly 700 km from Stonehenge and, by area, one of the larger sedimentary basins in Britain — stated by Clarke et al. (2026) to extend to some 10,000 km².
Bevins et al. (2024) subsequently investigated Mainland Orkney directly, applying portable XRF, automated SEM-EDS mineralogy, and clay XRD to field samples from the Stromness and Rousay Flagstone formations, and concluded these specific units do not match the Altar Stone — principally on abundant detrital K-feldspar and the near-absence of the diagenetic baryte cement and tosudite clay that characterise the Altar Stone. This is a well-evidenced exclusion of the sampled units. It is not evidence about the roughly 96–plus per cent of the basin's Devonian outcrop — across Caithness, Sutherland, the Moray Firth coast, Aberdeenshire outliers, and Shetland — that has not been sampled in the same way (the BGS Geology 625k bedrock map used in this study confirms that the great majority of preserved Old Red Sandstone outcrop lies well outside the limited, monument-adjacent ground sampled on Orkney).
This paper reports a systematic, reproducible, desk-based screen of that unsampled remainder, built entirely from free, nationally available geochemical and geological datasets, intended to rank locations for future field sampling rather than to substitute for it.
2. Data and Methods
2.1 Source records
Four records underpin the analysis.
- Altar Stone geochemistry: Bevins, R.E., Pearce, N.J.G., Ixer, R.A., Pirrie, D., Andò, S., Hillier, S., Turner, P., Power, M. (2023). “The Stonehenge Altar Stone was probably not sourced from the Old Red Sandstone of the Anglo-Welsh Basin: Time to broaden our geographic and stratigraphic horizons?” Journal of Archaeological Science: Reports, 51, 104215. https://doi.org/10.1016/j.jasrep.2023.104215.
- BGS G-BASE barium grid: national 500 m kriged stream-sediment grid. bgs.ac.uk/download/g-base-for-the-uk-barium_grid/ (confirmed).
- BGS G-BASE rubidium grid: as above. bgs.ac.uk/download/g-base-for-the-uk-rubidium_grid/ (confirmed).
- BGS Geology 625k: national bedrock, fault, and superficial-deposit polygon/line GIS layers. bgs.ac.uk/download/bgs-geology-625k-gis-line-and-polygon-data-shapefile-format/ (confirmed).
Full technical provenance, exact file names, grid parameters, and coordinate reference system details are given in Appendix A.
2.2 Geochemical threshold
Bevins et al. (2023) report that 105 of 106 pXRF analyses of the Altar Stone sensu lato — in-situ readings plus confirmed debitage fragments and the 2010 K240/Wilts 277 sample — exceed 1025 ppm Ba. The in-situ subset alone (n = 56) has a mean of 2758 ppm, confirmed directly against the raw supplementary data (Section 2.3). Sr correlates with Ba (Sr = 0.0092·Ba + 91, r = 0.71). The 1025 ppm value is used as an absolute floor throughout.
2.3 Ba/Rb ratio and threshold derivation
Checking the underlying raw pXRF data supplied with Bevins et al. (2023) directly (n=56 in-situ Altar Stone analyses in the supplementary dataset; mean Ba 2758 ppm, confirming the published figure) gives a mean Ba/Rb ratio for the Altar Stone of 108.5. The same dataset gives mean Ba/Rb ratios of 48.4 for sample WM-6 (which Bevins et al. 2023 describe as matching the Altar Stone on rubidium), against 22.5 for LORS-27 and 12.5 for LSF2-5504 (both described as excluded on rubidium grounds) — an ordering fully consistent with the paper's discriminant. One nuance is worth stating precisely rather than smoothing over: checked as simple sample means, only LSF2-5504's Rb (95.9 ppm) is close to the paper's ‘~3x higher’ description relative to the Altar Stone's mean Rb (26.1 ppm); LORS-27's mean Rb (27.6 ppm) is almost identical to the Altar Stone's. It is also worth noting that even the lowest of the Altar Stone's 56 in-situ Ba/Rb measurements (19.1) exceeds the basin-derived P95 threshold used for screening (13.76) — a small quantitative reassurance that the relative, stream-sediment-internal threshold, though not designed to reproduce the rock-level ratio, does not sit above the range of ratios the source rock itself actually displays. The paper's comparison is most likely made at matched Ba levels rather than as unconditional sample means, which this check cannot independently reproduce from summary statistics alone.
