A global classification dataset of daytime and nighttime marine low-cloud mesoscale morphology based on deep-learning methods

Marine low clouds tend to organize into larger mesoscale patterns with distinct morphological appearances over the ocean, referred to as mesoscale morphology. While previous studies have mainly examined the fundamental characteristics and shortwave radiative effects of these mesoscale morphologies, their behaviour in the nighttime marine boundary layer (MBL) remains underexplored due to limited observations. To address this, we established a global classification dataset of daytime and nighttime marine low-cloud morphology using a deep residual network model and infrared radiance data of 1° × 1° resolution from the Moderate Resolution Imaging Spectroradiometer (MODIS), with machine-learning-retrieved all-day cloud optical thickness aiding in model training. We analysed day-night contrasts in climatology, seasonal cycles, and cloud properties of different cloud morphology types in this study. Results show that the relative frequency of occurrence of closed mesoscale cellular convection (MCC) increases significantly at night, while that of suppressed cumulus (Cu) shows a remarkable decrease. Disorganized MCC and clustered Cu display a slight frequency increase at night. In addition, solid stratus and three MCC types exhibit distinct seasonal variations, whereas two cumuliform types show no clear seasonal cycle. Our dataset extends the study of mesoscale cloud morphologies from daytime to nighttime, and the 1° × 1° resolution makes it a better match with other climate datasets. It will provide an important foundation for further research on the interactions between cloud morphology and climate processes. The final cloud classification dataset and the model development datasets are open-access and available at https://doi.org/10.5281/zenodo.13801408 (Wu et al., 2024).