Bridging AI and conventional methods for comprehensive earthquake risk assessment in urban environments: Case study of Pohang, South Korea

초록

Urban areas are increasingly exposed to seismic hazards due to rapid urbanization, aging infrastructure, and growing population density. This study presents an integrated earthquake risk assessment for Pohang city, South Korea, by combining the conventional Risk-UE LM1 methodology with a data-driven Association Rule Learning (ARL) framework. Using a comprehensive citywide building inventory, the ARL model was employed to identify statistical relationships between key building attributes, including construction period, structural material, building height, and roof type, and the corresponding EMS-98 vulnerability classes. The derived vulnerability proxies were compared with classifications obtained from the Risk-UE LM1 approach to evaluate methodological consistency and reliability. The ARL-based model achieved an overall classification accuracy of approximately 73%, demonstrating its potential as a rapid and scalable alternative for urban-scale vulnerability characterization. The predicted vulnerability distributions were subsequently used to estimate earthquake damage under deterministic and probabilistic seismic scenarios representative of regional hazard conditions. Based on the predicted damage states, secondary impacts including human losses, economic losses, and debris generation were evaluated. Debris quantities were calculated using the standard HAZUS/FEMA methodology, which relates debris volume to structural damage ratios. The results indicate that older masonry and non-ductile reinforced concrete buildings concentrated in the historic urban districts exhibit the highest vulnerability to seismic damage. Under high-intensity scenarios, the analysis projects substantial impacts, including large volumes of debris, significant economic losses, and potential risks to human safety. While the debris estimation follows established empirical approaches, the principal contribution of this study lies in the application of ARL to improve the prediction of building vulnerability and damage distribution. The proposed framework demonstrates how data-driven techniques can complement conventional engineering models to support efficient urban earthquake risk assessment and resilience planning.

키워드