Storm Station

Live/Accuracy

Forecast accuracy, the day after

Every forecast Storm Station records is checked the next day against what the airfields reported, hour by hour. A reading is first checked against the stations around it, so a faulty instrument never counts against a forecast, and every miss of more than 10 F is explained.

Day scored
September 29, 2026
Hours of reports
24 of 24
Temperature forecasts scored
236,088
Readings checked
116,174
Readings that failed
154

Which forecast leads

The Storm Station correction is scored beside the blend every day, with only what was known before that day. It takes the lead for a figure only when the Diebold and Mariano test, over at least 14 days, finds it better with less than a 5 in 100 chance of luck. Until then the blend leads.

  • Temperature Blend of models2 days scored
  • Dew point Blend of models2 days scored
  • Wind Blend of models2 days scored
  • Sky cover Blend of models1 day scored

Temperature

Mean miss, F, by how far ahead the forecast was made
Forecast1 to 6 h7 to 12 h13 to 18 h19 to 25 hAllRanInside its bandHours
Blend of models2.22.22.42.52.3+0.152%236,088
Storm Station correction2.22.22.42.52.3+0.152%236,088
Forecaster’s grid2.02.12.2-2.1-0.3-2,073
Global model2.22.22.5-2.2-0.3-2,073
Storm Station lead1.91.92.1-1.9-0.3-2,073
Mean miss of the temperature forecast, F, each day from September 28, 2026 to September 29, 2026: the blend of models and the Storm Station correction0123September 28, 2026September 29, 2026
Blend of models Storm Station correction

Dew point

Mean miss, F, by how far ahead the forecast was made
Forecast1 to 6 h7 to 12 h13 to 18 h19 to 25 hAllRanInside its bandHours
Blend of models2.22.22.22.32.3+0.563%235,305
Storm Station correction2.22.22.22.32.3+0.563%235,305
Forecaster’s grid1.72.02.1-1.80.0-2,074
Global model2.52.82.9-2.6-0.8-2,074
Storm Station lead1.72.02.0-1.9-0.4-2,074
Mean miss of the dew point forecast, F, each day from September 28, 2026 to September 29, 2026: the blend of models and the Storm Station correction0123September 28, 2026September 29, 2026
Blend of models Storm Station correction

Wind

Mean miss, mph, by how far ahead the forecast was made
Forecast1 to 6 h7 to 12 h13 to 18 h19 to 25 hAllRanInside its bandHours
Blend of models2.42.52.52.52.5+0.353%235,184
Storm Station correction2.42.52.52.52.5+0.353%235,184
Forecaster’s grid2.62.73.0-2.7+0.2-2,066
Global model2.72.83.1-2.8-0.7-2,066
Storm Station lead2.42.62.8-2.5-0.3-2,066
Mean miss of the wind forecast, mph, each day from September 28, 2026 to September 29, 2026: the blend of models and the Storm Station correction0123September 28, 2026September 29, 2026
Blend of models Storm Station correction

Sky cover

Mean miss, %, by how far ahead the forecast was made
Forecast1 to 6 h7 to 12 h13 to 18 h19 to 25 hAllRanInside its bandHours
Blend of models25.726.126.427.026.3-15.245%88,103
Storm Station correction25.726.126.427.026.3-15.245%88,103
Forecaster’s grid18.620.422.0-19.7-13.2-1,734
Global model23.321.123.1-22.5-2.2-1,734
Storm Station lead17.716.919.6-17.7-7.7-1,734

Rain

The chance of rain in each hour is scored by the Brier score: the mean square of the chance less what happened, 0 when it did not rain and 1 when it did. Lower is better; skill compares it with always forecasting how often it rained. The amount is scored at 0.01 and 0.10 inch by the equitable threat score, which gives no credit for hits chance would have made.

ForecastBrier, 1 to 6 hBrier, 7 to 12 hBrier, 13 to 18 hBrier, 19 to 25 hSkill, 1 to 6 hThreat score, 0.01 inThreat score, 0.10 in
Blend of models0.0360.0370.0380.0400.380.310.10
Global model0.0340.0440.053-0.410.270.06
Storm Station lead0.0340.0440.053-0.410.290.12

Snow

-

Misses of more than 10 F, explained

  • 31The stationThe reading failed its check against its neighbours: the station was wrong, not the forecast.
  • 1The forecast, across the areaMost of the stations around it were missed the same way.
  • 317The forecast, at this stationIts neighbours were forecast well: its ground, its siting or a boundary the forecast did not place.
  • 39Cannot be toldToo few stations stand near it to say which was wrong.
AirfieldHourSaidReadIts neighboursHours aheadWho was wrong
PATA00 UTC46 F5.0 F-6Cannot be told
KOAR17 UTC63 F37.4 F62.2 F5The station
KOAR16 UTC59 F35.6 F57.7 F4The station
KGVW04 UTC63 F84.2 F78.7 F4The forecast, at this station
KGVW05 UTC63 F84.2 F78.2 F5The forecast, at this station
KGVW06 UTC63 F84.2 F77.8 F6The forecast, at this station
KGVW07 UTC63 F84.2 F78.2 F1The forecast, at this station
KGVW08 UTC63 F84.2 F76.7 F2The forecast, at this station
KGVW09 UTC63 F84.2 F76.7 F3The forecast, at this station
KGVW10 UTC63 F84.2 F77.1 F4The forecast, at this station
KGVW03 UTC64 F84.2 F79.8 F3The forecast, at this station
KOAR18 UTC66 F46.4 F68.0 F6The station
KGVW11 UTC63 F82.4 F75.6 F5The forecast, at this station
KGVW12 UTC63 F82.4 F75.2 F6The forecast, at this station
KGVW02 UTC65 F84.2 F80.7 F2The forecast, at this station
KGVW13 UTC65 F84.2 F78.5 F1The forecast, at this station
KOAR01 UTC60 F41.0 F64.1 F1The station
KOAR20 UTC67 F48.2 F79.1 F2The station
KOAR07 UTC56 F37.4 F56.8 F1The station
KSEE23 UTC82 F100.4 F77.3 F5The station
KOAR15 UTC54 F35.6 F53.2 F3The station
KGVW01 UTC66 F84.2 F82.4 F1The forecast, at this station
KC9922 UTC58 F41.0 F46.5 F4The forecast, at this station
KOAR02 UTC58 F41.0 F62.9 F2The station
KOAR19 UTC67 F50.0 F73.1 F1The station

How it is scored

What happened. For each hour, each airfield’s routine report from 15 minutes before the hour to 5 minutes after. Rain in the hour counts only where the station has a working gauge; a station without one has no zero to give.

The check. Each reading is compared with the stations within 60 miles and 2,000 ft of its height, their readings carried to its height. One further from the middle of them than four times their spread, and never less than 10 F, fails, and is neither scored nor learned from. With too few neighbours, the analysis of record is the second opinion.

The scores. The mean miss, and whether a forecast ran warm or cold. A forecast that gives its own spread is also scored on its band: an honest band holds about 68 in 100 readings. Units are F, miles an hour and per cent of the sky.

The same figures, as data: /api/accuracy, and the developer page for every field.