AIFS 2 against IFS and GFS: lower error in 24 of 28 tested comparisonsSample, sources and status
Scored against
Measurements from the national services of Germany, Austria, the Netherlands, Denmark, Switzerland and Finland
Sample
AIFS 2: 174 days, 23 May 2026 to 12 November 2026 for temperature and wind; solar energy 173 days
Result so far, AIFS 2
Against IFS and GFS: AIFS's error lower in 24 of 28 tested comparisons, higher in 0, not separated in 4
Status
Provisional: no full year yet. One pooled set with its own tests. A further 61 stations in 11 countries, from NOAA's coarser hourly reports, are scored as a set of their own.
The stations scored on this page. Countries read from their own weather service are shaded. Hover over or tap a point to see the site.
Reference
Forecasts
AIFS 2, up to 174 days to 12 November 2026
2 m temperature: skill against climatology
Raw forecasts, 174 days, 23 May 2026 to 12 November 2026. Bars are 95% intervals. The climatology is not adjusted for warming and ran 1.0 K too cold over these days, which raises every model's temperature skill against it.
2 m temperature: skill against persistence
Raw forecasts, 174 days, 23 May 2026 to 12 November 2026, on the 36,860 of 37,155 station-times where persistence can be scored as well. Bars are 95% intervals.
2 m temperature: skill against climatology
Bias-corrected forecasts, 144 days, 22 June 2026 to 12 November 2026. Bars are 95% intervals.
2 m temperature: skill against persistence
Bias-corrected forecasts, 144 days, 22 June 2026 to 12 November 2026, on the 30,442 of 30,737 station-times where persistence can be scored as well. Bars are 95% intervals.
AIFS's RMSE against each model, in %, over 174 days
Compared with
Day 1
Day 3
Day 5
Day 7
Day 10
IFS
▼−10%, AIFS lower error, tested
▼−12%, AIFS lower error, tested
▼−13%, AIFS lower error, tested
−9%, not separated
−9%, not separated
GFS
▼−28%, AIFS lower error, tested
▼−27%, AIFS lower error, tested
▼−24%, AIFS lower error, tested
▼−20%, AIFS lower error, tested
▼−14%, AIFS lower error, tested
AIFS's RMSE against each model, in %, over 144 days
Compared with
Day 1
Day 3
Day 5
Day 7
Day 10
IFS
−12%, not separated
−14%, not separated
−14%, not separated
−8%, not separated
−10%, not separated
GFS
−28%, not separated
−26%, not separated
−23%, not separated
−21%, not separated
−16%, not separated
10 m wind speed: skill against climatology
Raw forecasts, 174 days, 23 May 2026 to 12 November 2026. Bars are 95% intervals.
10 m wind speed: skill against persistence
Raw forecasts, 174 days, 23 May 2026 to 12 November 2026, on the 36,467 of 37,050 station-times where persistence can be scored as well. Bars are 95% intervals.
10 m wind speed: skill against climatology
Bias-corrected forecasts, 144 days, 22 June 2026 to 12 November 2026. Bars are 95% intervals.
10 m wind speed: skill against persistence
Bias-corrected forecasts, 144 days, 22 June 2026 to 12 November 2026, on the 30,010 of 30,583 station-times where persistence can be scored as well. Bars are 95% intervals.
AIFS's RMSE against each model, in %, over 174 days
Compared with
Day 1
Day 3
Day 5
Day 7
Day 10
IFS
▼−3%, AIFS lower error, tested
▼−5%, AIFS lower error, tested
▼−8%, AIFS lower error, tested
−6%, not separated
−1%, not separated
GFS
▼−15%, AIFS lower error, tested
▼−16%, AIFS lower error, tested
▼−16%, AIFS lower error, tested
▼−15%, AIFS lower error, tested
▼−8%, AIFS lower error, tested
AIFS's RMSE against each model, in %, over 144 days
Compared with
Day 1
Day 3
Day 5
Day 7
Day 10
IFS
−3%, not separated
−6%, not separated
−9%, not separated
−7%, not separated
−2%, not separated
GFS
−13%, not separated
−17%, not separated
−17%, not separated
−16%, not separated
−9%, not separated
Solar energy: skill against climatology
Raw forecasts, 173 days, 23 May 2026 to 12 November 2026. Bars are 95% intervals.
Solar energy: skill against persistence
Raw forecasts, 168 days, 23 May 2026 to 12 November 2026: those of the 173 days scored on which persistence can be scored as well. Bars are 95% intervals.
Solar energy: skill against climatology
Bias-corrected forecasts, 144 days, 22 June 2026 to 12 November 2026. Bars are 95% intervals.
Solar energy: skill against persistence
Bias-corrected forecasts, 140 days, 23 June 2026 to 12 November 2026: those of the 144 days scored on which persistence can be scored as well. Bars are 95% intervals.
AIFS's RMSE against each model, in %, over 173 days
Compared with
Day 1
Day 3
Day 5
Day 7
Day 10
IFS
▼−16%, AIFS lower error, tested
▼−17%, AIFS lower error, tested
▼−15%, AIFS lower error, tested
–
–
GFS
▼−23%, AIFS lower error, tested
▼−27%, AIFS lower error, tested
▼−24%, AIFS lower error, tested
▼−21%, AIFS lower error, tested
▼−14%, AIFS lower error, tested
ICON is not compared on solar: its forecasts come from Open-Meteo, which assembles the day differently.
AIFS's RMSE against each model, in %, over 144 days
Compared with
Day 1
Day 3
Day 5
Day 7
Day 10
IFS
−9%, not separated
−12%, not separated
−11%, not separated
–
–
GFS
−21%, not separated
−23%, not separated
−21%, not separated
−17%, not separated
−12%, not separated
ICON is not compared on solar: its forecasts come from Open-Meteo, which assembles the day differently.
▼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: minus means AIFS's was lower (better). Hover or tap a cell for the difference in units, its 95% interval and the p-value.
Bias, activity and eventsWhether a model runs high or low, how much it varies, and how it does in events
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. Hover an event score for its 95% interval.
2 m temperature: bias, in °CForecast minus measurement, on average. Plus: too high. Minus: too low. Zero is best.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
+0.06
+0.01
+0.06
+0.17
−0.20
IFS
+0.13
+0.17
+0.32
+0.44
+0.40
GFS
−0.04
−0.00
+0.04
+0.01
−0.19
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.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
0.95
0.94
0.95
0.94
0.93
IFS
0.95
0.94
0.97
0.97
0.99
GFS
0.94
0.95
0.96
0.96
0.98
2 m temperature of 0 °C or below: frequency biasObserved 121 times on 16 days in 37155 station-times (0.3%). Forecast events per observed event. 1.00 is best. Above 1 the model calls the event too often, below 1 too rarely.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
0.25
0.23
0.07
0.27
0.31
IFS
0.50
0.52
0.57
0.55
1.35
GFS
0.12
0.10
0.07
0.16
0.72
2 m temperature of 0 °C or below: equitable threat scoreHow well forecast and observed events coincide, beyond chance. 1 is perfect, 0 is chance, higher is better.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
0.14
0.15
0.02
0.05
0.05
IFS
0.17
0.15
0.19
0.10
0.18
GFS
0.07
0.06
0.00
0.04
0.03
2 m temperature: bias, in °CForecast minus measurement, on average. Plus: too high. Minus: too low. Zero is best.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
−0.03
−0.03
−0.07
−0.14
−0.07
IFS
+0.00
+0.01
+0.01
−0.02
−0.16
GFS
+0.17
+0.21
+0.25
+0.13
+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.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
0.93
0.91
0.93
0.92
0.93
IFS
0.94
0.94
0.99
1.01
0.99
GFS
0.95
0.98
0.98
1.00
0.99
2 m temperature of 0 °C or below: frequency biasObserved 121 times on 16 days in 30737 station-times (0.4%). Forecast events per observed event. 1.00 is best. Above 1 the model calls the event too often, below 1 too rarely.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
0.29
0.28
0.20
0.40
0.77
IFS
0.57
0.58
0.74
0.94
1.64
GFS
0.33
0.26
0.21
0.36
0.81
2 m temperature of 0 °C or below: equitable threat scoreHow well forecast and observed events coincide, beyond chance. 1 is perfect, 0 is chance, higher is better.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
