Physics-Informed Stochastic Modeling of Tornado-Induced Building Losses

Tornadoes are significant threats to the built environment, cascading into substantial economic losses and social disruption in hazard-prone communities. Understanding and quantifying these impacts requires robust modeling approaches; however, existing tornado loss models often lack the ability to generalize across diverse building configurations or fail to incorporate a physics-based representation of component-level damage progression. This study introduces a probabilistic framework for tornado-induced loss assessment of 1-story residential buildings with simple gable and hip roofs. The methodology integrates a standards tornado provision for wind force calculation with Monte Carlo simulations to evaluate sequential failures of building components. This stochastic framework incorporates variability in component capacities and demands, ensuring that each realization reflects potential randomness in building performance. Losses are mapped through probabilistic room layouts and cost functions, and aggregated results yield loss curves that represent the full distribution of possible outcomes rather than a single deterministic trajectory. To improve efficiency, adaptive sampling is employed, targeting input regions with the greatest influence on results, reducing redundant simulations while preserving accuracy. Input–output evaluations are conducted through collinearity diagnostics, bivariate exploration, multivariate pattern discovery, and predictive modeling to assess the influence of physical input parameters within the model. Finally, empirical calibration is performed using damage state thresholds to represent realistic loss conditions. This framework establishes a generalized basis for tornado loss modeling by linking physics-based component damage, probabilistic loss estimation, and data-driven interpretability, supporting both engineering design decisions and community resilience assessments.