AN APPLICATION OF VINE-BASED REGRESSION TO FLIGHT LANDING DATA

• 成果类型: Article
• 署名作者: Alnasser, Hassan; Czado, Claudia
• 署名单位: Technical University of Munich
• 刊物名称: ANNALS OF APPLIED STATISTICS
• ISSN/ISSBN: 1932-6157
• DOI: 10.1214/24-AOAS1997
• 发表日期: 2025
• 页码: 1623-1640
• 关键词: pair-copula constructions; bayesian network; accident risk; reliability

摘要

Runway overruns pose a significant challenge to aviation safety and represent the most common type of landing incident. Identifying the underlying factors contributing to runway overruns is vital for the development of effective prevention strategies. Since a few overruns are observed, the distance required for a pilot to achieve control of the aircraft, measured by the distance to decelerate to 80 knots, is used as a precursor. We propose vine copulabased distributional regression models for the distance to 80 knots, given a set of contributing factors, for the analysis of 711 flights sourced from the quick access recorder (QAR). This approach allows to accommodate nonlinear, non-Gaussian patterns observed in the data and to rank the impact of the contributing factors. Three vine-based regression models and two Gaussian benchmark models were investigated, revealing that D-vine regression model is the preferred model. This D-vine regression is then used to identify 41 high-risk flights, using estimated conditional probabilities for the distance to controllable speed exceeding a significant threshold. The analysis of the contributing factors for these flights reveals distinct marginal behaviors and dependent patterns that must be considered when designing risk prevention strategies.