This study conducted long-term observations using a volume scanning video disdrometer at five sites representative of the two contrasting climates in Japanese snow-covered areas to identify generalized patterns of solid precipitation particle size and velocity distributions (PSVDs). A novel framework combining the expectation-maximization (EM) algorithm and self-organizing maps successfully approximated complex distributions composed of multiple particle types through the superposition of generalized child distributions. This approach effectively eliminates the need for subjectively preconfiguring the number of child elements and mitigates the overfitting problem in standard EM algorithms. The child distributions were broadly classified into four groups: (S) a small and slow distribution serving as a fundamental component commonly present in all observed PSVDs; (A1) a large and slow distribution primarily corresponding to aggregated plate-type crystals; (A2) an intermediate distribution between A1 and G primarily corresponding to aggregated column-type crystals or their combinations with plate-type crystals; and (G) small and fast distributions corresponding to riming crystals. While the pattern consisting solely of S with low particle concentrations was the most frequent across all sites, superposition patterns of S+A1 and S+A2 (or solely A2) were the next most prevalent. These patterns were site-specific: S+A1 patterns predominated at sites facing the Sea of Japan, where precipitation is primarily driven by the winter monsoon, whereas S+A2 (or solely A2) patterns were more frequent at inland sites dominated by extratropical cyclones. As a generalizable framework, the findings of this study could potentially refine bulk microphysical schemes.