RNDiff: Rainfall nowcasting with Condition Diffusion Model

The Diffusion Models are widely used in image generation because they can generate high-quality and realistic samples. In contrast, generative adversarial networks (GANs) and variational autoencoders (VAEs) have some limitations in terms of image quality. We introduce a diffusion model to the precipitation forecasting task and propose a short-term precipitation nowcasting with condition diffusion model based on historical observational data, which is referred to as Rainfall nowcasting with Condition Diffusion Model(RNDiff). By incorporating an additional conditional decoder module in the denoising process, RNDiff achieves end-to-end conditional rainfall prediction. RNDiff is composed of two networks: a denoising network and a conditional encoder network. The conditional network is composed of multiple independent UNet networks. These networks extract conditional feature maps at different resolutions, providing accurate conditional information that guides the diffusion model for conditional generation. RNDiff surpasses GANs in terms of prediction accuracy, although it requires more computational resources. The RNDiff model exhibits higher stability and efficiency during training than GANs-based approaches, and generates high-quality precipitation distribution samples that better reflect future actual precipitation conditions. Compared to the current state-of-the-art GAN-based methods, our proposed approach achieves significant improvements on key evaluation metrics. Specifically, our method leads to improvements in the CSI, HSS, and FSS, which are increased by around 8%, 5%, and 6%, respectively. The experiment fully verified the advantages and potential of RNdiff in precipitation forecasting and provided new insights for improving rainfall forecasting. Our project is open source and available on GitHub at: https://github.com/ybu-lxd/RNDiff.