Beyond field surveys: Understanding the role of 3D spatial attributes for data-driven blast vulnerability assessment of masonry buildings

Unreinforced masonry (URM) buildings are highly vulnerable to blast loads, yet traditional post-disaster assessments often fail to capture important geometric and spatial attributes that govern structural behavior. This study leverages high-resolution 3D digital modeling and machine learning (ML) to extract and evaluate the predictive power of spatial features across 2042 historic URM buildings damaged in the 2020 Beirut explosion. By integrating 3D-derived attributes, including building orientation, aspect ratio, and façade opening ratio, into a supervised ML framework, we achieved over 90% accuracy in damage prediction. Unlike prior studies based on simulations or small-scale experiments — typically focused on seismic loading — this work offers a novel, empirical analysis of geometric predictors at urban scale under real-world blast conditions. This analysis reveals that while features like roof type and cladding (often prioritized in cultural heritage documentation) show negligible predictive value, urban morphology and building geometry emerge as dominant drivers of blast vulnerability. These findings provide real-world, physics-driven guidance for computational simulations, highlighting the need to prioritize 3D geometric interactions rather than material properties alone in blast modeling. For post-disaster reconnaissance, the results advocate rethinking field protocols to document orientation, opening distribution, and urban shielding effects, attributes previously overlooked but now shown to govern structural resilience. This workflow shifts from descriptive damage inventories to predictive, data-driven vulnerability assessments.