DiffSR: Learning Radar Reflectivity Synthesis via Diffusion Model from Satellite Observations

Xuming He,Zhiwang Zhou,Wenlong Zhang,Xiangyu Zhao,Hao Chen,Shiqi Chen,Lei Bai

2024-11-11

Abstract:Weather radar data synthesis can fill in data for areas where ground observations are missing. Existing methods often employ reconstruction-based approaches with MSE loss to reconstruct radar data from satellite observation. However, such methods lead to over-smoothing, which hinders the generation of high-frequency details or high-value observation areas associated with convective weather. To address this issue, we propose a two-stage diffusion-based method called DiffSR. We first pre-train a reconstruction model on global-scale data to obtain radar estimation and then synthesize radar reflectivity by combining radar estimation results with satellite data as conditions for the diffusion model. Extensive experiments show that our method achieves state-of-the-art (SOTA) results, demonstrating the ability to generate high-frequency details and high-value areas.

Image and Video Processing,Artificial Intelligence,Computer Vision and Pattern Recognition,Machine Learning

What problem does this paper attempt to address?

The problem that this paper attempts to solve is to synthesize weather radar reflectivity images using satellite observation data to fill in the data gaps in areas where ground - radar observation data are missing. Existing methods usually adopt reconstruction - based methods and use the mean - squared - error (MSE) loss to reconstruct radar data from satellite observations. However, these methods are prone to the over - smoothing problem, which hinders the generation of high - frequency details or high - value observation areas (related to convective weather). To solve this problem, the paper proposes a two - stage diffusion model method called DiffSR (Diffusion - based Synthetic Radar Reflectivity). Specifically:

1. **First stage**: Use Vision Transformer (ViT) to pre - train global - scale data to obtain a preliminary estimate of radar reflectivity.
2. **Second stage**: Combine the preliminary estimate results and satellite data as conditions, and use the diffusion model to further refine the synthesis of radar reflectivity.

Through this method, DiffSR can generate high - quality radar reflectivity images while retaining high - frequency details and high - value observation areas, thereby improving the detection accuracy of convective weather events. Experimental results show that DiffSR has achieved state - of - the - art performance on multiple quantitative indicators, especially in high - value areas.