8 matched pairs from 268 observations · rebuilt by the run of Sun 27 Sep 14:40 (Europe/Riga)
Read n first
Each row compares the forecast crosswind on 03/21 with what a station reported, by lead day. A row only shows up once an observation has caught up with the forecast that predicted it. D+0 fills within hours; every further lead day takes another day. A row built on a handful of pairs says next to nothing, so look at n before the rest.
p90 is the column to hold against your crosswind limit. It is roughly the margin worth keeping below the limit at that lead.
Crosswind error, open-air hours
EVVA · official METAR
| Source | Sector | Lead | n | mean | p90 | max | >2 kt | >3 kt | dir | speed | gust ok |
|---|---|---|---|---|---|---|---|---|---|---|---|
| ECMWF | other | D+0 | 0 | - | - | - | - | - | - | 0.3 (2) | 100% (2) |
| GFS | other | D+0 | 0 | - | - | - | - | - | - | 2.0 (2) | 100% (2) |
| ICON-EU | other | D+0 | 0 | - | - | - | - | - | - | 1.7 (2) | 100% (2) |
| TAF | other | D+0 | 1 | 1.3 | 1.3 | 1.3 | 0% | 0% | 20° | 1.5 (2) | - |
- n
- pairs with a crosswind error
- mean
- mean absolute crosswind error, kt
- p90
- 90th percentile crosswind error, kt
- max
- largest crosswind error, kt
- >2 kt
- share of pairs off by more than 2 kt
- >3 kt
- share of pairs off by more than 3 kt
- dir
- mean direction error, degrees
- speed
- mean wind speed error, kt
- gust ok
- share of pairs where the forecast gust agrees with the report
- (k)
- pairs behind that figure, when not n: calm and variable reports count for speed and gust
- -
- no value, e.g. every report in the row was calm or variable
How it is measured
Each report is paired with the model hour nearest to it, within 35 minutes, for leads up to D+5. Lead is counted in calendar days from the run, the same D+n as the day cards.
Only genuine forecasts are scored. The models return the whole day, so a midday run also carries hours that were already over when it was fetched. Those are hindsight, and pairing them would flatter D+0, the column this page tells you to read for margin. A pair whose forecast was issued after the hour it describes is dropped.
A calm or variable report has no direction, so there is no crosswind to get wrong. Those pairs are left out of n and the crosswind figures, and count only in the speed and gust columns, which show their own count in brackets when it differs from n.
With a calm or variable report, the forecast direction alone decides the sector. Other is everything outside the named sectors, and unknown means neither side had a direction.
A METAR gust group only appears when the peak beats the mean by 10 kt or more. A missing group means the gust stayed below mean + 10 kt: unknown, not zero. A forecast gust under that bound counts as consistent, because the report cannot contradict it. With a gust reported, the forecast has to land within 5 kt. Scoring a missing group as zero would make every model look like it overforecasts gusts, and would discount the gust warnings that were right.
The ceiling is not scored. The brief estimates it from model temperature and dew point, and that estimate is not checked here.
The lead-day confidence and the gust rules in the scoring were measured at EVAD and are applied to every airfield for now. The How to read page has the detail.