Generative AI to quantify uncertainty in weather forecasting
Google just deployed generative AI to forecast extreme weather 100x faster. SEEDS generates 256 ensemble forecasts in 3 minutes—what used to take supercomputers hours.
Why it matters
Google's SEEDS model uses diffusion-based generative AI to dramatically reduce the computational cost of probabilistic weather forecasting, enabling accurate prediction of rare extreme events (1-in-100 probability) that traditional physics-based ensembles can't capture efficiently. This has immediate applications for emergency management and energy trading.
The key facts
6 to knowSEEDS generates 256 ensemble members at 2° resolution per 3 minutes on Google Cloud TPUv3-32
Traditional methods require 10,000-member ensembles to forecast 1% probability events with <10% relative error
SEEDS matches or exceeds physics-based ensembles in RMSE, rank histogram, and continuous ranked probability score (CRPS)
Published in Science Advances; represents first application of probabilistic diffusion models to weather forecasting
Computational cost is negligible compared to supercomputer hours required by traditional methods
Tested on 2022 European heat waves; accurately captures spatial covariance and cross-field correlations that Gaussian models miss
Go to the source
Google Research Blogblog.research.google
Publisher excerpt: Posted by Lizao (Larry) Li, Software Engineer, and Rob Carver, Research Scientist, Google Research Accurate weather forecasts can have a direct impact on people’s lives, from helping make routine decisions, like what to pack for a day’s activities, to informing urgent actions, for example,…