Machine Learning Approaches to Identifying Tropical Waves That Develop into Hurricanes

African Easterly Waves (AEWs) are synoptic-scale atmospheric disturbances that serve as precursors to tropical cyclones (TCs) in the North Atlantic and North Africa. As climate changes, TC activities are increasingly frequent, leading to exponentially growing socio-economic losses. So understanding the physical mechanisms governing the tropical cyclogenesis (TCG) of AEWs remains a crucial problem. Competing theoretical frameworks, including baroclinic instability, barotropic instability, and moisture-vortex instability (MVI) have been proposed, but their relative importance and temporal evolution during storm development remain unclear. In this study, machine learning algorithms are used to empirically analyze the governing mechanisms of AEW development based on 40 years of reanalysis data (1979-2018). We develop a computer vision framework utilizing convolutional neural networks (CNNs) and transformer architectures to identify developing AEWs (DAEWs) from non-developing AEWs (NDAEWs) based on wave-centered composites of key thermodynamic and dynamic variables for storm development. The model results suggest that the MVI framework is a critical factor for high classification accuracy in distinguishing developers from non-developers.