0.19
0.20
0.10
0.09
0.18
IFS
0.27
0.22
0.26
0.22
0.20
GFS
0.18
0.16
0.06
0.12
0.08
10 m wind: bias, in m/sForecast minus measurement, on average. Plus: too high. Minus: too low. Zero is best.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
+0.06
−0.04
−0.11
−0.19
−0.11
IFS
+0.03
−0.02
−0.02
−0.07
−0.14
GFS
+0.42
+0.33
+0.33
+0.36
+0.29
10 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.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
0.98
0.94
0.92
0.91
0.96
IFS
0.97
0.96
0.95
0.94
0.93
GFS
1.07
1.04
1.01
1.04
1.04
10 m wind of 8 m/s or more: frequency biasObserved 1313 times on 138 days in 37050 station-times (4%). Forecast events per observed event. 1.00 is best. Above 1 the model calls the event too often, below 1 too rarely.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
0.80
0.70
0.59
0.55
0.72
IFS
0.58
0.58
0.55
0.48
0.49
GFS
1.50
1.33
1.24
1.42
1.27
10 m wind of 8 m/s or more: equitable threat scoreHow well forecast and observed events coincide, beyond chance. 1 is perfect, 0 is chance, higher is better.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
0.37
0.32
0.24
0.17
0.09
IFS
0.29
0.26
0.17
0.12
0.07
GFS
0.30
0.24
0.18
0.13
0.07
10 m wind of 10.8 m/s or more: frequency biasObserved 241 times on 57 days in 37050 station-times (0.7%). Forecast events per observed event. 1.00 is best. Above 1 the model calls the event too often, below 1 too rarely.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
0.64
0.60
0.41
0.39
0.60
IFS
0.54
0.49
0.44
0.38
0.37
GFS
1.43
1.25
1.05
1.18
1.16
10 m wind of 10.8 m/s or more: equitable threat scoreHow well forecast and observed events coincide, beyond chance. 1 is perfect, 0 is chance, higher is better.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
0.27
0.25
0.15
0.12
0.02
IFS
0.21
0.19
0.09
0.05
0.04
GFS
0.24
0.19
0.09
0.04
0.06
10 m wind: bias, in m/sForecast minus measurement, on average. Plus: too high. Minus: too low. Zero is best.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
+0.02
+0.03
+0.03
+0.05
+0.00
IFS
+0.01
+0.01
−0.01
+0.02
+0.00
GFS
+0.01
−0.00
0.00
+0.04
+0.02
10 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.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
0.98
0.95
0.93
0.93
0.95
IFS
0.97
0.95
0.94
0.95
0.90
GFS
1.07
1.04
1.01
1.06
1.00
10 m wind of 8 m/s or more: frequency biasObserved 1109 times on 112 days in 30583 station-times (4%). Forecast events per observed event. 1.00 is best. Above 1 the model calls the event too often, below 1 too rarely.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
0.93
0.86
0.85
0.90
0.93
IFS
0.81
0.82
0.78
0.77
0.74
GFS
1.12
1.08
0.95
1.11
1.13
10 m wind of 8 m/s or more: equitable threat scoreHow well forecast and observed events coincide, beyond chance. 1 is perfect, 0 is chance, higher is better.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
0.46
0.40
0.33
0.24
0.12
IFS
0.42
0.37
0.26
0.18
0.10
GFS
0.41
0.33
0.23
0.16
0.10
10 m wind of 10.8 m/s or more: frequency biasObserved 213 times on 50 days in 30583 station-times (0.7%). Forecast events per observed event. 1.00 is best. Above 1 the model calls the event too often, below 1 too rarely.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
0.61
0.60
0.46
0.60
0.69
IFS
0.49
0.49
0.47
0.40
0.44
GFS
0.99
0.94
0.78
0.81
0.95
10 m wind of 10.8 m/s or more: equitable threat scoreHow well forecast and observed events coincide, beyond chance. 1 is perfect, 0 is chance, higher is better.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
0.36
0.31
0.19
0.13
0.04
IFS
0.30
0.26
0.14
0.07
0.04
GFS
0.33
0.26
0.12
0.07
0.07
Daily solar energy: bias, in kWh/m²/dayForecast minus measurement, on average. Plus: too high. Minus: too low. Zero is best.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
+0.15
+0.12
+0.10
+0.09
+0.15
IFS
+0.34
+0.37
+0.37
–
–
GFS
+0.22
+0.28
+0.30
+0.32
+0.17
Daily solar energy: 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.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
0.77
0.76
0.76
0.71
0.65
IFS
0.80
0.77
0.77
–
–
GFS
1.00
0.99
0.94
0.90
0.89
Daily solar energy: bias, in kWh/m²/dayForecast minus measurement, on average. Plus: too high. Minus: too low. Zero is best.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
−0.01
−0.00
−0.02
−0.04
−0.01
IFS
−0.03
−0.05
−0.06
–
–
GFS
−0.00
−0.02
−0.04
−0.07
−0.04
Daily solar energy: 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.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
0.75
0.74
0.76
0.73
0.66
IFS
0.77
0.73
0.75
–
–
GFS
0.98
0.96
0.92
0.89
0.89
AIFS 1.1, the earlier version248 days, 7 September 2025 to 12 May 2026; other seasons, so not comparable with AIFS 2
2 m temperature: skill against climatology
Raw forecasts, 248 days, 7 September 2025 to 12 May 2026. Bars are 95% intervals.
2 m temperature: skill against persistence
Raw forecasts, 248 days, 7 September 2025 to 12 May 2026, on the 52,333 of 52,914 station-times where persistence can be scored as well. Bars are 95% intervals.
2 m temperature: skill against climatology
Bias-corrected forecasts, 218 days, 7 October 2025 to 12 May 2026. Bars are 95% intervals.
2 m temperature: skill against persistence
Bias-corrected forecasts, 218 days, 7 October 2025 to 12 May 2026, on the 45,907 of 46,476 station-times where persistence can be scored as well. Bars are 95% intervals.
AIFS's RMSE against each model, in %, over 248 days
Compared with
Day 1
Day 3
Day 5
Day 7
Day 10
IFS
▼−12%, AIFS lower error, tested
▼−13%, AIFS lower error, tested
▼−12%, AIFS lower error, tested
−11%, not separated
−7%, not separated
GFS
▼−31%, AIFS lower error, tested
▼−28%, AIFS lower error, tested
▼−23%, AIFS lower error, tested
▼−14%, AIFS lower error, tested
−9%, not separated
AIFS's RMSE against each model, in %, over 218 days
Compared with
Day 1
Day 3
Day 5
Day 7
Day 10
IFS
▼−13%, AIFS lower error, tested
▼−15%, AIFS lower error, tested
▼−12%, AIFS lower error, tested
−10%, not separated
−6%, not separated
GFS
▼−32%, AIFS lower error, tested
▼−28%, AIFS lower error, tested
▼−23%, AIFS lower error, tested
▼−13%, AIFS lower error, tested
−6%, not separated
10 m wind speed: skill against climatology
Raw forecasts, 248 days, 7 September 2025 to 12 May 2026. Bars are 95% intervals.
10 m wind speed: skill against persistence
Raw forecasts, 248 days, 7 September 2025 to 12 May 2026, on the 52,291 of 52,898 station-times where persistence can be scored as well. Bars are 95% intervals.
10 m wind speed: skill against climatology
Bias-corrected forecasts, 218 days, 7 October 2025 to 12 May 2026. Bars are 95% intervals.
10 m wind speed: skill against persistence
Bias-corrected forecasts, 218 days, 7 October 2025 to 12 May 2026, on the 45,791 of 46,459 station-times where persistence can be scored as well. Bars are 95% intervals.
AIFS's RMSE against each model, in %, over 248 days
Compared with
Day 1
Day 3
Day 5
Day 7
Day 10
IFS
▼−3%, AIFS lower error, tested
▼−5%, AIFS lower error, tested
▼−7%, AIFS lower error, tested
−3%, not separated
−6%, not separated
GFS
▼−14%, AIFS lower error, tested
▼−15%, AIFS lower error, tested
▼−16%, AIFS lower error, tested
▼−12%, AIFS lower error, tested
▼−16%, AIFS lower error, tested
AIFS's RMSE against each model, in %, over 218 days
Compared with
Day 1
Day 3
Day 5
Day 7
Day 10
IFS
▼−3%, AIFS lower error, tested
▼−6%, AIFS lower error, tested
▼−7%, AIFS lower error, tested
−3%, not separated
−6%, not separated
GFS
▼−13%, AIFS lower error, tested
▼−17%, AIFS lower error, tested
▼−17%, AIFS lower error, tested
▼−13%, AIFS lower error, tested
▼−17%, AIFS lower error, tested
Solar energy: skill against climatology
Raw forecasts, 248 days, 7 September 2025 to 12 May 2026. Bars are 95% intervals.
