How we predict red skies
An honest evening verdict for the Lake District. What this verdict forecasts is lit cloud — the undersides going orange and pink, which is the thing worth walking up a fell for. The colour itself comes from the light: near sunset it reaches you almost horizontally, through hundreds of kilometres of low atmosphere, and the air scatters the blue out of it long before it arrives. Cloud is the screen that catches what is left. That gives three things to get right, and they are the three this verdict checks: is there cloud up there to catch the light, is the light getting through, and will you be standing above the murk or inside it.
So a clear evening is not a colourless one. The sky can still go orange near the horizon with no cloud in it at all — the long path through the air does that on its own, and haze and dust deepen it. What a clear evening has is nothing for us to predict: there is no cloud field to score, so this verdict sits it out rather than calling it. A zero here means outside what we forecast, never "the sky cannot colour tonight".
Why there is no per-fell number
The Lake District is about 50 km across. The cloud that makes or breaks a sunset arrives on weather systems hundreds of kilometres wide. On most evenings, every summit in the district is under the same sky — so a per-fell sunset forecast would be two hundred near-identical answers, dressed up as two hundred independent ones.
So the evening verdict is regional: one for each of the seven Wainwright regions, and often all seven agree. When they do, that is the answer, not a failure of resolution.
There is something that genuinely differs from summit to summit, and we publish it — just not here. A fell with a high ridge to the west loses the sun minutes before its neighbour does, and that is geometry, computed from the terrain model for each of the 214 Wainwrights — a good estimate rather than an observed time, since it reads a 30 m elevation model with a single refraction coefficient and a real evening’s air does not hold still. You’ll find it as last light on each fell page, beside its first light. That is the per-summit fact. The weather is not one.
The four things we check
All four are read from the hour before sunset and the sunset hour itself, from the same UK Met Office UKV forecast (Open-Meteo / Met Office (opens in new tab)) that drives the walking scores on the conditions page, so the two are reading the same forecast rather than two different ones. They can still differ in detail if they pick up different runs of it.
| What | Why it matters | What kills it |
|---|---|---|
| Cloud aloft — the canvas | Mid and high cloud is the screen the light is projected onto. Without it there is nothing for this score to work with — the air can still colour, but we are not forecasting that. | A bare sky (no screen, so nothing we score) or a total overcast (no gaps, and the deck is too thick to glow). |
| The light path | The light arrives almost horizontally from the direction of the sunset. It has to get through a long, low corridor of air to reach you. | A bank of cloud sitting on the horizon in that direction — the classic sunset-killer. |
| Summits clear of cloud | You cannot watch a sunset from inside a cloud. | A cloud base below the tops. We estimate it from the dewpoint depression, the same way the walking score does. |
| Rain and visibility | Murk in the golden hour flattens everything, whatever is happening above it. | Rain, or visibility down in the low kilometres. |
Any one of these can end the evening on its own, so each can cap the verdict outright rather than being averaged away by the other three. When one does, that single reason is what the verdict shows you — there is no point telling you the summits are clear if the sun is setting into a wall of cloud.
Following the light back out to sea
This is the part most sunset forecasts skip. We work out the compass bearing of the setting sun for that evening — using the same solar-position code, checked against the same astronomy anchors, that computes first and last light on the fell pages — and then sample the cloud at three points out along that bearing: 50, 120 and 250 km from the middle of the district.
Where those points land changes through the year, and it is worth being precise about it rather than saying “out over the Irish Sea”, which is only true for part of the year. From the centre of the Lakes the sun sets at a bearing of roughly 315° at midsummer, swinging round to about 228° at midwinter — an arc of nearly 90°. So in June the light comes in over the Galloway hills and the Inner Hebrides; at the equinoxes it comes due west over the Isle of Man and Northern Ireland; in December it comes up the Irish Sea from the south-west. Same method, quite different geography.