No absolute Rb concentration for the Altar Stone is put forward in that paper as a standalone provenance criterion, and the rock-level ratios above are not directly transferable to the present study regardless: they are bulk pXRF measurements of solid rock, whereas the screen in this paper uses stream-sediment geochemistry, a physically different, diluted, catchment-averaged matrix with its own baseline. Rather than adopt an absolute rubidium cutoff imported from a different measurement matrix, a Ba/Rb ratio was calculated per grid cell from the stream-sediment grids themselves (ratio = Ba / Rb, valid cells only), with the screening threshold set at the 95th percentile of that ratio's own distribution within the chosen study extent (Section 2.6) — a threshold internal to the stream-sediment dataset, not an attempt to reproduce the Altar Stone's own rock-level ratio. This follows standard exploration-geochemistry practice of using element ratios, rather than raw concentrations, to reduce sensitivity to the ‘nugget effect’ — uneven local concentration of heavy minerals caused by hydraulic sorting in stream sediment.
The composite screening condition applied to every 500 m grid cell was therefore: Ba ≥ 1025 ppm AND (Ba/Rb) ≥ P95(ratio, within the study extent). Within the extent used here this evaluated to a ratio threshold of 13.76.
2.4 Clustering and cluster-level statistics
Cells meeting the composite condition were grouped by 8-connected connected-component labelling; clusters smaller than three cells (0.75 km²) were discarded as noise. For each surviving cluster, area, centroid coordinates, mean Ba, mean Rb, and mean ratio were recorded. Every result reported below is therefore a district-scale geochemical anomaly of 0.75 km² or larger, not a point location; the method ranks areas warranting field attention, not candidate quarry faces.
2.5 Bedrock verification
A geochemical anomaly indicates a location is of interest; it does not indicate what rock is present. Every individual grid cell within every cluster — not merely each cluster's centroid — was checked by point-in-polygon spatial join against the BGS Geology 625k bedrock layer, retrieving formation name (LEX_D), lithology description (RCS_D), and chronostratigraphic age (MAX_PERIOD/MIN_PERIOD). A cell was classified as genuine Old Red Sandstone if its age included Devonian and its lithology description did not contain any of: igneous, lava, tuff, schist, ultramafite, pyroclastic, metabreccia, felsic-rock, or gneiss — excluding Devonian-age igneous and metamorphic rock, which shares the age but is a different rock type. Each cluster's reported percentage is the proportion of its constituent cells meeting this test.
2.6 Study area
The analysis was run within an OSGB36 bounding box of easting 225,000–480,000, northing 790,000–1,219,700, covering the Moray Firth to Shetland. This is a pragmatic rather than a geologically principled boundary, and its edges produced two identifiable classification errors during development, both subsequently corrected by manual inspection (the Helmsdale/Brora Outlier locality and, more marginally, ground near Gamrie Bay and Pennan). Restricting to Great Britain as a whole would remove this edge-effect risk at the cost of also returning the already-independently-excluded Anglo-Welsh Basin outcrop; a dissolved outline of Devonian sedimentary polygons north of the Highland Boundary Fault would be a preferable extent for any future iteration of this method but was not implemented here.
2.7 Ancillary tests
Fault proximity. It was hypothesised that distance to a mapped fault could flag likely vein-hosted mineralisation (a known confound: see Section 3.4) as distinct from genuine sedimentary cementation. Distance from each cluster to the nearest BGS 625k fault line was calculated. The test did not discriminate as hypothesised: the two best-confirmed genuine Old Red Sandstone clusters (Helmsdale, 0.03 km from a mapped fault; Shetland, 1.15 km) sit closer to mapped faults than the confirmed vein locality at Yesnaby (2.46 km), because fault-bounded preservation is also the mechanism by which many Old Red Sandstone outliers survive erosion. This ancillary filter is reported as a negative result and was not incorporated into the final screen.
Superficial deposits and glacial till. The bedrock spatial join (Section 2.5) implicitly assumes stream sediment reflects the rock immediately beneath it. Checking the 31 clusters classified as 0% Old Red Sandstone against the BGS superficial deposits layer found 20 of the 31 are mapped as till-covered rather than exposed bedrock, meaning the underlying assumption does not hold for the majority of ‘excluded’ clusters. The superficial layer records deposit type only, not provenance, at this scale, so till origin could not be resolved directly. Published reconstructions of the last ice sheet's flow across this ground describe the dominant pattern as movement from the Sutherland/northwest Highlands basement northeastward onto the Caithness Old Red Sandstone lowlands, with a later phase moving northwest from the Moray Firth basin — both directions carrying material toward, not away from, the Old Red Sandstone ground, which argues against large-scale Old Red Sandstone-derived till having been transported into the basement clusters excluded here, though this cannot be confirmed at the level of any individual site without local striae or clast-lithology data not available for this study.
