On the information content of remotely-sensed observations of snowfall

Ground-based and space-borne collocated active and passive remote sensing measurements are now commonly used to simultaneously retrieve cloud and precipitation properties. By necessity, most retrievals assume aspects of the particle size distribution and particle shape are known, and employ Gaussian least-squares-based techniques to obtain a solution. In truth, the characteristics of the solution space are largely unknown, uncertainty associated with variability in PSD characteristics is poorly constrained, and it is not clear that a unique retrieval solution exists.

Markov chain Monte Carlo (MCMC) methods can be used to produce a robust estimate of the probability distribution of a retrieved quantity for nonlinear forward models and non-Gaussian statistics. It can flexibly accommodate additional sources of uncertainty and changes in assumed error magnitudes. In this work, an MCMC algorithm is used to explore the error characteristics of a surface-based cloud property retrieval for a case of orographic snowfall observed during the Storm Peak Laboratory Cloud Property Validation Experiment (StormVEX). It is found that a combination of passive microwave with radar reflectivity and Doppler velocity may be sufficient to constrain the liquid and ice particle size distributions (PSDs), but only if (1) the functional form of the PSD is specified, (2) the unconstrained PSD parameters and mass-area dimensional relationships are specified, and (3) the hydrometeor shape is known. If the PSD parameters and mass-area dimensional relationships are allowed to vary realistically, it is not possible to retrieve unique liquid and ice particle size distributions.