The words Britain uses for its hills, valleys and streams, mapped. I sorted 74,934 named features by their landscape word to find where beck gives way to brook, how sharp those lines are, and whether they follow the Vikings.
Walk north through England and the water changes its name. In Surrey a stream is a brook. In Yorkshire it is a beck. In Northumberland it becomes a burn, and in the Highlands an allt. The hills change too: downs in Wessex, tors on Dartmoor, fells in the Lake District, laws in the Borders, bens and cnocs further north.
These words are a fossil map of who named the land. I wanted to know where the lines between them fall, how sharp those lines are, and whether they match history: the Danelaw, the Norse settlement names that end in -by and -thwaite, and the borders with Wales and Scotland.
I took every named valley, hill and waterfall in Ordnance Survey's OS Open Names, and every named watercourse in OS Open Rivers, and sorted each one by its landscape word using explicit matching rules that are published with the data. Of 88,386 features, 74,934 (85%) carry a word I could recognise. I added 43,274 towns, villages and hamlets, checked for Norse elements such as -by, -thorpe and -thwaite.
Choose a family, then a word. Dots are the named features. The shading shows that word's share of all named features in its family, in hexagons 10 km across. Switch to Boundaries to see the lines I fitted between neighbouring words, with the approximate Danelaw line and the national borders. Or type a name to find it.
Beck comes from Old Norse bekkr. Brook and burn come from Old English brōc and burna.
The streams draw the clearest lines in Britain. OS Open Rivers names 32,548 separate watercourses (I joined 105,905 river links into single streams). 7,518 are burns, 5,924 are allts, 3,241 are brooks and 2,744 are becks.
The chart walks up England and southern Scotland in 20 km steps, leaving Wales aside. Brook rules the south. Then, in a band no wider than a county, beck takes over. A little further north, burn takes over from beck.
The beck/brook line runs across northern England, through Lancashire, the south Pennines and Nottinghamshire to Lincolnshire, at a median latitude of 53.5°N. It is remarkably clean. 99% of becks lie on the beck side of the line and 98% of brooks on the brook side. Across it, the share of becks falls from 80% to 20% in about 10.8 km.
The burn/beck line is just as sharp (11.7 km) but sits in the wrong place for a national border. 94% of its length is in England, through County Durham and northern Cumbria, a median 40 km south of the Scottish border. Northumberland's streams are burns, like Scotland's.
Northumberland's streams are burns, Yorkshire's are becks, Derbyshire's are brooks. The changes take about a day's walk.
Further north there is one more line, between Scots burn and Gaelic allt. It runs for 695 km, in 6 pieces, through Argyll and Bute, Perth and Kinross and Highland, and it is softer: the 80% to 20% change takes 26 km on average, and anywhere from 15.5 to 50 km on different stretches.
For each pair of neighbouring words I fitted a smooth surface of where each is more common, drew the line where they are equally common, and then measured how far you have to travel across it for one word to go from 80% of the pair to 20%. A short distance means a sharp line. The method is in the note at the end.
The sharpest lines are the stream lines: beck/brook (10.8 km) and burn/beck (11.7 km). The valley words dale and clough meet across Yorkshire and the Peak in 16.9 km. Welsh and English words meet at the Welsh border: nant and afon against brook in 20 km, cwm against dingle in 17 km.
The softest are the pairs where one word is rare. Combe and bottom blur over 50 km across the West Country, and only 62% of combes sit on the combe side. Tor and down, both Old English, mix over 40 km on the edge of Dartmoor.
Dale comes from Old English dæl, reinforced by Old Norse dalr. Combe is a Brittonic word, the same word as Welsh cwm. Fell comes from Old Norse fjall, down from Old English dūn and tor from Old English torr.
