This research explores the application of transformer models in predicting hurricane genesis, trajectory, and intensity changes. A three-model system was developed to address different stages of a hurricane's lifecycle: (1) a hurricane genesis model to estimate the likelihood of formation, (2) an initial conditions model to predict post-formation characteristics, and (3) a trajectory model to forecast the storm's lifespan and intensity changes until dissipation or landfall. These models were designed to work in a seamlessly pipeline. The study leveraged the NOAA’s HURDAT hurricane and the NOAA’s AVHRR Pathfinder oceanic conditions databases as training datasets for the TabTransformer models, implemented in PyTorch. While the combined system yielded mixed results, individual models demonstrated mixed to high accuracy. Despite these promising findings, the research identified areas for improvement, including refined data preprocessing and addressing possible sources of error. Future work aims to enhance model performance, generating more accurate and reliable hurricane predictions. This research’s results highlight the potential for further advancements in the use of transformers for weather forecasting applications and contribute to the growing body of literature on artificial intelligence in meteorology.
