How we test our forecast.

Calmaboard tells you when bumps are more likely during your flight. Here is how that forecast is made, how we checked it against a year of real flights, and where it falls short.

What the forecast looks at

Every 6 hours, the European Centre for Medium-Range Weather Forecasts (ECMWF) starts a new global weather forecast. Calmaboard reads its winds at cruising height along your route and computes a turbulence index from them: where the wind changes quickly with height and the air is stretched sideways, the ride tends to be bumpy. This kind of index has long been used in aviation weather forecasting.

Your route is drawn as the shortest line between the two airports and cut into 5-minute stretches. For each stretch of the cruise, Calmaboard keeps the worst value over three cruising heights and up to about 50 to 100 km around the line, because a plane never flies exactly on it. Each stretch then gets one of three levels, and the whole flight gets the line you read first:

  • Most likely a smooth flight.
  • Some light bumps possible.
  • Bumps more likely than usual.

The limits between the levels were set on past flights: about half of all flights fall in the lowest level, about 4 in 10 in the middle one, and about 1 in 10 in the top one.

Climb and descent are not covered. Neither are storms: our storm signal failed its test, so we removed it.

Two tests on real flights

We scored the forecast by ranking. Take one stretch where an aircraft met bumps and one where it didn't, and check which one our forecast rated higher. A score of 0.5 means no better than a coin flip; 1.0 means always right. It is not a percentage of correct forecasts. Both tests used forecasts made 42 to 54 hours before the weather they describe, about what you read 2 days before departure, over 12 months: October 2025 to September 2026.

Pilot reports, North America

Pilots report the bumps they meet, or a smooth ride. We used 336,360 reports made at cruising height, mostly over the United States, and compared clear bumps with smooth rides (light bumps were left out of this score). The method was written down before we computed any score.

The raw score in the summer months was 0.75, but it is flattered by geography: smooth reports cluster in places where our index is low anyway. Compared at the same place, the score is 0.67 to 0.70. That is the number we quote: about 2 times in 3, our forecast was higher where the bumps were.

Measurements by airliners

Some airliners measure the turbulence they fly through, automatically, every minute or so. NOAA keeps these measurements in a public archive. We counted a stretch as bumpy from a measured value of 0.20 (moderate) and as smooth below 0.10, and kept the index exactly as in the first test.

Bumpy measurements (for the North Atlantic: bumpy 5-minute stretches), and the score of the forecast. “Same place”: compared within the same small area and month. “Wrong week”: the same scoring with the weather of 7 days earlier.
Area Bumpy Score Same place Wrong week
Europe and the Mediterranean (test area up to 60° E) 1,136 0.813 0.762 0.542
North America 26,556 0.815 0.794 0.536
North Atlantic 2,759 0.796 0.756 0.525

With the wrong week's weather, the score falls close to a coin flip. So the skill comes from the day's weather, not from knowing which places are often bumpy. The North Atlantic was checked later, on 6 October 2026, with a method written after the first two tests. The European test area reached 60° E, but its Middle East part held only 5 bumpy measurements, so we don’t count the Middle East as tested.

Aircraft Based Observations (ABO) data provided by NOAA/NWS/NCEP/MADIS.

What happened on past flights

From the same airliner measurements, we rebuilt 410,910 flights and kept the 359,112 with at least 30 minutes of cruise. For each one, we looked at the worst stretch of its cruise: did the aircraft record light or stronger turbulence (a measured value of 0.10 or more)? Then we grouped the flights by the forecast each would have had.

Share of flights with light or stronger turbulence recorded during the cruise.
The forecast said October to May (236,107 flights) June to September (123,005 flights)
Most likely smooth 8.4 % 9.4 %
Light bumps possible 26 % 23 %
Bumps more likely than usual 42 % 35 %

In plain words: when the forecast said “most likely smooth”, about 9 in 10 flights recorded no light or stronger turbulence. When it said “bumps more likely than usual”, about 1 in 3 to 2 in 5 did. Strong turbulence was recorded on fewer than 1 flight in 100 in every group.

