Multifrequency retrieval of cloud ice particle size distributions

Abstract

There are many sources of uncertainty in remote sensing retrievals. This is particularly true where complex parameters such as liquid or ice hydrometeors must be retrieved. Many of the uncertainties are the direct result of assumptions made in the retrieval process to address the ill-posed nature of the inverse problem - namely that there are more variables than measurements. In this paper, an optimal estimation retrieval technique is applied to a multi-frequency data set from the Wakasa Bay AMSR-E validation experiment. First, airborne radar observations at 13.4, 35.6 and 94.9 GHz are integrated to retrieve all three parameters of a normalized gamma ice particle size distribution (PSD), N0*, μ, and Dm. This retrieved PSD was validated against the near-simultaneous coincident in situ cloud probe observations. The differences between the retrieved and in situ measured PSDs were explored through sensitivity analysis and the sources of uncertainty were found to be the ice particle density and the aspect ratio of the nonspherical particles modeled as oblate spheroids in the forward radiative transfer model. The optimal estimation technique was then applied to retrieve an optimal density and aspect ratio for the cloud under study through integration of the in situ and remote sensing observations. The optimal particle size-density relationship was found to be ρ(D) = 0.07*D-1.58 and the oblate spheroid aspect ratio was found to be 0.53. The use of these optimal values as improved assumptions in the PSD retrieval reduced the uncertainty in the retrieved reflectivity of the three radars from + /- 6 dB to + /- 2 dB. Next, the retrieval technique is expanded to include passive microwave observations and retrieve a full atmospheric column vertical hydrometeor profile. Eleven airborne passive microwave frequencies from 10.7 to 340 GHz are integrated with airborne radar observations at 13.4, 35.6 and 94.9 GHz to retrieve all three parameters of a normalized gamma ice particle size distribution (PSD): No*, μ, and Dm. The vertical profile retrieval is validated against a clear sky scene before being applied to the horizontal extent of an ice cloud. The PSD retrieval shows vertical structure consistent with cloud microphysical processes. The default density and shape retrieval is used as a baseline for comparison with the retrieval using the optimized model from the companion paper, which reveals an order of magnitude difference in ice water path between the two retrievals. This difference is explored and an information content analysis reveals that the optimized model improves on the information content of the retrieval by 287 more states resolved than the default model indicating a significant reduction in retrieval uncertainty.