Analysis of uncertainties in ice crystal size distributions, their representation as gamma distributions, and bulk properties derived from in-situ measurements during ARM field campaigns

Description

Few studies have adequately characterized uncertainties in the in-situ observations that serve as the basis of model parameterizations and remote sensing retrievals. Consequently how such uncertainties cascade to model predicted and retrieved fields has not been well determined. In this study, data from ARM field campaigns are used to quantify uncertainties in derived ice crystal size distribution (SDs), including those associated with artifacts generated by shattering of large ice crystals on probe tips and with varying assumptions used to convert raw data into processed SDs. SDs measured by different probes and processed by different algorithms are compared here to identify any source of discrepancies in SDs. Closure studies whereby bulk properties (e.g., total water content TWC) derived from SDs are compared against those measured in-situ provide further insight into the uncertainties. Because many model and retrieval applications require representations of SDs as gamma functions, uncertainties in the slope, intercept and shape parameter describing the SDs are derived through comparison of parameters obtained from different algorithms and by assessing their sensitivity to the choice of tolerance allowed on fit parameters. Results show that quantities dominated by higher order moments (e.g., TWC, extinction, mass weighted terminal velocity, scattering properties) were less impacted by shattering than quantities dominated by lower order moments (e.g., number concentration), suggesting use of higher order moments derived from probes with standard tips is still appropriate. Although there are large discrepancies in SDs from different processing algorithms, the sources of these differences were identified (i.e., techniques identifying artifacts, assumptions about size-dependent depth of field, treatment of partially imaged particles). Using a tolerance in fit parameters determined from the statistical uncertainty of SDs, it is shown a three-dimensional volume in N0-lambda-u phase space is required to represent SDs; the parameters in this volume of equally realizable solutions can vary substantially, with N0, in particular, spanning several orders of magnitude: the co-dependence of parameters must be taken into account in the application of this volume. Ramifications of these findings for the development of model and retrieval parameterization schemes and for calculations of microphysical process rates are discussed.