Solar energy: skill against persistence
Raw forecasts, 245 days, 7 September 2025 to 12 May 2026: those of the 248 days scored on which persistence can be scored as well. Bars are 95% intervals.
Solar energy: skill against climatology
Bias-corrected forecasts, 217 days, 8 October 2025 to 12 May 2026. Bars are 95% intervals.
Solar energy: skill against persistence
Bias-corrected forecasts, 212 days, 9 October 2025 to 12 May 2026: those of the 217 days scored on which persistence can be scored as well. Bars are 95% intervals.
AIFS's RMSE against each model, in %, over 248 days
Compared with
Day 1
Day 3
Day 5
Day 7
Day 10
IFS
▼−9%, AIFS lower error, tested
−10%, not separated
▼−11%, AIFS lower error, tested
–
–
GFS
▼−26%, AIFS lower error, tested
▼−28%, AIFS lower error, tested
▼−24%, AIFS lower error, tested
−15%, not separated
▼−19%, AIFS lower error, tested
ICON is not compared on solar: its forecasts come from Open-Meteo, which assembles the day differently.
AIFS's RMSE against each model, in %, over 217 days
Compared with
Day 1
Day 3
Day 5
Day 7
Day 10
IFS
−4%, not separated
−8%, not separated
−9%, not separated
–
–
GFS
▼−25%, AIFS lower error, tested
▼−28%, AIFS lower error, tested
▼−24%, AIFS lower error, tested
−16%, not separated
▼−18%, AIFS lower error, tested
ICON is not compared on solar: its forecasts come from Open-Meteo, which assembles the day differently.
▼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: minus means AIFS's was lower (better). Hover or tap a cell for the difference in units, its 95% interval and the p-value.
Bias, activity and events, 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. Hover an event score for its 95% interval.
2 m temperature: bias, in °CForecast minus measurement, on average. Plus: too high. Minus: too low. Zero is best.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
+0.14
+0.13
+0.11
+0.17
−0.06
IFS
−0.01
−0.04
−0.14
−0.20
−0.38
GFS
+0.30
+0.19
+0.14
+0.21
+0.20
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.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
0.95
0.94
0.95
0.95
1.00
IFS
0.93
0.94
0.95
0.99
0.99
GFS
0.93
0.92
0.94
0.93
0.98
2 m temperature of 0 °C or below: frequency biasObserved 6730 times on 160 days in 52914 station-times (13%). Forecast events per observed event. 1.00 is best. Above 1 the model calls the event too often, below 1 too rarely.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
0.90
0.92
0.97
0.91
0.97
IFS
0.95
0.96
1.03
1.01
1.07
GFS
0.78
0.77
0.83
0.81
0.80
2 m temperature of 0 °C or below: equitable threat scoreHow well forecast and observed events coincide, beyond chance. 1 is perfect, 0 is chance, higher is better.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
0.71
0.67
0.58
0.44
0.33
IFS
0.66
0.60
0.53
0.41
0.30
GFS
0.60
0.54
0.47
0.35
0.25
2 m temperature: bias, in °CForecast minus measurement, on average. Plus: too high. Minus: too low. Zero is best.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
0.00
−0.00
−0.02
−0.05
−0.07
IFS
−0.00
−0.01
−0.04
−0.10
−0.20
GFS
−0.06
−0.09
−0.12
−0.11
−0.29
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.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
0.94
0.93
0.94
0.93
0.97
IFS
0.94
0.95
0.97
1.01
0.99
GFS
0.95
0.93
0.95
0.94
0.99
2 m temperature of 0 °C or below: frequency biasObserved 6714 times on 157 days in 46476 station-times (14%). Forecast events per observed event. 1.00 is best. Above 1 the model calls the event too often, below 1 too rarely.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
0.96
0.96
1.02
1.02
1.11
IFS
0.94
0.95
0.99
0.94
1.09
GFS
1.00
0.99
1.01
1.01
1.06
2 m temperature of 0 °C or below: equitable threat scoreHow well forecast and observed events coincide, beyond chance. 1 is perfect, 0 is chance, higher is better.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
0.71
0.65
0.57
0.43
0.34
IFS
0.66
0.59
0.51
0.39
0.31
GFS
0.59
0.54
0.46
0.35
0.28
10 m wind: bias, in m/sForecast minus measurement, on average. Plus: too high. Minus: too low. Zero is best.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
−0.03
−0.11
−0.21
−0.23
−0.18
IFS
−0.04
−0.07
−0.14
−0.11
−0.08
GFS
+0.34
+0.28
+0.26
+0.32
+0.50
10 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.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
0.91
0.90
0.87
0.88
0.90
IFS
0.93
0.92
0.91
0.91
0.95
GFS
1.05
1.02
1.03
1.04
1.13
10 m wind of 8 m/s or more: frequency biasObserved 3226 times on 222 days in 52898 station-times (6%). Forecast events per observed event. 1.00 is best. Above 1 the model calls the event too often, below 1 too rarely.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
0.78
0.70
0.62
0.64
0.69
IFS
0.59
0.56
0.53
0.54
0.60
GFS
1.32
1.17
1.23
1.27
1.59
10 m wind of 8 m/s or more: equitable threat scoreHow well forecast and observed events coincide, beyond chance. 1 is perfect, 0 is chance, higher is better.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
0.40
0.33
0.23
0.15
0.08
IFS
0.32
0.27
0.19
0.11
0.06
GFS
0.35
0.29
0.21
0.13
0.06
10 m wind of 10.8 m/s or more: frequency biasObserved 597 times on 124 days in 52898 station-times (1%). Forecast events per observed event. 1.00 is best. Above 1 the model calls the event too often, below 1 too rarely.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
0.70
0.64
0.52
0.48
0.61
IFS
0.47
0.47
0.46
0.48
0.53
GFS
1.27
1.11
1.18
1.25
1.95
10 m wind of 10.8 m/s or more: equitable threat scoreHow well forecast and observed events coincide, beyond chance. 1 is perfect, 0 is chance, higher is better.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
0.29
0.25
0.16
0.08
0.03
IFS
0.21
0.18
0.13
0.07
0.03
GFS
0.28
0.20
0.16
0.08
0.04
10 m wind: bias, in m/sForecast minus measurement, on average. Plus: too high. Minus: too low. Zero is best.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
−0.02
−0.03
−0.04
−0.04
+0.01
IFS
−0.01
−0.00
+0.01
+0.03
+0.01
GFS
−0.01
+0.01
−0.02
+0.06
+0.05
10 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.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
0.89
0.87
0.84
0.88
0.90
IFS
0.91
0.90
0.90
0.91
0.93
GFS
1.03
1.01
1.02
1.06
1.13
10 m wind of 8 m/s or more: frequency biasObserved 2731 times on 198 days in 46459 station-times (6%). Forecast events per observed event. 1.00 is best. Above 1 the model calls the event too often, below 1 too rarely.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
0.73
0.68
0.66
0.76
0.80
IFS
0.71
0.72
0.72
0.80
0.83
GFS
1.02
0.97
1.00
1.17
1.24
10 m wind of 8 m/s or more: equitable threat scoreHow well forecast and observed events coincide, beyond chance. 1 is perfect, 0 is chance, higher is better.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
0.46
0.36
0.25
0.16
0.09
IFS
0.43
0.35
0.25
0.15
0.08
GFS
0.43
0.34
0.21
0.14
0.05
10 m wind of 10.8 m/s or more: frequency biasObserved 479 times on 112 days in 46459 station-times (1%). Forecast events per observed event. 1.00 is best. Above 1 the model calls the event too often, below 1 too rarely.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
0.45
0.39
0.39
0.62
0.62
IFS
0.41
0.44
0.52
0.62
0.75
GFS
0.85
0.81
0.81
1.17
1.52
10 m wind of 10.8 m/s or more: equitable threat scoreHow well forecast and observed events coincide, beyond chance. 1 is perfect, 0 is chance, higher is better.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
0.30
0.23
0.13
0.08
0.03
IFS
0.26
0.20
0.17
0.07
0.02
GFS
0.35
0.22
0.14
0.10
0.02
Daily solar energy: bias, in kWh/m²/dayForecast minus measurement, on average. Plus: too high. Minus: too low. Zero is best.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
+0.05
+0.05
+0.02
−0.01
−0.00
IFS
+0.18
+0.18
+0.15
–
–
GFS
+0.11
+0.12
+0.13
+0.10
−0.01
Daily solar energy: 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.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
0.82
0.79
0.77
0.74
0.62
IFS
0.85
0.85
0.82
–
–
GFS
1.03
0.99
0.96
0.91
0.88
Daily solar energy: bias, in kWh/m²/dayForecast minus measurement, on average. Plus: too high. Minus: too low. Zero is best.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
−0.01
−0.01
−0.01
−0.06
−0.08
IFS
−0.01
−0.01
−0.03
–
–
GFS
+0.00
−0.02
−0.05
−0.06
−0.11
Daily solar energy: 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.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
0.81
0.77
0.76
0.74
0.62
IFS
0.86
0.84
0.81
–
–
GFS
1.02
0.99
0.94
0.90
0.86
Every station read from a national weather service, pooled into one set: Austria (31), Germany (30), Netherlands (27), Denmark (13), Switzerland (5) and Finland (1). The German stations are part of this set and also have a page of their own.
