Oral C3 Al, Automation & Optimization

5:15 PM - 5:30 PM JST(8:15 AM - 8:30 AM UTC)

[GW9BIT7e-06] Enhanced Tropical Cyclone Track Prediction Using Ocean–Atmosphere Coupled Model Simulations and Hybrid Ensemble Deep Learning Approaches

*Amit Kumar Jena¹ (1. Japan Agency for Marine-Earth Science and Technology (Japan))

Accurate prediction of tropical cyclone trajectories remains a significant challenge due to the complex and nonlinear ocean–atmosphere interactions governing cyclone evolution, while conventional numerical weather prediction models are computationally intensive and often inadequate for real-time forecasting. In this study, Cyclone Amphan (2020) is simulated using both an ocean-atmosphere coupled WRF model and an uncoupled WRF model to investigate the impact of air–sea interactions on track prediction. The coupled model exhibits substantially improved performance over the uncoupled configuration by more realistically representing upper-ocean feedbacks, thereby reducing track errors. To further enhance prediction accuracy, a comprehensive deep learning framework is developed by implementing and systematically comparing CNN, LSTM, CNN-GRU, MLP, and Temporal Fusion Transformer (TFT) models using the outputs from both numerical simulations. The novelty of this study lies in integrating coupled model simulations with hybrid deep learning architectures that combine spatial feature extraction and temporal sequence learning, together with a dual evaluation framework based on RMSE and spatial track alignment. Results demonstrate that hybrid models, particularly CNN-GRU, achieve the highest prediction accuracy, while the coupled model consistently outperforms the uncoupled model by providing more realistic environmental conditions. The proposed framework offers a computationally efficient and accurate approach for tropical cyclone track prediction, with significant potential for improving operational forecasting, early warning systems, and disaster risk management.