The Sarsen Transport Routes - A Least Cost Model

A terrain model you can argue with. Open this to understand this post: https://sarsenroute.netlify.app/ 

---

Fifty of the fifty-two sarsens at Stonehenge came from West Woods, near Marlborough. That was settled in 2020 by matching their chemistry to the source. It leaves the interesting part open: how thirty-tonne stones travelled 25 km south, across the Vale of Pewsey, to the monument.

Several routes have been proposed over the years, mine among them. None has field evidence behind it. So rather than argue for another line on a map, I set out to ask a narrower question:

If the only things that mattered were the shape of the ground, what it is made of, and the water on it — where would the cheapest road run?

The answer is not a line. It is a corridor: 37 km², about 5% of the landscape searched. Within that band the terrain barely cares which way you go.

Doing it blind

I am one of the people with a published route, which is a problem. So the method was written down and committed before any data arrived, and the published proposals — including my own — were sealed in a file that stayed closed until the result was frozen. No proposed route, mine included, is scored anywhere in the paper.

---

https://sarsenroute.netlify.app/ 

What the ground says

Four things came out that I did not expect.

The vale crossings are on sand, not clay. National geological mapping shows the Vale of Pewsey floor as one great smear of Gault clay and greensand. Zoom in to 1:50,000 and the five places where roads actually cross the vale turn out to sit on greensand benches — firm ground — with the real clay lying to the west, between Stert and Urchfont. Across all 360 model runs, not one metre of route touches Gault. The clay is real. Nothing goes near it.

There are only five crossings in eleven kilometres of vale — and adding every footpath and bridleway to the metalled roads adds no new ones. The same five places. When footpaths don't create new crossings, the crossings are being fixed by the ground.

Better data made a question harder, not easier. Two ways down off the Marlborough Downs, 2.8 km apart, are what the model has to choose between. At 50 m resolution they differ by 1.7% in cost. At 1 m, measuring the gentlest line the ground allows, they differ by 1.2 percentage points of gradient. More resolution made them more alike. Huish is not steep ground.

What separates them isn't the slope. It's use. One descent — Walkers Hill — carries a worn way running continuously from 219 m down to 140 m at a steady 6%, with the dished profile of a holloway. Andrews and Dury drew a road down it in 1773 and none down the other. The Huish face has a narrow farm track and nothing else.

Andrews and Dury's map of 1773

The bit I find most interesting

With no thumb on the scale, the model preferred the unused descent.

That isn't a bug, it's a bias — and I suspect it's general. A 50 m terrain model averages a scarp face. A smooth, unwalked hillside averages nicely. A hillside with a deep worn track in it averages badly. So the model kept choosing the easier- looking slope precisely because nobody had ever used it.

Feed in the evidence that one descent was used and the other wasn't, and it moves. But the model needed telling. It could not work that out from the ground alone.

What it can't do

A least-cost path is not a route. It knows nothing about ownership, tradition, season, weather or who you were allowed to walk past. It says what the ground permits, not what people did.

And it doesn't reproduce my own route. I think the stones came down Walkers Hill and crossed the vale beside Honey Street, on a low rise you can see on the ground. The rise is real — 1 m laser survey and the terrain model both find it, standing a couple of metres above the ground either side. What isn't shown is that it was made, or that anyone used it. The model doesn't go that way, and I've left that disagreement in the paper rather than tidy it out. A model that agreed with me would be worth less.

Have a go yourself

The model is online as a set of sliders: sarsenroute.netlify.app

Three hundred and sixty routes, pre-solved. Turn up the fear of hills, or of wet feet, or of river crossings, and watch the corridor move. There is no correct setting — that's the point. The route is a function of what you think a hauling party was most afraid of, and the tool makes that dependence visible instead of hiding it behind one confident line.

The full technical paper, with the method, the data and the things that went wrong, is on https://www.researchgate.net/publication/413532460_Modelling_the_sarsen_transport_route_West_Woods_to_Stonehenge_A_blind_least-cost_analysis_over_layered_open_data and https://www.academia.edu/172184320/Modelling_the_sarsen_transport_route_West_Woods_to_Stonehenge_A_blind_least_cost_analysis_over_layered_open_data.

Data: OS Terrain 50, BGS Geology, Environment Agency LIDAR (all Open Government Licence), OpenStreetMap. Analysis and drafting with Claude and Grok; the errors are mine.