Individualizing an Agent-Based Flood Evacuation Model from Survey Data

Earlier agent-based models (ABMs) of flood evacuation by the author, first on an abstract grid and later extended to real 3D city-model geography, use a survey of Japanese residents to assign each simulated agent to one of a few discrete behavioral groups; within a group, the evacuation probability P and the neighbor-awareness radius L are, with one narrow exception, fixed at a single value shared by every agent. The same survey contains two individually answered questions – how much a respondent cares about neighbors’ evacuation status, and how many neighboring houses’ worth of information they want. The first was never used in that prior work, and the second only once, in isolation, on an abstract grid; both are natural candidates for individualizing P and L, respectively. This paper extends the model in four ways: whole respondent records are resampled, preserving the real correlation among a respondent’s evacuation threshold, concern level, and preferred radius; the preferred-neighbor count is converted into a metric radius using the actual local building density from real PLATEAU 3D city-model data for Toyohashi City; the persistent Evacuate-Now-Button regime and a direct-observation regime are unified within one model; and a real-geometry video visualization is added. Re-deriving the group distribution from the survey matches the previously published values to within 1.5 percentage points, and the observation regime reproduces the prior “no-system” rate (9.6% vs. ≈9%) as an emergent consequence rather than a calibrated target; an ablation shows that individualizing P alone, under an assumed concern-to-probability mapping, shifts the final evacuation rate by nearly 18 points (79.9% vs. 62.0%). Throughout, the parameters that remain researcher-declared assumptions are distinguished from those derived from data.

@article{Fukuda2026,
  title     = {Individualizing an Agent-Based Flood Evacuation Model from Survey Data},
  journal   = {International Journal of Advanced Computer Science and Applications},
  volume    = {17},
  number    = {9},
  year      = {2026},
  publisher = {The Science and Information Organization},
  author    = {Tatsuki Fukuda},
  doi       = {10.14569/IJACSA.2026.0170985},
  url       = {https://doi.org/10.14569/IJACSA.2026.0170985}
}