Bayesian data integration framework for the development of component-level fragilities derived from multiple post-disaster datasets

Abstract

Widespread, disaster-related damage to buildings often results in severe economic, environmental, and societal impacts for the affected regions. The design of targeted mitigation strategies and policies to reduce disaster-related losses at community scale would benefit from building-specific, component-level loss estimations across regional building inventories under realistic hazard scenarios. However, such highly granular loss estimations face a unique challenge in sourcing reliable fragilities to project likely damage to specific buildings in a given inventory, especially for the case of wind-vulnerable structures. Notably, increased accessibility to post-disaster datasets, particularly those that standardize damage quantifications at the component-level, are quickly advancing the utility of this data for the development of observation-informed fragilities. In response, this paper offers a Bayesian data integration framework to guide the development of component-level fragilities derived from the fusion of multiple post-disaster datasets. This framework advances the utility of Bayesian approaches to fragility model updating for applications to high-granularity regional loss estimation by contributing subroutines that (1) generalize the process of sample building selection and (2) formalize the query and integration of heterogeneous open disaster data. These contributions then enable automated model updating of component-level fragilities, taking advantage of an object-oriented data model to facilitate efficient building data management across the framework’s subroutines. Such a Bayesian approach to fragility development not only allows for continuous refinement of model parameters as post-disaster data becomes available, but also creates opportunities for the framework to support fragility descriptions for classes of buildings that comprise smaller proportions of the building inventory, and are thus unlikely to have large sample sizes. The framework is applied to two case studies in the state of Florida, utilizing datasets produced after Hurricanes Michael and Irma to update fragility models for roof cover components in residential and institutional buildings. The case studies not only demonstrate the proposed framework’s rational approach to generating observation-informed fragilities in support of regional damage assessments, but also how this data-driven Bayesian approach can in turn inform hazard intensities and building classes to be targeted in data collection after future hurricanes to further refine the fragilities for use cases/damage measures/wind speeds that have limited observations.