Application of Artificial Neural Networks on North Atlantic Tropical Cyclogenesis Potential Index in Climate Change

Abstract

Abstract A methodology using artificial neural networks is presented to project twenty-first-century changes in North Atlantic tropical cyclone (TC) genesis potential (GP) in a five-model ensemble of global climate models. Two types of neural networks—the self-organizing maps (SOMs) and the forward-feeding back-propagating neural networks (FBNNs)—were employed. This methodology is demonstrated to be a robust alternative to using GCM output directly for tropical cyclone projections, which generally require high-resolution simulations. By attributing the projected changes to the related environmental variables, Emanuel’s revised genesis potential index is used to measure the GP. Changes are identified in the first (P1) and second (P2) half of the twenty-first century. The early and late summer GP decreases in both the P1 and P2 periods over most of the eastern half of the basin and increases off the East Coast of the United States and the north coast of Venezuela during P1. The peak summer GP over the region of frequent TC genesis is projected to decrease more substantially in P1 than in P2. Vertical wind shear (850–200 hPa), temperature (600 hPa), and potential intensity are the most important controls of TC genesis in the North Atlantic basin (NAB) under the changing climate.