Weather and launch conditions

A launch in Vivapse goes from a real place into the weather that place has now. The wind the vehicle climbs through is the forecast wind over that pad, layer by layer up to about 31 km; the air it pushes through has that day's temperature, pressure and humidity; the drone ship downrange rides that day's sea.

The weather reaches the vehicle only through the air. Wind acts through aerodynamics and nothing else — a vehicle held on the pad in a gale does not drift, and one above the atmosphere feels nothing. That is the physics, and it is also the reason weather is a problem mainly in one part of the flight.

Three kinds of weather

• Live. A snapshot from Open-Meteo at the pad, the landing zone, a point halfway down the track to the drone ship, and the ship's station. It refreshes every ten minutes while the vehicle is on the pad and is frozen into the mission at launch, so a flight never changes weather halfway up. Lock the seed and the snapshot is held: every launch replays exactly those conditions. Move the pad more than 5 km and the snapshot is dropped rather than shown for the wrong place.
• Random. Rolled from the mission seed. On a real site it follows the latitude and the season: trade winds or westerlies near the ground, the jet stream's real direction and strength, stratospheric winds including the tropics' quasi-biennial oscillation, and the local sea-level pressure, humidity and tropopause.
• Custom. Surface wind and its direction, gusts, turbulence, the jet stream's speed and direction, temperature and sea state, as you set them.

If live weather is asked for and Open-Meteo cannot be reached, the flight falls back to random weather from the seed, and the launch-readiness card says so. The same seed always gives the same weather.

What a live snapshot contains

For each point, Open-Meteo returns the current conditions and one model hour's vertical profile — the hour nearest the current-conditions time:

• At the surface: 2 m temperature and humidity, surface and sea-level pressure, the 10 m wind and its gusts, total and low cloud cover, precipitation, the WMO weather code, visibility and CAPE (convective available potential energy, a measure of how readily the air overturns into cumulus and storms).
• Aloft: temperature, humidity, wind speed and direction and geopotential height at every one of its 35 pressure levels from 1 000 hPa to 10 hPa — roughly 0.1 to 31 km. Levels that fall below the ground are dropped.
• At sea, from Open-Meteo's marine service at the ship's station: significant wave height, mean and peak wave period, wave direction, swell and sea-surface temperature.

A four-point snapshot is about 16 KB, small enough to travel inside a share link. The data are Open-Meteo's, published under CC BY 4.0: Weather data by Open-Meteo.com.

From a snapshot to the air

A snapshot is a few columns of numbers. The flight needs air at every point it passes through, so the simulator interpolates.

Across the ground, the columns are blended by inverse distance squared, with a 3 km core so that a point on top of a column is not infinitely weighted to it. Columns with less than 2 % of the nearest one's weight are dropped.

Up each column, the wind is linear between pressure levels. Weather models report level heights as geopotential — energy per unit mass divided by standard gravity — so each is first converted to a geometric height with that point's own gravity; a tropical column stands about 0.26 % taller than standard gravity would place it. Below the first level the wind follows a surface layer:

where $u(z)$ is the wind at height $z$ above the surface, $u_{10}$ the reported 10 m wind, and $z_0$ the roughness length: 0.2 mm over open sea, 3 cm over open land. From 10 m to the first level the wind blends in the logarithm of height. Above the top level it decays with an 8 km scale and fades to nothing between 40 and 45 km; there is no wind above 45 km. The boundary layer is measured from the surface under the vehicle, not from the pad, so a drone ship at sea level below an elevated pad sees its own wind shear.

The air itself is built from the 2 m temperature and the level temperatures, blending into the US Standard Atmosphere over the 8 km above the top level. Humidity enters as virtual temperature — moist air is lighter than dry air at the same temperature and pressure — and the pressure is integrated hydrostatically upward from the measured surface pressure. The snapshot's own level pressures come back to about 0.1 % on average, 0.6 % at worst across the recorded snapshots. The whole atmosphere turns with Earth, and the wind is added on top of that rotation.

Gusts and turbulence

The forecast is a mean. What a vehicle actually meets is the mean plus turbulence, and on every real launch site that turbulence is a frozen three-dimensional Dryden field: the standard engineering model of atmospheric turbulence (MIL-F-8785C, MIL-HDBK-1797), built so that along any path, in any direction, the vehicle sees the right spectrum. Its energy spectrum is

where $k$ is the wavenumber, $\sigma$ the turbulence intensity (its RMS velocity) and $L$ its length scale. The field is synthesised from 63 Fourier modes on 21 logarithmic shells between $k=3.5\times 10^{-4}$ and 0.6 rad/m — wavelengths from about 10 m to 18 km.

• Near the ground (below 1 000 ft above the local surface) it follows the military low-altitude model: the vertical intensity is a tenth of the wind at 20 ft; the horizontal intensity starts at about twice that and falls to equal it by 1 000 ft; the length scales grow with height; and the vertical component dies away at the ground itself.
• Aloft (above 2 000 ft) it is isotropic, with a length scale of 1 750 ft (533 m) and an intensity from the standard probability-of-exceedance table, plus clear-air turbulence around the jet core. Between the two it blends linearly.
• Its strength comes from the forecast. The hour's peak three-second gust is taken as the mean wind plus three standard deviations, so ${\sigma }_u\approx (G-U)/3$ at 10 m, where $G$ is the reported gust and $U$ the mean wind. Shear, CAPE and a thunderstorm code raise a turbulence index, which adds to the gustiness near the ground and sets how rough the air is aloft.
• It moves with the wind. The pattern is carried along by the local mean wind — Taylor's frozen-turbulence hypothesis — so a vehicle hovering over the pad sees gusts arrive at the rate the wind brings them, and one crossing the field at 400 m/s sees it at the frequencies that speed implies. Real eddies do not live for ever, so two generations of the field, each lasting 120 s, are cross-faded.