Against ICON, up to 7 November 2026AIFS 2; a closed set that no longer grows
ICON's only archive of past runs was Open-Meteo's. Its last valid day read here is 7 November 2026, and the archive is no longer read. It is compared with AIFS here on the days it has, as a set of its own that no longer grows, while the other sets move forward each week. It returns as a new series from DWD's own files once enough runs have been collected.
2 m temperature: skill against climatology
Raw forecasts, 170 days, 23 May 2026 to 8 November 2026. This sample no longer grows.
2 m temperature: skill against persistence
Raw forecasts, 170 days, 23 May 2026 to 8 November 2026, on the 36,018 of 36,305 station-times where persistence can be scored as well. This sample no longer grows.
2 m temperature: skill against climatology
Bias-corrected forecasts, 140 days, 22 June 2026 to 8 November 2026. This sample no longer grows.
2 m temperature: skill against persistence
Bias-corrected forecasts, 140 days, 22 June 2026 to 8 November 2026, on the 29,600 of 29,887 station-times where persistence can be scored as well. This sample no longer grows.
2 m temperature, AIFS 2: how AIFS's error compared, over 170 days
Compared with
Day 1
Day 3
Day 5
Day 7
Day 10
ICON
▼−7%, AIFS lower error, tested
▼−13%, AIFS lower error, tested
▼−15%, AIFS lower error, tested
–
–
2 m temperature, AIFS 2: how AIFS's error compared, over 140 days
Compared with
Day 1
Day 3
Day 5
Day 7
Day 10
ICON
−10%, not separated
−14%, not separated
−16%, not separated
–
–
10 m wind speed: skill against climatology
Raw forecasts, 170 days, 23 May 2026 to 8 November 2026. This sample no longer grows.
10 m wind speed: skill against persistence
Raw forecasts, 170 days, 23 May 2026 to 8 November 2026, on the 35,644 of 36,214 station-times where persistence can be scored as well. This sample no longer grows.
10 m wind speed: skill against climatology
Bias-corrected forecasts, 140 days, 22 June 2026 to 8 November 2026. This sample no longer grows.
10 m wind speed: skill against persistence
Bias-corrected forecasts, 140 days, 22 June 2026 to 8 November 2026, on the 29,187 of 29,747 station-times where persistence can be scored as well. This sample no longer grows.
10 m wind speed, AIFS 2: how AIFS's error compared, over 170 days
Compared with
Day 1
Day 3
Day 5
Day 7
Day 10
ICON
+3%, not separated
−3%, not separated
▼−10%, AIFS lower error, tested
–
–
10 m wind speed, AIFS 2: how AIFS's error compared, over 140 days
Compared with
Day 1
Day 3
Day 5
Day 7
Day 10
ICON
−1%, not separated
−8%, not separated
−14%, not separated
–
–
Solar energy: skill against climatology
Raw forecasts, 168 days, 23 May 2026 to 7 November 2026. This sample no longer grows.
Solar energy: skill against persistence
Raw forecasts, 163 days, 23 May 2026 to 7 November 2026: those of the 168 days scored on which persistence can be scored as well. This sample no longer grows.
Solar energy: skill against climatology
Bias-corrected forecasts, 139 days, 22 June 2026 to 7 November 2026. This sample no longer grows.
Solar energy: skill against persistence
Bias-corrected forecasts, 135 days, 23 June 2026 to 7 November 2026: those of the 139 days scored on which persistence can be scored as well. This sample no longer grows.
▼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)
2 m temperature: bias, in °CForecast minus measurement, on average. Plus: too high. Minus: too low. Zero is best.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
+0.06
+0.01
+0.07
+0.17
−0.21
ICON
+0.11
+0.07
+0.12
–
–
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.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
0.95
0.94
0.95
0.94
0.93
ICON
0.94
0.94
0.96
–
–
2 m temperature of 0 °C or below: frequency biasObserved 82 times on 13 days in 36305 station-times (0.2%). Forecast events per observed event. 1.00 is best. Above 1 the model calls the event too often, below 1 too rarely.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
0.32
0.32
0.10
0.21
0.39
ICON
0.23
0.12
0.18
–
–
2 m temperature of 0 °C or below: equitable threat scoreHow well forecast and observed events coincide, beyond chance. 1 is perfect, 0 is chance, higher is better.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
0.17
0.20
0.03
0.05
0.04
ICON
0.12
0.03
0.02
–
–
2 m temperature: bias, in °CForecast minus measurement, on average. Plus: too high. Minus: too low. Zero is best.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
−0.04
−0.03
−0.06
−0.15
−0.07
ICON
+0.01
+0.02
+0.01
–
–
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.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
0.93
0.91
0.93
0.92
0.94
ICON
0.94
0.93
0.97
–
–
2 m temperature of 0 °C or below: frequency biasObserved 82 times on 13 days in 29887 station-times (0.3%). Forecast events per observed event. 1.00 is best. Above 1 the model calls the event too often, below 1 too rarely.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
0.35
0.32
0.26
0.39
0.77
ICON
0.37
0.18
0.35
–
–
2 m temperature of 0 °C or below: equitable threat scoreHow well forecast and observed events coincide, beyond chance. 1 is perfect, 0 is chance, higher is better.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
0.23
0.23
0.13
0.11
0.12
ICON
0.27
0.10
0.10
–
–
10 m wind: bias, in m/sForecast minus measurement, on average. Plus: too high. Minus: too low. Zero is best.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
+0.06
−0.04
−0.12
−0.20
−0.11
ICON
−0.19
−0.18
−0.17
–
–
10 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.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
0.98
0.94
0.92
0.91
0.96
ICON
0.93
0.93
0.93
–
–
10 m wind of 8 m/s or more: frequency biasObserved 1301 times on 135 days in 36214 station-times (4%). Forecast events per observed event. 1.00 is best. Above 1 the model calls the event too often, below 1 too rarely.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
0.80
0.70
0.59
0.56
0.72
ICON
0.81
0.83
0.85
–
–
10 m wind of 8 m/s or more: equitable threat scoreHow well forecast and observed events coincide, beyond chance. 1 is perfect, 0 is chance, higher is better.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
0.37
0.32
0.24
0.17
0.09
ICON
0.38
0.33
0.23
–
–
10 m wind of 10.8 m/s or more: frequency biasObserved 235 times on 56 days in 36214 station-times (0.7%). Forecast events per observed event. 1.00 is best. Above 1 the model calls the event too often, below 1 too rarely.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
0.64
0.61
0.42
0.40
0.62
ICON
0.84
0.86
0.81
–
–
10 m wind of 10.8 m/s or more: equitable threat scoreHow well forecast and observed events coincide, beyond chance. 1 is perfect, 0 is chance, higher is better.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
0.27
0.25
0.15
0.12
0.02
ICON
0.32
0.29
0.17
–
–
10 m wind: bias, in m/sForecast minus measurement, on average. Plus: too high. Minus: too low. Zero is best.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
+0.02
+0.03
+0.03
+0.04
+0.00
ICON
+0.01
−0.00
+0.00
–
–
10 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.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
0.98
0.95
0.93
0.93
0.95
ICON
0.93
0.93
0.93
–
–
10 m wind of 8 m/s or more: frequency biasObserved 1097 times on 109 days in 29747 station-times (4%). Forecast events per observed event. 1.00 is best. Above 1 the model calls the event too often, below 1 too rarely.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
0.93
0.87
0.85
0.91
0.93
ICON
0.86
0.83
0.88
–
–
10 m wind of 8 m/s or more: equitable threat scoreHow well forecast and observed events coincide, beyond chance. 1 is perfect, 0 is chance, higher is better.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
0.46
0.40
0.33
0.24
0.12
ICON
0.44
0.37
0.26
–
–
10 m wind of 10.8 m/s or more: frequency biasObserved 207 times on 49 days in 29747 station-times (0.7%). Forecast events per observed event. 1.00 is best. Above 1 the model calls the event too often, below 1 too rarely.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
0.62
0.61
0.46
0.62
0.70
ICON
0.76
0.79
0.78
–
–
10 m wind of 10.8 m/s or more: equitable threat scoreHow well forecast and observed events coincide, beyond chance. 1 is perfect, 0 is chance, higher is better.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
0.37
0.32
0.19
0.13
0.04
ICON
0.36
0.31
0.18
–
–
Daily solar energy: bias, in kWh/m²/dayForecast minus measurement, on average. Plus: too high. Minus: too low. Zero is best.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
+0.15
+0.12
+0.10
+0.09
+0.15
ICON
−0.14
−0.15
–
–
–
Daily solar energy: 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.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
0.77
0.76
0.75
0.70
0.65
ICON
0.91
0.91
–
–
–
Daily solar energy: bias, in kWh/m²/dayForecast minus measurement, on average. Plus: too high. Minus: too low. Zero is best.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
−0.01
−0.00
−0.02
−0.04
−0.01
ICON
+0.04
+0.04
–
–
–
Daily solar energy: 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.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
0.75
0.74
0.76
0.73
0.66
ICON
0.91
0.92
–
–
–
Country by countryAIFS 2; a check that one country does not carry the pooled figure
A check that the pooled figure is not carried by one country. Countries with fewer than 10 stations are in the pooled set only. These are estimates without verdicts: AIFS's raw RMSE, and beside it how it compares with IFS's at the same stations and times. The differences were not tested, so their colour shows the sign only.
