SNOWFALL REGIME CLASSIFICATION: APPLICATION OF MACHINE LEARNING CLASSIFIER TO PASSIVE MICROWAVE OBSERVATIONS

L. Milani^1,2, V. Petkovic¹

¹Earth System Science Interdisciplinary Center, University of Maryland, College Park, MD, ²NASA Goddard Space Flight Center, Greenbelt, MD

Several studies demonstrated the importance of snowfall regime identification when retrieving snowfall rate from Passive Microwave (PMW) observations. Different environmental conditions affect the ice microphysics consequently impacting the passive microwave response observed by satellite sensors. The availability of complete and representative datasets taking into account the specific snowfall regime characteristics is crucial to correctly detect and quantify snowfall from space. Within the Global Precipitation Measurement (GPM) mission, the Goddard PROFiling (GPROF) algorithm snowfall retrieval is investigated. A combined CloudSat-GPM dataset is used to build a training dataset for an eXtreme Gradient Boost (XGB) model in which the GPM Microwave Imager (GMI) brightness temperatures are associated with a Cloud Profiling Radar (CPR) snowfall regime, classifying the observed scene into ‘dry’ (no precipitation detected), ‘shallow convective’, ‘deep stratiform’ or ‘other’ snowfall class. The Machine Learning (ML) approach is crucial to interpret strong but complex relationships between PMW signals within the atmosphere and snowfall regimes at the surface. The ML classifier training is performed using a CloudSat classifying technique, based on snowing profiles and cloud classification, and applied to GPROF. A couple of case studies will be presented to show the benefits of classifying the snowfall regime for PMW snowfall retrievals.