Objectively Identifying Transverse Cirrus Bands in Tropical Cyclones using a Convolutional Neural Network

Transverse cirrus bands (TCBs) are bands of upper-level clouds that are regularly seen in
mesoscale and synoptic-scale weather systems. In tropical cyclones, their appearance has
been subjectively linked to intensification and the diurnal cycle, but these hypothesized
relationships have not been rigorously tested due to the difficulty of objectively identifying TCBs
in satellite images. This presentation describes a machine learning technique that successfully
identifies TCBs in imagery from the GOES-16 Advanced Baseline Imager. The technique uses a
convolutional neural network (CNN) that assigns a probability of each pixel in the image being
associated with a TCB. Using the CNN, a database of TCBs from 2018 to 2022 was developed
for the Atlantic basin. Statistics using this database will be presented, including the relationship
between TCBs and deep-layer vertical wind shear, intensity change, and the TC diurnal cycle.