FLOODABM — coupled ABM × catastrophe flood model

Problem

Flood-adaptation models typically treat households as rule-following automatons with behavior asserted rather than measured. Adaptation and insurance outcomes depend on how real owners and renters actually decide.

Why it matters

Policy conclusions and equity analyses drawn from uncalibrated behavior are conclusions about the modeler's assumptions. Grounding agent behavior in primary data is what makes the simulation's claims testable.

Approach

My role

Sole code owner; the companion manuscript's co-authors are paper collaborators, not committers. I also designed and fielded the survey that calibrates the model.

Method

A 937-household flood-adaptation survey (Passaic River Basin, NJ) feeds a Bayesian calibration pipeline; the calibrated agent-based model couples to a catastrophe flood model so that individual adaptation decisions and basin-scale losses interact in both directions.

What was built

A 52,141-household agent-based model across 27 census tracts with National Flood Insurance Program premium, payout, and deductible mechanics, tenure-differentiated adaptation actions, damage-to-threat-perception feedback, and income-normalized equity analysis of who bears flood losses.

Key challenge

Carrying measured psychology into simulation honestly — turning survey constructs into calibrated agent parameters without overfitting, and validating simulated losses against observed insurance-claims data rather than declaring plausibility.

Validation & limitations

Simulation outcomes validated against observed NFIP claims data (OpenFEMA), and a candid known-limitations register that names the model's silent-failure modes — published judgment, not marketing.

Results & status

Research prototype — archived companion code to a first-author manuscript under review. Zenodo-archived with citation metadata (CITATION.cff) and published seed lists.

Links

Transferable relevance

Academic The empirical anchor of the dissertation arc — instrument to inference to simulation in one chain.

Industry Catastrophe-risk vocabulary in working code — AEP curves, NFIP mechanics, loss validation — legible to risk-analytics employers.