Use the buttons above the walking-north chart to switch to valleys or hills. The valleys sort into neat bands. Bottoms fill Hampshire, Dorset and Wiltshire, combes Somerset and Devon, cloughs Lancashire and the south Pennines, dales Yorkshire, cleughs the Borders and Dumfries and Galloway. Clough and cleugh are the same word in English and Scots spelling, and the dales sit between them. They never meet closely enough for me to draw a line.
Hills are different. Hill itself is everywhere: 20,259 of the 42,089 named hills in the family are just hills. The local words are islands in that sea. Down (1,473) is southern England, led by Wiltshire, Devon and Cornwall. Fell (745) is Cumbria, the Pennines and Galloway. Tor (142) is Devon and Cornwall, with a few in the Peak. Law (483) is the Scottish Borders and Northumberland. Fell gives way to law across the Scottish border over 33 km.
-by, -thorpe, -thwaite and -toft come from Old Norse words for a farm or settlement, an outlying farm, a clearing and a plot. Some -thorpe names may instead be Old English throp.
The obvious test for beck is Norse settlement. In every hexagon I took the share of streams that are becks and the share of towns, villages and hamlets with a Norse element in their name. Across 626 English hexagons with enough of both, the two go together strongly: a rank correlation of 0.63 (95% interval 0.43 to 0.75, from a bootstrap that resamples 100 km blocks, because neighbouring hexagons are not independent).
The link is strong but not one to one. Where no settlement has a Norse name, 20% of beck-or-brook streams are still becks. Where at least a fifth of settlements do, 94% are. On the beck side of the beck/brook line, 15% of settlements have a Norse element. On the brook side, 2%.
Other words line up too. Across English hexagons, dale's share of valleys correlates with Norse names at 0.49, beck at 0.59, gill at 0.23 and fell at 0.19. Brook (-0.43) and down (-0.35) go the other way.
The treaty between Alfred and Guthrum set a boundary up the Thames, up the Lea to its source, straight to Bedford, then up the Ouse to Watling Street. I drew an approximate version of that line and continued it along Watling Street and then straight to Chester. It is for orientation only.
The beck/brook line does not follow it. The whole of the beck/brook line lies north of it, a median 84 km away and never closer than 53 km. In the Danelaw south of 53.2°N, brooks outnumber becks by 685 to 82.
The settlement names are a different story. North of 53.2°N almost everything is north-east of the line, so I tested only the part of England the line actually crosses, south of that latitude. There, 27% of all OS named features lie north-east of the line. So do 90% of settlements with a Norse element, 91% of -by names and 90% of -thorpe names, against 32% of -ton names and 33% of brooks.
Even here the change is a gradient rather than a wall. In the 20 km strip just south-west of the line, 0.9% of settlements have a Norse element; in the strip just north-east, 2.8%; and 11.4% at 60 to 80 km. Becks only appear in numbers more than 60 km into the Danelaw, in Lincolnshire and Norfolk.
The Danelaw line separates the place-names. The stream names change much further north.
I derived the borders of England with Wales and Scotland from the country recorded for every OS name. The Welsh words stop almost exactly there. The line between nant or afon and brook runs a median 10.6 km from the Welsh border, and 81% of it lies within 20 km. The cwm/dingle line is a median 6.4 km from the border. A second, separate nant/brook line crosses south Pembrokeshire, where the English-speaking south of the county meets the Welsh-speaking north.
The Scottish border is not a word boundary at all. The burn/beck line runs a median 40 km inside England, and the fell/law line a median 35 km from the border, wandering either side of it. The words of the Borders belong to both countries.
For each county or unitary area: its commonest stream, valley and hill word, and in brackets its most distinctive word, the one most over-represented there compared with Britain as a whole (with at least 5 examples). The commonest stream word is brook in 52 areas, burn in 29, nant in 14 and beck in 11.