3. Results
3.1 Regional barium contrast
Screened on the barium floor alone, Orkney returns a 0.03% hit rate against valid grid cells (a single cell, at Yesnaby — itself independently attributable in the literature to vein-hosted baryte rather than diagenetic cement), against 10.8% for mainland Caithness/Sutherland/Moray/Black Isle and 2.8% for Shetland: a roughly 350-fold contrast between Orkney and the mainland basin.
3.2 Composite Ba/Rb ratio screen
Applying the composite condition (Section 2.3) across the full study extent yielded 45 clusters of area ≥0.75 km². Bedrock verification (Section 2.5) classified these as follows:
| Category | Clusters | Genuine ORS area |
|---|---|---|
| ≥50% of pixels on real Devonian sedimentary rock | 2 | 65.5 km² |
| Partial mix — real ORS present, under half the cluster | 6 | 39.75 km² |
| 0% — confirmed basement, granite, or metasediment | 37 | 0 km² |
3.3 Bedrock-verified candidates
East Caithness, near Sarclet. A 42.5 km² cluster, 98.2% genuine Middle Old Red Sandstone, mean Ba 1453 ppm (maximum 1577 ppm), mean ratio 18.2, centred at 58.33°N, 3.26°W. Strachan et al. (2021) report the precise National Grid Reference of the zircon sample (RS-ORS-18-17) as ND 3470 4270, at Sarclet Harbour — approximately 9.7 km from our cluster's centroid, with the nearest cell 6.0 km away and the farthest 13.4 km away. For their own ice-flow modelling, Clarke et al. (2026) combine Sarclet and Braemore into a single named region (“Caithness”), treating that part of their analysis at a regional rather than a point scale; the present comparison uses the more precise point location for completeness. This is the same general stretch of East Caithness coast as Sarclet, not the identical outcrop, and that distinction is kept explicit throughout this section. This cluster did not register under an earlier, absolute-rubidium-threshold version of the screen; Caithness's background rubidium runs sufficiently above a Sutherland-calibrated absolute cutoff to fail that version despite a strong underlying Ba/Rb relationship, and only emerged once the ratio formulation (Section 2.3) was adopted. An independent replication of the grid-screening component (Appendix B) reproduced the same threshold, cell count, and cluster count from the same source data; see Section 3.5 and Appendix B.6 for a discrepancy in which specific cluster that replication identified as ‘Sarclet’, and why it illustrates the bedrock-verification step's necessity rather than undermining it.
Sarclet is independently documented as GCR site 17 in the JNCC Geological Conservation Review of the Old Red Sandstone of Great Britain (Barclay et al. 2005), selected for preserving one of the best-exposed Lower Devonian sections in the northern Orcadian Basin. The exposed succession there begins with the Sarclet Conglomerate Formation — poorly sorted alluvial-fan conglomerate containing clasts of granite, schist, quartzite, and basalt derived from the local Moine basement and contemporaneous Caledonian igneous activity — resting close to basement and passing upward, over a few tens of metres, into the fluvial/aeolian Sarclet Sandstone Formation, dated to the late Emsian by spore assemblage (Collins and Donovan 1977). This basal sequence is itself a poor facies match to the Altar Stone, for the same reason set out for Loch Duntelchaig in Section 4.1. Critically, however, the same coastal belt continues up-sequence, within a few kilometres at Wick, into the fine-grained, rhythmically laminated lacustrine Caithness Flagstone Group — the facies that includes the fish-bearing Achanarras-type horizons characteristic of the Orcadian Basin's Lake Orcadie system, and considerably closer in grain size and bedding style to the Altar Stone's described fine ripple-laminated sandstone than the basal conglomerate is. Diagenetic cementation, including sulphate minerals in some beds, and clay mineralogy are both reported to vary with facies and stratigraphic position across this sequence — precisely the kind of within-basin variability that sampling confined to the Stromness and Rousay Flagstone formations on Orkney (Bevins et al. 2024) would not have captured. This does not establish which specific facies underlies the geochemical anomaly reported here, which sits roughly 10 km from the GCR site itself — that requires the field verification recommended throughout this paper — but it establishes that the right facies is documented as present within the same short stretch of coast, which is not true of the Loch Duntelchaig candidate.
