<p>Accurate and timely prediction of tropical cyclone (TC) trajectories is crucial for disaster prevention and mitigation, particularly in coastal regions that are highly vulnerable to extreme weather events. Real-time TC forecasts enable emergency responders to implement effective precautionary measures. However, existing TC prediction methods typically rely on multistep temporal inputs, significantly increasing computational demands and hindering real-time forecasting applications. Moreover, these approaches exhibit limited capability in extracting complex correlation features from high-resolution meteorological data, which often leads to suboptimal prediction accuracy. Their heavy reliance on historical trajectory patterns further restricts adaptability to the rapidly evolving atmospheric conditions associated with TCs. To address these challenges, we propose a novel multimodal TC path prediction model that utilizes single-time-step physical data to capture movement-related trends more efficiently and precisely. Our model employs advanced attention mechanisms to process multimodal meteorological reanalysis data, thereby enhancing feature extraction and achieving higher prediction accuracy. Furthermore, by incorporating knowledge distillation techniques, we develop a lightweight network with reduced parameters, significantly improving computational efficiency without compromising performance. Experimental results demonstrate that our approach outperforms existing deep learning-based methods, offering a promising avenue for more accurate and timely tropical cyclone path predictions in dynamic weather environments.</p> Graphical abstract <p></p>