Flood-pedestrian simulator: an agent-based modelling framework for urban evacuation planning

Agent-Based Modelling (ABM) is an increasingly used approach for characterisation of human behaviour in evacuation simulation modelling. ABM-based evacuation models used in flood emergency are developed mostly for vehicular scenarios at regional scale. Only a few models exist for simulating evacuations of on-foot pedestrians responding to floods in small and congested urban areas. These models do not include the heterogeneity and variability of individuals’ behaviour influenced by their dynamic interactions with the floodwater properties. This limitation is due to the modelling restrictions pertaining to the computational complexity and the modelling flexibility for agent characterisation. This PhD research has aimed to develop a new ABM-based pedestrian evacuation model that overcomes these challenges through an ABM platform called Flexible Large-scale Agent Modelling Environment for the Graphics Processing Units (FLAME GPU). To achieve this aim, a hydrodynamic model has been integrated into a pedestrian model within the FLAME GPU framework. The dynamic interactions between the flood and pedestrians have been formulated based on a number of behavioural rules driving the mobility states and way-finding decisions of individuals in and around the floodwaters as well as the local changes in the floodwater properties as a result of pedestrians’ crowding. These rules have been progressively improved and their added value has been explored systematically by diagnostically comparing the simulation results obtained from the base setup and the augmented version of the model applied to a synthetic test case. A real-world case study has been further used to specifically evaluate the added value of rules relating the individuals’ way-finding mechanism to various levels of flood-risk perception. The findings from this research have shown that increasing the level of pedestrians’ heterogeneity and the effect of pedestrians’ crowding on the floodwater hydrodynamics yield to a considerably different prediction of flood risk and evacuation time. Besides, accounting for pedestrians’ various levels of flood-risk perception has been found to be one determinant factor in the analysis of flood risk and evacuation time when there are multiple destinations. Finally, the sensitivity analysis on the simulation results have shown that the deviations in the simulation outcomes increases in line with the increase in the sophistication of human behavioural rules.