Clemson University
Theory-Informed Generative Agents for Human Behavioral Modeling in Disasters
Abstract
dc:description.abstractThis dissertation develops a theory-informed generative-agent framework for modeling human behavioral decisions in disasters. Existing flood and disaster preparedness models often emphasize physical hazards, infrastructure exposure, or statistical correlations, but they struggle to capture the heterogeneous and evolving choices households make. This limitation is especially important for climate-related hazards, where future damage depends not only on changes in rainfall, inundation, and urban development, but also on decentralized protective actions such as house elevation, flood insurance, evacuation, and early preparedness. The dissertation integrates two empirical studies: a flood-risk study in Charleston, South Carolina, and a household disaster-preparedness study across hurricane contexts. The first study develops a coupled framework that integrates cellular-automata urban development, HEC-RAS flood inundation outputs, empirical building-elevation data, and LLM-based generative agents that simulate homeowner elevation decisions. The study shows that flood damage is increasingly concentrated in low-risk areas, despite substantial elevation adoption in moderate-risk and high-risk areas, as defined by the FEMA-derived 100-year inundation-depth classification. It further demonstrates that more conservative elevation-policy threshold assumptions can reduce projected damage more consistently than subsidy-only or risk-perception-only interventions. The second study develops and evaluates a behavioral theory-informed LLM framework for predicting the timing of household disaster preparedness across hurricane contexts. The framework uses constructs from Protection Motivation Theory and Social Cognitive Theory to convert survey responses into structured cognitive representations and tests whether these representations improve transferability across Hurricane Florence, Hurricane Harvey, and Hurricane Michael relative to linear regression and transfer-learning baselines. Across both studies, the central contribution is a shift from hazard-centric prediction toward socio-environmental modeling in which households are represented as adaptive agents whose decisions reshape risk over time. The dissertation contributes an integrative account of behavior-informed disaster modeling and shows how LLM-based agents can support policy evaluation, risk communication, and adaptive planning under changing climate and urban conditions.
Degree
thesis:*- Name thesis:degree_name
- Doctor of Philosophy (PhD)
- Level thesis:degree_level
- Dissertation
- Discipline thesis:degree_discipline
- Civil Engineering
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Lu, Liming
- Contributors dc:contributor
-
- Professor Chao Fan
- Professor Yuyuan “Lance” Ouyang
- Professor Chung-Yi Lin
- Professor M.Z. Naser
- Professor Siyu Huang
Subjects
dc:subject × 10Identifiers
dc:identifier.*- Repository record dc:identifier
- https://open.clemson.edu/all_dissertations/4339