2 m temperature, AIFS 2, raw forecasts: AIFS's RMSE in °C and its difference from IFS's, 174 days to 12 November 2026
Country
Day 1
Day 3
Day 5
Day 7
Day 10
Austria (31)
1.52−12% (lower than IFS, better)
1.78−12% (lower than IFS, better)
2.32−9% (lower than IFS, better)
3.18−6% (lower than IFS, better)
4.17−9% (lower than IFS, better)
Germany (30)
1.38−10% (lower than IFS, better)
1.60−13% (lower than IFS, better)
2.13−13% (lower than IFS, better)
2.91−9% (lower than IFS, better)
3.93−11% (lower than IFS, better)
Netherlands (27)
1.26−12% (lower than IFS, better)
1.51−14% (lower than IFS, better)
2.02−18% (lower than IFS, better)
2.67−13% (lower than IFS, better)
3.67−8% (lower than IFS, better)
Denmark (13)
1.28−3% (lower than IFS, better)
1.46−5% (lower than IFS, better)
1.72−10% (lower than IFS, better)
2.21−10% (lower than IFS, better)
3.02−10% (lower than IFS, better)
10 m wind speed, AIFS 2, raw forecasts: AIFS's RMSE in m/s and its difference from IFS's, 174 days to 12 November 2026
Country
Day 1
Day 3
Day 5
Day 7
Day 10
Austria (31)
1.25−6% (lower than IFS, better)
1.30−7% (lower than IFS, better)
1.49−7% (lower than IFS, better)
1.70−7% (lower than IFS, better)
1.92−4% (lower than IFS, better)
Germany (30)
1.20−1% (lower than IFS, better)
1.31−3% (lower than IFS, better)
1.50−7% (lower than IFS, better)
1.80−5% (lower than IFS, better)
2.190% (equal to IFS)
Netherlands (27)
1.23−8% (lower than IFS, better)
1.36−10% (lower than IFS, better)
1.62−12% (lower than IFS, better)
1.99−7% (lower than IFS, better)
2.42−5% (lower than IFS, better)
Denmark (13)
1.69+2% (higher than IFS, worse)
1.78−2% (lower than IFS, better)
2.00−8% (lower than IFS, better)
2.39−3% (lower than IFS, better)
3.04+11% (higher than IFS, worse)
−8% AIFS's RMSE 8% lower than IFS's: better. +8% 8% higher: worse. Not tested, so violet and yellow-brown show the sign only, not the green or red of a verdict.
How the stations were chosenThe rules, and how well the model grids fit the stations
The rules follow the German ones, with two differences outside Germany: one station per square of 0.25° by 0.375°, against 0.5° by 0.75° in Germany, and a nominal anemometer height of 10 m where a service publishes none (64 of 77 stations). The lists were fixed without reading any forecast value; the full rules and later removals are in the station files in the repository.
The height test was run at selection with the grid heights Open-Meteo reported. Recomputed from each model's own surface height for AIFS, IFS and GFS: at the nearest land point the largest gap is 223 m and six stations exceed 100 m for at least one model; under interpolation it is 229 m and 12 stations exceed 100 m for at least one model. The station list was not changed. Temperature is moved to the station's height before scoring, which removes most of the effect of such a gap but not all of it.
The 107 stations, by countryName, height and position
The stations of this set: name, country, height and position
Station
Country
Height
Lat, lon
Litschau
Austria
558 m
48.95, 15.04
Raabs an der Thaya
Austria
459 m
48.85, 15.50
Retz
Austria
320 m
48.76, 15.94
Laa an der Thaya Kläranlage
Austria
184 m
48.73, 16.39
Allentsteig
Austria
599 m
48.69, 15.37
Poysdorf Ost
Austria
198 m
48.67, 16.64
Schöngrabern
Austria
253 m
48.60, 16.06
Gars am Kamp
Austria
273 m
48.59, 15.70
Reichenau im Mühlkreis
Austria
689 m
48.46, 14.35
Zwerndorf
Austria
144 m
48.34, 16.83
Langenlebarn
Austria
175 m
48.32, 16.12
Wien Stammersdorf
Austria
191 m
48.31, 16.41
Wien Hohe Warte
Austria
198 m
48.25, 16.36
Ranshofen Braunau
Austria
382 m
48.22, 13.03
Ried im Innkreis
Austria
427 m
48.22, 13.47
Enns
Austria
317 m
48.20, 14.45
Wels-Schleißheim
Austria
309 m
48.16, 14.07
Ostermiething
Austria
412 m
48.05, 12.83
Lilienfeld-Tarschberg
Austria
696 m
48.03, 15.59
Bruckneudorf
Austria
166 m
48.01, 16.84
Mattsee
Austria
502 m
47.98, 13.11
Seibersdorf
Austria
185 m
47.98, 16.50
Wiener Neustadt Flugplatz
Austria
275 m
47.83, 16.23
Mattersburg
Austria
284 m
47.74, 16.39
Lutzmannsburg
Austria
201 m
47.47, 16.65
Gleisdorf
Austria
377 m
47.12, 15.71
Güssing
Austria
215 m
47.06, 16.32
Graz Straßgang
Austria
357 m
47.05, 15.41
Fürstenfeld
Austria
271 m
47.03, 16.08
Feldbach
Austria
323 m
46.95, 15.88
Leibnitz-Wagna
Austria
267 m
46.77, 15.55
Sankt Peter-Ording
Germany
5 m
54.33, 8.60
Dörnick
Germany
27 m
54.17, 10.35
Ueckermünde
Germany
1 m
53.74, 14.07
Norderney
Germany
12 m
53.71, 7.15
Hamburg-Fuhlsbüttel
Germany
11 m
53.63, 9.99
Bremen
Germany
4 m
53.05, 8.80
Diepholz
Germany
38 m
52.59, 8.34
Braunschweig
Germany
81 m
52.29, 10.45
Lindenberg
Germany
98 m
52.21, 14.12
Lügde-Paenbruch
Germany
258 m
51.87, 9.27
Leinefelde
Germany
356 m
51.39, 10.31
Görlitz
Germany
239 m
51.16, 14.95
Dresden-Klotzsche
Germany
228 m
51.13, 13.75
Köln/Bonn
Germany
91 m
50.86, 7.16
Aachen-Orsbach
Germany
231 m
50.80, 6.02
Gießen/Wettenberg
Germany
203 m
50.60, 8.64
Nürburg-Barweiler
Germany
485 m
50.36, 6.87
Lautertal-Oberlauter
Germany
344 m
50.31, 10.97
Würzburg
Germany
268 m
49.77, 9.96
Mannheim
Germany
98 m
49.51, 8.56
Nürnberg
Germany
314 m
49.50, 11.05
Waldmünchen
Germany
499 m
49.39, 12.68
Saarbrücken-Ensheim
Germany
319 m
49.21, 7.11
Stuttgart (Schnarrenberg)
Germany
314 m
48.83, 9.20
Fürstenzell
Germany
476 m
48.55, 13.35
Augsburg
Germany
462 m
48.43, 10.94
Weihenstephan-Dürnast
Germany
477 m
48.40, 11.69
Mühldorf
Germany
406 m
48.28, 12.50
Chieming
Germany
551 m
47.88, 12.54
Konstanz
Germany
428 m
47.70, 9.13
Lauwersoog
Netherlands
3 m
53.41, 6.20
Hoorn Terschelling
Netherlands
1 m
53.39, 5.35
Leeuwarden Airport
Netherlands
1 m
53.22, 5.75
Nieuw Beerta
Netherlands
0 m
53.20, 7.15
Groningen Airport Eelde
Netherlands
5 m
53.13, 6.58
De Kooy Airport
Netherlands
1 m
52.93, 4.78
Stavoren
Netherlands
-1 m
52.90, 5.38
Hoogeveen
Netherlands
16 m
52.75, 6.57
Marknesse
Netherlands
-3 m
52.70, 5.89
Berkhout
Netherlands
-2 m
52.64, 4.98
Lelystad Airport
Netherlands
-4 m
52.46, 5.52
Heino
Netherlands
4 m
52.44, 6.26
Schiphol Airport
Netherlands
-3 m
52.32, 4.79
Twenthe Airport
Netherlands
35 m
52.27, 6.89
Voorschoten
Netherlands
-1 m
52.14, 4.44
De Bilt
Netherlands
2 m
52.10, 5.18
Hupsel
Netherlands
29 m
52.07, 6.66
Deelen Airport
Netherlands
48 m
52.06, 5.87
Cabauw
Netherlands
-1 m
