ECMWF IFS timed tomorrow's rain best at Lakenheath; 3 days ahead, no model scored above 34.8.
Rainfall totals and rain timing are different tests. This report checks the models on both.
From 8 September to 14 September 2026, Lakenheath recorded 19.1 millimetres of rain. At least one millimetre fell on 3 of those days; 13 September 2026 was the wettest with 9.7 millimetres. Over the whole period from 8 September to 29 September 2026, the gauge collected 20.6 millimetres, with at least a millimetre on 3 of its 22 days.
Every model was scored on the same 22 days, at every notice its runs reach out to seven days ahead, for timing and for amount. One day ahead, ECMWF IFS timed the rain best, with a score of 57.5; GFS followed at 56.3.
The table below reads each model from a week out down to the day before. One day ahead, the forecasts for those 22 days ranged from 7.3 millimetres by ICON to 29.9 by ECMWF AIFS, against 20.6 measured. With only 28 wet hours in the sample, one shower caught or missed moves a score by several points, so cells a few points apart do not separate two forecasts.
| Notice | ECMWF IFS Europe | GFS United States | UKMO United Kingdom | ECMWF AIFS Europe, AI | ICON Germany |
|---|---|---|---|---|---|
| 7 days ahead | 31.564.6 mm | 17.87.3 mm | beyond horizon | 30.924.5 mm | beyond horizon |
| 6 days ahead | 58.826.9 mm | 9.913.4 mm | 24.826.0 mm | 47.722.4 mm | 22.121.0 mm |
| 5 days ahead | 36.616.1 mm | 58.629.2 mm | 19.814.7 mm | 27.728.8 mm | 4.327.1 mm |
| 4 days ahead | 41.910.2 mm | 21.617.1 mm | 30.023.2 mm | 34.429.6 mm | 16.925.1 mm |
| 3 days ahead | 34.822.8 mm | 22.419.3 mm | 25.915.9 mm | 29.930.6 mm | 20.522.5 mm |
| 2 days ahead | 33.333.6 mm | 30.626.4 mm | 44.928.7 mm | 37.229.3 mm | 28.113.6 mm |
| 1 day ahead | 57.523.2 mm | 56.312.7 mm | 47.123.0 mm | 40.629.9 mm | 26.17.3 mm |
The wettest day, hour by hour
On 13 September 2026, Lakenheath measured 9.7 millimetres across 8 of the 24 hours it could read. The strip below sets each model's run from the day before against the gauge, hour by hour. ECMWF IFS caught 8 of those 8 wet hours, with 4 false alarms.
| Model | 00 | 01 | 02 | 03 | 04 | 05 | 06 | 07 | 08 | 09 | 10 | 11 | 12 | 13 | 14 | 15 | 16 | 17 | 18 | 19 | 20 | 21 | 22 | 23 | Day |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Gauge | Dry | Dry | Dry | Dry | 0.8 | Dry | 2.8 | 1.5 | 2.0 | 0.3 | Dry | 1.0 | 1.0 | 0.3 | Dry | Dry | Dry | Dry | Dry | Dry | Dry | Dry | Dry | Dry | 9.7 mm |
| ECMWF IFS | Dry | False alarm | Dry | Within an hour | Caught | Within an hour | Caught | Caught | Caught | Caught | Within an hour | Caught | Caught | Caught | Within an hour | False alarm | False alarm | False alarm | Dry | Dry | Dry | Dry | Dry | Dry | 6.2 mm |
| ECMWF AIFS | False alarm | False alarm | False alarm | Within an hour | Caught | Within an hour | Caught | Caught | Caught | Caught | Within an hour | Caught | Caught | Caught | Within an hour | False alarm | False alarm | False alarm | False alarm | False alarm | Dry | Dry | Dry | Dry | 7.0 mm |
| UKMO | False alarm | Dry | Dry | Dry | Caught | Within an hour | Caught | Caught | Caught | Caught | Within an hour | Caught | Missed | Missed | Dry | Dry | Dry | Dry | Dry | Dry | Dry | Dry | Dry | Dry | 11.7 mm |
| GFS | False alarm | Dry | Dry | Dry | Caught | Within an hour | Caught | Caught | Caught | Caught | Dry | Missed | Missed | Missed | Dry | Dry | Dry | Dry | Dry | Dry | Dry | Dry | Dry | Dry | 0.8 mm |
| ICON | Dry | Dry | Dry | Dry | Missed | Dry | Missed | Missed | Missed | Missed | Dry | Missed | Missed | Missed | Dry | Dry | Dry | Dry | Dry | Dry | Dry | Dry | Dry | Dry | 0.0 mm |
Which model timed tomorrow's rain best?
Across all 22 days, ECMWF IFS offered the best balance between missed rain and false alarms, with a score of 57.5. GFS followed at 56.3. ICON finished last at 26.1: it caught 6 of the 28 wet hours, with 10 false alarms. How ready a model is to forecast rain shows in the hours it called wet, from 15 to 154 over the same 509 hours. For rain at Lakenheath one day ahead, ECMWF IFS was the best choice in this test.
| Model | Rain score | Caught | Missed | False alarm | Hours called wet |
|---|---|---|---|---|---|
| ECMWF IFS Europe | 20 | 8 | 27 | 52 | |
| GFS United States | 20 | 8 | 23 | 43 | |
| UKMO United Kingdom | 16 | 12 | 27 | 45 | |
| ECMWF AIFS Europe, AI | 26 | 2 | 114 | 154 | |
| ICON Germany | 6 | 22 | 10 | 15 |
About this comparison
The period's wet spell runs from 8 September 2026 to 14 September 2026; its wettest day, 13 September 2026 with 9.7 millimetres, is drawn hour by hour. The other two figures cover every day of the period. Forecasts come from the public models' 00 UTC runs and are compared with the gauge at Lakenheath.
How rain is judged
An hour counts as wet from 0.1 millimetres. A forecast may be 1 hour early or late; when the gauge has no reading, that hour is left out for every model. This is a test at one gauge in Lakenheath, not a verdict on a whole country.
A day enters the sample only when the gauge could be read and all 7 runs before it are in the archive, so every notice is scored on the same 22 days. A model whose run does not cover every one of them at some notice has no score there, rather than a score on fewer days. Each notice is judged on its own. There is no row for how far ahead a model first had the rain: showers do not reduce to one number that a forecast either lands near or misses.
At Lakenheath, the scheduled METAR at 55 minutes past the hour carries a P group for rain during the preceding hour. It is compared with the model hour ending five minutes later. No P group in a scheduled report means dry; no scheduled report means unknown, and that hour is left out. A P0000 group reports a trace, rain too small to measure, and counts as dry.
We compare archived model output with independent station observations. Data is at data.json, the full method at methodology and corrections at corrections.
The live version of this report is an app
After The Forecast comes out on a schedule. VeriSky keeps the rolling scores current for the places you care about and shows how models change with forecast distance.
Free standings · No account · No cookies