Strachan et al. (2021) independently describe the same succession in more detail, consistent with the account above: the Sarclet Sandstone Group has a minimum exposed thickness of c. 435 m, comprising the 70 m-thick basal Sarclet Conglomerate (rounded clasts of volcanic, metamorphic, and granitic material) succeeded by the 85 m-thick Sarclet Sandstone, both deposited by a fluvial system flowing northwest. These are overlain by the Ulbster/Riera Geo Mudstones (lacustrine) and the Ulbster/Ires Geo Sandstone (a return to fluvial conditions, dated to the Late Emsian by microspores), with the whole group passing, apparently conformably, into Middle Old Red Sandstone strata — independent literature confirmation of the Middle ORS classification this study's own bedrock spatial join returned for the same locality.
The zircon sample itself (RS-ORS-18-17, collected at Sarclet Harbour) is described by Strachan et al. (2021) as a fine-grained, moderately sorted lithic arkose with sub-angular quartz and feldspar in roughly equal proportions, muscovite flakes below 0.5 mm, a reddish colour attributed to oxide coating on grain surfaces, and accessory pyrite, zircon, and rutile. This description, compiled for the purpose of zircon separation rather than diagenetic characterisation, does not report baryte, kaolinite, or calcite cement — the three minerals that define the Altar Stone's diagnostic signature — nor does it specify whether the feldspar present is dominantly plagioclase or K-feldspar, the ratio central to the Altar Stone's own defining characteristics. Their absence from a brief petrographic note is not evidence of their absence in the rock; it means this specific point has not yet been examined for the minerals that actually matter, which remains the outstanding task regardless of how well the geochemical and geochronological signals align.
Shetland (Melby/Walls). 85% genuine Middle Old Red Sandstone, the most consistently confirmed locality across every version of this screen.
Nairn/Elgin corridor and Helmsdale. Both partial: genuine Old Red Sandstone (the Brora Outlier, in Helmsdale's case) is present but accounts for well under half of the connected geochemical anomaly, the remainder of which extends onto adjacent, non-matching ground.
Loch Duntelchaig / Great Glen. A cluster splitting almost evenly between genuine Middle Old Red Sandstone breccio-conglomerate and an adjacent Devonian-age mafic igneous intrusion — geochemically indistinguishable, bedrock-distinguishable.
3.4 Confirmed exclusions
Yesnaby (Orkney) and a locality near Tongue (North Sutherland) both register strongly on barium but are confirmed, respectively, as vein-hosted baryte mineralisation and Lewisian Complex orthogneiss — igneous basement with no relationship to Devonian sedimentation. A cluster at Kinbrace, Sutherland, resolves to an Ordovician–Silurian granite intrusion. The bulk of the remaining Sutherland interior anomaly resolves to Lewisian gneiss, Moine psammite, and Cambrian Durness Group carbonate. None of the three small, isolated, published Old Red Sandstone outliers at Tomintoul, Cabrach, and Rhynie, nor ground near Aberdeen, produced a single composite hit, despite being genuine Devonian sedimentary outcrop — attributable either to a genuine absence of the target geochemical signature or to the limited resolution of a 500 m grid against outcrops of comparable or smaller scale.
3.5 Independent convergence with detrital zircon geochronology
Clarke et al. (2026) independently tested a small number of previously-sampled Orcadian Basin localities for detrital zircon age-spectrum similarity to the Altar Stone using Kolmogorov–Smirnov statistical comparison. Sarclet returned the strongest match of any locality tested (p = 0.96), with Braemore, Kirtomy, and Portskerra also statistically compatible and the New Aberdour outlier compatible but weaker (p = 0.08). This sits within about 10 km of the cluster identified independently in Section 3.3 above — the same short stretch of East Caithness coast, not the identical point — by a method (regional stream-sediment element ratios, screened continuously across the whole basin with no dependence on pre-existing sample locations) sharing no data or assumptions with detrital zircon geochronology (the isotopic dating of zircon crystals eroded from basement source terranes hundreds of millions of years before Old Red Sandstone deposition).
