Google announced a new AI model for accurate weather forecasting

Google's new SEEDS generative AI model could reduce uncertainty in weather forecasting. It uses other probabilistic models to control diffusion noise, according to a Google Research blog post.

Tech Apr 2, 2024 0 209 Add to Reading List

Google announced a new AI model for accurate weather forecasting

Google Research has announced a new generative AI model that can improve accuracy and reduce uncertainty in weather forecasting. It's called the Scalable Ensemble Envelope Diffusion Sampler, or SEEDS.

Google already has experience working on weather forecasting models, including GraphCast, which can predict weather 10 days in advance, and MetNet-3, a high-resolution model for 24-hour forecasts. What makes SEEDS special, however, is that it makes forecasting more accurate and less expensive.

SEEDS uses other probabilistic models to control diffusion noise, according to a Google Research blog post. The AI model was trained based on metrics such as rank histogram, root mean square error (RMSE), and continuous ranked probability score (CRPS). The paper states that SEEDS improves the accuracy of the initial forecast by requiring fewer forecast generations over a given period of time.

The research team also provided examples of using the model and found that it provided greater reliability than the Gaussian model. SEEDS has not yet been peer-reviewed, but could soon be developed for commercial purposes.