Spatiotemporal Dynamics of Operational Seismic Resilience: AI-Assisted Evacuation Modeling in Dense Mediterranean Urban Canyons
9th International Conference of Contemporary Affairs in Architecture and Urbanism, İstanbul, Turkey, 7 - 08 May 2026, pp.75, (Summary Text)
- Publication Type: Conference Paper / Summary Text
- City: İstanbul
- Country: Turkey
- Page Numbers: pp.75
- Dokuz Eylül University Affiliated: Yes
Abstract
Standard seismic evacuation models relying on static cadastral maps frequently overestimate the capacity of dense Mediterranean street canyons by ignoring dynamic street clutter. This study addresses this gap by introducing an AI-assisted workflow to quantify "phantom capacity" in İzmir, Türkiye. By coupling Monocular Depth Estimation with the Segment Anything Model (SAM 3), we transform 2D street-view imagery into metric 3D navigation meshes, simulating pedestrian flow under "social force" panic dynamics. Results indicate that unmodeled street furniture and temporal commercial occupation reduce effective evacuation flow rates by 40–60% during peak hours, generating critical micro-scale bottlenecks invisible to traditional GIS analysis. These findings challenge static urban resilience paradigms, demonstrating that the cultural vibrancy of the "living street" creates a measurable spatiotemporal trade-off with operational safety, necessitating dynamic, sensordriven digital twins for realistic risk assessment.