Google’s GenCast AI weather model is better than today’s top forecast system 

Google has been working on AI weather models for the past few years, and GenCast is its latest forecasting system with “state-of-the-art accuracy.” 

GenCast is a diffusion model — like what powers image, video, and music generation tools — but “adapted to the spherical geometry of the Earth, and learns to accurately generate the complex probability distribution of future weather scenarios when given the most recent state of the weather as input,” which will continue to be supplied by today’s traditional approaches.  

Google’s previous model (GraphCast from 2023) “provided a single, best estimate of future weather.” In comparison, GenCast “comprises an ensemble of 50 or more predictions, each representing a possible weather trajectory.”

An ensemble forecast expresses uncertainty by making multiple predictions that represent different possible scenarios. If most predictions show a cyclone hitting the same area, uncertainty is low. But if they predict different locations, uncertainty is higher. GenCast strikes the right balance, avoiding both overstating or understating its confidence in its forecasts.

In practice, Google says GenCast “provides better forecasts of both day-to-day weather and extreme events than the top operational system, the European Centre for Medium-Range Weather Forecasts’ (ECMWF) ENS, up to 15 days in advance.” 

Google trained GenCast on historical weather up to 2018 and then tested it on 2019 data. Of 1320 combinations, “GenCast was more accurate than ENS on 97.2% of these targets, and on 99.8% at lead times greater than 36 hours.” It’s also better at predicting extreme heat and cold, as well as high wind speeds:

Now consider tropical cyclones, also known as hurricanes and typhoons. Getting better and more advanced warnings of where they’ll strike land is invaluable. GenCast delivers superior predictions of the tracks of these deadly storms.

Notably, it takes 8 minutes on a single Google Cloud TPU v5 to create a 15-day forecast with all the possibilities. Existing systems take “hours on a supercomputer with tens of thousands of processors.”

Google has made GenCast an open model with code and weights available today. 

We’ll soon be releasing real-time and historical forecasts from GenCast, and previous models, which will enable anyone to integrate these weather inputs into their own models and research workflows.