Two qualifications are necessary. First, Clarke et al. (2026) are explicit that the Orcadian Basin's zircon coverage is sparse relative to its size (up to 10,000 km²); their result identifies the strongest match among available sampled localities, not a result tested against the full basin, which is the same limitation the present screen carries in reverse — continuous basin-wide coverage of geochemistry, but no capacity to test isotopic age at any location. Second, the ice-flow modelling in the same paper finds no viable direct glacial pathway from Caithness to Stonehenge: modelled ice flow from the favoured Caithness sources disperses predominantly north and east, and only specific, less-favoured sensitivity scenarios permit transport as far as Dogger Bank, still some 400 km short of Stonehenge. The paper treats this partial pathway as speculative rather than established, citing the absence of any corroborating erratic or detrital evidence at Dogger Bank itself, a multi-millennium timing gap before any subsequent human transport, and the coarse spatial and temporal resolution of the ice-flow model. A companion study (Clarke and Kirkland, 2026) found no glacial detrital zircon–apatite signature from northeast Scotland in Salisbury Plain sediments at all. Both lines of evidence support human rather than glacial transport for the Altar Stone's journey, and both are downstream of, and independent from, the zircon match itself.
4. Discussion
4.1 Formation identity is not facies identity
The Loch Duntelchaig result (3.3) demonstrates a limit intrinsic to this method that no refinement of it removes. That locality is, in part, unambiguously genuine Middle Old Red Sandstone by formation name, age, and lithology description — and is also, on the published description of its depositional setting, almost certainly the wrong facies: coarse, basin-margin breccio-conglomerate deposited as an alluvial fan against an active fault scarp, where the Altar Stone is fine-grained, ripple-laminated, and mica-rich, indicating deposition by slow water at a distance from any such scarp. A bedrock polygon records formation and age, not depositional texture. Every result in this paper should be read as identifying locations warranting facies-level (petrographic, sedimentological) assessment, not as identifying a lithological match in itself.
Sarclet illustrates the same point from the opposite direction. The specific GCR site is itself dominated by basal alluvial-fan conglomerate and fluvial/aeolian sandstone — by the logic above, likely as poor a facies match as Loch Duntelchaig at the outcrop scale of a single site description. But because the published stratigraphy documents this basal sequence passing, within the same coastal belt, into fine-grained lacustrine flagstone facies close in character to the Altar Stone, the surrounding area cannot be excluded on facies grounds the way Loch Duntelchaig can. The distinction is not that Sarclet has been shown to have the right facies where Loch Duntelchaig has the wrong one; it is that Sarclet's documented stratigraphy leaves the question open, for a cluster roughly 10 km along the same coast, in a way that is worth resolving in the field, where Loch Duntelchaig's does not.
4.2 Failed and unresolved refinements
The fault-proximity test (2.7) is reported as a negative result rather than omitted, on the view that a plausible-sounding refinement that does not survive contact with data is informative and should not be quietly dropped. The glacial till question (2.7) is reported as unresolved rather than dismissed: a majority of clusters classified as bedrock-excluded are till-covered, the available data cannot establish till provenance directly, and the directional argument offered in mitigation is a regional generalisation, not a site-specific confirmation.
4.3 Geographic scope as a stated choice
The bounding-box approach to defining the Orcadian Basin's extent (2.6) is acknowledged as the least satisfactory methodological element of this study, having produced at least one classification error (Helmsdale) during development. Any replication of this method should either use the same coordinates for direct comparability or adopt a geologically defined extent (a dissolved union of mapped Devonian sedimentary polygons) and report the change explicitly, as cluster counts and the ratio threshold itself (13.76 in this instance) are both extent-dependent.
4.4 The value of methodological independence
The convergence with Clarke et al. (2026) at Sarclet (3.5), with the two results roughly 10 km apart on the same stretch of coast, is offered as evidence worth taking seriously in proportion to the independence of the two methods, not as confirmation in itself. Two approaches sharing no input data, no statistical framework, and no common set of prior assumptions arriving within 10 km of one another, in a basin of some 10,000 km², is a stronger form of corroboration than the same method applied twice, though it remains short of the direct petrographic and sedimentological comparison that would be required to establish an actual source.
