Research on Radar Echo Extrapolation Method Integrating Terrain Features

In recent years, severe convective weather has caused significant harm to lives and property, leading to flooding and geological disasters. Consequently, nowcasting and early warning of such events are crucial. Advances in artificial intelligence technologies have provided critical support for these efforts through radar echo extrapolation based on deep learning. However, most current deep learning methods primarily rely on historical radar reflectivity factors or vertically integrated liquid water content for extrapolation. Despite the significant influence of orographic effects on meteorological processes, the impact of terrain feature data on radar echo extrapolation remains underexplored. Therefore, this paper proposes a radar echo extrapolation model that integrates terrain feature data. By incorporating both radar and terrain feature data at the input stage, the model more effectively captures the spatiotemporal characteristics of meteorological processes, thereby enhancing its predictive capability. Additionally, a terrain-guided composite loss function is introduced to further constrain the model, leading to more accurate predictions. Experimental results demonstrate that this approach can improves prediction accuracy.