Snowflakes have very complex shapes, and modeling their properties requires vast computing power. We produced a large number of realistic snowflakes and modeled their average properties by leveraging their fractal structure. Our approach allows modeling the properties of big ensembles of snowflakes, taking into account their natural variability, at a much lower cost. This enables the usage of remote sensing instruments, such as radars, to monitor the evolution of clouds and precipitation.