Detecting Organization of Shallow Cumulus Clouds in the Central Pacific using Artificial Intelligence

Shallow cumulus clouds are the dominant cloud type over subtropical ocean regions, and account for substantial variability in climate change predictions. Because of their importance, much more work must be done to understand these clouds on the microphysical and mesoscale level. In this study, a machine-learning algorithm was developed for the purpose of detecting four prevalent cloud patterns in a large dataset to simplify identification of temporal and spatial trends in mesoscale-level patterns of cumulus clouds. A convolutional neural network was trained to identify these patterns on 2019 and 2020 June-July-August daily satellite images from NASA worldview covering a 10°x 20° area northeast of the Hawaiian islands. The model was then applied to daily JJA screen captures from 2000-2018. The neural network recorded the frequency of each detected pattern, and based on the proximity and breadth of formations, categorized a certain day as dominated by one of the four patterns. This data was compared with weather radar, rain gauge, and radiosonde data to verify the environmental profiles for each pattern type and identify correlations between the four convective organizations and subsequent weather over Hawaii. This algorithms increased capacity for image data processing facilitates streamlined detection of temporal and spatial trends in the mesoscale organization of shallow cumulus clouds and their contribution to global climate.