Frequency biases in predictions of aviation icing occurrence: What can we learn from climatologies?

The prediction of locations where aircraft are susceptible to the accumulation of ice due to the presence of supercooled liquid water is an important part of the hazard forecasts provided to aviators. We describe a method for quantitatively assessing the performance of different icing algorithms by using of model-derived climatologies and comparing them to climatologies derived from observations. This method helps identify frequency biases in potential new icing algorithms and helps in their subsequent development prior to testing in an NWP setting. Our icing-climatology evaluation method is illustrated by comparing the frequency-of-occurrence biases of a new icing algorithm to those of the index used operationally. Using the notion that the actual liquid water content (LWC) encountered by aircraft can be described using an probability density function describing the fluctuations of the LWC about the model-predicted grid-box-mean value our new index quantifies the likelihood of encountering local supercooled liquid water contents above any threshold. Our evaluation using climatologies then shows that the new index can be tuned to give unbiased predictions of icing occurrence from a climatological point-of-view. The new index is shown to capture the variations in icing frequency as a function of month, height and geographical location as well as the control icing index. However, the new index has the advantage that its formulation allows for the development of bespoke icing indices that decouple icing severity and risk. Additionally, the new index can be interpreted as a probability, allowing it to be incorporated into an ensemble-based probabilistic forecast.