Bayesian Geospatial Analysis of Wind Damage to Residential Buildings During Hurricane Harvey

Abstract

Following severe winds, a good understanding of how various building-related factors interplay in determining the final damage state is essential to building restoration and reconstruction efforts. Recent advancements in data collection technology and data processing algorithms have significantly improved the efficiency of post-disaster damage assessment. However, few research studies have addressed the uncertainties associated with the damage assessment process (e.g., rapid changes in building damage status and inconsistencies in data collection). The various sources of uncertainty, if not modeled properly, will lead to unreliable damage assessment. To facilitate more accurate model representations and interpretations, this research quantifies uncertainties by incorporating geospatial correlation and parametric uncertainty through the framework of Bayesian generalized linear geostatistical models. Specifically, the integrated nested Laplace approximation (INLA) and stochastic partial differential equation (SPDE) are applied due to their computational efficiency. This research investigates wind damage to residential buildings in Key Allegro following Hurricane Harvey. A systematic post-disaster damage assessment is conducted for obtaining the true building damage condition and extracting detailed building attributes. Based on the processed data, different building damage scenarios are studied and evaluated to understand how building attributes influence building performance under severe winds. In summary, this research demonstrates the improved model performance in building damage estimations after quantifying uncertainties by incorporating geospatial correlation and parametric uncertainty. The improved model performance is expected to facilitate rapid post-disaster building damage assessment. Additionally, attributes that significantly impact building performance are summarized, and changes in relationships between building attributes under different damage scenarios are explained. These findings provide valuable information for home builders, insurance companies, and government agencies to support efficient rebuilding efforts following severe wind impacts.

Keywords: Wind damage, building damage assessment, uncertainty quantification, geospatial dependency, INLA-SPDE