Improving the Nowcast of Unstable Approaches

The approach and landing is a complex phase of flight, in which flight crews need to conduct a series of procedures in a condensed time frame to make the approach stable. Unstable approaches can lead to events with reduced safety margins and additional operating cost. To ensure flights have stable approaches, airlines have established stabilized approach criteria to make sure the landing flight is on track, within appropriate range of speed and rate of descent. If the criteria are not satisfied at stabilization altitudes, a go-around must be executed. Nowcasting the likelihood of unstable approaches before reaching the stabilization altitudes may assist flight crews in taking actions to correct the flight trajectory to avoid a potential unstable approach. A previous study shows the feasibility of detecting unstable approaches using historical trajectory data and the feasibility of nowcasting unstable approaches using state variables. This paper describes improvements in the nowcast performance through the modification of existing features and the addition of more features. For the Nowcast at 6 nm for unstable approaches after reaching 1000’ AGL, the accuracy was improved from 71.3% to 74.8%, the recall was improved from 60.8% to 64.7%, and the precision was improved from 74.0% to 78.4%. The improvements are discussed. Keywords—unstable approach; prediction; supervised learning