AI models are becoming a bigger part of weather forecasts

In short: More weather forecasts are now being made with AI, and they are often faster and more accurate than older methods, especially for hurricane paths.

What's going on

Weather forecasting is increasingly using machine learning, a type of AI that learns patterns from large amounts of past data, instead of relying only on traditional physics-based models. You can think of it like a veteran forecaster who has studied millions of past weather maps and learned what usually happens next.

Major weather groups have put these systems into real use. The European Centre for Medium-Range Weather Forecasts, or ECMWF, began running its AI Forecasting System in 2025. ECMWF said its “AIFS Single” system beat its older physics-based approach by up to 20% on measures such as predicting tropical cyclone tracks, which is the projected path of storms.

In the US, NOAA deployed three global AI weather models in December 2025. One goal was better “ensemble” forecasting, which means producing many slightly different forecasts to show uncertainty (like giving several plausible routes a storm might take, not just one line). NOAA’s approach can generate 51 variations efficiently.

Google DeepMind released WeatherNext 2 in November 2025 and says it can produce hundreds of forecast scenarios in under a minute on specialized chips. Wired reports it has been integrated into Google products such as Search and Maps, so these improvements can show up directly in consumer weather tools.

What to watch

Experts still report weak spots. Forecasts for the 2 to 6 week range remain hard, and some studies find AI can miss parts of a storm’s internal structure, such as wind patterns, even when it gets the track right. Many agencies are leaning toward hybrid approaches that combine AI with traditional models and human review.

Source: Wired