Capturing the diversity of mesoscale trade wind cumuli using complementary approaches from self-supervised deep learning

At the mesoscale, trade wind clouds organize with a wide variety of spatial arrangements, which influences their effect on Earth’s energy budget. Past studies used high-resolution satellite measurements and clustering/labeling techniques to classify trade wind clouds into distinct classes. However, these methods only capture a part of the observed organization variability. This work proposes an integrated framework using a continuous followed by discrete self-supervised deep learning approach based on cloud optical depth from geostationary satellite measurements. The neural network learns the semantics of cloud system structure and distribution, verified through visualizations of different layers. Our analysis compares classes defined by human labels with machine-identified classes, aiming to address the uncertainties and limitations of both approaches. Additionally, we illustrate a case study of sugar-to-flower transitions, a novel aspect not covered by existing methods.