Track synthesis, materials and catalysis as it is published.
Search papers, labs, and topics across Lattice.
This paper introduces Just Weather Scoring (JWS), an end-to-end distributional diffusion model for probabilistic precipitation nowcasting that forecasts directly in radar space, eliminating the need for separate compression or deterministic components. By integrating Masked Asynchronous Diffusion with a scoring-rule objective, JWS enables efficient few-step generation and achieves state-of-the-art performance on SEVIR and MeteoNet benchmarks with significantly reduced computational cost. The proposed approach demonstrates competitive forecasting accuracy even with substantially fewer parameters and over 17x faster inference compared to existing methods.
End-to-end diffusion models can achieve state-of-the-art precipitation nowcasting with over 17x faster inference by forecasting directly in radar space.
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. 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. Radar-space modeling greatly simplifies training and inference and eliminates uncertainty arising from lossy compression. 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. On the SEVIR and MeteoNet benchmarks, JWS achieves state-of-the-art probabilistic forecasting performance at reduced training and inference cost. Even our smallest model remains competitive using substantially fewer parameters and more than 17x faster inference.