5. Conclusion
This screen, built entirely from freely available national datasets, ranks a stretch of the East Caithness coast approximately 10 km from Sarclet, and the Melby/Walls area of Shetland, pending facies-level verification, as the two highest-priority targets for field-based follow-up within the Orcadian Basin, with the Nairn/Elgin corridor and the Helmsdale area (Brora Outlier) as secondary candidates. No location discussed in this paper has been confirmed as an actual quarry site, and none of the results reported here substitute for direct petrographic, heavy-mineral, or geochronological sampling. The method's principal value is in triage: narrowing a roughly 10,000 km² basin, at negligible cost and using only public data, to a small number of specific, geologically-verified localities, one of which independently corroborates the strongest result yet published by an unrelated geochronological method.
Data and Code Availability
The composite screening code, full per-cell and per-cluster CSV output underlying Sections 3.2–3.3 and Appendix B.6, README, and licence are archived at: https://github.com/TimDaw37/Altar-Stone-Source-Screening (CC BY 4.0). Source data are public and cited in full in Section 2.1 and Appendix A.
|
Locality |
BGS
sample ID |
NGR |
Thin
section |
|
Cliff
near Sarclet |
S13937 |
ND
336 428 |
|
|
Stack
of Ulbster |
S13938 |
ND
335 413 |
|
|
Borrowston
Quarry, Wick–Lybster line |
S27114 |
ND
326 433 |
|
|
Gillyvoan
Quarry, Latheron |
S27115 |
ND
199 341 |
|
|
Borrowstone
Quarry, Thrumster (hand specimen only) |
MC5544 |
ND
326 423 |
No |
References
Bevins, R.E., Pearce, N.J.G., Ixer, R.A., Pirrie, D., Andò, S., Hillier, S., Turner, P., Power, M. (2023). The Stonehenge Altar Stone was probably not sourced from the Old Red Sandstone of the Anglo-Welsh Basin: Time to broaden our geographic and stratigraphic horizons? Journal of Archaeological Science: Reports, 51, 104215. https://doi.org/10.1016/j.jasrep.2023.104215.
Bevins, R.E. et al. (2024). Was the Stonehenge Altar Stone from Orkney? Investigating the mineralogy and geochemistry of Orcadian Old Red sandstones and Neolithic circle monuments. Journal of Archaeological Science: Reports, 58.
Barclay, W.J., Browne, M.A.E., McMillan, A.A., Pickett, E.A., Stone, P. & Wilby, P.R. (2005). The Old Red Sandstone of Great Britain. Geological Conservation Review Series No. 31, JNCC, Peterborough.
Clarke, A.J.I. et al. (2024). A Scottish provenance for the Altar Stone of Stonehenge. Nature.
Clarke, A.J.I., Veness, R.L.J., Kirkland, C.L., Clark, C.D., Gandy, N., Emery, A. et al. (2026). From Highlands to Henge: Refining the Provenance and Transport Pathways of Stonehenge's Altar Stone. Journal of Quaternary Science, 1–8. https://doi.org/10.1002/jqs.70080
Clarke, A.J.I. and Kirkland, C.L. (2026). Detrital zircon–apatite fingerprinting challenges glacial transport of Stonehenge's megaliths. Communications Earth & Environment, 7, Article 54. https://doi.org/10.1038/s43247-025-03105-3
Strachan, R.A., Olierook, H.K.H. and Kirkland, C.L. (2021). Evidence from the U-Pb-Hf signatures of detrital zircons for a Baltican provenance for basal Old Red Sandstone successions, northern Scottish Caledonides. Journal of the Geological Society, 178, jgs2020–241. https://doi.org/10.1144/jgs2020-241
British Geological Survey. G-BASE Geochemical Baseline Survey of the Environment, UK stream sediment geochemistry grids. bgs.ac.uk
British Geological Survey. BGS Geology 625k, GIS bedrock, fault, and superficial deposit line and polygon data. bgs.ac.uk
Daw, T. (2026). Altar-Stone-Source-Screening [code and data repository]. https://github.com/TimDaw37/Altar-Stone-Source-Screening
Appendix A — Data Sources and Replication Notes
This appendix reproduces, in full, the standalone replication document prepared alongside the main analysis, so that the method in Sections 2–4 above can be independently rerun and checked.
A.1 Source records
Full citations and confirmed source URLs for all four records are as given in Section 2.1 and the References. Grid parameters for both G-BASE grids: NCOLS 1310, NROWS 2428, XLLCORNER 250, YLLCORNER 5750, CELLSIZE 500, NODATA_VALUE −9999, coordinate reference system OSGB36 / British National Grid (EPSG:27700, false easting 400000, false northing −100000, central meridian −2°, scale factor 0.9996012717), the same CRS as the BGS Geology 625k layer.
