Identifying determinants of evacuation reliability in flooded suburban railway stations: A large-scale virtual reality approach

This study investigates the determinants of evacuation reliability in flooded suburban railway stations using a large-scale virtual reality (VR) experiment. The researchers developed a VR simulation based on a typical underground station on Shanghai’s Airport Link Line, where 987 participants performed flood evacuation tasks and completed pre- and post-experiment surveys. The methodology included immersive VR scenarios with varied environmental stress, emergency broadcasting, and levels of staff guidance. Data were analyzed using binary logistic regression and an ensemble learning algorithm (LightGBM), combined with SHAP (Shapley Additive explanations) to assess global and local sensitivity. The study evaluated participants’ route choices and evacuation efficiency amid flooding, covering individual characteristics, spatial familiarity, and intervention effects.

The findings show that manual guidance by station staff is the key factor in improving evacuation reliability, with high staff deployment dramatically enhancing optimal route selection and reducing evacuation time and path length. Emergency broadcasts with detailed instructions further synergize with staff guidance, prompting effective wayfinding. Education level and spatial familiarity were significant internal factors, influencing participants’ speed and route accuracy. Environmental stress had minimal impact on decision reliability when robust guidance was present. The authors recommend prioritizing staff deployment and clear emergency broadcasting, especially for unfamiliar passengers, and suggest integrating VR and ensemble modeling into flood emergency planning. Future research should explore real-time sensor-driven digital twins, richer environmental variables, and collaborative evacuation behavior to refine strategies under complex, multi-hazard conditions.