An agent-based model to simulate human responses to flash flood warnings for improving evacuation performance

Flash floods, as a common disaster, cause serious casualties and economic losses every year. Flash flood warnings have been taken as one of the most important ways to prevent flash food disasters through triggering evacuation. The flash flood warnings will not be effective without positive human responses to the warnings. However, few studies have been conducted to simulate the human responses to the flash flood warnings and find ways to improve the evacuation performance. As the response processes involve the interactions between social and natural systems, this study proposed an agent-based model (ABM) suitable for solving the complex system problem to simulate the human responses to the flash flood warnings. There are three sub-modules, namely early warning, social, and flood sub-modules, and their interactions in the ABM. The ABM was applied to Liulin Town, China, which experienced a severe flash flood disaster on August 12, 2021. The results highlight that the longer lead time for flash flood warnings in the early stages does not guarantee the better performance in evacuation. A low trust level in early warning information refrains residents from evacuating through making them tend to wait for the evacuation of others, especially when no pioneer evacuates. People-centered risk communication is more effective than the one-size-fits-all approach, even this top-down communication approach can reach more residents. And risk communication is more robust than engineering measures to reduce casualties under extreme rainstorms. These findings demonstrate that the proposed ABM is an effective tool to understand the human responses to the flash flood warnings for improving the evacuation performance.