Treatment of and sensitivity to epistemic uncertainty in seismic risk assessment of infrastructures

Seismic risk assessment of an infrastructure, intended as a system of systems including buildings, lifelines and critical facilities, is typically affected by several sources of uncertainty, classified as aleatory or epistemic. Momentum to this work came from the need not only to properly take into account all these uncertainties, but also to provide the confidence in the estimate and quantify the contribution of the employed models and parameters to the total uncertainty. After a brief overview about treatment of and sensitivity to epistemic uncertainty, this paper focuses on some critical issues, advocates the use of parallel models arranged in a so-called logic tree, and demonstrates the applicability of (modified) ANOVA to evaluate sensitivity of the total variance in the risk to each component of the input epistemic uncertainty, with reference to a “synthetic” city composed of buildings and a water network. Results show how the methodology can give the analyst a clear indication on which models or parameters are the most influential and thus deserve increased “knowledge” in order to reduce the total epistemic uncertainty in the problem.