Case study 02 · Coupled simulation
FLOODABM
A 52,141-household agent-based model coupled to catastrophe, insurance, and financial mechanics across 27 New Jersey census tracts.
- 52,141 households
- 27 census tracts
- 2011–2023
- 50 stochastic runs per scenario
Why this problem matters
Flood adaptation decisions change losses unevenly across households. A useful model must connect behavioral pathways to hazard, property, insurance, and out-of-pocket outcomes over time.
How I approached it
I designed and fielded the 937-household survey, translated pathways into agent parameters, built the household simulation, and coupled it to catastrophe and NFIP mechanics.
What it contributes
Coupling behavior to physical and financial feedback makes assumptions auditable and exposes where a seemingly reasonable action creates uneven consequences.
For AI teams, FLOODABM is a stateful evaluation environment: decisions update an external system, consequences accumulate, and model assumptions remain visible.
System artifact
Inspect the research logic
The AGU 2025 poster and archived repository document the coupled architecture and aggregate outcomes.
Evidence enters. Decisions stay inspectable.
- Agent decision
- Install a temporary barrier
- Contents protection
- Not applied
- Hazard level
- 3Unchanged by household action
- Resource tokens
- 6
A homeowner can protect the building; the river hazard stays unchanged.
Levels and tokens are synthetic units; not FLOODABM outputs.
Interactive controls require JavaScript. The complete flow and sources remain below.
Evaluation
- Human evidence
Measured constructs and reported choices establish the empirical reference.
Published evidence - Context
A label-blind synthetic profile carries only approved attributes into the prompt.
Illustrative example - LLM proposal
Repeated model runs produce a choice and a concise rationale for comparison.
Illustrative example - Validation / repair
Checks compare direction, subgroup behavior, and run-to-run stability without exposing respondent records.
Public artifact - Consequence
The result becomes an evaluation finding, not a claim that the model represents a person.
Illustrative example
Governance
- Human evidence
Behavioral and physical constraints define what a plausible action must respect.
Published evidence - Context
A synthetic agent state limits the actions the model may propose.
Illustrative example - LLM proposal
The LLM returns a structured proposal rather than mutating the simulation directly.
Illustrative example - Validation / repair
Validators reject invalid fields, request targeted repair, and retain an audit trace.
Public artifact - Consequence
Only an accepted proposal updates the coupled model state.
Illustrative example
Simulation
- Human evidence
Survey evidence and public flood-risk sources ground household behavior.
Published evidence - Context
Household tenure, risk perception, resources, and exposure form the decision context.
Published evidence - Agent decision
Agents choose adaptation, insurance, or no action under modeled constraints.
Public artifact - Validation / repair
Model checks keep behavioral, financial, and physical state transitions consistent.
Public artifact - Consequence
Flood losses and adaptation alter the household and environmental state for the next cycle.
Published evidence
Research context and method
Illustrative system architecture · not a reported study result
Human track
- 01Perceive risk
- 02Decide and adapt
- 03Update resources
Environment track
- 01Hazard state
- 02Damage and loss
- 03Next environment
Outcomes feed back into later perception, capacity, and exposure.
- 01Household state
- 02Adaptation choice
- 03Flood event
- 04Damage, payout, and out-of-pocket update
- 05Next annual step
What changes in this lensOwner state includes property adaptation, insurance, damage, payout, and out-of-pocket feedback.
Method in detail
- Survey-ground behavioral parameters through SEM and Bayesian calibration.
- Advance household adaptation and insurance decisions over annual time steps.
- Couple agents to flood hazard, property damage, NFIP premium/payout/deductible, and financial state.
- Run 50 stochastic realizations for each scenario and report aggregate distributions.
Validation and limitations
Validation
- Survey-grounded behavioral parameters
- Scenario ensembles
- Multi-level diagnostics
- Archived code and citation metadata
Limitations
- Calibration and survey evidence are place- and period-specific.
- Household agents simplify decision processes and social context.
- Hazard and financial outputs inherit uncertainty from coupled models.
- Scenario results should not be treated as forecasts for an individual household.
Evidence slice
From question to consequence
- 01Problem
- Individual adaptation choices can change long-run flood losses, insurance, and household capacity.
- 02My role
- I developed and tested the household decision layer and coupled it to catastrophe-model state transitions.
- 03Input
- Survey-grounded behavior, household context, flood hazard, damage, insurance, and financial state.
- 04Decision
- Households evaluate protection, insurance, or no action under tenure and resource constraints.
- 05Validation
- Behavioral calibration, state invariants, repeated scenarios, and comparison with public evidence.
- 06Consequence
- Each annual decision changes exposure, loss, and the resources available for the next cycle.
Contact
Build evaluations that connect model behavior to real evidence
Interested in evaluating AI decisions or building evidence-linked research systems? I welcome research and open-source collaboration, and Summer 2027 internship conversations.
More about my research path