Google’s new AI weather model can make 15-day forecasts more accurate

Google’s new AI weather model could redefine how we predict and prepare for changing conditions.

Google DeepMind, the company’s AI research division focused on solving scientific problems, announced on December 4 a new AI weather model called GenCastthat promises to provide more accurate odds on different weather conditions up to 15 days in advance.

While weather forecasting typically uses physics-based models, which can take hours to calculate on massive supercomputers, Google says it aims to achieve greater accuracy in just minutes. GenCast can generate a 15-day forecast scenario in just eight minutes using a Google Cloud Tensor processing unit chip, according to the company.

GenCast is part of Google’s growing AI-powered weather model suite, which enhances Google Search and Maps, among other things, with improved forecasts for precipitation, wildfires, flooding and extreme heat.

Details of the latest tests were published in the journal Nature.

While Google’s AI model provided a single best estimate of future weather, GenCast uses probability-based forecasts with 50 or more predictions about how the weather might change and rates the likelihood of those scenarios. The technology runs on a diffusion model, similar to the machine learning models used in generative AI.

“We trained it based on 40 years of historical data [the European Centre for Medium-Range Weather Forecasts]which included variables such as temperature, wind speed and pressure at different heights – allowing it to learn,” the company explained in a tweet.

When predicting extreme heat, cold and high wind speeds, Google said GenCast outperformed current forecast models. It also noted that it provides “superior forecasts” of the path of tropical cyclones up to five days in advance.