Abstract
Floods in an urban environment are often associated with casualties among pedestrians who move to reach a safe area, due to body stability issues. Pedestrian instability is a random event that depends on the interaction between individual movement and floodwater. Randomness is however often overlooked by current methods, which assess instability risk by deterministic models and provide simplified thresholds that decision-makers then use in emergency planning. Here, instability risk is estimated by a logistic regression model where the toppling probability is a function of water depth and flow speed. The proposal depends on parameters that can be rigorously estimated by experimental data using standard statistical methods. It can be straightforwardly extended to allow for multiple instability mechanisms and subject-specific biometrical information. It includes previous proposals as particular cases, providing a general framework where different thresholds can be compared. It therefore provides a novel rigorous probabilistic integration of instability mechanisms into a unified logistic regression framework, calibrated with experimental data, and applied at urban scale.