MeteoSignal

Cabauw tower, the Netherlands.

AIFS 2 against IFS and GFS: lower error in 7 of 20 tested comparisonsSample, sources and status
Scored against
Measurements on KNMI's Cabauw tower
Sample
AIFS 2: 174 days, 23 May 2026 to 12 November 2026 for wind; temperature 172 days
Result so far, AIFS 2
Against IFS and GFS: AIFS's error lower in 7 of 20 tested comparisons, higher in 0, not separated in 13
Status
Provisional: no full year yet. A case study at one site, never pooled with the stations. Some months use measurements KNMI has not yet verified.

Last reviewed 16 November 2026Tower measurements to 12 November 2026How verdicts are reachedData access

Cabauw
The Cabauw tower, between Utrecht and Rotterdam.
Reference
Forecasts

AIFS 2, up to 174 days to 12 November 2026

Provisional. The Cabauw tower measurements for June 2026, September 2026 and November 2026 are KNMI's unvalidated records: KNMI has not verified those months yet. The unvalidated records carry 70 of the 172 days scored for 2 m temperature (41%) and 72 of the 174 days scored for 100 m wind (41%). The tower results may change once it publishes the validated data.

2 m temperature
-250255075▲ violet: lower error than climatologyDay 1Day 3Day 5Day 7Day 10Skill, %AIFSIFSGFS

Raw forecasts against climatology. 172 days, 23 May 2026 to 12 November 2026. Bars are 95% intervals.

100 m wind speed (estimate)
-50-2502550▲ violet: lower error than climatologyDay 1Day 3Day 5Day 7Day 10Skill, %AIFSIFSGFS

Raw forecasts against climatology. 174 days, 23 May 2026 to 12 November 2026. Bars are 95% intervals.

2 m temperature, AIFS 2: how AIFS's error compared, over 172 days
Compared withDay 1Day 3Day 5Day 7Day 10
IFS−12%, not separated−14%, not separated−16%, not separated−13%, not separated−6%, not separated
GFS−45%, AIFS lower error, tested−39%, AIFS lower error, tested−32%, AIFS lower error, tested−29%, not separated−16%, not separated
  • AIFS lower error, tested (better)
  • AIFS higher error, tested (worse)
  • Not separated: no clear difference, too few days, or the verdict depends on how the station is read or on the height correction (hover says which)

Each number is AIFS's RMSE minus the other model's, as a percentage of the other model's RMSE: negative means AIFS's was lower. Hover or tap a cell for the difference in °C, its 95% interval and the p-value. A tint and a mark (▼ lower, ▲ higher) appear only where the difference was tested and holds.

100 m wind speed (estimate), AIFS 2: how AIFS's error compared, over 174 days
Compared withDay 1Day 3Day 5Day 7Day 10
IFS−10%, not separated−5%, not separated−10%, not separated−8%, not separated−6%, not separated
GFS−22%, AIFS lower error, tested−23%, AIFS lower error, tested−22%, AIFS lower error, tested−15%, not separated−19%, AIFS lower error, tested
  • AIFS lower error, tested (better)
  • AIFS higher error, tested (worse)
  • Not separated: no clear difference, too few days, or the verdict depends on how the station is read or on the height correction (hover says which)

Each number is AIFS's RMSE minus the other model's, as a percentage of the other model's RMSE: negative means AIFS's was lower. Hover or tap a cell for the difference in m/s, its 95% interval and the p-value. A tint and a mark (▼ lower, ▲ higher) appear only where the difference was tested and holds.

Bias and activityWhether a model runs high or low, and how much it varies

Bold and underlined marks the value nearest the ideal at each lead, as printed. It is not a verdict: these values are not tested against each other. Strong-wind events are defined for 10 m wind only, so the tower has none for 100 m wind.

2 m temperature: bias, in °CForecast minus measurement, on average. Plus: too high. Minus: too low. Zero is best.
ModelDay 1Day 3Day 5Day 7Day 10
AIFS−0.02−0.03+0.12+0.12−0.09
IFS−0.01+0.09+0.23+0.15−0.08
GFS+0.43+0.51+0.28+0.38−0.11
2 m temperature: activityVariability of the forecast divided by that of the measurements, both as departures from climatology, with each station centred on its own mean at each valid hour first. 1.00 is best. Below 1 the forecast is too smooth, which flatters RMSE.
ModelDay 1Day 3Day 5Day 7Day 10
AIFS0.950.950.990.960.94
IFS0.900.910.960.900.94
GFS1.011.031.031.071.01

2 m temperature of 0 °C or belowObserved 0 times on 0 days in 344 station-times (0.0%). Too few events to score: at least 30, on at least 10 days, are needed.

