Forecast crosswind on 15/33 against what was actually observed, by lead time. Crosswind is scored rather than wind speed because crosswind is the number that decides the flight. The margin worth keeping below the 10 kt limit is about the p90 column.
Rows only appear once an observation has caught up with the forecast that predicted it, so D+0 fills within hours of first collection and each further lead day takes another day to appear. A row with a small n says almost nothing yet — read the n column before reading the rest. Sector clean is wind reaching the field across open ground; other is everything outside both named sectors.
| Source | Station | Sector | Lead | n | mean |xw err| | p90 | max | dir MAE | >3 kt | gust consistent |
|---|---|---|---|---|---|---|---|---|---|---|
| ECMWF | EVAD | other | D+0 | 9 | 4.4 | 4.4 | 4.4 | 41 | 100% | 0% |
| ECMWF | EVRA | other | D+0 | 43 | 2.7 | 5.0 | 5.0 | 22 | 44% | 79% |
| GFS | EVAD | clean | D+0 | 4 | — | — | — | — | — | — |
| GFS | EVAD | other | D+0 | 4 | 6.2 | 6.2 | 6.2 | 41 | 100% | 100% |
| GFS | EVRA | other | D+0 | 36 | 3.7 | 5.6 | 5.6 | 20 | 78% | 100% |
| ICON-EU | EVAD | other | D+0 | 8 | 5.8 | 5.8 | 5.8 | 46 | 100% | 0% |
| ICON-EU | EVRA | other | D+0 | 36 | 2.4 | 4.7 | 4.7 | 21 | 33% | 89% |
| TAF | EVRA | other | D+0 | 9 | 1.8 | 3.4 | 3.4 | 11 | 22% | — |
These are hours with wind in the 200-270 true sector at EVAD, where the on-field sensor sits roughly 139 m downwind of the hangar row. The sensor reads a floor there, not a measurement, so an apparent model bias in this table is most likely the sensor. Pooling these rows with the ones above would show every model running fast from the southwest and would calibrate you into flying an at-limit crosswind. They are kept because the size of the shelter effect is itself worth knowing.
Sheltered sector — no pairs yet. Verification needs observations to catch up with the forecasts that predicted them, so this fills in over the days after first run.
A METAR gust group only appears when the peak exceeds the mean by 10 kt or more, so a missing group means the gust was somewhere in [mean, mean+10) — unknown, not zero. A forecast gust inside that band is counted consistent because the observation cannot contradict it. Scoring a missing group as zero would make every model look like it wildly over-forecasts gusts and would discount the gust warnings that were right.
Ceiling over EVAD. There is no observation of it, and EVRA 15 nm away does not stand in for the field. The ceiling rows compare a model estimate against EVRA only, and they say how well the model does at EVRA — not what the cloud was doing over the runway.