Just Weather Scoring: Efficient End-to-end Nowcasting with Distributional Diffusion

Key points

• Generative diffusion models are well-suited for probabilistic precipitation nowcasting, but existing approaches often rely on separately trained compression or deterministic forecasting components and remain costly at inference due to iterative denoising.1
• Radar-space modeling greatly simplifies training and inference and eliminates uncertainty arising from lossy compression.1
• JWS combines Masked Asynchronous Diffusion, a timestep-sampling scheme that preserves clean context while adapting diffusion training to high-dimensional spatio-temporal data, with a simple scoring-rule objective that aligns training with probabilistic forecasting and unlocks few-step generation.1
• On the SEVIR and MeteoNet benchmarks, JWS achieves state-of-the-art probabilistic forecasting performance at reduced training and inference cost.1

Sources (1)

• [1]Just Weather Scoring: Efficient End-to-end Nowcasting with Distributional Diffusion
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 8, 03:50 PM

  We introduce Just Weather Scoring (JWS), a single-stage, end-to-end diffusion model which addresses both issues by forecasting directly in radar space and enabling few-step generation.

  Generative diffusion models are well-suited for probabilistic precipitation nowcasting, but existing approaches often rely on separately trained compression or deterministic forecasting components and remain costly at inference due to iterative denoising.

Extractive summary: sentences quoted from the sources.