Investigating the Key Drivers of Hurricane Wind Damage in Commercial Buildings Using Causal Inference

Hurricanes are among the most destructive and costly natural-hazard related disasters. Post-hurricane field surveys provide crucial real-world observations of building damage and are key to better understanding relationships between structural characteristics and hurricane hazard intensity. However, most existing related studies and readily available datasets primarily focus on residential structures, such that a significant gap remains in the study of commercial building vulnerability to hurricanes. To address this limitation, we develop a dataset capturing wind-related damage caused by Hurricane Ian (2022) to commercial buildings. This dataset integrates property records, satellite and street-level imagery, post-event damage assessments, and estimated hurricane wind speeds, which are spatially linked at the individual building level. It covers commercial buildings in Lee County, Florida, one of the most severely impacted area by Hurricane Ian, and includes 344 unique building records.

Using this dataset, we investigate causal relationships between different building features and wind-induced damage, by employing the Double/Debiased Machine Learning (DML) causal inference framework. Results indicate that building shape, number of stories, roof cover material, building material, and roof shape are, in descending order, the most influential factors affecting damage. For example, buildings with an elongated rectangular shape are associated with an average increase of approximately 34 percentage points in the probability of damage.  In contrast, low-rise buildings are associated with an average reduction of approximately 25 percentage points in the probability of damage, relative to mid-rise buildings. These findings provide an important foundation for evaluating and improving hurricane wind vulnerability models and, therefore, hurricane catastrophe risk assessments.