The field is a pure function of position, time, the day and the seed: the same flight meets the same gusts. On the classic equatorial range the turbulence is the original model's sum of sine waves instead, which fidelity and its limits explains.

Why wind matters most around max-Q

The aerodynamic load that breaks rockets is the bending of the airframe, and it scales with the product of dynamic pressure and angle of attack:

where $\rho$ is the air density, $V$ the speed through the air, and $\alpha$ the angle between the nose and the oncoming air. A crosswind $w$ across a vehicle climbing at speed $v$ tilts the oncoming air by roughly $\alpha \approx \arctan (w/v)$ — so if the vehicle keeps flying its planned attitude, the wind shows up directly as angle of attack. The flight computer reads the product as fc.qAlpha, and above fc.limits.qAlpha, 250 kPa·°, the vehicle breaks up.

Near the ground the wind is a large fraction of the vehicle's speed, but the speed is low: at 1 km the sketch below is doing 100 m/s at under 6 kPa. High up the vehicle is fast, but the air is too thin to push. The load is worst in between, at maximum dynamic pressure, and that is where the jet stream lives: between roughly 9 and 14 km, exactly the altitudes at which a launcher passes max-Q.

At its defaults the figure's stage meets a 40 m/s jet core with 225 kPa·° — 90 % of the limit. At 50 m/s, the board's marginal threshold for upper winds, it breaks up. That is the unrelieved case: a vehicle that holds its pitch programme and lets the wind do what it likes.

A real vehicle does not. It relieves the load by turning into the relative wind, and in Vivapse that is one line: steer along fc.airPrograde and fc.airProgradeYaw, the direction of the oncoming air, and the steady part of the crosswind stops producing angle of attack. The cost is a trajectory that bends with the wind, which guidance has to correct later. What load relief cannot remove is what it does not see coming:

• Shear. A vehicle passing max-Q climbs at about 400 m/s vertically and crosses a kilometre of height in under three seconds. Through a shear of 16 m/s per km — the board's no-go threshold — the crosswind changes by 16 m/s in that time, about two degrees of angle of attack at max-Q speeds, some 70 kPa·° of load before the vehicle can turn.
• Gusts. The autopilot's aerodynamic feed-forward is built from the forecast mean wind, never the gust, so the gust is fought by feedback alone. Expect tenths of a degree of pointing error in gusty air at max-Q.

The launch-weather board

Before a live launch the Mission panel grades the snapshot against a set of rules modelled on real range practice: the Lightning Launch Commit Criteria of the US Eastern Range as applied by the 45th Weather Squadron, and the wind, shear and recovery limits launch providers add to them. The thresholds are chosen to sit in the range of the real ones; they are not any provider's exact numbers.

A layer counts as cloud where the relative humidity at consecutive levels is 90 % or more, extended half-way to the dry levels either side; the thick-cloud rule exists because a vehicle climbing through such a layer can trigger lightning. The cloud base is estimated as 125 m for every degree between the temperature and the dew point. The verdict is the worst rule's status, and a rule with no data behind it is shown as unknown and ignored. The recovery rules only apply to the platforms the mission actually uses.

Recorded on 18 September 2026 at about 16:00 UTC, all four reference sites came out marginal, each for a different reason: Starbase for CAPE of 3 260 J/kg along the track, Kourou for 13.6 m/s gusts at the pad, Baikonur for a 706 m cloud layer in the freezing band, and Vandenberg for a marine-layer ceiling of about 360 m.

What the board does and does not do

The board advises; it never refuses a launch. Whether to fly is your decision, and the physics then decides what happens. It is worth being precise about which rules the physics enforces for you:

• Wind, shear, gusts and sea state have physical consequences. They reach the vehicle through the aerodynamics and the drone ship's motion, and they can break it up, tip it over on the deck, or push it off the landing zone.
• CAPE and thunderstorms act indirectly. They raise the turbulence, so the air is rougher; there is no lightning.
• Precipitation, icing, cloud, visibility and ceiling have no physical effect. Rain does not load the airframe, ice does not form, and nobody needs to see the vehicle. Those rules are on the board because a real range would scrub for them, and a no-go there is a lesson rather than a threat.

The sea

The drone ship heaves, rolls, pitches, surges and sways. With live weather its motion comes from the significant wave height and the peak period at its station: seas on the beam roll it, seas on the bow pitch it, and its tilt follows the slope of the waves. Without marine data the sea is derived from the wind. Landings on the ship are judged relative to its moving deck, and a stage that has landed rides the deck and still feels the wind: the Aster booster, nearly empty, is blown over at about 26 m/s of surface wind. Coming back down covers the landing itself.

On the Moon and Mars

The Moon has no weather and no air. Mars has a mean CO₂ atmosphere and a light, steady westerly that rises from nothing at the ground to 5 m/s by about 5 km and holds there, with no gusts, no turbulence and no dust storms. The launch-weather board is about the launch, which is always from Earth.

What the flight computer sees

There is no wind sensor on fc. The program sees the wind's effect instead: fc.airspeed against fc.surfaceSpeed, and the direction of the oncoming air as fc.airPrograde and fc.airProgradeYaw against the ground-relative fc.prograde and fc.progradeYaw. fc.airDensity, fc.pressure, fc.dynamicPressure, fc.mach, fc.aoa and fc.qAlpha give the rest. The predictions — fc.impact, fc.predict(), fc.stopPoint() — include drag in the forecast mean wind, and like a real flight computer they cannot know the gusts. With realistic sensors the air data carry errors too: 1 % on dynamic pressure, density and pressure, 0.5 % on airspeed and Mach, 0.2° on angle of attack.