Over the past few decades, residential buildings along the coastal areas of the United States have suffered enormous structural damage and economic losses due to hurricane strikes. The significant variations in the building characteristics of residential buildings lead to distinctive building-level vulnerabilities under extreme winds. Therefore, an accurate representation of the building inventory is critical for quantifying regional hurricane risk.
In this dissertation, a regional wind risk assessment framework is developed to evaluate hurricane-induced structural damage and economic losses for residential communities. Unlike existing loss models that represent the building stock by archetype models with limited variations in building characteristics, the proposed framework applies site-specific risk assessments on every house in the region of interest based on parcel-based building inventories. A sensitivity analysis is conducted to investigate the effects of different building features on building vulnerability to identify the most critical features and explore the means of simplifying the building modeling process. To apply site-specific damage assessments at regional level, an automatic building modeling workflow is integrated into the framework, which is supported by property-specific characteristics extracted through machine learning-aided data collection approaches.
The framework is applied to residential communities in New Hanover County, North Carolina. Through site-specific risk assessments on 1,746 realistic building models, the overall variance in building-level damage and loss results among single-family houses is evaluated. The damage results reveal significant differences in wind vulnerability due to variations in architectural features. Furthermore, a comparative study shows that the aggregated regional loss calculated based on refined building models is substantially higher than that derived from building archetypes used in existing regional loss models. The building inventory generation and building modeling modules integrated into the framework largely reduce the inherent uncertainties of hurricane risk prediction. The high-resolution damage and loss results produced by the framework offer insights into local risk conditions, which facilitate the improvement of hazard risk mitigation and post-disaster management strategies.