51.97, 4.93
Rotterdam Airport
Netherlands
-4 m
51.96, 4.45
Herwijnen
Netherlands
1 m
51.86, 5.15
Volkel
Netherlands
22 m
51.66, 5.71
Gilze-Rijen Airport
Netherlands
15 m
51.57, 4.94
Eindhoven Airport
Netherlands
23 m
51.45, 5.38
Vlissingen
Netherlands
8 m
51.44, 3.60
Westdorpe
Netherlands
2 m
51.23, 3.86
Maastricht Airport
Netherlands
114 m
50.91, 5.76
Tylstrup
Denmark
13 m
57.19, 9.95
Mejrup
Denmark
53 m
56.38, 8.67
Isenvad
Denmark
60 m
56.09, 9.18
Hvide Sande
Denmark
2 m
56.01, 8.14
Borris
Denmark
25 m
55.96, 8.62
Askov
Denmark
62 m
55.47, 9.11
Årslev
Denmark
49 m
55.31, 10.44
Tystofte
Denmark
12 m
55.25, 11.33
Assens/Torø
Denmark
2 m
55.24, 9.89
Nexø Vest
Denmark
23 m
55.06, 15.10
Store Jyndevad
Denmark
16 m
54.90, 9.13
Abed
Denmark
7 m
54.83, 11.33
Gedser
Denmark
3 m
54.57, 11.94
Bischofszell / Sitterdorf
Switzerland
507 m
47.51, 9.27
Aadorf / Tänikon
Switzerland
539 m
47.48, 8.90
Zürich / Kloten
Switzerland
426 m
47.48, 8.54
Bern / Zollikofen
Switzerland
553 m
46.99, 7.46
Fribourg / Grangeneuve
Switzerland
651 m
46.77, 7.11
Helsinki-Vantaa lentoasema
Finland
47 m
60.33, 24.97
Further countries: 61 stations from NOAA's hourly reports.
Hourly synoptic reports gathered by NOAA (GHCN hourly), for 11 countries whose own service is not read directly. They are coarser than the national services' data and are scored as a set of their own, without solar and without ICON.
AIFS 2, up to 101 days to 10 November 2026
Only 94 days are available here, too few to call any difference between models.
2 m temperature: skill against climatology
Raw forecasts, 101 days, 23 May 2026 to 10 November 2026. Bars are 95% intervals. The climatology is not adjusted for warming and ran 0.7 K too cold over these days, which raises every model's temperature skill against it.
2 m temperature: skill against persistence
Raw forecasts, 61 days, 23 May 2026 to 7 September 2026: those of the 101 days scored on which persistence can be scored as well. Bars are 95% intervals.
2 m temperature: skill against climatology
Bias-corrected forecasts, 37 days, 22 June 2026 to 2 October 2026. Bars are 95% intervals.
2 m temperature: skill against persistence
Bias-corrected forecasts, 28 days, 23 June 2026 to 20 July 2026: those of the 37 days scored on which persistence can be scored as well. Bars are 95% intervals.
AIFS's RMSE against each model, in %, over 101 days
Compared with
Day 1
Day 3
Day 5
Day 7
Day 10
IFS
−8%, not separated
−9%, not separated
−8%, not separated
−7%, not separated
−5%, not separated
GFS
−24%, not separated
−25%, not separated
−24%, not separated
−19%, not separated
−13%, not separated
AIFS's RMSE against each model, in %, over 37 days
Compared with
Day 1
Day 3
Day 5
Day 7
Day 10
IFS
−9%, not separated
−10%, not separated
−8%, not separated
−7%, not separated
−5%, not separated
GFS
−26%, not separated
−25%, not separated
−26%, not separated
−22%, not separated
−17%, not separated
10 m wind speed: skill against climatology
Raw forecasts, 94 days, 23 May 2026 to 9 November 2026. Bars are 95% intervals.
10 m wind speed: skill against persistence
Raw forecasts, 61 days, 23 May 2026 to 7 September 2026: those of the 94 days scored on which persistence can be scored as well. Bars are 95% intervals.
10 m wind speed: skill against climatology
Bias-corrected forecasts, 34 days, 22 June 2026 to 27 September 2026. Bars are 95% intervals.
10 m wind speed: skill against persistence
Bias-corrected forecasts, 28 days, 23 June 2026 to 20 July 2026: those of the 34 days scored on which persistence can be scored as well. Bars are 95% intervals.
AIFS's RMSE against each model, in %, over 94 days
Compared with
Day 1
Day 3
Day 5
Day 7
Day 10
IFS
0%, not separated
−3%, not separated
−4%, not separated
−4%, not separated
−2%, not separated
GFS
−13%, not separated
−14%, not separated
−15%, not separated
−14%, not separated
−8%, not separated
AIFS's RMSE against each model, in %, over 34 days
Compared with
Day 1
Day 3
Day 5
Day 7
Day 10
IFS
−1%, not separated
−4%, not separated
−6%, not separated
−6%, not separated
−4%, not separated
GFS
−14%, not separated
−17%, not separated
−19%, not separated
−15%, not separated
−11%, 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: minus means AIFS's was lower (better). Hover or tap a cell for the difference in units, its 95% interval and the p-value.
Bias, activity and eventsWhether a model runs high or low, how much it varies, and how it does in events
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. Hover an event score for its 95% interval.
2 m temperature: bias, in °CForecast minus measurement, on average. Plus: too high. Minus: too low. Zero is best.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
0.00
−0.02
+0.02
−0.01
−0.07
IFS
−0.09
−0.06
−0.04
+0.01
+0.01
GFS
−0.14
−0.13
−0.08
−0.10
−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.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
0.91
0.90
0.91
0.91
0.93
IFS
0.91
0.90
0.91
0.93
0.94
GFS
0.95
0.94
0.94
0.94
0.96
2 m temperature of 0 °C or below: frequency biasObserved 48 times on 16 days in 11976 station-times (0.4%). Forecast events per observed event. 1.00 is best. Above 1 the model calls the event too often, below 1 too rarely.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
0.85
0.81
1.00
0.65
0.73
IFS
0.83
0.88
0.98
0.38
0.67
GFS
0.58
0.56
0.65
0.67
0.75
2 m temperature of 0 °C or below: equitable threat scoreHow well forecast and observed events coincide, beyond chance. 1 is perfect, 0 is chance, higher is better.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
0.56
0.58
0.47
0.36
0.20
IFS
0.49
0.45
0.40
0.14
0.16
GFS
0.41
0.41
0.32
0.31
0.15
2 m temperature: bias, in °CForecast minus measurement, on average. Plus: too high. Minus: too low. Zero is best.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
+0.01
−0.02
−0.06
−0.14
−0.23
IFS
+0.10
+0.08
+0.02
+0.21
+0.20
GFS
+0.24
+0.27
+0.28
+0.37
+0.34
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.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
0.90
0.90
0.90
0.94
0.91
IFS
0.91
0.93
0.96
0.98
0.95
GFS
0.97
0.95
0.97
0.97
0.95
2 m temperature of 0 °C or belowObserved 1 times on 1 days in 4386 station-times (0.0%). Too few events to score: at least 30, on at least 10 days, are needed.
