Deep-learning climate emulator ACE2 reveals a global decrease in tropical cyclone 5 frequency in the 15th Century under an El Niño-like sea surface temperature pattern

Mu-Ting Chien , Wenchang Yang , Eric D Maloney, Gabriel Vecchi, Elizabeth Barnes

The relatively short modern observational record limits our understanding of the relationship between global tropical cyclone (TC) frequency and sea surface temperature (SST), resulting in uncertain future TC projections. Using novel deep-learning-based past-millennium simulations with the Ai2 Climate Emulator version 2 (ACE2), we provide insight into the connections between SSTs and TCs. ACE2 simulations reveal that the Atlantic and global tropical cyclone numbers decreased in the 15th Century under an El Niño-like SST pattern, consistent with the high-resolution dynamical climate model HiRAM, and Atlantic sediment proxies. Global TC frequency decreased mainly due to stronger vertical wind shear and the drier midtroposphere as the Hadley circulation strengthened. Conversely, global TC frequency would increase under a La Niña-like SST pattern, as shown in idealized SST experiments with ACE2. This study demonstrates an exciting opportunity to examine TC-climate interactions using deep-learning-based paleoclimate simulations and highlights the need to constrain future projections of SST patterns to accurately estimate TC frequency.