Improving active remote sensing retrievals of snowfall at microwave wavelengths: an emphasis on the global precipitation measurement mission’s dual-frequency precipitation radar

Even though snowfall at the surface is often constrained to higher latitudes or altitudes, the contribution of solid-phase hydrometeors to the hydrologic cycle is not trivial and can be related to more than 50% of all surface rain events. Furthermore, the quantification of snow and ice in the atmospheric column is required to understand the Earth’s outgoing thermal radiative budget. Thus, the retrieval of snowfall from spaceborne radars that can sample remote regions of the world is invaluable for both atmospheric and climate sciences. One spaceborne radar capable of measuring snowfall is the Global Precipitation Measurement mission’s Dual-frequency Precipitation Radar (GPM-DPR). Initial evaluations of the retrieval of near-surface snowfall from GPM-DPR against the common global snowfall reference (i.e., CloudSat) showed large discrepancies between the two radar retrieval estimates. The large discrepancy between the CloudSat and GPM-DPR snowfall retrieval served as the main motivation for the work conducted here. Three tasks were formulated and conducted in this dissertation: (1) Evaluate the assumptions within the current GPM-DPR retrieval of snowfall; (2) Create an alternative retrieval for GPM-DPR; (3) Compare the new retrieval to the old retrieval methods. Task 1 is found in Chapter 2, Task 2 is in Chapter 3 and Task 3 is in Chapter 4. For Task 1, the investigation of ground-based measurements of both rain and snow and their particle size distributions allowed for the assessment of the main microphysical assumption of the GPM-DPR retrieval, which assumes that all hydrometeors obey the same empirical relationship between the precipitation rate (R) and the mass-weighted mean diameter (D_m). Rainfall observations showed that the default R-D_m relation for rainfall is plausible and shows general consistency with a Pearson ρ correlation coefficient of 0.63. However, snowfall observations showed that the R-D_m relation does not apply well for snowfall resulting in the underestimation of R. Furthermore, the low correlation between the log⁡〖(R〗) and D_m (ρ=0.23) suggests that an R-D_m retrieval is not optimal for snowfall retrievals and other methods should be explored. Motivated from the results of Task 1, an alternative retrieval for GPM-DPR was designed in Task 2 using a neural network, state-of-the-art particle scattering models and measured particle size distributions. The main result from Task 2 is that the neural network retrieval significantly improves (p