Multihazard Hurricane Fragility Model for Wood Structure Homes Considering Hazard Parameters and Building Attributes Interaction

Carol C Massarra,J Casey Dietrich,Carol J Friedland,Brian D Marx

doi:10.3389/fbuil.2020.00147

Abstract

Predicting building damage as a function of hurricane hazards, building attributes, and the interaction between hazard and building attributes is a key to understanding how significant interaction reflect variation hazard intensity effect on damage based on building attribute levels. This paper develops multi-hazard hurricane fragility models for wood structure homes considering interaction between hazard and building attributes. Fragility models are developed for ordered categorical damage states (DS) and binary collapse/no collapse. Exterior physical damage and building attributes from rapid assessment in coastal Mississippi following Hurricane Katrina (2005), high-resolution numerical hindcast hazard intensities from the Simulating WAves Nearshore and ADvanced CIRCulation (SWAN+ADCIRC) models, and base flood elevation values are used as model input. Leave-one-out cross-validation (LOOCV) is used to evaluate model prediction accuracy. Eleven and forty-nine combinations of global damage response variables and main explanatory variables, respectively, were investigated and evaluated. Of these models, one DS and one collapse model met the rejection criteria. These models were refitted considering interaction terms. Maximum 3-second gust wind speed and maximum significant wave height were found to be factors that significantly affect damage. Interaction between maximum significant wave height and number of stories was the significant interaction term for the DS and collapse models. For every 0.3 m (0.98 ft) increase in maximum significant wave height, the estimated odds of being in a higher damage state rather than lower damage state for DS model were found to be 1.95 times greater for one-story buildings rather than two-story buildings. For every 0.3 m (0.98 ft) increase in maximum significant wave height, the estimated odds of collapse were found to be 2.23 times greater for one-story buildings rather than two-story buildings. Model prediction accuracy was 84% and 91% for DS and collapse models, respectively. This paper does not consider the full hazard intensity experienced in Hurricane Katrina; rather, it focuses on single-family homes in a defined study area subjected to wind, wave, and storm surge hazards. Thus, the findings of this paper are not applicable for events with hazards that exceed those experienced in the study area, from which the models were derived.

Highlights

Data-based fragility models account for a range of variables (Nateghi et al, 2011; Pitilakis et al, 2014), model damage as a function of multiple hazard parameters and building attributes, consider variability in building and environmental attributes, and use field data to predict future damage and validate model performance
Data-based fragility models have been widely implemented to estimate the probability of collapse or being in or exceeding a specified damage state for buildings subjected to tsunami (e.g., Reese et al, 2007, 2011; Koshimura et al, 2009; Suppasri et al, 2012, 2013; Charvet et al, 2014a,b, 2015; Muhari et al, 2015) and earthquake (Porter et al, 2007; Tang et al, 2012; Lallemant et al, 2015)
Physical damage of one- and two-story woodframed residential buildings built on slab and elevated foundations in coastal areas was statistically modeled as a function of maximum 3-s gust wind speed, maximum significant wave height, maximum surge depth, maximum water speed, foundation type, number of stories, and two-factor interactions of hazard and building attribute variables

Summary

Introduction

Data-based fragility models account for a range of variables (Nateghi et al, 2011; Pitilakis et al, 2014), model damage as a function of multiple hazard parameters and building attributes, consider variability in building and environmental attributes, and use field data to predict future damage and validate model performance. If field data are representative of a range of hazard parameters, building attributes, and building damage data, data-based models will effectively predict damage, and identify variables that significantly contribute to damage. In addition to their simplicity, data-based models are more realistic than simulation-based models, as the models are developed based on observed data. The availability of data and presence of missing data due to severity of damage may be a common issue for these models. Not specific to building, Padgett et al (2012) empirically modeled damage to coastal bridges along the US Gulf Coast using multivariate logistic regression models, Reed et al (2016) developed a logit fragility model to predict damage for power systems, and Kameshwar and Padgett (2018) developed a wind buckling and storm surge flotation fragility models for oil storage tanks

Methods

Results

Conclusion