USA
The National Weather Service uses many models, now including AI ones. For warnings, rely on the National Weather Service and your local TV or radio forecast, not an app you do not recognise.
Real World: Weather
How AI Is Changing Your Weather Forecast
Weather centres now run AI forecast models alongside their traditional ones. The European centre, based in Reading in England, switched its AI model on in February 2025. The US weather agency NOAA followed in December 2025. Both say AI forecasts use far less computer power and can track storms better.
Last checked 16 October 2026. Every figure on this page links to its source below.
How Were Forecasts Made Before AI?
Traditional forecasts come from physics. Supercomputers take millions of readings from satellites, weather balloons, ships and ground stations, then solve the equations for how air, heat and water move. This works well but needs huge amounts of computing power.
What Does An AI Weather Model Do Differently?
An AI model learns from decades of past weather how one day's conditions tend to turn into the next. It starts from the same up-to-date readings, then uses what it has learned to predict the coming days. It does this much faster and with far less electricity.
Which Weather Centres Use AI?
The European Centre for Medium-Range Weather Forecasts (ECMWF), which is based in Reading and supplies forecasts to 35 member and co-operating countries including the UK, put its Artificial Intelligence Forecasting System (AIFS) into operation on 25 February 2025. It runs side by side with the centre's physics-based model.
In the USA, NOAA launched three AI-driven global models on 17 December 2025. Its main AI model was built from Google DeepMind's GraphCast and then trained further on NOAA's own data.
What the weather centres say AI models achieve
- About 1,000 timesless energy per forecast (ECMWF)
- Up to 20%better tropical cyclone tracks (ECMWF)
- 0.3%of the computing used by NOAA's traditional model
- About 40 minutesfor a 16-day forecast (NOAA)
Are AI Forecasts Better?
For some things, yes. ECMWF says its AI model outperforms physics-based models on many measures, including storm tracks. NOAA says its AI model cuts errors in where tropical storms will go at longer ranges.
They are not better at everything. NOAA says its first AI model is worse at predicting how strong a tropical storm will become. ECMWF's AI model also uses a coarser grid, 28 km, against 9 km for its physics model. That is why both centres run AI and physics models together, and forecasters compare them.
NOAA has also built a "hybrid" set of forecasts that mixes 31 AI runs with 31 physics runs. It says this combined set beats either kind on its own across most of its checks. Running many slightly different forecasts like this is how forecasters judge how sure they can be, for example whether rain at the weekend is likely or only possible.
What It Means For You
UK
Your forecast may draw on European centre data that now includes AI. For severe weather, follow the Met Office warnings. Forecasts beyond a week or so are still uncertain, AI or not.
Common Questions
Is the weather forecast on my phone made by AI?
Possibly in part. Forecasts often blend several models, and weather centres in Europe and the USA now run AI models alongside traditional ones. The app maker decides which data it uses.
Are AI forecasts more accurate?
For some things, such as storm tracks, the centres say yes. NOAA says its first AI model is weaker at predicting storm strength, so experts use both kinds together.
Why does using less computer power matter?
It means forecasts arrive faster and cost less to run. NOAA says a 16-day AI forecast finishes in about 40 minutes.
Does the Met Office use AI?
The Met Office is part of the UK's membership of ECMWF, which runs an AI model. We have not covered the Met Office's own AI work here.
New to all this? Our free AI Starter Guide walks you through the basics, and you can print it.
Read Next
Sources
Last checked 16 October 2026.
Updated 19 October 2026