  • October to May are the months we used to set the levels, so those figures are not an independent check. We also chose where to split the levels (half of the flights, then 9 in 10) after seeing the rates. June to September were kept aside and checked afterwards.
  • Most of these flights are over North America: 9.66 million of the 11.46 million cruise stretches. For flights that start their cruise over Europe, the June to September figures are 6 %, 16 to 24 % and 31 to 43 %.
  • A flight's area here is the area of its first cruise stretch, so “Europe” is not a set of European routes.

Aircraft Based Observations (ABO) data provided by NOAA/NWS/NCEP/MADIS.

Where we tested

The app calls a route tested when its whole cruise stays inside these areas:

  • North America: 20° N to 72° N, 170° W to 50° W.
  • The North Atlantic: 25° N to 72° N, 50° W to 15° W.
  • Europe and the northern Mediterranean: 36° N to 72° N, 15° W to 28° E.

If part of the cruise is outside, the app says: “Part of this route is outside the areas where we tested our forecast.” If all of it is outside: “We have not tested our forecast in this area yet.” Routes that fly too close to the North or South Pole get no forecast at all.

The European area stops at 36° N and 28° E because almost all the bumpy 5-minute stretches over Europe lie inside it: 861 of the 872 found between 30° N and 72° N, 15° W and 36° E. South of 36° N there were 4, east of 28° E there were 7, too few to call those places tested. The wider Mediterranean, from 30° N to 46° N and 6° W to 36° E, scored 0.862, but on only 139 bumpy measurements.

Not tested yet: North Africa south of 36° N, the eastern Mediterranean east of 28° E, the Middle East, the rest of Africa, Asia, South America and Oceania.

Aircraft Based Observations (ABO) data provided by NOAA/NWS/NCEP/MADIS.

How far ahead

Your first forecast arrives about 2 days before departure. By the end of a long cruise, the weather it describes can be up to about 83 hours after the forecast was made, longer than in our two tests. So we scored the same stretches again with forecasts made 57 to 81 hours ahead:

Score, with its 95 % range over whole days.
Area 42 to 54 hours ahead 57 to 81 hours ahead
North Atlantic 0.796 (0.777 to 0.815) 0.777 (0.755 to 0.797)
Europe and the northern Mediterranean 0.795 (0.765 to 0.821) 0.768 (0.739 to 0.795)
North America 0.804 (0.797 to 0.810) 0.787 (0.780 to 0.794)
All areas 0.802 (0.796 to 0.809) 0.785 (0.778 to 0.792)

This table uses one smooth stretch in 10 and the narrower European area, so its first column differs slightly from the scores above. Made further ahead, the forecast is slightly less sharp, and still far above chance. This check was done after the first tests, not planned before them. Forecasts made less than 42 hours ahead, like the update on the day of your flight, were not tested. Weather forecasts usually improve closer to the time, but we haven't measured it for ours.

Aircraft Based Observations (ABO) data provided by NOAA/NWS/NCEP/MADIS.

What it doesn't measure

  • What you feel in your seat. Our tests use measurements from aircraft; they do not measure what each passenger feels.
  • Storms. That part of our method failed its test (it scored 0.38, worse than a coin flip), so we dropped it.
  • Climb and descent.
  • Every airline. The airliner measurements come mostly from US airlines' aircraft, on routes across North America, the Atlantic and northern Europe.
  • Summer as reliably as winter. Summer is the hardest season for the forecast: in June to September the scores were 0.700 over the North Atlantic, 0.734 over Europe and 0.765 over North America.
  • Bumps are rare in the data, and the minutes of one bumpy patch are counted one by one. The ranges above are computed over whole days to allow for this.

Calmaboard is an estimate based on weather forecasts, not a safety tool. Keep your seatbelt fastened whenever you're seated, and always follow the crew's instructions.

Sources: ECMWF open forecast data (credit below); pilot reports from the Iowa Environmental Mesonet archive; airliner measurements from NOAA's MADIS archive.

Aircraft Based Observations (ABO) data provided by NOAA/NWS/NCEP/MADIS.