On the Value of Time-Lag-Ensemble Averaging to Improve Numerical Model Predictions of Aircraft Icing Conditions

The High-Resolution Rapid Refresh (HRRR) model with its hourly updating cycles provides multiple weather forecasts valid at any given time. A logical combination of these individual deterministic forecasts is postulated to show more skill than any single forecast for predicting clouds containing supercooled liquid water (SLW), an aircraft icing threat. To examine the potential value of using multiple HRRR forecasts for icing prediction, a time-lag-ensemble (TLE) averaging method of combining a number of HRRR forecasts was implemented for a multiple month real-time test during the winter of 2016/17. The skills of individual HRRR and HRRR-TLE aircraft icing predictions were evaluated using icing pilot reports (PIREPs) and surface weather observations and compared with the operational Forecast Icing Product (FIP) using the Rapid Refresh (RAP) model. The HRRR-TLE was found to produce a higher capture rate of icing PIREPs and surface icing conditions of freezing drizzle or freezing rain than single deterministic HRRR forecasts. As a trade-off, the volume of airspace warned in HRRR-TLE increased, resulting in a higher false detection rate than in the deterministic HRRR forecasts. Overall, the HRRR-TLE had similar probability of detection and volume of airspace warned for icing as the operational FIP prediction for the icing probability of 25% or greater. Alternative techniques for composing TLE from multiple HRRR forecasts were tested in postseason rerun experiments. The rerun tests also included a comparison of the skills of HRRR and HRRR-TLE to the skills of RAP and RAP-TLE.