[Paper] SEEDS: Emulation of Weather Forecast Ensembles with Diffusion Models

[hchen19]

Arxiv/Blog/Paper Link

Blog Arxiv Science Paper Code

Detailed Description

SEEDS can generate a large ensemble conditioned on as few as one or two forecasts from an operational numerical weather prediction system. The generated ensembles not only yield plausible real-weather–like forecasts but also match or exceed physics-based ensembles in skill metrics such as the rank histogram, the root-mean-squared error (RMSE), and the continuous ranked probability score (CRPS).

Context

Goal:

• improve the computational efficiency of ensemble weather forecasting, not to replace physics-based models

  Advantages:

• Efficient: The learned models are highly scalable with respect to high-performance computing accelerators and can sample hundreds to tens of thousands of realistic weather forecasts at low cost
• Flexible: flexible enough to enable direct generation of debiased ensembles
• Forecasting: able to forecast extreme events and assign meaningful likelihoods to them

[jacobbieker]

Oh, this looks really interesting!