Google’s new AI model can predict cyclones up to 15 days in advance

AI model can forecast cyclones formation, path, intensity, size, and shape, up to fifteen days in advance
An undated image of Google AI weather forecast app. — Google Blog
An undated image of Google AI weather forecast app. — Google Blog 

Google is making progress in AI weather forecasting with a new web application "Weather Lab," developed by Google DeepMind and Google Research. 

Alphabet-owned tech giant revealed on Thursday, June 12, 2025, that "Weather Lab" provides an interactive version of Google's AI that forecasts the future of tropical cyclones, which could also be called hurricanes and typhoons, depending on where the storm occurs.

Cyclones are dangerous storms that can cause flooding, heavy winds, and catastrophic damage. The new AI model can forecast a cyclones formation, path, intensity, size, and shape, up to fifteen days in advance. 

It creates 50 unique probabilistic futures that scientists can look through to get a sense of what may occur when the AI uses stochastic neural networks.

Google has a partnership with the US National Hurricane Center (NHC) to evaluate the suitability of this model. 

According to Google the 5-day cyclone track forecasts are typically around 140 kilometres more accurate than ECMWF ENS, the leading physics-based model, which is the best accuracy the ENS obtains in just 3.5 days suggesting Google's AI is effectively operates 1.5 days ahead of one of the best models, and increases the normally required adjust that would take years.

Although powerful, "Weather Lab" is not for public use. It’s designed to support meteorologists and emergency response teams, helping them issue faster and better warnings. 

The tool allows users to compare live and past cyclone predictions from multiple models, including ECMWF HRES and Google’s own WeatherNext Gen.

A note on the website reminds users to rely on official forecasts from national weather services.