CMFPN: A Cross-Modal Fusion Prediction Network for Precipitation Nowcasting

Precipitation nowcasting is crucial and fundamental for various public services, ranging from heavy rain warnings to flight safety. In order to further improve the prediction accuracy of the spatiotemporal sequence forecasting problem, we propose cross-modal fusion prediction network. By design the correction module and multi-scale feature extraction module, the features of multiple data sources are mined and cross-fused, thereby enhancing the overall performanc. Experiments in a real-world radar echo data-set of South China show that compared with the ingenious PredRNN++, TrajGRU methods, and newly proposed Rainformer and simVP, our network performs better on the precipitation nowcasting metrics, as well as occupies small GPU memory.