How close each model's forecasts have been to what the station measured. Pick a window and how far ahead the forecasts were made.

“Forecast made” picks which run is scored: “1 day ahead” is the forecast made about 24 hours before each hour, what you'd have looked at the day before. “Latest run” is each hour's most recent forecast, close to a nowcast, so it always scores best. “Daytime only” leaves out the night hours you wouldn't ride.

Solid = observed hourly average (shaded to gust), dashed = what each model forecast that far ahead. Show one model to see its gusts and a second row of its direction arrows under the observed ones. Drag to pan, pinch/scroll to zoom.

Each model's forecast against the station's measured wind, averaged per hour.

± (the big number) is the average miss: how far off it was, whether too high or too low. Lower is better; the best at each lead is in bold. This is the one to judge a model by.

The small number is the bias: the same misses with their sign kept, so it shows whether the model leans one way. “+2” means it forecasts about 2 more than you get; near 0 means its misses go both ways.

Read them together: ±3 with a +2.8 bias is a model that's almost always high, so take about 3 off and it's good. ±3 with no bias is just erratic. Tap a cell for how many hours it's based on.

Each model's forecast gust against the strongest gust the station measured in that hour. ± is the average miss and the small number the bias, as in the base wind table.

Models define gusts differently, so expect bigger misses than for the base wind and a steadier bias. That makes the bias the useful part: “−4” means its gusts run about 4 weaker than what you get, so add 4 to them.

Models without a gust forecast (AIFS) aren't listed; the NWS forecast isn't archived. Tap a cell for how many hours it's based on.

± is the average angle between the forecast and observed direction, for hours averaging over 5 kt (lighter wind is too fickle to score).

Underneath is the lean: “cw” means the model tends to forecast the wind clockwise of what happened (SE when it was E), “ccw” the other way, “no lean” within 3°. A station whose vane is off by a few degrees makes every model lean the same way. Tap a cell for the share within 30°.

Every model at the chosen lead, best first. Avg error and Bias are the base wind's ± and bias (see the base wind table).

Rideable called: of the hours the model said would be in your rideable range (Settings), the share that really were. Low means false alarms: you'd go and find it too light or too strong.

Rideable caught: of the hours that really were rideable, the share the model saw coming. Low means missed sessions. High called with low caught is a cautious model (believe its “go”, but it misses some); the reverse is an optimist. These can rank models differently from ±, since what matters is being right at the edges of your range.

Within ±3 kt: the share of hours it was that close. Hours: how many were scored; with only a few (a light-wind week, or a new model like RRFS), one hour swings the percentages.

HRRR and NAM only forecast ~2 days out. The NWS forecast isn't archived, so it isn't scored. Past forecasts from Open-Meteo's Previous Runs archive.