REWT
Rivers of England and Wales, Temporally
Routing
A cost surface for coastal navigation: where a vessel could go, on the water, between the mouths of the rivers. It is the sea half of a network whose other half is the modern drainage, and it exists so that port to port is computable — down one river, along the coast, and up another.
It makes no historical claim, and that is a rule rather than a caveat. A hexagon in the Irish Sea does not know what year it is. What has been built is geometry, depth, and what can be seen from where; weights, seasons, hulls and dates are all deferred to later stages. Nothing on this page requires knowing when.
Every figure below is printed from the router’s own artefacts rather than typed here — the
files under docs/router/data/ carry their method in their own properties, and this page
is generated from them.
Sight of land decides the surface
A cell is in the network only if land can be seen from it. Not preferred, not penalised — included. Water out of sight of land is not part of the surface at all, so there is no third state to explain and no threshold anyone can tune: a route either keeps land in view or it does not exist.
The range at which land drops below the horizon is sqrt(2Rkh), which for the refraction
coefficient this uses is
3.7945 × √(height in metres) kilometres, with a refraction coefficient of 1.13.
Not 3.86, which is the figure most
readers will supply from memory and which corresponds to a refraction coefficient of
1.17. The difference is small and it is
systematic, which is the combination that survives a sanity check. The constant on this page
is read out of the layer’s own formula string when the page is built, and checked against
sqrt(2Rk) for the coefficient it names, so the prose here cannot drift from the code that
drew the map — and if the two ever disagree, the build stops rather than publishing either.
A sighted range has two terms, and the difference between them decides what any number here means. Land goes below the horizon at a distance set by its height; an observer sees over the horizon by a distance set by theirs; and a coast is in sight when the two together span the gap. So every reach quoted from this layer is the land’s term, and the water actually in sight extends a further 12.0 km beyond it.
The observer is not at sea level, and that is a choice with a citation rather than a default: 10 m, a lookout at the masthead. After Alvarez-Palau and Dunn (2019) as reported by Litvine, Lewis and Starzec (2024), doi:10.1057/s41599-024-02906-9 (CC BY 4.0), whose coastal routing model for England and Wales takes the observer at 10 m on the ship’s mast.
So these ranges are not floors. They assume someone is aloft. A lookout on deck at 2 m has an observer term of 5.4 km and therefore sees 6.6 km less than this surface allows — which is the figure that matters, and is not the same as that observer’s own horizon.
Choosing a height is choosing a kind of ship, which is why it is the one parameter here that had to be argued rather than set: it is the point at which geometry starts to describe somebody’s vessel.
Each cell carries the height that governs it: the tallest land that reaches this cell — the band that first marked it. Absent where nothing reaches. So the surface is not a buffer round a coastline. A high headland projects sight much further than the flat shore beside it, and the reach of a stretch of water is set by the highest thing that can see it.
What two voyages do and do not show
Two passages carry the calibration — London to Skagen, at the northern tip of Denmark, and Falmouth to Bordeaux — each described as about a week’s sailing in July, from eRutter.
Measured on a surface that permitted a stretch of blind sailing, the two came out 48% apart. Measured on a surface where land must stay in sight, they agree to 6%. The same two routes, the same two durations, and a difference that is entirely a change in the geometry they were measured over.
That is consistent with the rule and is not evidence for it, and the distinction is worth keeping sharp because the stronger claim is the tempting one. eRutter says of itself that it is approximate — better for comparing routes than for settling times — so these are not two independent measurements converging. They are two soft figures, each “about a week”, implying similar distances. Two approximate durations agreeing tests very little. What has actually been shown is that a constraint changes the geometry in the direction coastwise sailing would predict, which is a reason to keep the constraint rather than a reason to believe it.
