Google’s new AI model outperforms world’s best weather forecast system

Google's DeepMind has introduced an advanced AI-driven weather prediction model named GenCast, which outperforms the world's top operational forecasting system.

According to a study published in Nature, the GenCast model surpasses the European Centre for Medium-Range Weather Forecasts (ECMWF), recognized as the leading global forecasting system.

“GenCast advances the prediction of weather uncertainties and risks, offering faster and more accurate forecasts up to 15 days ahead,” stated Google in a press release.

The new AI model represents a significant evolution in weather forecasting. Unlike Google's earlier deterministic model, which provided a single best estimate for future weather, GenCast generates an ensemble of 50+ predictions, each representing a potential weather trajectory.

As a diffusion model, GenCast leverages generative AI techniques similar to those used in image, video, and music generation. However, it is uniquely adapted to Earth's spherical geometry and excels in generating complex probability distributions for future weather scenarios based on current data.

Improved predictions of extreme weather events could prove crucial in saving lives, preventing damage, and reducing costs. For instance, GenCast offers enhanced forecasts for the paths of tropical cyclones, hurricanes, and typhoons, providing vital advanced warnings for affected areas.

Google plans to release real-time and historical forecasts from GenCast and previous models, enabling researchers and developers to integrate these inputs into their workflows.

This innovation is part of Google's expanding suite of next-generation AI-based weather models, including Google DeepMind’s deterministic medium-range forecasts and tools like NeuralGCM, SEEDS, and flood prediction models.