Comparison of epistemic uncertainty captured in probabilistic seismic hazard assessments for critical infrastructure

Site-specific probabilistic seismic hazard assessments (PSHAs) are fundamental for designing and evaluating critical infrastructure, e.g. nuclear power plants. However, PSHAs involve significant epistemic uncertainties. This study compares epistemic uncertainties from thirty site-specific PSHAs, half of which followed the Senior Seismic Hazard Analysis Committee (SSHAC) guidelines. Uncertainties were estimated for three intensity measures (IMs) at four annual frequencies of exceedances (AFEs), and for some sites, at both bedrock and target horizons. Four uncertainty metrics were used: mean-to-median ratio (R); log-ratio of the 84th and 16th fractiles (Unc); standard deviation of the fractiles assuming a lognormal distribution (σhaz); and a standard deviation of a R-based relationship (σµ). Results show strong correlations amongst Unc, σhaz and σµ. R correlates well with σµ but only moderately with the others, likely due to asymmetry in the fractile distribution. Unexpectedly, uncertainties are often constant across AFEs and ground-motion levels but vary across IMs. SSHAC level has limited impact on uncertainty levels, though non-SSHAC studies show greater dispersion. Results provide a benchmark for assessing appropriateness of uncertainty levels in PSHAs; however, the risk of ‘anchoring’ to previous studies must be considered. Routine estimation and documentation of total epistemic uncertainty are recommended, especially for critical infrastructure.