A critical look into the discretization of epistemic variables and treatment of their correlation in seismic hazard and risk assessment

Epistemic uncertainty in seismic hazard and risk models is widely captured and modelled through the use of logic trees. The latter are used to represent different views on a certain model parameter or component, such as the definition of seismogenic sources and their associated seismicity rates or the prediction of ground motion intensity. In principle such logic tree branching can be applied to a multitude of parameters, however one will quickly run into a problem of combinatorial explosion, if the design of the logic tree is not done with caution. Some of the strategies often employed to tackle this problem include the discretization of continuous epistemic variables, the implicit assumption of perfect correlation between certain variables, and the modelling of some epistemic variables as aleatoric. Here, we discuss some of these practices and highlight the need to further understand their impact on hazard and risk estimates