<![CDATA[Tropical cyclones in the Southern Indian Ocean pose severe threats to coastal infrastructure and socio-economic stability, yet predicting their recurving trajectories and intensity remains a significant meteorological challenge. This study evaluates the performance of a particle-motion-based machine learning framework, utilizing the Trackpy library, to forecast cyclone behavior. Leveraging historical data from 2018 to 2025 (JTWC and IBTrACS), the model treats cyclones as physical particles with temporal inertia, employing a multi-lag feature to capture movement momentum. Evaluation using a dataset of 115 cyclones (78:22 train/test ratio) reveals that the Trackpy framework achieves high spatial precision, with Mean Squared Error (MSE) values of 0.1728 for latitude (±33.3 km) and 1.0250 for longitude (±53.2 km). While the intensity prediction yielded a higher MSE of 47.7544 (approximately 6.9-knot deviation), the model successfully captured major strengthening and weakening phases across prominent cyclones, including TC Wallace and TC Neville. These findings demonstrate that integrating temporal inertia is highly effective for maintaining trajectory consistency, establishing Trackpy as a robust architectural foundation for operational forecasting. Further optimization via hybrid models and additional meteorological variables is recommended to enhance intensity accuracy.]]>
Copyrights © 2026
