Application of Bayesian Networks to Assess the Seismic Hazard for an Infrastructure System in the Central and Eastern United States

Abstract

Earthquakes are often considered by the public to be a hazard restricted to the Western United States (WUS). However, earthquakes occur in the Central and Eastern United States (CEUS) as observed with the August 23, 2011 Mineral, Virginia earthquake, which damaged approximately 600 residential properties, resulted in 200-300 million dollars in economic losses, and initiated a shutdown of the North Anna nuclear power plant (Horton et al., 2015). Since earthquakes occur more frequently in the WUS, most seismic research is performed to support the WUS tectonic regime. This is also true when developing methods to assess the seismic risk to infrastructure systems, and current practices involve performing a large number of simulations of potential earthquake events, ground motion fields, structural performance, and failure consequences. These simulations can require significant computational resources, and it may be difficult to convince stakeholders to assess the seismic risk of their infrastructure system in the CEUS since earthquakes occur less often and perceived risks are lower. However, this risk must be assessed, given the density and age of infrastructure in the CEUS. Additionally, ground motion attenuation is lower in the region, so infrastructure distributed across greater distances may be impacted during an earthquake event. As a first step in developing a method that is tailored to assess system risk in the CEUS, this research proposes a Bayesian Network (BN) framework to estimate multi-site seismic hazards. Importantly, this framework utilizes existing products from a Probabilistic Seismic Hazard Analysis (PSHA), which reduces computational burdens and allows a user to incorporate the epistemic uncertainty characterized by experts as part of previously performed large-resource efforts. Additionally, the framework incorporates sources of hazard correlation between sites in a transparent and computationally tractable manner. An example problem is provided to validate this framework against a simulation that reflects the current state of practice in the WUS. Applications of the framework are then explored to assess when various input parameters may influence hazard results and identify when more or less resource-intensive assessments may be appropriate. This includes evaluating the impact of the ground motion within-event residual correlation and site separation distance. A scenario is also presented to illustrate how the BN can be used to make hazard-informed decisions in the context of the operation of two dams. The framework is then expanded to illustrate how failure modes can be characterized to understand system performance better. Since hazard correlation is an important aspect of the multi-site hazard, within-event residual correlation in the CEUS is also investigated. Empirical models are available to estimate ground motion within-event residual correlation in the WUS, but these may not be appropriate for the CEUS, given the lower attenuation. Earthquake recordings available from the NGAEast database (Goulet et al., 2014) and applicable CEUS ground motion models are used to calculate ground motion residuals. Correlation between the residuals at different sites is analyzed and compared against models developed for the WUS. Insights from this analysis and the proposed framework are provided to aid practitioners in assessing seismic risk for an infrastructure system in the CEUS.