A Comprehensive Machine Learning Study to Classify Precipitation Type over Land from Global Precipitation Measurement Microwave Imager (GPM-GMI) Measurements

Precipitation type is a key parameter used for better retrieval of precipitation characteristics as well to understand the cloud–convection–precipitation coupling processes. Ice crystals and water droplets inherently exhibit different in regimes (e.g., convection, stratiform), which reflect on satellite remote sensing measurements that help us distinguish them. The Global Measurement (GPM) Core Observatory’s microwave imager (GMI) dual-frequency radar (DPR) together provide ample information global characteristics. As an active sensor, DPR provides accurate assignment, while passive sensors such GMI are traditionally only empirical understanding regimes. Using collocated flags from “truth”, this paper employs machine learning (ML) models train test predictability accuracy using GMI-only observations with ancillary reanalysis surface emissivity products. Out six ML models, four simple ones (support vector machine, neural network, random forest, gradient boosting) 1-D convolutional network (CNN) model identified produce 90–94% prediction globally five types (convective, stratiform, mixture, no precipitation, other precipitation), much more robust than previous similar effort. One novelty work introduce data augmentation (subsampling bootstrapping) handle extremely unbalanced samples each category. A careful evaluation impact matrices demonstrates polarization difference (PD), brightness temperature (Tc) at high-frequency channels dominate decision process, consistent physical polarized radiative transfer over types, snow liquid clouds microphysical properties. Furthermore, view-angle dependency artifact DPR’s flag bears does not propagate into conical-viewing retrievals. This new promising way future physics-based algorithm development.