Flexible discriminant techniques for forecasting clear-air turbulence

The problem of forecasting clear-air turbulence is addressed by an integrated approach which departs from the traditional solutions.

The current approach to the problem relies on the evaluation of single indicators derived from numerical weather model output, but this univariate strategy does not meet the standards agreed upon for effective forecast value.

An alternative is the use of multidimensional adaptive regression techniques coupled with flexible discriminant analysis (MARS, FDA, Neural Networks), that can enhance the predicting skills of the indices considered in isolation.

We analise records of turbulence encountered by aircraft over the continental United States over several days during the past winter, and evaluate the performance of our approach. Results are close to the standards required from an effective operational forecasting system.