City-scale evacuation management in flood scenarios, implementation and comparison of a multi-agent based approach and a traffic assignment approach

Abstract

Flood evacuation models can provide an effective mechanism to analyse flood risks and evacuation response actions. In this study, practical applications of evacuation simulation models are presented to help examine flood-related evacuation scenarios. Simulation frameworks are demonstrated through the case studies of Brisbane City, Queensland, Australia, which has a long history of floods and has experienced major flooding events in 2011 and 2013. These case studies were investigated to demonstrate feasible applications of flood extent prediction, network bottleneck estimation, evacuees’ behaviour and shelter demand, which contribute to flood risk mitigation and evacuation planning. The proposed flood evacuation models are proven to help increase community resilience in at-risk areas in Brisbane. Effective flood emergency management needs an integrated operation of interacting with human and technological systems. In this study, firstly, spatial toolkits are employed to analyse the shelter assignment and routing strategies based on network calculations. Simulation results indicate that the nearest shelters and routing directions can be determined based on the unique location of each household. To analyse the temporal flood risks and identify dangerous areas, an inundation model is proposed to provide flooding information for a dynamic risk analysis study. Test results of the inundation model show that it is able to predict the flood inundation extent at an accuracy of 66.9% which is higher than or comparable with the existing studies. A large-scale inundation can affect the endangered areas progressively and the temporal aspect of the incident should be captured in evacuation planning. By integrating the simulated flood dynamics, a city-scale microscopic evacuation model is built through an agent-based approach; different flood stages associated with departure times and various behaviour rules are tested for the Brisbane evacuation scenarios. The inclusion of flood dynamics in the evacuation model is essential for identifying temporally critical locations in the network as it provides a perspective to observe the dynamic interactions between evacuees and floodwater. In the agent-based flood evacuation scenarios, more than 12,000 evacuees are simulated and less than 7% of evacuees were still moving towards shelters after 120 minutes since the evacuation started. Evacuees are more evenly distributed in the pre-determined six shelters when more complex behaviour such as evacuee density detection is considered. A static traffic assignment approach is also implemented towards simulating the urban evacuation scenarios. The results of the traffic assignment approach show a much less network clearance time compared to the agent-based approach, which reveals the difference between these two models in terms of input, modelling algorithms and output. Suggestions are provided for the choice of modelling tools based on this comparison analysis. Overall, the flood evacuation model enables explicit interactions among evacuees and between human response and floods to be captured as the flood incident evolves. It can be easily adapted to simulate a wide range of flood scenarios. Case studies of Brisbane have demonstrated the model’s capability to examine flood risks and support for flood evacuation planning.