Convolutional neural networks for specific and merged data sets of optical array probe images: compatibility of retrieved morphology-dependent size distributions

Abstract

This study addresses the challenges of ice particle morphology classification from images of optical array probes (OAPs) and proposes a more refined processing to enable better interpretation of observational data. The convolutional neural network methodology is applied to train classification tools for hydrometeor images from optical array probes. Two models were developed in a previous work for the Precipitation Imaging Probe (PIP) and 2D Stereo (2D-S). In addition, three new models are introduced in this study: one for the Cloud Imaging Probe (CIP), one for the High Volume Particle Spectrometer (HVPS), and a global model trained on a data set that merges all available data from the above four instruments. The methodology of retrieving morphology-specific size distributions from OAP data is provided. Size distributions for each morphological class, obtained using both the specific and global classification models, are compared for the data set of the ICE-GENESIS (Creating the next generation of 3D simulation means for icing) project, where all four probes were operated simultaneously. The reliability and coherence of these newly obtained machine learning classification tools are clearly demonstrated. The analysis shows significant advantages of using the global model compared to the specific ones. The presented methodology retrieves morphology-specific crystal size distributions that effectively allow systematic identification of microphysical growth processes from OAP data sets mainly collected on research aircraft during measurement campaigns. By combining the quantitative reliability of OAP-derived total number and mass size distributions with advanced machine learning morphological individual crystal classification, this approach establishes a foundation for investigations on past OAP data sets to be reinterpreted.