Wednesday, 10 January 2018: 9:45 AM
Room 19AB (ACC) (Austin, Texas)
Aviation is sensitive to low-visibility conditions at airports because capacity-reducing procedures have to be taken to ensure aviation safety for arriving or departing aircraft. These low-visibility procedures have several ordered categories, which typically depend on horizontal and vertical visibility thresholds. As accurate probabilistic forecasts of these low-visibility categories are essential for economic air traffic regulation, we develop a probabilistic nowcasting system based on a statistical regression model. Generally, there is an excess of observations with good visibility and a relatively low number of observations falling into the low-visibility categories of main interest. To account for this characteristic, a statistical hurdle model is proposed where the first part (the "hurdle") estimates the probability of any visibility constraints (vs. good visibility). A second part estimates the extent of the low visibility, given that the "hurdle" is crossed, by using an ordered logistic regression. Subsequently, combining the probabilities from the first and the second model part leads to the final forecast. The hurdle model is tested in a nowcasting setup at Vienna International Airport for lead times up to 2 hours. It is based on meteorological point measurements from the airport and its vicinity. The hurdle model outperforms persistence, which is a benchmark for short lead times and also improves the forecasts of a basic ordered logistic regression model. It is even better than predictions from human forecasters at Vienna International Airport. The advantage of the hurdle model results from its ability to represent the low-visibility formation and dissipation processes and the processes responsible for the intensity of low-visibility events separately. Overall, the statistical hurdle framework is easy to apply, computationally fast, and provides accurate forecasts supporting the airport decision makers.
- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner