Residential building damage assessment and evaluation for the July 2021 flood in Westport, New Zealand

Reliable flood damage models are informed by detailed damage assessments. Damage models are a critical flood risk assessment component representing flood exposed object vulnerabilities. This study evaluates residential building flood damage information collected following the July 2021 flood in Westport, New Zealand. We report hazard, exposure and damage characteristics observed and recorded from 247 damaged buildings then develop and evaluate the predictive performance of six univariable and two multivariable damage models. Building characteristic homogeneity and component damage caused primarily from water depth resulted in similar relative damage prediction precision and reliability for different univariable models based on water depth above ground alone. Compared to univariable models the Random Forest algorithm improved precision by 10% when the most important explanatory damage variables for relative damage were considered. Our results suggest that building heterogeneity should be identified for exposed objects when determining univariable or multivariable damage model selection in flood risk assessment. Residential building damage homogeneity in Westport indicates the comparatively simpler univariable damage models could be applied in similar flood risk contexts.