Spatial Risk Characterization and Influencing Factors of Aviation Accidents Under Extreme Weather Conditions

The increasing frequency of extreme weather events poses severe challenges to aviation safety. To systematically investigate the spatial association between aviation accidents and extreme weather, this study integrates aviation accident records with radar detection data on four categories of extreme weather, namely hail, mesocyclones, storm structure, and tornado vortex signatures. Specifically, through spatiotemporal matching, an integrated dataset linking accident locations with extreme weather events in their vicinity is constructed. First, a grid partitioning method is adopted to divide the study area into regular units. Subsequently, a nonlinear composite weighted risk index that integrates accident density, severity, and extreme weather exposure is developed to spatially characterize aviation accident risk at the grid level. Based on this spatially characterized risk quantification, a TPE-XGBoost machine learning model is further developed to explore the associations between extreme weather characteristics and spatial aviation accident risk, with model interpretability analysis employed to identify key contributing features. Finally, the results reveal spatial variations in aviation accident risk, with regional differences observed between coastal and inland areas. SHAP analysis indicates that extreme weather features contribute differently to model predictions, with weather intensity, spatial configuration, and interactions among meteorological variables showing varying degrees of contribution. These findings support regionally differentiated aviation safety management and spatial risk early warning.