To predict aircraft icing conditions, the Current Icing Potential (CIP) is an expert system model that uses observations and NWP output to infer information about cloud physics and behavior. Statistical models have been used to improve the model performance. These statistical models have in the past used pilots' reports as predictands. For many reasons, PIREPs are not an ideal dataset to measure icing. Most notably, they rely on the pilots subjective appraisal of icing conditions. Icing information collected by research aircraft is more consistent and of higher quality than PIREPs. In the winter of 2002 and 2003, the NASA Twin Otter was flown around the Cleveland area to collect data in icing environments. Instrumentation on this aircraft collects information on the occurrence of icing, droplet size and shape, and liquid and total water content. These data providean alternative to PIREPs for creating statistical models.

Two statistical models are discussed. The first uses PIREP data as predictands while the second uses icing information from the research aircraft. For both models, data fields produced by the CIP provide the predictors. These include meteorological parameters such as cloud cover, vertical velocity, temperature, relative humidity and icing potential. The two statistical models are used to create icing predictions at each of the CIP coordinates (in this case, the RUC40 pressure-level grid). The aircraft's position is matched to the nearest CIP coordinate.

Several comparisons between the two statistical models are presented, including relative operating characteristic (ROC) plots. To some degree, one would expect that the model created using PIREP data would match the PIREPs better, and the model based on aircraft data would model the aircraft data better. Differences in the performance of the two statistical models, as well as the original CIP algorithm, are summarized.