Answers are grounded in this paper and cite their sources.
Natural disasters displace millions of people each year, and effective disaster response requires accurate models of population movement. This study presents DFlee, an agent-based model for flood-induced evacuation built on the Flee framework for conflict-driven displacement. DFlee introduces three key extensions: (i) dynamic integration of evolving, location-specific flood conditions into agent decision-making; (ii) risk-sensitive destination selection accounting for flood severity, shelter inundation risk, and accessibility; and (iii) recession-driven short-term return-to-origin. Synthetic scenarios with recurrence intervals ranging from 10 to over 100 years represent minor and major riverine floods, coastal storm surges, and compound flooding. Validation combines empirical comparison against the 2022 Bauchi flood in Nigeria with behavioural validation of three stylised facts: preference for nearby safe destinations, short-term return to origin as flood conditions improve, and heterogeneous evacuation timing. Across scenarios, approximately 84–91% of evacuees select the nearest safe destination, while returns to origin begin promptly as flood conditions recede, with cumulative returns within the simulated window limited by late-peaking displacement dynamics. These behaviours persist across a tenfold increase in population size, and ARCHER2 experiments simulate 16 million agents in under one minute on a single compute node and around five seconds on 32 nodes, supporting uncertainty quantification, scenario exploration, and humanitarian planning.
No takes yet. Share an insight, caveat, or question.
Jahani et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: