<![CDATA[Tropical cyclones are extreme weather phenomena that generate strong winds, heavy rainfall, high waves, and infrastructure damage, necessitating rapid and accurate prediction methods to support early warning systems. This study aims to develop a Temporal Convolutional Network (TCN) model to forecast the short-term track and intensity of tropical cyclones at lead times of 6, 12, 18, and 24 hours within World Meteorological Organization (WMO) Regional Association V (RA V), covering waters south of Indonesia. Best-track data from the Australian Bureau of Meteorology (BOM) for 1973–2026 were used, comprising longitude, latitude, central pressure, and maximum wind speed. Cyclones Seroja, Cempaka, Dahlia, Anggrek, and Savanna served as independent test data, while the remaining data were chronologically split into 90% training (1973–2021) and 10% validation (2022–2026). At a 6-hour lead time, the RMSE for longitude, latitude, maximum wind speed, and central pressure were 1.87°, 0.33°, 2.50 knots, and 3.37 hPa, respectively, with R² values of 0.947–0.994. At 24 hours, RMSE increased to 2.70°, 1.25°, 6.76 knots, and 8.95 hPa; R² for longitude and latitude remained high (0.955 and 0.947), while R² for maximum wind speed and central pressure declined to 0.616 and 0.658. Spatially, the model captured the main trajectory pattern for most cyclones, though performance for Cempaka was relatively low. TCN proved more reliable for track than intensity prediction. Incorporating additional environmental variables is recommended to improve intensity prediction accuracy and strengthen Indonesia's tropical cyclone early warning system.]]>
