Reproduce “SEEDS: Emulation of Weather Forecast Ensembles with Diffusion Models” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Lizao Li, Rob Carver, Ignacio Lopez-Gomez, Fei Sha, John Anderson
Leveraging deep generative diffusion models, SEEDS provides a computationally efficient framework to create large, high-fidelity weather forecast ensembles, demonstrating superior reliability and extreme event prediction capabilities over traditional operational systems, particularly for near-surface temperatures and rare occurrences.
Key insights
- Diffusion models can effectively learn and reproduce the complex spatial and multivariate correlation structures inherent in high-dimensional atmospheric data.
- Conditioning generative models on a few physics-based forecasts enables the efficient generation of large, plausible weather ensembles, capturing the phase space of atmospheric attractors.
- Integrating historical reanalysis data during generative model training allows for simultaneous debiasing and uncertainty quantification, creating more reliable forecasts.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.