Preliminary results indicate that model errors are largest with active and intensifying weather systems such as troughs, low pressure systems, and jet streams; near the surface and regions of topography; the western edge of the geographical domain; and in gradient regions. For example, during the winter months for forecasts verifying at 12 UTC, peak errors of 5-10,000 ft occur daily in 3 h RUC and 6 h Eta forecasts of the freezing level height. These errors occur with only 1-3 C errors in model temperature sounding profiles in inversions aloft and larger temperature errors at and near the surface. This result suggests that large uncertainty exists in the forecasting of aviation weather phenomena, even when model errors are small. Diagnosis of cases of extreme aviation weather events suggests that model errors can: 1) be persistent with run time for a particular model, 2) be persistent with respect to a synoptic feature as it travels and evolves, and 3) be the same for an ensemble of forecast models (AVN, Eta, NGM, RUC, UK) over many forecast cycles. The error analyses are providing evidence of model biases that vary by model, vertical level, synoptic system, and changes in model formulation.
Initial results for aviation forecasting suggest that model uncertainty information is being applied by forecasters to: 1) support the choice of a model or blend of models, 2) support efforts to refine and correct model output in the case of large model errors that are persistent over forecast cycles and multiple models, 3) modify forecaster confidence in the occurrence of aviation weather phenomena, and 4) tailor the size and valid times of forecast areas in an effort to maximize forecast skill.