Constructing Deep Learning Datasets to Reveal Climate Trends of Tropical Cyclone Intensity and Structure Extremes

Climate change has been linked to tropical cyclone (TC) poleward migration, TC extreme precipitation, and an increased proportion of major hurricanes. Understanding past TC trends and variability is critical for projecting future TC impacts on human society. However, past trends in TC structure and energy remain uncertain due to limited observations; subjective analyses and inherent spatiotemporal heterogeneity of best-track data have led to reduced confidence in the assessed TC responses to the changing climate. This study uses deep learning to reconstruct past “observations”, constructing objective radial profiles of surface winds for global TCs during 1981–2020 and facilitating a comprehensive examination of TC intensity (V max ), structure, and integrated kinetic energy (IKE). The proposed model converts multichannel satellite imagery to a 0-750-km radial profile of axisymmetric surface winds; it is trained based on labeled data integrating best tracks and numerical model reanalysis of 2004–2018 TCs. Furthermore, the model performance is verified to be sufficient for climate analyses according to independent satellite-derived radar surface winds. The new homogenized dataset shows that, first, the major TC proportion has increased by ~13% in the past four decades, and second, the proportion of extreme-IKE TCs (IKE > 135 TJ, the 95 th percentile) has increased by ~25%, along with an increasing trend (> one standard deviation of the 40-y variability) of the mean IKE of high-IKE TCs (IKE > 79 TJ, the 80 th percentile). This new deep learning dataset reveals novel trends regarding TC structure extremes and may help verify climate simulations regarding TCs.