Some words are tiny local dialects. Every rife in the data is in West Sussex. Lincolnshire has 16 of the 17 eaus, a fenland name for a watercourse. Lodes are in the fens of Cambridgeshire and Lincolnshire. Foss is a North Yorkshire waterfall word; elsewhere in the north a waterfall is a force, from Old Norse fors, while Scotland has its linns and Wales its pistylls and rhaeadrs.
I also gave every English town and village a Norse score: of the Norse-linked words within 10 km (beck, gill, fell, force, how, and the settlement endings), and English-linked words (brook, bourne, burn, down, dean, clough, bottom, -ton, -ham, -worth, -ley and a few more), what share are Norse? Places with fewer than 25 such words nearby are left out.
The most Norse place in England on this measure is Applethwaite, with 151 Norse-linked words and 1 English-linked one within 10 km. Among towns, Keswick leads. The least Norse is Tintwistle in Derbyshire, high in the Peak, with 102 cloughs and 45 brooks within 10 km and not one beck. It is only 18.9 km from the beck/brook line.
Stream and valley words also hide inside settlement names: Ashbourne, Blackburn, Holbeck, Ilfracombe. They follow the same order from north to south: burn, then dale and beck, then brook, bourne and combe.
One caution runs through all of this. OS Open Names is a modern map, not a historical record. Many names are recent: drains cut in the fens, farm-named hills, cloughs named after the mill above them. A name being on the map tells us what people call a place now, or called it when the surveyors came, not when it was first named. The patterns are still striking, because local words for landscape are exactly the kind of thing new names inherit from old ones.
Features. OS Open Names, filtered to 267,744 named features. Valleys are the local types Valley and Cirque Or Hollow (13,012). Hills are Hill Or Mountain, Hill Or Mountain Ranges, Cliff Or Slope, Other Landform and Spot Height, plus Other Landcover features named fell or down, because OS records most named downs and fells as open land rather than hills (42,089). Waterfalls are the type Waterfall (737). Settlements are all populated places (43,274).
Streams. OS Open Rivers splits each river into links between junctions. Counting links would weight long, fragmented rivers more than short ones, so I joined same-named links that share a node or come within 2 km into one watercourse and placed it at the vertex nearest its length-weighted centre. That gives 32,548 watercourses from 105,905 links.
Words. Each name gets at most one landscape word: its generic. In English the generic usually comes last (Mill Beck); in Welsh and Gaelic it comes first (Afon Conwy, Allt Beag, Cwm Idwal); in "Burn of Tafts" it comes before "of". Compounds count too (Borrowdale, Troutbeck, Scafell). The full rules are regular expressions in code/words.py, with tests. Where the main name has no match I tried the Welsh or Gaelic alternative name (549 matches).
Checking. I read 150 random matches, 30 from each family, and judged 0 to be wrong. A second check of 120 matches drawn only from the rarer words found 0 wrong and 1 uncertain. Earlier checks caught Easter and Wester being read as Norse -ster names, and Dunkirk as kirk; the rules now exclude them.
Shares. 3,206 hexagons 10 km across on the British National Grid. A word's share is its count divided by all named features of its family in the hexagon, including those with no recognised word.
Boundaries. For each pair of words I counted both on a 2 km grid and smoothed each with a Gaussian kernel, choosing the kernel width (6 to 20 km) by leave-one-out likelihood. The line is where the two smoothed surfaces are equal, drawn only on land and where at least 5 kernel-weighted features of either word are nearby. For sharpness, I took every feature of the two words within 60 km of the line, measured its signed distance to the line, and fitted a logistic regression of word on distance. The 80% to 20% distance is 2 ln 4 divided by the slope. Intervals come from 300 bootstrap resamples. The local thickness on the map repeats the fit on 60 km stretches of line.
Danelaw and borders. The Danelaw line is hand-placed from the treaty description and is approximate. The national borders come from the nearest OS name's country field on a 2 km grid. Correlations are Spearman rank correlations between hexagons with at least 5 settlements and 5 features of the family, with 95% intervals from a 100 km block bootstrap.