| Satellite and ground-based instrumentation, such as radar, lidar and radiometers, have the capability of remotely detecting water, graupel, ice and snow particles in the atmosphere. The detection and characterization of liquid water in the atmosphere benefits from the strong spherical symmetry and indistinguishability of droplets. This thesis focuses on issues concerning the detection of ice and snow particles. These particles are not spherically symmetric, may take preferred orientations and obey different particle size / number density relationships in a shape-dependent fashion. Early studies attempted to treat ice in a similar manner as liquid water but yielded poor results. Work within the past decade has focused extensively on examining purely shape-dependent behavior through the use of several scattering algorithms, notably the T-Matrix method (TMM) and the Discrete Dipole Approximation (DDA). The Discrete Dipole Approximation gives accurate results for complex aggregate particles, but has large memory and processing requirements, whereas the T-matrix method is much faster and uses less memory, but is generally restricted to surfaces of revolution or spherical aggregates.;This study examines the use of the T-matrix method as a proxy for DDA calculations for sector snowflakes, bullet rosettes and 6-point bullet rosette aggregates. Limits on the domain of interest are determined, and candidate mappings for T-matrix to DDA results are discussed. Mappings are nonlinear and are strongly dependent on effective medium refractive index, volume fraction, frequency and effective radius. Sector snowflakes are found to have a nearly 1:1 relationship between T-matrix and DDA results. Bullet rosettes and rosette aggregates in using the T-matrix approach yield scattering and differential backscatter cross-sections that are a factor of ten smaller than expected. For these flakes, low volume fractions of ice produce refractive indices near unity, making the T-matrix ellipsoid too similar to the ambient medium. However, the relationships with DDA results are still well-correlated over frequency and particle size. As such, conversions are found to be feasible for use in an operational product, possibly using a Bayesian PDF or neural net approach. |