100 m wind: bias, in m/sForecast minus measurement, on average. Plus: too high. Minus: too low. Zero is best.
ModelDay 1Day 3Day 5Day 7Day 10
AIFS−0.48−0.59−0.69−0.71−0.83
IFS−0.65−0.60−0.56−0.70−0.89
GFS+0.06−0.09+0.01−0.01−0.31
100 m wind: activityVariability of the forecast divided by that of the measurements, both as departures from climatology, with each station centred on its own mean at each valid hour first. 1.00 is best. Below 1 the forecast is too smooth, which flatters RMSE.
ModelDay 1Day 3Day 5Day 7Day 10
AIFS0.960.920.880.900.91
IFS0.910.900.910.890.82
GFS1.111.111.060.981.06
AIFS 1.1, the earlier version245 days, 7 September 2025 to 12 May 2026; other seasons, so not comparable with AIFS 2

Provisional. The Cabauw tower measurements for January 2026 are KNMI's unvalidated records: KNMI has not verified that month yet. The unvalidated records carry 31 of the 244 days scored for 2 m temperature (13%) and 31 of the 245 days scored for 100 m wind (13%). The tower results may change once it publishes the validated data.

2 m temperature
-250255075▲ violet: lower error than climatologyDay 1Day 3Day 5Day 7Day 10Skill, %AIFSGFSIFS

Raw forecasts against climatology. 244 days, 7 September 2025 to 12 May 2026. Bars are 95% intervals.

100 m wind speed (estimate)
-50050▲ violet: lower error than climatologyDay 1Day 3Day 5Day 7Day 10Skill, %AIFSIFSGFS

Raw forecasts against climatology. 245 days, 7 September 2025 to 12 May 2026. Bars are 95% intervals.

2 m temperature, AIFS 1.1: how AIFS's error compared, over 244 days
Compared withDay 1Day 3Day 5Day 7Day 10
IFS−11%, not separated−8%, not separated−6%, not separated−7%, not separated−10%, not separated
GFS−33%, AIFS lower error, tested−24%, AIFS lower error, tested−17%, AIFS lower error, tested−11%, not separated−6%, not separated
  • AIFS lower error, tested (better)
  • AIFS higher error, tested (worse)
  • Not separated: no clear difference, too few days, or the verdict depends on how the station is read or on the height correction (hover says which)

Each number is AIFS's RMSE minus the other model's, as a percentage of the other model's RMSE: negative means AIFS's was lower. Hover or tap a cell for the difference in °C, its 95% interval and the p-value. A tint and a mark (▼ lower, ▲ higher) appear only where the difference was tested and holds.

100 m wind speed (estimate), AIFS 1.1: how AIFS's error compared, over 245 days
Compared withDay 1Day 3Day 5Day 7Day 10
IFS−9%, not separated−4%, not separated−2%, not separated−7%, not separated−6%, not separated
GFS−15%, AIFS lower error, tested−18%, not separated−14%, not separated−14%, not separated−18%, AIFS lower error, tested
  • AIFS lower error, tested (better)
  • AIFS higher error, tested (worse)
  • Not separated: no clear difference, too few days, or the verdict depends on how the station is read or on the height correction (hover says which)

Each number is AIFS's RMSE minus the other model's, as a percentage of the other model's RMSE: negative means AIFS's was lower. Hover or tap a cell for the difference in m/s, its 95% interval and the p-value. A tint and a mark (▼ lower, ▲ higher) appear only where the difference was tested and holds.

Bias and activity, AIFS 1.1

Bold and underlined marks the value nearest the ideal at each lead, as printed. It is not a verdict: these values are not tested against each other. Strong-wind events are defined for 10 m wind only, so the tower has none for 100 m wind.

2 m temperature: bias, in °CForecast minus measurement, on average. Plus: too high. Minus: too low. Zero is best.
ModelDay 1Day 3Day 5Day 7Day 10
AIFS+0.23+0.16−0.01−0.03−0.48
IFS−0.01−0.07−0.26−0.34−0.74
GFS+0.32+0.21+0.28+0.26+0.09
2 m temperature: activityVariability of the forecast divided by that of the measurements, both as departures from climatology, with each station centred on its own mean at each valid hour first. 1.00 is best. Below 1 the forecast is too smooth, which flatters RMSE.
ModelDay 1Day 3Day 5Day 7Day 10
AIFS0.980.981.001.011.08
IFS0.960.981.001.021.03
GFS0.960.950.950.930.99

2 m temperature of 0 °C or belowObserved 21 times on 16 days in 488 station-times (4%). Too few events to score: at least 30, on at least 10 days, are needed.

100 m wind: bias, in m/sForecast minus measurement, on average. Plus: too high. Minus: too low. Zero is best.
ModelDay 1Day 3Day 5Day 7Day 10
AIFS−0.59−0.75−0.92−0.83−0.84
IFS−0.74−0.75−0.88−0.90−0.81
GFS−0.08−0.08−0.15−0.14+0.21
100 m wind: activityVariability of the forecast divided by that of the measurements, both as departures from climatology, with each station centred on its own mean at each valid hour first. 1.00 is best. Below 1 the forecast is too smooth, which flatters RMSE.
ModelDay 1Day 3Day 5Day 7Day 10
AIFS0.960.960.970.951.01
IFS0.900.890.910.900.99
GFS1.091.061.121.121.24

The 100 m wind is an estimate: a power law through the 10-minute measurements at 80 m and 140 m. KNMI validates the tower data month by month, some months late. For the months it has not validated yet, its unvalidated records from the same instruments are used and marked provisional. In May and August 2026, where both exist, the two agreed on every record they share when checked on 11 November 2026. The climatology uses validated months only. ICON is left out because its 100 m wind is derived from other heights. Temperature is the tower's own 2 m reading, scored at the same two times.