RTCP-Net: Tropical cyclone generation prediction model based on multi-source information fusion

Tropical cyclones are among the most destructive extreme weather phenomena in nature, and accurately predicting whether a tropical cloud cluster will develop into a tropical cyclone is crucial for disaster prevention and mitigation. Considering the insufficient extraction of numerous key features in tropical cloud cluster data by previous deep learning-based tropical cyclogenesis prediction studies, in response, this paper proposes the Real Time Tropical Cyclogenesis Prediction-Net (RTCP-Net) based on multi-source information fusion. The model computes convective core maps and polar coordinate representations from infrared images of tropical cloud clusters and employs ResNet along with self-attention mechanism to extract their spatiotemporal features. Experimental results demonstrate that the proposed model achieves high accuracy and stability, it attains a detection rate of 99.4\% and a false alarm rate of 0.36\% when predicting the formation of tropical cyclones 24 hours in advance. Notably, the model not only ensures potential of real-time prediction capabilities from satellite data but also surpasses the accuracy of models that utilize reanalysis data.