10 m wind: bias, in m/sForecast minus measurement, on average. Plus: too high. Minus: too low. Zero is best.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
−0.02
−0.11
−0.15
−0.18
−0.20
IFS
+0.17
+0.15
+0.15
+0.14
+0.04
GFS
+0.56
+0.50
+0.47
+0.44
+0.40
10 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.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
0.91
0.88
0.87
0.89
0.90
IFS
0.94
0.93
0.94
0.93
0.92
GFS
1.09
1.07
1.07
1.06
1.05
10 m wind of 8 m/s or more: frequency biasObserved 666 times on 94 days in 11177 station-times (6%). Forecast events per observed event. 1.00 is best. Above 1 the model calls the event too often, below 1 too rarely.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
0.65
0.58
0.52
0.55
0.53
IFS
0.87
0.87
0.84
0.82
0.77
GFS
1.15
1.09
1.06
0.95
0.93
10 m wind of 8 m/s or more: equitable threat scoreHow well forecast and observed events coincide, beyond chance. 1 is perfect, 0 is chance, higher is better.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
0.33
0.28
0.21
0.17
0.09
IFS
0.37
0.32
0.24
0.18
0.10
GFS
0.32
0.27
0.21
0.12
0.09
10 m wind of 10.8 m/s or more: frequency biasObserved 175 times on 70 days in 11177 station-times (2%). Forecast events per observed event. 1.00 is best. Above 1 the model calls the event too often, below 1 too rarely.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
0.51
0.43
0.42
0.38
0.39
IFS
0.83
0.84
0.87
0.69
0.67
GFS
1.22
1.19
1.11
0.94
1.03
10 m wind of 10.8 m/s or more: equitable threat scoreHow well forecast and observed events coincide, beyond chance. 1 is perfect, 0 is chance, higher is better.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
0.27
0.25
0.17
0.11
0.06
IFS
0.30
0.26
0.18
0.09
0.05
GFS
0.33
0.26
0.19
0.10
0.06
10 m wind: bias, in m/sForecast minus measurement, on average. Plus: too high. Minus: too low. Zero is best.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
+0.03
+0.05
+0.07
+0.06
+0.11
IFS
+0.03
+0.03
+0.05
+0.02
+0.13
GFS
+0.09
+0.10
+0.08
+0.09
+0.18
10 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.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
0.91
0.89
0.89
0.91
0.91
IFS
0.95
0.92
0.94
0.94
0.91
GFS
1.06
1.05
1.03
1.03
1.04
10 m wind of 8 m/s or more: frequency biasObserved 213 times on 34 days in 4055 station-times (5%). Forecast events per observed event. 1.00 is best. Above 1 the model calls the event too often, below 1 too rarely.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
0.87
0.85
0.81
0.89
0.82
IFS
0.91
0.81
0.89
0.85
0.93
GFS
1.10
1.12
0.96
0.92
1.16
10 m wind of 8 m/s or more: equitable threat scoreHow well forecast and observed events coincide, beyond chance. 1 is perfect, 0 is chance, higher is better.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
0.50
0.44
0.31
0.27
0.16
IFS
0.47
0.37
0.29
0.22
0.11
GFS
0.41
0.38
0.26
0.18
0.13
10 m wind of 10.8 m/s or more: frequency biasObserved 57 times on 27 days in 4055 station-times (1%). Forecast events per observed event. 1.00 is best. Above 1 the model calls the event too often, below 1 too rarely.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
0.53
0.44
0.46
0.54
0.49
IFS
0.56
0.60
0.67
0.70
0.63
GFS
1.10
0.95
0.86
0.70
1.19
10 m wind of 10.8 m/s or more: equitable threat scoreHow well forecast and observed events coincide, beyond chance. 1 is perfect, 0 is chance, higher is better.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
0.26
0.24
0.20
0.10
0.07
IFS
0.21
0.24
0.13
0.11
0.08
GFS
0.28
0.21
0.17
0.06
0.06
AIFS 1.1, the earlier version240 days, 7 September 2025 to 12 May 2026; other seasons, so not comparable with AIFS 2
2 m temperature: skill against climatology
Raw forecasts, 240 days, 7 September 2025 to 12 May 2026. Bars are 95% intervals.
2 m temperature: skill against persistence
Raw forecasts, 215 days, 15 September 2025 to 12 May 2026: those of the 240 days scored on which persistence can be scored as well. Bars are 95% intervals.
2 m temperature: skill against climatology
Bias-corrected forecasts, 210 days, 8 October 2025 to 12 May 2026. Bars are 95% intervals.
2 m temperature: skill against persistence
Bias-corrected forecasts, 158 days, 13 October 2025 to 12 May 2026: those of the 210 days scored on which persistence can be scored as well. Bars are 95% intervals.
AIFS's RMSE against each model, in %, over 240 days
Compared with
Day 1
Day 3
Day 5
Day 7
Day 10
IFS
▼−14%, AIFS lower error, tested
▼−15%, AIFS lower error, tested
▼−13%, AIFS lower error, tested
▼−10%, AIFS lower error, tested
−6%, not separated
GFS
▼−28%, AIFS lower error, tested
▼−27%, AIFS lower error, tested
▼−23%, AIFS lower error, tested
▼−18%, AIFS lower error, tested
▼−11%, AIFS lower error, tested
AIFS's RMSE against each model, in %, over 210 days
Compared with
Day 1
Day 3
Day 5
Day 7
Day 10
IFS
▼−13%, AIFS lower error, tested
▼−14%, AIFS lower error, tested
▼−12%, AIFS lower error, tested
▼−9%, AIFS lower error, tested
▼−6%, AIFS lower error, tested
GFS
▼−28%, AIFS lower error, tested
▼−26%, AIFS lower error, tested
▼−22%, AIFS lower error, tested
▼−17%, AIFS lower error, tested
▼−9%, AIFS lower error, tested
10 m wind speed: skill against climatology
Raw forecasts, 240 days, 7 September 2025 to 12 May 2026. Bars are 95% intervals.
10 m wind speed: skill against persistence
Raw forecasts, 215 days, 15 September 2025 to 12 May 2026: those of the 240 days scored on which persistence can be scored as well. Bars are 95% intervals.
10 m wind speed: skill against climatology
Bias-corrected forecasts, 210 days, 8 October 2025 to 12 May 2026. Bars are 95% intervals.
10 m wind speed: skill against persistence
Bias-corrected forecasts, 158 days, 13 October 2025 to 12 May 2026: those of the 210 days scored on which persistence can be scored as well. Bars are 95% intervals.
AIFS's RMSE against each model, in %, over 240 days
Compared with
Day 1
Day 3
Day 5
Day 7
Day 10
IFS
0%, not separated
▼−4%, AIFS lower error, tested
▼−6%, AIFS lower error, tested
▼−5%, AIFS lower error, tested
▼−5%, AIFS lower error, tested
GFS
▼−13%, AIFS lower error, tested
▼−17%, AIFS lower error, tested
▼−18%, AIFS lower error, tested
▼−14%, AIFS lower error, tested
▼−13%, AIFS lower error, tested
AIFS's RMSE against each model, in %, over 210 days
Compared with
Day 1
Day 3
Day 5
Day 7
Day 10
IFS
▼−3%, AIFS lower error, tested
▼−7%, AIFS lower error, tested
▼−8%, AIFS lower error, tested
▼−6%, AIFS lower error, tested
▼−7%, AIFS lower error, tested
GFS
▼−17%, AIFS lower error, tested
▼−21%, AIFS lower error, tested
▼−21%, AIFS lower error, tested
▼−16%, AIFS lower error, tested
▼−15%, 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: minus means AIFS's was lower (better). Hover or tap a cell for the difference in units, its 95% interval and the p-value.
Bias, activity and events, 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. Hover an event score for its 95% interval.
