Verification of WAFC in-flight icing forecasts using satellite observations

In-flight icing constitutes a major hazard to aviation, and so it is vital to be able to forecast the risk of icing accurately. The Met Office is one of two World Area Forecast Centres (WAFC) to routinely produce and validate icing forecasts. The existing verification methodology of Bowyer and Gill (2018) evaluates each WAFC forecast against satellite-derived icing potential. However, the methodology currently evaluates the full Global forecast data, whereas satellite-derived icing potential is only available during daytime. At night, the presence of cloud is reported, and so only correct rejections and false alarms are possible during nocturnal hours.

We first present an extension to the existing verification methodology by restricting the analysis to only include daytime data and demonstrate that this significantly reduces the degree of over-forecasting previously reported. In addition, we examine the performance of the new WAFC icing severity forecasts and compare against the routine product, during Winter 2020/2021.

There are two major challenges when comparing forecast icing severity and forecast icing potential. The first is that we are comparing the potential for an icing event to occur with its predicted intensity. The second challenge is that the new severity forecasts are on a 0.25º grid, compared with 1.25º for icing potential.

We present results both under the assumption that moderate icing potential is the same as moderate icing severity, and as an independent comparison with a new satellite-derived icing severity product. We also test the sensitivity to the choice of verification grid by re-gridding to both 0.25º and 1.25º.

The results show that WAFC icing severity is over-predicted when compared with satellite-derived severity, especially over the tropics. However, icing events are likely to be too infrequent in the observations, and so the magnitude of the over-prediction is over-estimated. All forecasts show regional skill at predicting icing severity and the results are not sensitive to the choice of verification grid. However, the performance of the higher resolution icing severity forecast is likely to influenced by the double penalty problem.