Machine Learning Classification of Snowflakes to Enhance Microphysical and Scattering Characterization of Snow

No two snowflakes are alike. However, they can be classified into over a hundred different categories based on their geometry. Recent developments in machine learning algorithms have led to the possibility of automatically classifying more classes of snowflakes more accurately and efficiently. We present supervised and unsupervised learning approaches to snowflake classification coupled with dimensional reduction techniques. Proper automatic and detailed classification of ice and snow hydrometeors enables advanced characterization of geometrical, microphysical, and scattering properties of particles, which, in turn, is essential for development of radar-based quantitative precipitation estimation of snow.