Integrated data imputation of national inventories for bridging information gaps and outcome risk uncertainty in community resilience modeling

Reliable natural hazard resilience modeling requires the systematic synthesis of building inventories; however, data scarcity and incomplete records often render public datasets unsuitable for quantitative risk analysis. This paper addresses these limitations by establishing a link between data gaps and uncertainty in risk metrics. We propose a data imputation approach that uses stratified Monte Carlo sampling to impute missing exposure attributes in public national and local datasets, leveraging statistical relationships from regional reference data. Multiple complete building-inventory realizations are generated to propagate uncertainty arising from missing exposure data into community-level risk metrics. At each realization, missing exposure attributes are probabilistically assigned, which in turn determines the structural classification of individual buildings and the corresponding fragility curves used for vulnerability assessment. Each imputed inventory, therefore, represents a distinct realization of the community’s vulnerability model, rather than simply a completed dataset. These realizations are propagated through a seismic risk workflow to quantify how uncertainty in fragility assignment induced by data gaps affects regional damage and loss estimates. The methodology is demonstrated and validated using three seismic real-world community testbeds by comparing risk outcomes from National Structure Inventory (NSI) data (incomplete) against tax assessor data (high-resolution). The testbeds include Seaside, OR, Shelby County, TN, and Salt Lake County, UT. The analysis shows that uncertainty in risk estimates stabilizes as the number of inventory realizations increases, while inadequate sampling leads to biased risk estimates. This framework allows practitioners to quantify the bounds of risk uncertainty caused by data incompleteness, providing a pathway to use national datasets when local high-resolution data is unavailable.