DuoCast: Duo-Probabilistic Diffusion for Precipitation Nowcasting

Penghui Wen, Mengwei He, Patrick Filippi, Na Zhao, Feng Zhang, Thomas Francis Bishop, Zhiyong Wang, Kun Hu

arXiv preprint · 02 December 2024 · Fudan University (FuXi)

In one sentence. DuoCast splits precipitation nowcasting into low- and high-frequency diffusion branches in latent subspaces, outperforming single-branch U-Net diffusion baselines on four radar benchmarks.

Abstract

Accurate short-term precipitation forecasting is critical for weather-sensitive decision-making in agriculture, transportation, and disaster response. Existing deep learning approaches often struggle to balance global structural consistency with local detail preservation, especially under complex meteorological conditions. We propose DuoCast, a dual-diffusion framework that decomposes precipitation forecasting into low- and high-frequency components modeled in orthogonal latent subspaces. We theoretically prove that this frequency decomposition reduces prediction error compared to conventional single branch U-Net diffusion models. In DuoCast, the low-frequency model captures large-scale trends via convolutional encoders conditioned on weather front dynamics, while the high-frequency model refines fine-scale variability using a self-attention-based architecture. Experiments on four benchmark radar datasets show that DuoCast consistently outperforms state-of-the-art baselines, achieving superior accuracy in both spatial detail and temporal evolution.

Labels

Labels and summary are generated by Claude and reviewed weekly. Wrong? Edit this entry.