A beam of light is blocked by whichever part of it is blocked, so the path score takes the worst of the three samples rather than their average — with a little more benefit of the doubt allowed the further out the sample is, since those readings are both less certain and less decisive. An early version averaged them, and a solid bank 50 km out was politely outvoted by two clear readings behind it. That is exactly backwards.
Why the far sample reads a different cloud layer
Read that table as the illustration it is. It gives the curvature term alone — how far the Earth falls away beneath a level ray — and not the beam’s height above the ground it is crossing, which also depends on where the light started, how high the sun still is, and what the land does in between. Nor is the ray that reaches you necessarily the ray lighting the cloud you are looking at. It is the reasoning behind a deliberate choice of which cloud layer to read at which range; it is not a proof.
Because the Earth curves away underneath a horizontal beam. Light that arrives level with you at a summit was, further back along its path, well above the ground it was passing over. With refraction folded in at the same coefficient our panorama engine uses (k = 0.13), the beam sits at roughly:
| Distance out | How far the ground drops away | So we read |
|---|---|---|
| 50 km | about 170 m | low cloud |
| 120 km | about 1.0 km | low cloud |
| 250 km | about 4.3 km | mid cloud |
At 250 km the ground has dropped more than four kilometres below a level ray, which is the middle cloud layer (Open-Meteo (opens in new tab) splits cloud into low, mid and high). Low cloud out there is very likely to sit under the light rather than across it. Reading it would have been scoring a variable with no bearing on the answer.
What the verdict words mean
The evening verdict uses the same four words as the walking score — Cracking, Decent, Marginal, Save it — deliberately, rather than inventing a second vocabulary for the same page. They mean different things about different questions, and the column headings say which is which: one is about whether it is a good day to be on a fell, the other about whether the sky is likely to colour. A Marginal walk and a Marginal sunset are not the same claim.
Why there is no percentage
It would be easy to print one, and a number looks more authoritative than a word. We don’t, for two reasons.
The first is false precision. Nothing in the inputs supports a figure given to the percentage point. The cloud forecast for a particular hour is not that accurate, and the step from “this much cloud at these levels” to “this likely to go red” is a judgement about how sunsets work, not a measurement of one.
The second is that we have nothing to calibrate against. A probability is a claim about long-run frequency — that among evenings forecast like this one, some definite fraction actually produce a red sky. Establishing that needs a record of what the sky did on a lot of evenings, and no such record exists for the Lakes. Any percentage we printed would be a number we had no way to check, which is a worse answer than an honest word.
So no number reaches the verdict. There is arithmetic behind the band, of course, but it is used to choose the word and then discarded; the code refuses to publish a verdict or a reason containing a digit. That rail exists because a number sitting next to a sunset verdict gets read as a probability whether or not anyone intended it to be one.
What we don’t claim
- Modelled, never observed. No accuracy rate, now or later, because there is no observation record to compute one from.
- Regional, not per-summit. The weather half of this is shared across the district. Only the last-light time on each fell page is a per-summit fact.
- A view forecast, never a safety one. Staying out for a sunset means coming down in fading light. Plan the descent, take a torch, and treat the walking conditions on the conditions page as the thing that decides whether to go.
- Tonight and tomorrow only. Cloud detail beyond about two days is not good enough to be worth a verdict.
- A quiet answer is a real answer. Most evenings are not red-sky evenings, and this will say so.
Sources
- Open-Meteo — Weather Forecast API — hourly cloud cover by layer (low, mid, high), visibility and precipitation; the variables this verdict is built from (opens in new tab)
- Open-Meteo / Met Office — UK Met Office UKV 2 km deterministic model, served through Open-Meteo — the short-range model behind the walking scores and this evening verdict (opens in new tab)
- Copernicus / ESA — Copernicus DEM Product Handbook (v5.0) — GLO-30 at 30 m, the terrain model behind the per-summit last-light times this page defers to (opens in new tab)