Acquiring signal…
MaxQ // Weather Center
Weather Center
Loading the Weather Center…
MaxQ // Weather Center
How it works
The Weather Center checks each upcoming launch against the weather rules its rocket must meet, the same Launch Commit Criteria the range forecasters use, and turns that into one number: Weather GO, the chance the weather lets it fly. The card, the briefing, the map and the launch page all show the same number for the same launch. No AI writes any of it: every sentence is built from the numbers.
The number
We work out the scrub odds from the weather rules (below), then print them the other way round: scrub odds of 12% read as Weather GO 88%. Three bands, each with a word, a shape and a sentence, so colour never carries the meaning alone:
A range is always printed. It comes from how much the forecast inputs disagree, and it's never narrower than ±2 points at T-0, ±8 a day out, ±10 two days out and ±15 five days out — until we've measured our own errors well enough to replace that floor.
Other people's numbers stay theirs. Launch Library 2 publishes its own weather number and the 45th Weather Squadron publishes a probability of violation for Florida launches. We show each one with its name and time, in plain ink, never in our colours, never inside our dial and never averaged with ours. The launch provider's status (“Go for Launch”) is a separate pill: it's about the rocket, not the weather.
Clocks and tiers
Launch odds are always for the launch's own T-0. Readings at a site (“Right now”) are what its weather station observes this hour, and never carry odds. How far out a launch is sets its tier:
| Tier | When | Stale after | Refreshed every |
|---|---|---|---|
| Nowcast | From 30 minutes before T-0, and through a hold past the launch time for up to 6 hours. Leans on what the pad is observing right now. | 15 min | 5 min at US pads, 10 abroad |
| Short range | 30 minutes to 48 hours out. A forecast for T-0 itself. | 45 min | 10 min at US pads, 15 abroad |
| Early look | 48 to 120 hours (five days) out. Drawn lighter with a wider range: forecasts this far out move a lot. | 3 h | 30 min |
| Forecast opens | More than 5 days out: the card says when odds start (five days before T-0). | — | — |
Past the stale limit a number is marked “may be out of date”; at twice the limit it's hidden (“Forecast temporarily unavailable”) rather than shown on old data. A number is published only when the nearest surface forecast and the upper-air profile are each within 90 minutes of T-0; otherwise the card says “Forecast for T-0 isn't ready yet”.
No odds at all for a launch with no firm time yet (only a day or a month), one that has already flown, or one from a pad or on a rocket we don't model. Those say so in words; they never borrow another rocket's limits.
The rules
These are the Launch Commit Criteria the range applies to every launch (each rocket sets its own wind and temperature limits). The briefing shows each one with its reading against its limit.
Rules have their own words, deliberately not GO and NO-GO, so “a rule is NO-GO” is never read as “the launch is NO-GO”:
Each reading also says where it came from:
“+N pts if it cleared” on a concern is how much Weather GO would rise if that one rule cleared and nothing else changed. The points for different concerns don't add up: the rules combine as “any one of them can stop it” (below), so clearing two rules is worth less than the sum of each alone.
Rule-based or calibrated
Rule-based estimate. Each rule's status becomes a small chance it stops the launch — a rule observed over its limit about 88%, one forecast over it about 60% (forecast stand-ins err on the cautious side), a near-limit rule more for the heavier rules, a rule with no data 2%, a rule that's OK 0.4% — and they combine as the chance that at least one of them stops it.
That has a ceiling. Two rules can't be measured from public data (the range's electric-field mills and smoke), so each always counts its 2%, and every OK rule adds its 0.4%. On a flawless day that leaves a rule-based estimate at about 91% Weather GO. It isn't a forecast of trouble; it's what we can't see.
Calibrated model. Where the crew have tagged enough real launch outcomes for a rocket, a model fitted on those launches replaces the hand-set chances (“Calibrated model, fitted on N launches”), and clear days read 95–99%. Either way, a launch rule over its limit puts a floor under the scrub odds, so no model can print a high GO next to a rule that would stop the launch. How the calls have gone is on the track record.
Sources and credits
The Forecaster Desk also shows SPC convective outlooks, NDBC buoys, RainViewer radar and GOES / NASA GIBS satellite imagery. Looking at a page doesn't call these services: every launch's odds are worked out once and shared by everyone who looks, refreshed on the schedule above. The one exception is the live sky over your own place, fetched once per roughly 11 km square and shared by everyone in it.
Blind spots
Source health
Checking the board's sources…
When a source fails, the numbers that depend on it are marked rather than guessed: a launch can read “Forecast temporarily unavailable” and its briefing lists what's missing, while the rest of the page carries on.
Your place and privacy
Every “your sky” module (Tonight, the YOU markers, the live dome) reads one place, chosen in this order:
Any of the 8 launch sites (Cape Canaveral, Vandenberg, Starbase, Wallops, Kodiak, Mahia, Kourou, Andøya) can be your place. Choosing one never changes your home base, which also aims your aurora, launch and re-entry alerts. “View the sky here” on the map shows a site's sky without changing your place at all.
wx_place cookie holds only what you chose: a launch site, a ZIP or city at that grid square, or just “home base” or “current location”. Never your device's coordinates.More in the privacy policy.