Hybrid AI framework for comparative urban seismic vulnerability and risk assessment across multiple cities

초록

This study presents a hybrid AI-analytical framework for comparative seismic vulnerability and risk assessment across three urban systems with contrasting characteristics: Pohang (South Korea), Jammu (India), and Muscat (Oman). The methodology integrates association rule learning (ARL) with the Risk-UE LM1 model to enable robust classification of EMS-98 vulnerability classes and multi-dimensional loss estimation. A harmonized building inventory incorporating construction period, building height, structural material, and roof type was utilized to ensure cross-city consistency. The framework captured physical damage, human losses, economic losses, and debris generation. Results indicated distinct seismic risk signatures, with Jammu exhibiting fragility-dominated behavior, Muscat showing resilience at moderate intensities but nonlinear escalation at higher intensities, and Pohang demonstrating intermediate performance with improved class separability. Central urban zones contribute over 60-70% of total losses, highlighting the influence of urban density. The proposed approach enhances predictive reliability and scalability, offering a transferable tool for data-constrained regions and supporting risk-informed urban planning, resilience enhancement, and disaster mitigation strategies at a global scale.

키워드