From warning to action: qualitatively informed agent-based modelling (ABM) to strengthen flood early warning and evacuation

Accurate and timely flood warnings do not always translate into effective evacuations. This study integrates semi-structured interviews with stakeholders, from the federal and local government agencies to flood-affected residents, with a spatially explicit Agent-Based Model (ABM) to examine gaps in Flood Early Warning Systems (FEWS) and evacuation during the 2021 Taman Sri Muda flood in Shah Alam, Malaysia. Qualitative findings were systematically translated into model rules representing how agents receive warnings, perceive risk, interpret information, and decide when and where to evacuate as flood conditions evolve. The findings highlight three key challenges: (i) gaps in warning dissemination and coverage, (ii) a disconnect between technically accurate warnings and actionable community responses, and (iii) heterogeneous risk perceptions that influence compliance with evacuation advice. The model was subsequently used as a policy-testing environment to evaluate how variations in warning coverage, message actionability, response time, and shelter guidance affect evacuation outcomes in a dense urban setting. Simulation results indicate that improvements in message clarity, actionability, and timeliness, together with stronger public trust in FEWS, produce the greatest improvements in evacuation performance. The study demonstrates the value of integrating qualitative evidence with ABM to capture behavioural heterogeneity and evaluate people-centred flood-warning strategies before their implementation in practice.