Snowflake Models for Ice Microphysics Retrievals with Multi-frequency Radars

Microphysical retrievals of cloud and precipitation properties by radar requires understanding of the structure of the scattering hydrometeors and their interaction with the electromagnetic pulse from the radar. Unlike liquid hydrometeors, which assume near-spherical shapes, atmospheric ice particles have a complex morphology driven by several different processes, such as deposition growth, aggregation and riming. Thus, the interpretation of radar signals from snow can be expected to be more complex than that of radar returns from liquid water. Indeed, studies over the last decade have shown that, in particular with millimeter-wavelengths, detailed snowflake models are needed to accurately model radar signals from snowfall. These models need to be coupled to numerical scattering simulation methods that can handle arbitrarily shaped scatterers. Of these, the discrete dipole approximation (DDA) is the most popular.

The large variety of different types of snowflakes means that a retrieval algorithm must select the correct model from many possibilities. It is not always immediately obvious which model is appropriate for a given case. In order to facilitate the selection of the correct snowflake shapes, we have created a new dataset of snowflakes that is based on a cloud model driven by observational data. The snowflake shape models are created using a 3D snowflake generation program, which in turn is controlled by the output from a single-column cloud model that simulates ice microphysics by tracking Lagrangian particles called "simulated ice particles" (SIPs). The deposition growth, aggregation and riming of the SIPs can be recorded and used to drive the 3D snowflake generation model. Inputs to the single-column model are driven by CloudSat observations of surface snowfall rate and the corresponding thermodynamical profiles from the ECMWF model. The model uses the ECMWF profile as a starting point and is then tuned to give the correct snowfall rate at the surface. With this method, we can generate snowflake models that correspond more closely to natural snowflakes than ones from previous studies. The radar and radiometer scattering properties of these can be computed with DDA for use in snowfall retrieval algorithms.

Melting snowflakes present an even greater challenge to particle models. In addition to all the complexity of ice particles, the melting on the ice surface is driven by heat transfer from the surrounding atmosphere, and the behavior of the meltwater is controlled by surface tension of the liquid water on the ice. This leads to considerable inhomogeneity of the liquid water distribution especially in the initial stage of melting.

To improve the understanding of snowflake melting and the corresponding radar signals, we have created a snowflake melting model using the Smoothed Particle Hydrodynamics (SPH) method. This fully Lagrangian framework for fluid dynamics simulation is well suited for modeling the melting behavior of liquid water on ice. Surface tension is simulated using pairwise particle forces, and heat transfer from the environment to the snowflake is implemented using a Monte Carlo simulation. This can reproduce the key features of single-particle melting phenomena. Melting is observed starting at the extremities of snowflakes; after this, meltwater is pulled into the concave regions of the snowflake surface. Once these regions are filled with liquid water, the remaining meltwater forms a liquid layer around the snowflake surface. Eventually, the water forms a near-spherical shell around an ice core.

The melting model can use the results of the 3D snowflake generation model as an input, so it can simulate the differences in the melting behavior between different types of particles. For example, the model predicts that breakup occurs more often in unrimed snowflakes than in rimed ones. Since the melting model only simulates single-particle behavior, it does not currently reproduce phenomena such as aggregation of melting particles or collisional breakup.

The SPH particle output can be converted quite easily for use in DDA, and thus we expect this model to also improve the understanding of radar signals from the melting layer, in particular the nature of the well-known brightband feature exhibited by this region of the precipitation column.