Natural Hazards Research Summit 2022: Efficient propagation of aleatoric uncertainties in the hazard description using stochastic emulation

One of the most important steps in hazard estimation and consequently in risk assessment efforts is the accurate and detailed description of the hazard itself. This task entails the propagation of all known sources of uncertainty mainly by using a numerical model, and can be distinguished between two main categories: the parametric and aleatoric uncertainties. The former ones can be typically addressed in a systematic way, whether the latter ones pose a bigger challenge, representing commonly higher-dimensional uncertainties or uncertainties without a clear mathematical description, and requiring substantial computational burden to probabilistically characterize their impact on the output of interest. In order to reduce this burden computational statistics tools can be used to promote a fast, yet accurate, risk assessment. A novel framework based on stochastic emulation is developed that allows for the efficient approximation of the full distribution for the output of interest, circumventing challenges associated with the representation and characterization of aleatoric uncertainties. Seismic risk assessment is one example where the developed framework can be applied, addressing the aleatoric uncertainty that stems from ground motion to ground motion variability within the chosen excitation model. The objective here is to provide directly the full engineering demand parameter (EDP) distribution (under the impact of aleatoric uncertainties) within a PBEE vulnerability quantification setting.