Towards multi-variable tsunami damage modeling for coastal roads: Insights from the application of explainable machine learning to the 2011 Great East Japan Event

• Novel machine-learning approach for assessing tsunami-induced road damage. • Multi-variable nature of tsunami damage mechanisms to roads. • Traditional fragility functions may not fully capture complex damage mechanisms. • Geospatial analysis can support in retrieving features to enhance model accuracy. The accurate assessment of tsunami-induced damage to coastal roads is crucial for effective disaster risk management. Traditional approaches, reliant on univariate fragility functions, often fail to capture the complex interplay of variables influencing road damage during tsunami events. This study addresses this limitation by employing machine learning techniques on an extensive dataset compiled after the 2011 Great East Japan tsunami. The dataset, enriched with additional explicative variables accounting for the hydraulic features of the event and the physical characteristics at roads’ location, enables a comprehensive analysis of road damage mechanisms. Results indicate that while inundation depth remains a significant predictor, factors such as wave approach angle, road orientation and potential overflow from inland watercourses also play critical roles.