And the week is an order of magnitude, not a threshold. The surface is trimmed at roughly a week’s sailing of England and Wales — roughly, in the same breath as the trim, because a rounded estimate of a voyage becomes a boundary condition the moment it is written without that word. Nothing here computes duration: sailing time needs wind, current and a hull, and weights are deferred. This calibrates a DISTANCE against an attested duration; it does not claim to compute duration.
A week is not a radius — measured along the water, so Norway is as far as the route round the Skagerrak makes it, not as near as a straight line suggests
And the two anchors still disagree, which is kept rather than averaged away: the two attested week-long voyages imply different route distances. That is information, not noise: route distance is a poor proxy for time, and the difference is what wind and season would explain. The LARGER is taken so that both voyages fall inside the bound.
What is not published here, and why
No percentage of “sea that can see land”. Such a figure moves with the extent of the model, the resolution of its grid and the height of its observer, without anything about the world changing — it is a property of the model rather than a measurement of the sea. It is deliberately absent from the map’s own panel for the same reason.
No cell, join or trace counts. The grid’s banding changed recently and the joins depend on work not yet landed, so those numbers are still moving. They live in the artefacts, where a reader can fetch today’s, rather than in a sentence that will be quoted back in a month.
And nothing here is a route a vessel could follow. The bathymetry beneath it is research
data published DO NOT USE FOR NAVIGATION — a constraint on purpose, not on
redistribution. A topological surface is not navigation, and the naming of everything built on
it is chosen to keep that boundary rather than merely to acknowledge it.
How it was checked
exact WGS84 geodesics on 250 sample points: 246 agree, and all 4 disagreements are conservative (the raster misses visibility, never invents it)
The distances are computed in EPSG:32630, whose worst projection error over this extent is 0.33% — smaller than the 2.0 km bands the surface is built in, so it cannot move a cell from one band to another.
Cells beyond sight are absent rather than shaded. A thing that is not there is visibly not there; a thing drawn in a colour meaning unknown looks like an answer. That is the same rule the audit works to — an unknown must fail towards the visible fault.
So what is the edge of the surface? A boundary drawn on a map asserts that something changes there, and it is worth being exact about what changes at this one. The layer says so about itself: NO BLIND SAILING (Stephen, 4 Sep 2026): the buffer is zero. A cell is in the surface only if land is in sight from it, and a landmass joins only when its sighted water TOUCHES sighted water already admitted — two blind zones meeting is not a route. The trimmed edge is therefore the limit of sight itself, not a routing decision and not a coast. A crossing that leaves sight of land — Land’s End to Cork among them — is therefore NOT in this surface. The blind span of that crossing was measured on the superseded buffered surface and is not restated here, because a figure derived from a surface that no longer exists is worse than no figure.
No blind sailing at all, then — the edge is the limit of sight and nothing else, and a landmass enters the network only where its sighted water touches sighted water already in it, because two blind zones meeting is not a route. The consequence is concrete and worth stating because it is the kind a reader would otherwise assume away: Land’s End to Cork is not in this surface. A crossing that goes 56 km out of sight of land is not a passage this network can make, whoever in fact made it.
Looking at it
The sightline surface on the map — that link opens the viewer on this layer alone, with the river network and every other overlay switched off, because the default view does not illustrate anything on this page.
The map draws the same file this page is generated from, so a figure quoted here and a cell clicked there cannot disagree. What the map adds is the thing prose is worst at: the shape of the surface — where sight reaches far out and where it clings to the shore — which is one picture and several paragraphs.
Contains EMODnet Bathymetry data. EMODnet Bathymetry Consortium (2024): EMODnet Digital Bathymetry (DTM 2024), licensed CC BY 4.0.
Generated from docs/router/data/sightline2_r6.geojson, docs/router/data/reach_summary.json by tools/docs/routing_page.py.
To correct a figure, correct the artefact and regenerate — an edit to this page is overwritten
at the next build.
| Methodology | how the river network is made traversable, and why the sea is a test rather than a definition |
| The rules | every general rule the build works to, in the order they apply |
| Evidence | what each source can and cannot say, and on what terms |