Data-driven Mesoscale Weather Forecasting Combining Swin-Unet and Diffusion Models

In particular, diffusion models represent fine-scale details wit hout spatial smoothing, which is crucial for mesoscale predictions, such as heavy rainfall fo recasting. However, the applications of diffusion models to mesoscale prediction remain limited. T o address this gap, this study proposes an architecture that combines a diffusion model with Swin-Unet as a deterministic model, achieving mesoscale predictions while maintain ing flexibility. The proposed architecture trains the two models independently, allowin g the diffusion model to remain unchanged when the deterministic model is updated. Comp arisons using the Fractions Skill Score and power spectral analysis demonstrate th at incorporating the diffusion model leads to improved accuracy compared to predictions with out it. These findings underscore the potential of the proposed architecture to enha nce mesoscale predictions, particularly for strong rainfall events, while maintaining flexibility.