A.2 Bedrock join fields and filter
Fields used from the 625k bedrock layer: LEX_D (formation name), RCS_D (lithology description), MAX_PERIOD and MIN_PERIOD (chronostratigraphic age). A cell is classified as genuine Old Red Sandstone if age includes Devonian and RCS_D does not contain IGNEOUS, LAVA, TUFF, SCHIST, ULTRAMAFIT, PYROCLASTIC, METABRECCIA, FELSIC-ROCK, or GNEISS.
A.3 Geographic scope
Study extent: OSGB36 easting 225,000–480,000, northing 790,000–1,219,700. This is a stated methodological choice, not a geological boundary; see Section 2.6 and 4.3 for the edge-effect risk this carries and an alternative approach not yet implemented.
A.4 What this method does and does not show
- Does not establish depositional facies (grain size, sorting, bedding style) — formation-name and age matching only (Section 4.1).
- Does not distinguish diagenetic cement from vein-hosted mineralisation with certainty — the Ba/Rb ratio reduces but does not remove this ambiguity.
- Does not resolve glacial till provenance at excluded localities (Section 2.7).
- Does not establish that any candidate location was ever quarried or could physically have yielded a block the size of the Altar Stone.
This is a desk-based screening and triage method, intended to rank locations for further attention and rule out others with reasonable confidence — not a substitute for petrographic analysis, heavy-mineral work, detrital zircon geochronology, or direct field examination.
Appendix B — Independent Replication (Grok, xAI)
The following independent replication was performed by Grok (xAI), using the exact source files and method parameters set out in Appendix A, without reference to the authors' own implementation code. It is reproduced here in full as a robustness check on the core screening result.
B.1 Verification of source records
All four foundational records were checked against the live sources and the published literature. The Altar Stone pXRF signature (105/106 analyses >1025 ppm Ba; mean Ba >2750 ppm; Sr = 0.0092·Ba + 91, r = 0.71; Anglo-Welsh look-alikes at ~3× higher Rb; no absolute Rb value published) was confirmed in the source text. The supplied G-BASE ASCII grids matched the documented headers (NCOLS 1310, NROWS 2428, XLLCORNER 250, YLLCORNER 5750, CELLSIZE 500 m, NODATA −9999) and OSGB36 CRS. The supplied BGS Geology 625k shapefile set contained the required fields (LEX_D, RCS_D, MAX_PERIOD/MIN_PERIOD) with the correct CRS.
B.1a Editorial note: Rb availability, reconciled with Section 2.3
B.1 above records that the source paper's prose contains "no absolute Rb value published" and describes Anglo-Welsh look-alikes as being at "~3× higher Rb" — an accurate summary of the source paper's own narrative at the time this replication was performed. Section 2.3, however, checks the raw supplementary pXRF dataset accompanying that same paper directly, and finds it does contain per-sample Rb values: a mean of 26.1 ppm (range 18.9–37.1 ppm) across the 56 in-situ Altar Stone analyses. The "no absolute Rb value published" statement is therefore true of the source paper's text, but not of its underlying data. Similarly, the "~3× higher Rb" description holds for one of the two comparison samples checked (LSF2-5504, 95.9 ppm) but not the other (LORS-27, 27.6 ppm, essentially identical to the Altar Stone's own mean) — see Section 2.3 for the sample-level detail. Both figures in B.1 are left as originally reported for transparency about what the replication could establish from the text alone; readers should treat Section 2.3 as the more complete and current account.
B.2 Implementation
Grid processing was executed in Python (numpy + scipy.ndimage) in an independent sandbox environment: both ASCII grids were loaded and a joint valid-cell mask created; the Ba/Rb ratio was computed on valid cells only; the analysis was clipped to the same Scotland/Orcadian Basin box specified in Appendix A.3; the 95th percentile of the ratio distribution within the box was independently calculated at 13.758, matching the value used in the main analysis to three decimal places; the composite screen (Ba ≥ 1025 ppm AND ratio ≥ P95) was applied; 8-connected component labelling was performed with clusters below 3 cells discarded; per-cluster statistics were generated; and every passing cell (approximately 3,528 rows) plus per-cluster centroids and summary statistics were exported for downstream use.
