Weather

1 Where it lives, and who can change it

Open Boats. Each boat’s row ends with a set of links — Polar, Targets, Calibration, Sails, Race plans, Weather. Weather is last, and it sits after Race plans because it serves them: what it decides is fetched for the plans on the page before it.

Any role can read this page. Changing it takes an admin or an owner — deliberately a higher bar than the race plan beside it, which any member can author. The two are different kinds of thing. A plan is an intention, written for one race by whoever is navigating it. This is the boat’s programme: it spends the organisation’s bandwidth and storage on a schedule, several times a day, for every routed plan the boat holds. It is also changed roughly never, so the extra friction lands on nobody’s race day.

The page is not behind the paywall. It renders for a boat with no active coverage and tells you exactly what it would fetch, because you should be able to see what the product does before you decide to pay for it.

2 What the platform does with it

One sentence, and the whole page follows from it: Nordic Stars fetches these models ashore, on each model’s own run schedule, for every race plan this boat holds that has a route. Nobody presses anything, and there is nothing here to press.

• There is no “fetch now”, on purpose. A button cannot conjure a model run that does not exist. At nine in the morning there is no 12z ECMWF, and all an on-demand fetch could do is re-cut a run the platform already has — which is what the schedule did when it fetched it.
• A plan with no route never gets weather. The area to fetch is derived from the route, so a windward/leeward plan, and any coastal or offshore plan whose route is still empty, has nothing to cut a forecast to. That is arithmetic rather than a rule, and you can predict it from the plan in front of you.
• A boat with many saved plans is not fetched for all of them. The platform keeps the fivemost recently updated routed plans of each boat current — which are the plans somebody touched last, and so the ones somebody is about to sail. Update a plan before the regatta and it is one of the five being kept fresh.

The job runs every ten minutes. Each run walks the routed plans of every covered boat, up to five per boat, and builds a corridor for any it does not already hold one for. So a plan saved now has its files about ten minutes later, plus however long the build takes — and a plan whose route you edit outside the corridor already built for it is rebuilt on the same schedule. What a run cannot do is invent a model run that has not been published yet.

This is what gets fetched, not what gets shown: which model a crew looks at is picked on the boat, from the bundles actually in hand. There is no display setting anywhere on this page, and there should not be. A stored “show me ECMWF” goes stale the moment a run fails — you would arrive at an empty overlay because a document told the tablet to draw a file it does not have. A picker over what is actually aboard cannot do that.

3 The models, and what each one cannot do

Six models, listed with the organisation that publishes each. The line under a model is a capability fact — something you would otherwise find out on deck. It is never a grade and never a recommendation, and where a model carries no line it is because there is no such fact to state, not because we like it more.

The two southern numbers are about two different models and are not two spellings of one limit. 37.5–55.4° N is AROME’s whole domain. 29.5° N is where ICON’s high-resolution European nest ends — global ICON carries on south of it. A Canaries race is served by the global models; a Solent race has AROME available to it and a Caribbean one does not.

AIFS having no gusts is worth reading twice. It is not a gap in our fetch and not something a later release fills: the model does not produce a gust field, so a bundle from it carries wind and no gusts. We will not manufacture one from the mean wind, because a gust number is what a crew reefs on and a derived one would carry our guess with the model’s authority.

The set, and what it costs

• A boat nobody has configured starts on ECMWF and GFS, with a 30 nm margin. The page never labels them “default”, because the platform does not record whether a setting was seeded or chosen, and a screen claiming otherwise would be describing our history rather than your boat.
• Four is the cap. Each model is a scheduled fetch, a stored file per run, and a download the crew stands on a pontoon waiting for. At four, the boxes that are not ticked stop accepting a tick and the page says why; the ticked ones are never disabled, so the way out of the cap is never itself capped. Untick one to add another.
• No models at all is a valid setting, not an unfinished form. The list reads No models — weather fetching is off for this boat, and that is exactly what it does: it stops fetching weather for this boat, and nothing else.
• A model we have not heard of is shown, by name and ticked, with a small newer than this page chip beside it. That happens when the platform is running ahead of this screen. It is listed rather than hidden because saving replaces the whole set — an invisible model would be dropped by the next thing you did, and unticking it ought to be something you chose.

4 The corridor, and what the margin buys

A forecast for a route is not a forecast for a line. It is a corridor: the box around the route, widened by a margin you set here, and the model run cut down to that. The margin is the only number on this page.

Margin (nm) is how far outside the route’s bounding box the corridor reaches, between 5 and 200 nautical miles, and 30 to begin with. It exists because a boat does not sail the rhumb line: a beat takes you twenty miles off it, a headland gets a wide berth, and a forecast that stops at the drawn route is a forecast that runs out exactly where a navigator starts making decisions.

It drives the download roughly quadratically. Doubling the margin is nearer four times the bytes, so this is the field that decides whether the pre-fetch at the dock is three minutes or twenty. Widen it for an ocean crossing where you might be a long way off plan; leave it near the default for a coastal race down a familiar shore.

Editing the route later does not invalidate anything by itself, and nothing on the boat has to guess: each fetched bundle states the box it actually covers, so the tablet can say this weather does not cover your route before it spends marina wifi rather than after, and before the wind field runs off the edge at a headland.

5 Getting it to the boat

The platform fetching a forecast puts it on the platform. Getting it onto the tablet is one deliberate act, done at the dock, and it is the whole reason this feature exists.

