A method for assessing urban flood impacts by combining hydrodynamic models with entity semantic evolution

Urban flood impact assessment is crucial for effective disaster management. These impacts, driven by changes in entity semantics (function, status), are essential for understanding the true vulnerabilities of urban systems, while current assessment models suffer from low intelligence and poor accuracy in this regard. To address this challenge, this research proposes a novel framework integrating hydrodynamic simulation, knowledge graphs, and semantic reasoning with Large Language Models (LLMs). The framework simulates flood physical processes, combines urban entity knowledge graphs, and utilizes LLMs to automatically reason the dynamic evolution of functional attributes and cross-domain impact chains. Key results demonstrate that this method effectively identifies both known and unknown cascade impact pathways, with LLM reasoning showing high logical consistency (LC = 0.91) in cross-model validation, and automatically inferred impact chains rated as highly reasonable by experts (average score 3.99/5). By introducing dynamic analysis at the semantic level, this research significantly enhances the automation and depth of flood impact assessment, providing a powerful tool for revealing potential functional vulnerabilities in urban systems, which has important implications for improving disaster resilience.