Probabilistic validation of community-scale tornado damage models for historic unreinforced masonry buildings

Communities with rich cultural heritage are increasingly vulnerable to tornado hazards, yet the structural resilience of older unreinforced masonry (URM) buildings remains understudied. State-of-the-art platforms for predicting damage at a community scale, such as IN-CORE (Interdependent Networked Community Resilience Modeling Environment), use a single random wind speed on the Enhanced Fujita (EF) scale rather than a continuous spectrum, thereby introducing inherent stochasticity into model predictions. To bridge these gaps, we propose an analytical probabilistic reformulation of IN-CORE’s stochastic damage output. Rather than validating a single sampled damage state against deterministic field observations, we integrate damage-state probabilities across each EF wind-speed interval for each archetype and compare the resulting ordinal probability forecasts with observed damage using the Ranked Probability Skill Score (RPSS). We apply this framework to 185 older URM buildings affected by the 2020 Nashville and 2021 Quad-State tornadoes. As a secondary diagnostic step, Model Reliance (MR) and Mutual Information (MI) are used to examine which building and hazard attributes are associated with the measured discrepancies.