Prediction and Interpretation of Low-Level Wind Shear Criticality Based on Its Altitude above Runway Level: Application of Bayesian Optimization–Ensemble Learning Classifiers and SHapley Additive exPlanations

Abstract

Low-level wind shear (LLWS) is a rare occurrence and yet poses a major hazard to the safety of aircraft. LLWS event occurrence within 800 feet of the runway level are dangerous to approaching and departing aircraft and must be accurately predicted. In this study, first the Bayesian Optimization–Ensemble Learning Classifiers (BO-ELCs) including Adaptive Boosting, Light Gradient Boosting Machine, Categorical Boosting, Extreme Gradient Boosting, and Random Forest were trained and tested using a dataset of 234 LLWS events extracted from pilot flight reports (PIREPS) and weather reports at Hong Kong International Airport. Afterward, the SHapley Additive exPlanations (SHAP) algorithm was utilized to interpret the best BO-ELC. Based on the testing set, the results revealed that the Bayesian Optimization–Random Forest Classifier outperformed the other BO-ELCs in accuracy (0.714), F1-score (0.713), AUC-ROC (0.76), and AUR-PRC (0.75). The SHAP analysis found that the hourly temperature, wind speed, and runway 07LA were the top three crucial factors. A high hourly temperature and a moderate-to-high wind speed made Runway 07LA vulnerable to the occurrence of critical LLWS events. This research was a first attempt to forecast the criticality of LLWS in airport runway vicinities and will assist civil aviation airport authorities in making timely flight operation decisions.

Keywords

• low-level wind shear
• ensemble learning classifiers
• Bayesian optimization
• SHapley Additive exPlanations