How WeatherTomorrow compares
Our aim is to beat the best forecasts there are. Every forecast on this page, ours and the others, is scored the same way: against what each airport measured.
Our aim is to beat the best forecasts there are. Every forecast on this page, ours and the others, is scored the same way: against what each airport measured.
To see where we stand against current practice, we held back a full year that our model never trained on, 1 November 2025 to 24 October 2026, and scored our published forecast beside:
Everything is scored at 319 airports on 358 days, all issued 18:00 local, the evening before from what was public by then, against what each airport measured. Our high was off by 0.92° on average. The next best, ECMWF AIFS (AI model), corrected per airport, was off by 1.00°.
| Forecast | High, average error | High within 2° | Low, average error |
|---|---|---|---|
| WeatherTomorrow | 0.92° | 90% | 0.95° |
| ECMWF AIFS (AI model), corrected per airport | 1.00° | 88% | 1.08° |
| ECMWF IFS, corrected per airport | 1.09° | 85% | 1.13° |
| NOAA GFS, corrected per airport | 1.18° | 83% | 1.39° |
| WT-1 alone, our neural network | 1.29° | 79% | 1.10° |
| ECMWF IFS, raw | 1.47° | 73% | 1.35° |
| NOAA GFS, raw | 1.58° | 71% | 1.72° |
| ECMWF AIFS (AI model), raw | 1.68° | 67% | 1.60° |
| Normal for the date | 2.84° | 49% | 2.44° |
| Same as today | 2.98° | 49% | 2.59° |
In the US, forecasters start from MOS: a physics model corrected against each station's history. On 97 US airports, the same days and the same issue time:
| Forecast | High, average error | High within 2° | Low, average error |
|---|---|---|---|
| WeatherTomorrow | 0.98° | 88% | 0.95° |
| ECMWF AIFS (AI model), corrected per airport | 1.06° | 86% | 1.10° |
| GFS MOS (US guidance), corrected per airport | 1.25° | 81% | 1.17° |
| ECMWF IFS, corrected per airport | 1.26° | 81% | 1.21° |
| NOAA GFS, corrected per airport | 1.33° | 78% | 1.50° |
| GFS MOS (US guidance) | 1.35° | 76% | 1.31° |
| WT-1 alone, our neural network | 1.39° | 76% | 1.15° |
| NAM MOS (US guidance), corrected per airport | 1.42° | 76% | 1.25° |
| ECMWF IFS, raw | 1.53° | 72% | 1.37° |
| NOAA GFS, raw | 1.62° | 70% | 1.81° |
| ECMWF AIFS (AI model), raw | 1.69° | 66% | 1.43° |
| NAM MOS (US guidance) | 1.79° | 64% | 1.31° |
| Normal for the date | 3.80° | 37% | 3.28° |
| Same as today | 3.93° | 40% | 3.37° |
The US National Blend of Models, the official US guidance that blends several corrected models, could be checked from July to October 2026 on 105 US airports. Its next-day high was off by 1.22° (1.02° with its bias removed on the scored days, which favours it), ours by 0.87°.
ECMWF data: CC BY 4.0, www.ecmwf.int. NOAA data: public domain.
MOSMIX is the station forecast of Germany's national weather service (DWD). Every day we score ours and MOSMIX's at the 224 airports where DWD forecasts for a station within 5 km: the same days, the same airport measurements and the same moment of issue. The count for our current forecast, WT-1 combined with ECMWF since 27 October 2026, starts as soon as its first days have been measured.
Before that, from 23 October 2026 to 27 October 2026, our forecast was WT-1 without ECMWF. On 553 airport-days its high was off by 1.31° and MOSMIX's by 0.99°; its low by 1.14° and MOSMIX's by 1.06°.
MOSMIX is the right yardstick because it chases the same goal by other means. It takes DWD's ICON model and ECMWF's IFS, which simulate the whole atmosphere, and corrects them with years of each station's own measurements. We start from ECMWF too, and improve it with WT-1, our neural network trained on GPUs on 26 years of reports from 6,047 airports, calibrated for each airport. MOSMIX never feeds our forecast.
MOSMIX figures: based on data from Deutscher Wetterdienst (opendata.dwd.de), own elements added: we reduce DWD's hourly forecasts to daily highs and lows. Licence: CC BY 4.0.