1. On the tablet, with a connection — marina wifi, the pontoon, the car park — open SETTINGS → WEATHER AND DATA → WEATHER. It sits beside REPLAY, because both are documents you fetch for a race while standing still, and neither is touched on the water.
2. The card is about one plan: the one loaded, or else the newest routed one. Any other plans are counted and named rather than quietly left out.
3. Each row is the newest run of one model, with its size and whether it is already aboard. Press to take them. The progress bar counts bytes across the whole run, not files, so it does not sit at 25% for four minutes because one model is far bigger than the next.
4. Then leave. Everything the wind screens do afterwards reads off the tablet’s own storage, with the antenna down.

Three kinds of file come down that button, not one. The models above are the first. Beside them the platform builds, for the same corridor:

• The sea model — the depth grid, the coastline, the seamarks and the hazards along the route. It is what the depth check in the routing reads, and what tints the water on the chart. It is not a nautical chart, and the tablet says so.
• An official chart extract, where a national hydrographic office publishes one on terms that allow it. The tablet draws it above the sea model and names which of the two you are looking at, in three places: the chart legend, the card you get by tapping the water, and the row beside the corridor on this card.

Neither is affected by the model list on the boat’s weather page. Unticking every model stops the weather fetching and nothing else.

Three things about that screen behave the way you would want and are worth knowing in advance:

• A second press costs nothing for what is already held. A run is immutable — a new run is a new file, never a changed one — so a bundle already on the tablet is not re-checked and not re-downloaded.
• Cancelling is not a failure. Rows in flight go back to missing and the next press starts them again. Nothing is left half-written: a download only takes its real name once it has arrived whole and its size matches what the list promised.
• A bundle that does not arrive is amber, not red, and it is a bundle to ask for again. A run that has aged out of retention is ordinary rather than broken, and it costs that one file rather than the rest of the press.

Weather is fetched by hand and there is no background sync for it. That is the point: multi-megabyte downloads belong to a person standing at a dock with a decision to make, not to a tablet that might start one at T−4:00.

6 The staleness line means what it says

Wherever the tablet shows you weather it shows you this beside it:

ECMWF 12z · fetched 09:41 · to Thu 18:00

Three facts, because they answer three different questions and a crew needs all three:

It is always on screen where the wind is. A wind field with no provenance beside it reads as the wind; what it actually is is a forecast from a run that may be eighteen hours old.

7 Reading it on the water

The forecast draws over the course on NAVIGATION, under the marks and the laylines, with a transport for walking the forecast hours and a picker for which model you are looking at. The rest of that screen is Race Control on the boat. What belongs here is what the wind layer will and will not claim.

• A missing barb is missing data, never a calm. Limited-area models do not cover a whole corridor, and where the model said nothing the tablet draws nothing rather than interpolating across the hole. A calm circle is the one glyph that would say the opposite while looking entirely deliberate — so a real calm gets a circle and an absence gets bare water.
• Barbs are read to the nearest five knots. A pennant is 50, a full barb 10, a half barb 5. That rounding is the meteorological convention rather than an approximation we chose — a barb has never been able to say 13 knots — and we are not going to invent a two-fifths feather to pretend otherwise. Below 2.5 knots it is a circle, not a bare shaft, because a shaft with no feathers claims a direction the field does not have.
• Gust windows differ, and not only between models. A gust is not an instantaneous value: it is the maximum over a window, and the window is not uniform across a single run — a model may publish a 1-hour maximum at short leads and 3- or 6-hour maxima further out. Gusts from steps with different windows are not comparable with each other, and comparing two models’ gusts compares two conventions as well as two forecasts. The bundle carries the window per step so the tablet is never guessing at it. And AIFS carries no gusts at all.
• The forecast hours stop where the model spoke. You can walk to the steps the file actually carries and not between them, because a smooth continuum would let you land on an instant whose gust statistic covers no window at all. The wind between steps is interpolated for reading a position; where you may stop is not.
• The model picker is built from what is aboard. A run that failed to fetch is simply not offered — there is no row that breaks when you tap it, and no stored preference pointing at a file that never arrived.
• The wind at the next waypoint is a projection and says so. It reads the forecast at your projected arrival, which assumes you hold your current speed made good to that mark — and a boat in a dying breeze will not. Past the end of the forecast there is no line at all, because nothing extrapolates.

None of this is navigation. It is a forecast drawn over a course, and the ship’s plotter remains the navigator’s authority — that line stays true of every layer we ever add.

8 Changing it later

A change takes effect at the next scheduled fetch. Nothing is rebuilt and nothing already downloaded is deleted — bundles held for a model you remove keep serving until retention reclaims them. The page prints when the setting was last replaced, so you can tell a change that has not been picked up yet from one that has.

Practically: change it ashore, days before, not on the morning of a race. Adding a model at eight o’clock does not produce a file to download at half past — it produces one when that model next publishes a run and the platform next fetches. If a forecast has to be aboard on a particular morning, the thing to get right the evening before is the fetch at the dock, not this page.

9 Where to go next

• Race plans — drawing the route this weather is fetched for, which is the thing that has to exist first.
• Race Control on the boat — the tablet from the licence to the first gun, including the navigation screens the wind draws over.
• Getting started — if the boat is not in the list yet.