2 m temperature: bias, in °CForecast minus measurement, on average. Plus: too high. Minus: too low. Zero is best.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
+0.19
+0.16
+0.15
+0.11
+0.11
IFS
−0.09
−0.13
−0.15
−0.11
−0.15
GFS
+0.10
+0.04
−0.03
−0.04
+0.08
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.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
0.94
0.92
0.93
0.94
0.97
IFS
0.94
0.95
0.96
0.97
0.96
GFS
0.94
0.94
0.94
0.96
0.98
2 m temperature of 0 °C or below: frequency biasObserved 4515 times on 207 days in 28833 station-times (16%). Forecast events per observed event. 1.00 is best. Above 1 the model calls the event too often, below 1 too rarely.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
0.94
0.94
0.95
0.95
0.95
IFS
0.96
0.97
0.97
0.99
0.99
GFS
0.91
0.90
0.91
0.92
0.89
2 m temperature of 0 °C or below: equitable threat scoreHow well forecast and observed events coincide, beyond chance. 1 is perfect, 0 is chance, higher is better.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
0.78
0.74
0.68
0.58
0.46
IFS
0.75
0.71
0.64
0.55
0.43
GFS
0.71
0.66
0.60
0.51
0.39
2 m temperature: bias, in °CForecast minus measurement, on average. Plus: too high. Minus: too low. Zero is best.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
+0.01
+0.01
+0.00
−0.03
−0.07
IFS
−0.01
−0.03
−0.04
−0.06
−0.11
GFS
−0.03
−0.04
−0.05
−0.06
−0.10
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.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
0.94
0.92
0.94
0.94
0.96
IFS
0.94
0.95
0.97
0.97
0.95
GFS
0.92
0.92
0.93
0.94
0.94
2 m temperature of 0 °C or below: frequency biasObserved 4477 times on 204 days in 24834 station-times (18%). Forecast events per observed event. 1.00 is best. Above 1 the model calls the event too often, below 1 too rarely.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
0.97
0.97
0.97
0.97
0.98
IFS
0.98
0.98
0.98
1.00
0.99
GFS
0.98
1.00
1.00
0.98
0.98
2 m temperature of 0 °C or below: equitable threat scoreHow well forecast and observed events coincide, beyond chance. 1 is perfect, 0 is chance, higher is better.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
0.80
0.75
0.68
0.56
0.44
IFS
0.76
0.70
0.63
0.54
0.41
GFS
0.70
0.66
0.58
0.50
0.39
10 m wind: bias, in m/sForecast minus measurement, on average. Plus: too high. Minus: too low. Zero is best.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
−0.01
−0.10
−0.17
−0.26
−0.24
IFS
+0.19
+0.16
+0.16
+0.12
+0.06
GFS
+0.50
+0.48
+0.47
+0.35
+0.42
10 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.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
0.89
0.87
0.85
0.84
0.84
IFS
0.95
0.94
0.94
0.95
0.93
GFS
1.09
1.09
1.09
1.06
1.07
10 m wind of 8 m/s or more: frequency biasObserved 2791 times on 240 days in 28829 station-times (10%). Forecast events per observed event. 1.00 is best. Above 1 the model calls the event too often, below 1 too rarely.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
0.71
0.66
0.61
0.55
0.56
IFS
0.87
0.87
0.86
0.82
0.78
GFS
1.11
1.11
1.09
0.99
1.01
10 m wind of 8 m/s or more: equitable threat scoreHow well forecast and observed events coincide, beyond chance. 1 is perfect, 0 is chance, higher is better.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
0.42
0.37
0.28
0.20
0.15
IFS
0.45
0.40
0.31
0.23
0.15
GFS
0.41
0.35
0.25
0.18
0.12
10 m wind of 10.8 m/s or more: frequency biasObserved 961 times on 219 days in 28829 station-times (3%). Forecast events per observed event. 1.00 is best. Above 1 the model calls the event too often, below 1 too rarely.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
0.63
0.53
0.51
0.47
0.48
IFS
0.91
0.87
0.86
0.86
0.79
GFS
1.21
1.23
1.18
1.03
1.09
10 m wind of 10.8 m/s or more: equitable threat scoreHow well forecast and observed events coincide, beyond chance. 1 is perfect, 0 is chance, higher is better.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
0.33
0.27
0.20
0.13
0.08
IFS
0.38
0.30
0.22
0.16
0.11
GFS
0.35
0.28
0.19
0.12
0.08
10 m wind: bias, in m/sForecast minus measurement, on average. Plus: too high. Minus: too low. Zero is best.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
+0.01
+0.01
−0.00
−0.00
+0.01
IFS
+0.01
+0.02
+0.01
+0.01
+0.04
GFS
+0.01
+0.01
−0.02
+0.01
+0.03
10 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.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
0.87
0.86
0.84
0.82
0.84
IFS
0.94
0.94
0.93
0.93
0.91
GFS
1.08
1.08
1.07
1.04
1.05
10 m wind of 8 m/s or more: frequency biasObserved 2392 times on 210 days in 24812 station-times (10%). Forecast events per observed event. 1.00 is best. Above 1 the model calls the event too often, below 1 too rarely.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
0.78
0.79
0.76
0.74
0.75
IFS
0.82
0.83
0.81
0.81
0.82
GFS
0.95
0.95
0.92
0.92
0.94
10 m wind of 8 m/s or more: equitable threat scoreHow well forecast and observed events coincide, beyond chance. 1 is perfect, 0 is chance, higher is better.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
0.51
0.46
0.35
0.26
0.19
IFS
0.50
0.45
0.33
0.25
0.17
GFS
0.46
0.38
0.27
0.20
0.14
10 m wind of 10.8 m/s or more: frequency biasObserved 821 times on 191 days in 24812 station-times (3%). Forecast events per observed event. 1.00 is best. Above 1 the model calls the event too often, below 1 too rarely.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
0.73
0.72
0.74
0.71
0.73
IFS
0.84
0.87
0.84
0.86
0.85
GFS
1.08
1.11
1.08
1.02
1.06
10 m wind of 10.8 m/s or more: equitable threat scoreHow well forecast and observed events coincide, beyond chance. 1 is perfect, 0 is chance, higher is better.
Model
Day 1
Day 3
Day 5
Day 7
Day 10
AIFS
0.44
0.38
0.26
0.18
0.13
IFS
0.43
0.36
0.24
0.18
0.14
GFS
0.40
0.32
0.21
0.14
0.10
Country by countryAIFS 2; a check that one country does not carry the pooled figure
A check that the pooled figure is not carried by one country. Countries with fewer than 10 stations are in the pooled set only. These are estimates without verdicts: AIFS's raw RMSE, and beside it how it compares with IFS's at the same stations and times. The differences were not tested, so their colour shows the sign only.
2 m temperature, AIFS 2, raw forecasts: AIFS's RMSE in °C and its difference from IFS's, 101 days to 10 November 2026
Country
Day 1
Day 3
Day 5
Day 7
Day 10
Romania (13)
1.67−13% (lower than IFS, better)
1.89−12% (lower than IFS, better)
2.38−11% (lower than IFS, better)
3.18−5% (lower than IFS, better)
3.84−5% (lower than IFS, better)
Sweden (11)
1.51−7% (lower than IFS, better)
1.73−10% (lower than IFS, better)
2.17−8% (lower than IFS, better)
2.72−8% (lower than IFS, better)
3.730% (equal to IFS)
Norway (10)
1.23−7% (lower than IFS, better)
1.35−9% (lower than IFS, better)
1.64−8% (lower than IFS, better)
2.02−12% (lower than IFS, better)
2.680% (equal to IFS)
10 m wind speed, AIFS 2, raw forecasts: AIFS's RMSE in m/s and its difference from IFS's, 94 days to 9 November 2026
Country
Day 1
Day 3
Day 5
Day 7
Day 10
Romania (13)
1.25−2% (lower than IFS, better)
1.27−5% (lower than IFS, better)
1.40−7% (lower than IFS, better)
1.63−2% (lower than IFS, better)
1.75−6% (lower than IFS, better)
Sweden (11)
1.65+1% (higher than IFS, worse)
1.72−3% (lower than IFS, better)
1.93−5% (lower than IFS, better)
2.19−6% (lower than IFS, better)
2.42−3% (lower than IFS, better)
Norway (10)
1.88−7% (lower than IFS, better)
2.15−9% (lower than IFS, better)
2.73−8% (lower than IFS, better)
3.26−5% (lower than IFS, better)
3.60−6% (lower than IFS, better)
−8% AIFS's RMSE 8% lower than IFS's: better. +8% 8% higher: worse. Not tested, so violet and yellow-brown show the sign only, not the green or red of a verdict.
Why these stations are kept apartThe rules, and how well the model grids fit the stations
They are kept out of the European set above for two reasons. NOAA's files have lost 20 to 50% of reports since July 2026: for AIFS 2, wind can be scored on 94 days there, against 174 at the German stations, and persistence on 61, and pooling would bring those gaps into the European set. And the reports are coarser: wind is the 10-minute mean before the hour, in whole m/s at many stations. One station per square of 1.5° by 2.25°, at least 80% complete.
Each station had to lie within 100 m of the grid height of AIFS, IFS and GFS at the nearest land point, taken from the models' own surface-height files. So at the nearest land point the largest gap is 127 m and one station exceeds 100 m for at least one model; under interpolation it is 206 m and six stations exceed 100 m for at least one model. Temperature is moved to the station's height before scoring, which removes most of the effect of such a gap but not all of it.
The 61 stations, by countryName, height and position
The stations of this set: name, country, height and position