An Approach for Evacuation Vulnerability Assessment with Consideration of Predicted Evacuation Time

Heavy rainfall is a frequent and widespread severe weather hazard that may cause flood damage and human casualties. Since heavy rainfall is a progressive disaster, its scale and hazardous areas can be foreseen beforehand. Therefore, evacuating people from hazardous buildings to shelters in advance is an efficient effort to reduce casualties, but a scientific basis is still required. This paper proposes an approach for assessing each building’s evacuation vulnerability based on predicted evacuation time, aiming to support evacuation decision-making under heavy rainfall. As such, this paper applies Dijkstra’s algorithm to find the evacuation route from each building to accessible shelters. Moreover, a prediction model based on the random forest algorithm is developed to estimate their time-varying evacuation time. Road spatial and temporal characteristics that may affect evacuation time are used when developing the model. Finally, the proposed approach is implemented in Joso City, Japan, to verify its feasibility. As a result, the proposed approach accurately predicts and visualizes the evacuation time between each building and its optimal evacuation shelter. It also visually identifies the hard-to-evacuate buildings. The results indicate that the proposed approach can effectively reflect evacuation vulnerability and support heavy rainfall evacuation decision-making, which proves its validity and practicality.