How our turbulence forecast works

The weather model

Every six hours the European Centre for Medium-Range Weather Forecasts (ECMWF) publishes a new run of its global model, IFS, on a grid of about 25 km. It is one of the most accurate weather models in the world and its data is open (CC BY 4.0). We download only the fields we need: wind, temperature and height at the levels where airliners cruise (400 to 150 hPa, roughly 24,000 to 44,000 feet), the vertical motion of the air, storm energy (CAPE) and rain.

Our turbulence indices

Weather models do not forecast turbulence directly; they forecast the conditions that make it. For each point of the grid and three height bands (around 27,000, 34,000 and 39,000 feet) we compute:

• Wind shear and deformation (the Ellrod indices TI1 and TI2): where wind speed or direction changes quickly with height and distance, the air tends to break into eddies. This is the main source of clear-air turbulence near the jet stream.
• Stability (the Richardson number): when the air is only weakly stable for the amount of shear, it overturns more easily.
• Storms: energy for rising air (CAPE) where it is actually raining, a sign of convective clouds.
• Mountain waves: strong wind blowing across high terrain, which can make waves in the air far above the mountains.

Each index is turned into a score from 0 to 1, and the scores are combined into one ride score. The ride score is shown on the Cup Scale: Smooth, Ripples, Bumpy or Seatbelt on. The thresholds and weights are public, in our rules.json.

Your flight

For a flight we draw the shortest path between the airports (the great circle), estimate the flight time from the distance and a typical cruising height from how long the flight is, and add a climb and a descent. Then we read the ride score every two minutes along the way, at the right height and at the right time, interpolating between the model's three-hour steps. Short blips of under four minutes are merged unless they are bumpy, because a forecast this precise would be false precision.

What we show is always labelled: confidence (high under 24 hours before departure, medium to 72 hours), the time of the model run, and what was estimated. Close to the ground, where our indices do not apply, we hold the value from the lowest height we cover instead of guessing.

Checking ourselves every day

Pilots send reports of what they feel, with a turbulence intensity, height and position. Every day we take every report from the day before between about 25,000 and 45,000 feet over North America and compare it with the forecast we had published for that time, place and height. We count how often moderate or stronger bumps were inside an area we flagged, and how often smooth reports were in areas we called Smooth. The results, good or bad, are on our accuracy page.

We tune the weights on past days and check them on days the tuning never saw, so we do not fool ourselves. Each day stays scored with the settings that were used that day.

What we don't do

• We don't promise a smooth flight. Weather changes, and airlines choose routes and heights we can't see.
• We are not a tool for pilots or dispatchers. Forecast for passengers' peace of mind. Not for flight operations. Airlines and pilots decide.
• We don't use ads or trackers, and you don't need an account.

Glossary for the curious

• EDR (eddy dissipation rate): the standard way aircraft measure turbulence intensity. Not in our open data, so we use the indices above.
• Flight level (FL): cruising height in hundreds of feet; FL350 is about 35,000 feet.
• PIREP: a pilot report.
• Jet stream: a river of fast wind high up, often where clear-air turbulence forms.