Linking Building Attributes and Tornado Vulnerability Using a Logistic Regression Model

Abstract

The residential building stock in the United States suffers a disproportionately large amount of tornado damage, which is linked to vulnerability of the building and its components. Whereas the building stock exhibits a wide diversity in ages, construction type, sizes, and other building attributes, it is unclear to what extent such attributes impact the vulnerability of residential buildings to extreme wind hazards. This study relates physical damage levels observed in single-family residential structures in the aftermath of the 211 Joplin, MO tornado with common building attributes and estimates of the peak wind speed that occurred at each home location during the tornado. The building attributes considered include year built, living area, appraised value, and number of stories. Peak wind speed estimates for each building are obtained from a near-surface wind field model of the Joplin tornado conditioned to tree-fall patterns. The influences of wind speed and building attributes on the observed damage levels to residential buildings are quantified using a multinomial logistic regression model for ordinal response, specifically the proportional odds model. With all else being equal, the study finds that in Joplin, newer homes, and homes with lower value per unit living area increased the likelihood of experiencing increased tornado damage. The number of stories weakly correlated with increasing likelihood of experiencing higher levels of tornado damage. Ultimately, the study presents a procedure for assessing the relative influence of common building attributes or other factors contributing to tornado damage vulnerability.