I hope this makes our Google AI weather models even more accessible and useful, and we'd love to hear your feedback, either direct to weather-lab@google.com or through the site's feedback button in the upper right!

By popular demand, we've updated our interactive Google Weather Lab platform to show WeatherNext 2 ensemble mean gridded forecasts alongside our tropical cyclone predictions, including point forecasts for: temp, precip, wind speed, and sea level pressure: deepmind.google.com/science/weat...

Actual rollout schematic animation here:

Looks like Bluesky has a separate button for videos that I missed 🙃 Actual Milton animation here:

Cyclone max wind speeds are still underestimated, but this performance on tracks is really promising. One recent devastating cyclone was Hurricane Milton, which caused >$85 billion in damages. GenCast predicted ~70% probability of landfall in Florida 8.5 days before it struck.

We also extracted cyclone tracks from GenCast and ENS and compared them with ~100 cyclones observed in 2019. GenCast's ensemble mean cyclone track has a 12-hour position error advantage over ENS out to 4 days, and more actionable track probability fields out to 7 days.

For example, we created a dataset of simulated wind power data at wind farm sites across the globe, and found that GenCast outperforms ENS by 10–20% up to 4 days ahead. This is promising, because better weather forecasts can reduce renewable energy uncertainty and accelerate decarbonisation.

GenCast uses diffusion to generate multiple 15-day forecast trajectories for the atmosphere. It assigns more accurate probabilities to possible weather scenarios than the SoTA physics-based ensemble system from ECMWF, across a 2019 evaluation period.

Quite happy with IceNet getting its 200th citation a little after its 3rd birthday ❄️🎂💻 IceNet began around the time when applying ML to Earth sciences was the stuff of hushed conversations in corridors. A few years later and its potential to revolutionise the field is indisputable. Lots more to do!