Tsunami evacuation risk assessment and probabilistic sensitivity analysis using augmented sample-based approach

Abstract

Tsunami evacuation is an effective way to save lives from the near-field earthquake-induced tsunami. To accurately assess tsunami evacuation risk, various uncertainties in evacuation need to be considered. For risk mitigation, it is also important to identify critical parameters (or risk factors) that contribute more to the evacuation risk to guide more effective tsunami evacuation. Probabilistic sensitivity analysis can be used for the latter. However, both risk assessment and sensitivity analysis require a large number of model evaluations and entail significant computational challenges, especially for the expensive evacuation model. This paper proposes an efficient augmented sample-based approach to address the above challenges. It only requires one set of samples/simulations (hence the high efficiency) to estimate the evacuation risk and calculate the sensitivity measures for all uncertain parameters. The approach is applied to estimate the tsunami evacuation risk for Seaside, Oregon, where a novel agent-based tsunami evacuation model is used to simulate the evacuation process more realistically. Various uncertainties in the evacuation process are explicitly quantified by properly selected probability distribution models. Besides the evacuation risk, critical risk factors are identified using probabilistic sensitivity analysis. The results provide important insights on tsunami evacuation and critical information for guiding effective evacuation risk mitigation.