1.2 Discrimination Between Winter Precipitation Types Based on Spectral-bin Microphysical Modeling

Wednesday, 13 January 2016: 8:45 AM
Room 344 ( New Orleans Ernest N. Morial Convention Center)
Heather D. Reeves, CIMMS/Univ. of Oklahoma and NOAA/NSSL, Norman, OK; and A. Ryzhkov, J. Krause, and W. Bartolini

The prediction of the surface precipitation type is an important task for aviation safety, yet it remains a formidable challenge, particularly for freezing rain and ice pellets (FZRA and PL). Most modeling systems employ one or more postprocessing algorithms that attempt to parameterize the effects of melting and refreezing in order to discriminate between these classes, but to poor effect. Herein, a new classification algorithm is presented in which melting and refreezing are explicitly calculated for a spectrum of hydrometeors: the Spectral Bin Classifier (SBC). The SBC diagnoses six categories of precipitation: rain (RA), snow (SN), a RASN mix, FZRA, PL, and a FZRAPL mix. It works by calculating the liquid-water fraction of falling hydrometeors for a spectrum of sizes for a given temperature (T) and relative humidity (RH) at each model level from the cloud top to the surface. Demonstrations of the SBC output for individual T and RH profiles shows that it provides reasonable estimates of the liquid-water fraction of various-sized hydrometeors for the different categories of precipitation. Consideration of the plan views of precipitation type for select events shows the SBC faithfully represents the horizontal distribution of precipitation type in as much as the model analysis used to create the distribution is accurate. When applied to a collection of observed soundings, the PODs for FZRA and PL are considerably higher than what one obtains using other algorithms, but are still lower than for RA and SN. This difference is due to various uncertainty effects that are difficult to reconcile. These include uncertainty in the drop-size distribution, amount of riming, and ice nucleation temperature. Other sources of uncertainty, such as model and observational error, compound the problem further still. These effects on the detection of FZRA and PL are quantified herein and some thoughts on probabilistic forecasting are provided.
- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner