A Comparison of RUC-derived and RAP-derived CIP and FIP Icing Products

This paper presents a comparison of icing analysis and forecast elements derived using input data from both the Rapid Update Cycle (RUC) weather prediction system and the newly-developed WRF Rapid Refresh (RAP) model. Since CIP and FIP algorithm output are valuable sources of supplemental information for in-flight icing conditions, the Aviation Weather Research Program's Quality Assessment Product Development Team (AWRP / QAPDT) was tasked with quantifying the degree of similarity and difference between RUC-derived and RAP-derived icing algorithm output prior to operational implementation of the new RAP model. An important goal is to assist the primary users of these products with understanding fundamental differences in icing algorithm output derived from the RUC and the RAP so that they can quickly and effectively adapt to using newly implemented icing products for making important decisions.

This paper summarizes results over a time period for which algorithm output was concurrently available from each model: 10-25 September 2011. Results show that output derived from the RAP model yields aggregate performance (for MOG severity) that is similar to that of the RUC model while providing more efficient use of the available airspace.

Additionally, using data obtained from this time period, the paper will describe analysis methods and present results to show:

For CIP– RAP production of icing is less than that of the RUC, especially at night and in mid-to-upper vertical levels. For CIP probability and severity, grid-to-grid agreement and correlation are highest in low-to-mid vertical levels. CIP SLD agreement is high throughout all vertical levels. While the RAP detection rate for MOG icing events is slightly lower than for the RUC, the RAP detection rate for non-icing events is slightly higher than for the RUC.

For FIP– RAP production of icing probability and severity is similar to that of the RUC, but RAP production of probability is greater at the higher probability values. RAP SLD production is much less than it is for the RUC, likely due to differences in the way the forecast models handle convection in the humid southeast. For FIP SLD, grid-to-grid agreement and correlation are poor at upper levels. As forecast lead time increases, so do differences in the amount of SLD produced by the two models. While the RAP detection rate for MOG icing events is slightly higher than for the RUC, the detection rate for non-icing events is nearly identical.