B.3 Results
The replication reproduces the core outputs of the method. The independently calculated P95 ratio threshold (13.758) matches the main analysis. 46 clusters remained after the minimum-size filter, against 45 in the main analysis — a difference attributed to normal floating-point or clipping variation and considered immaterial. The largest and highest-barium clusters occur in the East Caithness / broader Caithness Flagstone belt and in the northern part of the study box; in particular, one cluster (228 cells, mean Ba 1775 ppm, maximum 3232 ppm, mean ratio 21.7) lies in the East Caithness Middle Old Red Sandstone province, spatially consistent with the locality reported in Section 3.3 as ‘near Sarclet’ and with the geological setting discussed in Clarke et al. (2026). This hotspot was identified from the geochemical grids and the published Altar Stone signature alone, independent of the detrital zircon results in Clarke et al. (2026), which the replication treats as independent corroboration rather than an input to the screen.
B.4 Limitations of the replication
The final per-cell point-in-polygon join to the 625k bedrock layer, the Devonian sedimentary lithology filter, and the calculation of percentage genuine Old Red Sandstone per cluster could not be executed within the replication sandbox, as GIS libraries were unavailable in that environment. The exported cell- and cluster-level data were provided as a complete, ready-to-use dataset for this step in any standard GIS package, sufficient in principle to reproduce the bedrock-verified categorisation and percentages reported in Section 3.2–3.3.
B.5 Assessment
The grid-screening component of the method is judged fully reproducible, yielding hotspots matching the main analysis in style and location, including the East Caithness anomaly and a correspondingly weak signal in the limited, previously-sampled zone around the Ring of Brodgar and Stones of Stenness on Orkney. Minor differences in exact cluster footprint or cell count between the two independent implementations are expected from floating-point and boundary-clipping variation and do not affect scientific interpretation. This replication is treated as validating both the general workflow and the specific East Caithness result as a geochemically-derived lead independent of, and consistent with, the zircon evidence in Clarke et al. (2026), subject to the same caveats set out in Section 4 and Appendix A.4 above.
B.6 Editorial note: a cluster misidentification, and why it matters
Cross-checking this replication against the bedrock-verified output described in Section 3.3 identified a discrepancy worth reporting in full rather than silently correcting. The 228-cell cluster (mean Ba 1775 ppm, mean ratio 21.7) that Section B.3 above identifies as ‘the East Caithness Middle Old Red Sandstone province, spatially consistent with … near Sarclet’ is not, in fact, the same cluster verified in Section 3.3 as the genuine Sarclet hotspot. Its coordinates (58.43°N, 4.36°W) place it in North Sutherland, near Tongue, roughly 85 km from Sarclet; and when checked against the bedrock layer, it resolves to only 1.3% genuine Old Red Sandstone, dominantly Ordovician–Silurian basement. The correctly located and bedrock-confirmed East Caithness cluster is a separate, 170-cell (42.5 km²) feature at 58.33°N, 3.26°W, 98.2% genuine Middle Old Red Sandstone, with with mean Ba 1453 ppm and mean ratio 18.2, sitting 9.7 km from the precise zircon sample coordinate reported by Strachan et al. (2021) at Sarclet Harbour (Section 3.3), or approximately 10 km from Sarclet village by the same proxy measure used earlier in this project — present in this replication's own cluster set but not the one selected for narrative description in Section B.3. That 10 km separation is worth stating plainly here too: this section corrects an 85 km misattribution in Section B.3, and it would be inconsistent to let this section imply pinpoint coincidence with Sarclet village that the main analysis does not itself claim (see Section 3.3).
The cause is instructive rather than a simple error. The replication's grid-screening arithmetic — the threshold, the cell count, the cluster count — is independently correct, as Section B.3 shows. What was not, and could not be, independently verified in that sandbox was which specific cluster corresponds to a real, named, geologically confirmed location, because the bedrock spatial join (Section 2.5) could not be executed without GIS libraries. In their absence, the largest and highest-barium cluster was assumed to be the significant one and matched to Sarclet by approximate regional geography rather than by confirmed coordinates or bedrock. This is the same distinction argued throughout Section 4.1: a geochemical anomaly indicates a location is of interest, not what is actually there, or, in this case, not even reliably which named place it is. The error was caught only because the per-cluster output was checked coordinate-by-coordinate against the bedrock-verified results in Section 3.3 before this paper's data-availability materials were finalised — underscoring the value of publishing full per-cluster coordinates and statistics (as both Appendix A and this replication do) rather than summary claims alone.
No comments:
Post a Comment
Comments welcome on fresh posts - you just need a Google account to do so.