What Might a Probabilistic Forecast of Turbulence Look Like?

Turbulence forecasts, by their very nature, call for a probabilistic format. The size of turbulent eddies that affect aviation range from tens of meters to roughly 1 km (depending on the size of the aircraft). A model resolution of meters to a few tens of meters is thus required to fully resolve these eddies. The highest-resolution operational forecast model in the United States currently is the 3-km High-Resolution Rapid Refresh (HRRR) model. Consequently, turbulent eddies, themselves, cannot be forecast. Rather, turbulence is forecast according to the likelihood of the presence of turbulent eddies given the properties (i.e., divergence, stability, etc.) of larger-scale fields that are resolved by the models. In other words, the forecast is inherently probabilistic, which is then cast as a deterministic forecast for dissemination. In addition, because of the small spatial and temporal scale of the turbulent eddies relative to the forecast grids and the poor sampling of the majority of US airspace, aircraft may, by chance, fly through a turbulence region without encountering a significant turbulent eddy, thereby incorrectly penalizing the forecast.

However, strong turbulence occurs rarely (Allan Murphy defined extreme events as those with an event frequency less than 5%; moderate-or-greater turbulence occurs in less than 1% of in situ observations), so forecast values will be dominated by low probabilities that are both likely to be ignored by unsophisticated users and difficult to verify. It is thus necessary to develop strategies for communicating and evaluating probabilistic turbulence. This presentation will not fully develop those strategies, but it will attempt to explain the nature of the problem, making use of six years of in situ eddy dissipation rate (EDR) data, and provide suggestions for further investigation.